Release 2.1.0 (#1998)

* bump version
* update release notes

.github/ISSUE_TEMPLATE/bug_report.md

@@ -0,0 +1,48 @@
+---
+name: Bug report
+about: Create a report to help us improve
+title: ''
+labels: ''
+assignees: ''
+
+---
+
+**Describe the bug**
+A clear and concise description of what the bug is.
+
+**To Reproduce**
+Steps to reproduce the behavior:
+
+**Expected behavior**
+A clear and concise description of what you expected to happen.
+
+**Screenshots**
+If applicable, add screenshots to help explain your problem.
+
+**Environment**
+- Patroni version:
+- PostgreSQL version:
+- DCS (and its version):
+
+**Patroni configuration file**
+```
+Please copy&paste your Patroni configuration file here
+```
+
+**patronictl show-config**
+```
+Please copy&paste the output of "patronictl show-config" command here
+```
+
+**Have you checked Patroni logs?**
+Please provide a snippet of Patroni log files here
+
+**Have you checked PostgreSQL logs?**
+Please provide a snippet here
+
+**Have you tried to use GitHub issue search?**
+Maybe there is already a similar issue solved.
+
+
+**Additional context**
+Add any other context about the problem here.

.github/workflows/install_deps.py

@@ -0,0 +1,186 @@
+import inspect
+import os
+import shutil
+import subprocess
+import stat
+import sys
+import tarfile
+import time
+import zipfile
+
+
+def install_requirements(what):
+    old_path = sys.path[:]
+    w = os.path.join(os.getcwd(), os.path.dirname(inspect.getfile(inspect.currentframe())))
+    sys.path.insert(0, os.path.dirname(os.path.dirname(w)))
+    try:
+        from setup import EXTRAS_REQUIRE, read
+    finally:
+        sys.path = old_path
+    requirements = ['mock>=2.0.0', 'flake8', 'pytest', 'pytest-cov'] if what == 'all' else ['behave']
+    requirements += ['psycopg2-binary', 'coverage']
+    for r in read('requirements.txt').split('\n'):
+        r = r.strip()
+        if r != '':
+            extras = {e for e, v in EXTRAS_REQUIRE.items() if v and any(r.startswith(x) for x in v)}
+            if not extras or what == 'all' or what in extras:
+                requirements.append(r)
+
+    subprocess.call([sys.executable, '-m', 'pip', 'install', '--upgrade', 'pip'])
+    r = subprocess.call([sys.executable, '-m', 'pip', 'install'] + requirements)
+    s = subprocess.call([sys.executable, '-m', 'pip', 'install', '--upgrade', 'setuptools'])
+    return s | r
+
+
+def install_packages(what):
+    from mapping import versions
+
+    packages = {
+        'zookeeper': ['zookeeper', 'zookeeper-bin', 'zookeeperd'],
+        'consul': ['consul'],
+    }
+    packages['exhibitor'] = packages['zookeeper']
+    packages = packages.get(what, [])
+    ver = versions.get(what)
+    subprocess.call(['sudo', 'sed', '-i', 's/pgdg main.*$/pgdg main {0}/'.format(ver),
+                     '/etc/apt/sources.list.d/pgdg.list'])
+    subprocess.call(['sudo', 'apt-get', 'update', '-y'])
+    return subprocess.call(['sudo', 'apt-get', 'install', '-y', 'postgresql-' + ver, 'expect-dev', 'wget'] + packages)
+
+
+def get_file(url, name):
+    try:
+        from urllib.request import urlretrieve
+    except ImportError:
+        from urllib import urlretrieve
+
+    print('Downloading ' + url)
+    urlretrieve(url, name)
+
+
+def untar(archive, name):
+    with tarfile.open(archive) as tar:
+        f = tar.extractfile(name)
+        dest = os.path.basename(name)
+        with open(dest, 'wb') as d:
+            shutil.copyfileobj(f, d)
+            return dest
+
+
+def unzip(archive, name):
+    with zipfile.ZipFile(archive, 'r') as z:
+        name = z.extract(name)
+        dest = os.path.basename(name)
+        shutil.move(name, dest)
+        return dest
+
+
+def unzip_all(archive):
+    print('Extracting ' + archive)
+    with zipfile.ZipFile(archive, 'r') as z:
+        z.extractall()
+
+
+def chmod_755(name):
+    os.chmod(name, stat.S_IRUSR | stat.S_IWUSR | stat.S_IXUSR |
+             stat.S_IRGRP | stat.S_IXGRP | stat.S_IROTH | stat.S_IXOTH)
+
+
+def unpack(archive, name):
+    print('Extracting {0} from {1}'.format(name, archive))
+    func = unzip if archive.endswith('.zip') else untar
+    name = func(archive, name)
+    chmod_755(name)
+    return name
+
+
+def install_etcd():
+    version = os.environ.get('ETCDVERSION', '3.3.13')
+    platform = {'linux2': 'linux', 'win32': 'windows', 'cygwin': 'windows'}.get(sys.platform, sys.platform)
+    dirname = 'etcd-v{0}-{1}-amd64'.format(version, platform)
+    ext = 'tar.gz' if platform == 'linux' else 'zip'
+    name = '{0}.{1}'.format(dirname, ext)
+    url = 'https://github.com/etcd-io/etcd/releases/download/v{0}/{1}'.format(version, name)
+    get_file(url, name)
+    ext = '.exe' if platform == 'windows' else ''
+    return int(unpack(name, '{0}/etcd{1}'.format(dirname, ext)) is None)
+
+
+def install_postgres():
+    version = os.environ.get('PGVERSION', '12.1-1')
+    platform = {'darwin': 'osx', 'win32': 'windows-x64', 'cygwin': 'windows-x64'}[sys.platform]
+    name = 'postgresql-{0}-{1}-binaries.zip'.format(version, platform)
+    get_file('http://get.enterprisedb.com/postgresql/' + name, name)
+    unzip_all(name)
+    bin_dir = os.path.join('pgsql', 'bin')
+    for f in os.listdir(bin_dir):
+        chmod_755(os.path.join(bin_dir, f))
+    subprocess.call(['pgsql/bin/postgres', '-V'])
+    return 0
+
+
+def setup_kubernetes():
+    get_file('https://storage.googleapis.com/minikube/k8sReleases/v1.7.0/localkube-linux-amd64', 'localkube')
+    chmod_755('localkube')
+
+    devnull = open(os.devnull, 'w')
+    subprocess.Popen(['sudo', 'nohup', './localkube', '--logtostderr=true', '--enable-dns=false'],
+                     stdout=devnull, stderr=devnull)
+    for _ in range(0, 120):
+        if subprocess.call(['wget', '-qO', '-', 'http://127.0.0.1:8080/'], stdout=devnull, stderr=devnull) == 0:
+            break
+        time.sleep(1)
+    else:
+        print('localkube did not start')
+        return 1
+
+    subprocess.call('sudo chmod 644 /var/lib/localkube/certs/*', shell=True)
+    print('Set up .kube/config')
+    kube = os.path.join(os.path.expanduser('~'), '.kube')
+    os.makedirs(kube)
+    with open(os.path.join(kube, 'config'), 'w') as f:
+        f.write("""apiVersion: v1
+clusters:
+- cluster:
+    certificate-authority: /var/lib/localkube/certs/ca.crt
+    server: https://127.0.0.1:8443
+  name: local
+contexts:
+- context:
+    cluster: local
+    user: myself
+  name: local
+current-context: local
+kind: Config
+preferences: {}
+users:
+- name: myself
+  user:
+    client-certificate: /var/lib/localkube/certs/apiserver.crt
+    client-key: /var/lib/localkube/certs/apiserver.key
+""")
+    return 0
+
+
+def main():
+    what = os.environ.get('DCS', sys.argv[1] if len(sys.argv) > 1 else 'all')
+
+    if what != 'all':
+        if sys.platform.startswith('linux'):
+            r = install_packages(what)
+            if r == 0 and what == 'kubernetes':
+                r = setup_kubernetes()
+        else:
+            r = install_postgres()
+
+        if r == 0 and what.startswith('etcd'):
+            r = install_etcd()
+
+        if r != 0:
+            return r
+
+    return install_requirements(what)
+
+
+if __name__ == '__main__':
+    sys.exit(main())

.github/workflows/mapping.py

@@ -0,0 +1 @@
+versions = {'etcd': '9.6', 'etcd3': '14', 'consul': '13', 'exhibitor': '12', 'raft': '11', 'kubernetes': '14'}

.github/workflows/run_tests.py

@@ -0,0 +1,50 @@
+import os
+import shutil
+import subprocess
+import sys
+import tempfile
+
+
+def main():
+    what = os.environ.get('DCS', sys.argv[1] if len(sys.argv) > 1 else 'all')
+
+    if what == 'all':
+        flake8 = subprocess.call([sys.executable, 'setup.py', 'flake8'])
+        test = subprocess.call([sys.executable, 'setup.py', 'test'])
+        version = '.'.join(map(str, sys.version_info[:2]))
+        shutil.move('.coverage', os.path.join(tempfile.gettempdir(), '.coverage.' + version))
+        return flake8 | test
+    elif what == 'combine':
+        tmp = tempfile.gettempdir()
+        for name in os.listdir(tmp):
+            if name.startswith('.coverage.'):
+                shutil.move(os.path.join(tmp, name), name)
+        return subprocess.call([sys.executable, '-m', 'coverage', 'combine'])
+
+    env = os.environ.copy()
+    if sys.platform.startswith('linux'):
+        from mapping import versions
+
+        version = versions.get(what)
+        path = '/usr/lib/postgresql/{0}/bin:.'.format(version)
+        unbuffer = ['timeout', '600', 'unbuffer']
+        args = ['--tags=-skip'] if what == 'etcd' else []
+    else:
+        path = os.path.abspath(os.path.join('pgsql', 'bin'))
+        if sys.platform == 'darwin':
+            path += ':.'
+        args = unbuffer = []
+    env['PATH'] = path + os.pathsep + env['PATH']
+    env['DCS'] = what
+
+    ret = subprocess.call(unbuffer + [sys.executable, '-m', 'behave'] + args, env=env)
+
+    if ret != 0:
+        if subprocess.call('grep . features/output/*_failed/*postgres?.*', shell=True) != 0:
+            subprocess.call('grep . features/output/*/*postgres?.*', shell=True)
+        return 1
+    return 0
+
+
+if __name__ == '__main__':
+    sys.exit(main())

.github/workflows/tests.yaml

@@ -0,0 +1,177 @@
+name: Tests
+
+on:
+  pull_request:
+  push:
+    branches:
+    - master
+    tags:
+    - v.*
+
+jobs:
+  unit:
+    runs-on: ${{ matrix.os }}-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ubuntu, windows, macos]
+
+    steps:
+    - uses: actions/checkout@v1
+    - name: Set up Python 2.7
+      uses: actions/setup-python@v2
+      with:
+        python-version: 2.7
+      if: matrix.os != 'windows'
+    - name: Install dependencies
+      run: python .github/workflows/install_deps.py
+      if: matrix.os != 'windows'
+    - name: Run tests and flake8
+      run: python .github/workflows/run_tests.py
+      if: matrix.os != 'windows'
+
+    - name: Set up Python 3.5
+      uses: actions/setup-python@v2
+      with:
+        python-version: 3.5
+    - name: Install dependencies
+      run: python .github/workflows/install_deps.py
+    - name: Run tests and flake8
+      run: python .github/workflows/run_tests.py
+
+    - name: Set up Python 3.6
+      uses: actions/setup-python@v2
+      with:
+        python-version: 3.6
+    - name: Install dependencies
+      run: python .github/workflows/install_deps.py
+    - name: Run tests and flake8
+      run: python .github/workflows/run_tests.py
+
+    - name: Set up Python 3.7
+      uses: actions/setup-python@v2
+      with:
+        python-version: 3.7
+    - name: Install dependencies
+      run: python .github/workflows/install_deps.py
+    - name: Run tests and flake8
+      run: python .github/workflows/run_tests.py
+
+    - name: Set up Python 3.8
+      uses: actions/setup-python@v2
+      with:
+        python-version: 3.8
+    - name: Install dependencies
+      run: python .github/workflows/install_deps.py
+    - name: Run tests and flake8
+      run: python .github/workflows/run_tests.py
+
+    - name: Set up Python 3.9
+      uses: actions/setup-python@v2
+      with:
+        python-version: 3.9
+    - name: Install dependencies
+      run: python .github/workflows/install_deps.py
+    - name: Run tests and flake8
+      run: python .github/workflows/run_tests.py
+
+    - name: Combine coverage
+      run: python .github/workflows/run_tests.py combine
+
+    - name: Install coveralls
+      run: python -m pip install coveralls
+
+    - name: Upload Coverage
+      env:
+        COVERALLS_FLAG_NAME: unit-${{ matrix.os }}
+        COVERALLS_PARALLEL: 'true'
+        GITHUB_TOKEN: ${{ secrets.github_token }}
+      run: python -m coveralls --service=github
+
+    - name: Run codacy-coverage-reporter
+      uses: codacy/codacy-coverage-reporter-action@master
+      env:
+        SECRETS_AVAILABLE: ${{ secrets.CODACY_PROJECT_TOKEN != '' }}
+      with:
+        project-token: ${{ secrets.CODACY_PROJECT_TOKEN }}
+        coverage-reports: coverage.xml
+      if: ${{ matrix.os == 'ubuntu' && env.SECRETS_AVAILABLE == 'true' }}
+
+  behave:
+    runs-on: ${{ matrix.os }}-latest
+    env:
+      DCS: ${{ matrix.dcs }}
+      ETCDVERSION: 3.3.13
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ubuntu]
+        python-version: [2.7, 3.5, 3.8]
+        dcs: [etcd, etcd3, consul, exhibitor, kubernetes, raft]
+
+    steps:
+    - uses: actions/checkout@v1
+    - name: Set up Python
+      uses: actions/setup-python@v2
+      with:
+        python-version: ${{ matrix.python-version }}
+    - name: Add postgresql apt repo
+      run: sudo sh -c 'echo "deb http://apt.postgresql.org/pub/repos/apt $(lsb_release -cs)-pgdg main" > /etc/apt/sources.list.d/pgdg.list'
+    - name: Install dependencies
+      run: python .github/workflows/install_deps.py
+    - name: Run behave tests
+      run: python .github/workflows/run_tests.py
+    - uses: actions/setup-python@v2
+      with:
+        python-version: 3.9
+    - name: Install coveralls
+      run: python -m pip install coveralls
+    - name: Upload Coverage
+      env:
+        COVERALLS_FLAG_NAME: behave-${{ matrix.os }}-${{ matrix.dcs }}-${{ matrix.python-version }}
+        COVERALLS_PARALLEL: 'true'
+        GITHUB_TOKEN: ${{ secrets.github_token }}
+      run: python -m coveralls --service=github
+
+  behavem:
+    runs-on: ${{ matrix.os }}-latest
+    env:
+      DCS: ${{ matrix.dcs }}
+      ETCDVERSION: 3.3.13
+      PGVERSION: 12.1-1  # for windows and macos
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [macos]  #, windows]
+        python-version: [3.7]
+        dcs: [etcd, etcd3, raft]
+
+    steps:
+    - uses: actions/checkout@v1
+    - name: Set up Python
+      uses: actions/setup-python@v2
+      with:
+        python-version: ${{ matrix.python-version }}
+    - name: Install dependencies
+      run: python .github/workflows/install_deps.py
+    - name: Run behave tests
+      run: python .github/workflows/run_tests.py
+    - name: Install coveralls
+      run: python -m pip install coveralls
+    - name: Upload Coverage
+      env:
+        COVERALLS_FLAG_NAME: behave-${{ matrix.os }}-${{ matrix.dcs }}-${{ matrix.python-version }}
+        COVERALLS_PARALLEL: 'true'
+        GITHUB_TOKEN: ${{ secrets.github_token }}
+      run: python -m coveralls --service=github
+
+  coveralls-finish:
+    name: Finalize coveralls.io
+    needs: [unit, behave, behavem]
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/setup-python@v2
+    - run: python -m pip install coveralls
+    - run: python -m coveralls --service=github --finish
+      env:
+        GITHUB_TOKEN: ${{ secrets.github_token }}

.travis.yml

@@ -1,157 +0,0 @@
-sudo: true
-dist: trusty
-language: python
-addons:
-  apt:
-    packages:
-      - expect-dev # for unbuffer
-env:
-  global:
-  - ETCDVERSION=3.0.17 ZKVERSION=3.4.11 CONSULVERSION=0.7.4
-  - PYVERSIONS="2.7 3.5 3.6"
-  - EXCLUDE_BEHAVE="3.5"
-  - BOTO_CONFIG=/doesnotexist
-matrix:
-  include:
-  - python: "3.5"
-    env: TEST_SUITE="python setup.py"
-  - python: "3.6"
-    env: DCS="etcd" TEST_SUITE="behave"
-  - python: "3.6"
-    env: DCS="exhibitor" TEST_SUITE="behave"
-  - python: "3.6"
-    env: DCS="consul" TEST_SUITE="behave"
-  - python: "3.6"
-    env: DCS="kubernetes" TEST_SUITE="behave"
-branches:
-  only:
-  - master
-  - /^v\d+\.\d+(\.\d+)?$/
-cache:
-  directories:
-  - $HOME/mycache
-before_cache:
-- |
-  rm -fr $HOME/mycache/python*
-  for pv in $PYVERSIONS; do
-      if [[ $TEST_SUITE != "behave" || $pv != $EXCLUDE_BEHAVE ]]; then
-          fpv=$(basename $(readlink $HOME/virtualenv/python${pv}))
-          mv $HOME/virtualenv/${fpv} $HOME/mycache/${fpv}
-      fi
-  done
-install:
-  - |
-    set -e
-
-    if [[ $TEST_SUITE == "behave" ]]; then
-        function get_consul() {
-            CC=~/mycache/consul_${CONSULVERSION}
-            if [[ ! -x $CC ]]; then
-                curl -L https://releases.hashicorp.com/consul/${CONSULVERSION}/consul_${CONSULVERSION}_linux_amd64.zip \
-                    | gunzip > $CC
-                [[ ${PIPESTATUS[0]} == 0 ]] || return 1
-                chmod +x $CC
-            fi
-            ln -s $CC consul
-        }
-
-        function get_etcd() {
-            EC=~/mycache/etcd_${ETCDVERSION}
-            if [[ ! -x $EC ]]; then
-                curl -L https://github.com/coreos/etcd/releases/download/v${ETCDVERSION}/etcd-v${ETCDVERSION}-linux-amd64.tar.gz \
-                    | tar xz -C . --strip=1 --wildcards --no-anchored etcd
-                [[ ${PIPESTATUS[0]} == 0 ]] || return 1
-                mv etcd $EC
-            fi
-            ln -s $EC etcd
-        }
-
-        function get_kubernetes() {
-            wget -O localkube "https://storage.googleapis.com/minikube/k8sReleases/v1.7.0/localkube-linux-amd64"
-            chmod +x localkube
-            sudo nohup ./localkube --logtostderr=true --enable-dns=false > localkube.log 2>&1 &
-
-            echo "Waiting for localkube to start..."
-            if ! timeout 120 sh -c "while ! curl -ks http://127.0.0.1:8080/ >/dev/null; do sleep 1; done"; then
-                sudo cat localkube.log
-                echo "localkube did not start"
-                exit 1
-            fi
-            echo "Check certificate permissions"
-            sudo chmod 644 /var/lib/localkube/certs/*
-            sudo ls -altr /var/lib/localkube/certs/
-
-            echo "Set up .kube/config"
-            mkdir ~/.kube
-            echo -e "apiVersion: v1\nclusters:\n- cluster:\n    certificate-authority: /var/lib/localkube/certs/ca.crt\n    server: https://127.0.0.1:8443\n  name: local\ncontexts:\n- context:\n    cluster: local\n    user: myself\n  name: local\ncurrent-context: local\nkind: Config\npreferences: {}\nusers:\n- name: myself\n  user:\n    client-certificate: /var/lib/localkube/certs/apiserver.crt\n    client-key: /var/lib/localkube/certs/apiserver.key\n" > ~/.kube/config
-        }
-
-        function get_exhibitor() {
-            ZC=~/mycache/zookeeper-${ZKVERSION}
-            if [[ ! -d $ZC ]]; then
-                curl -L http://www.apache.org/dist/zookeeper/zookeeper-${ZKVERSION}/zookeeper-${ZKVERSION}.tar.gz | tar xz
-                [[ ${PIPESTATUS[0]} == 0 ]] || return 1
-                mv zookeeper-${ZKVERSION}/conf/zoo_sample.cfg zookeeper-${ZKVERSION}/conf/zoo.cfg
-                mv zookeeper-${ZKVERSION} $ZC
-            fi
-            $ZC/bin/zkServer.sh start
-            # following lines are 'emulating' exhibitor REST API
-            while true; do
-                echo -e 'HTTP/1.0 200 OK\nContent-Type: application/json\n\n{"servers":["127.0.0.1"],"port":2181}' \
-                    | nc -l 8181 &> /dev/null
-            done&
-        }
-
-        attempt_num=1
-        until get_${DCS}; do
-            [[ $attempt_num -ge 3 ]] && exit 1
-            echo "Attempt $attempt_num failed! Trying again in $attempt_num seconds..."
-            sleep $(( attempt_num++ ))
-        done
-    fi
-
-    for pv in $PYVERSIONS; do
-        if [[ $TEST_SUITE != "behave" || $pv != $EXCLUDE_BEHAVE ]]; then
-            fpv=$(basename $(readlink $HOME/virtualenv/python$pv))
-            if [[ -d ~/mycache/${fpv} ]]; then
-                mv ~/virtualenv/${fpv} ~/virtualenv/${fpv}.bckp
-                mv ~/mycache/${fpv} ~/virtualenv/${fpv}
-            fi
-            source ~/virtualenv/python${pv}/bin/activate
-            # explicitly install all needed python modules to cache them
-            for p in '-r requirements.txt' 'psycopg2-binary behave codacy-coverage coverage coveralls flake8 mock pytest-cov pytest setuptools'; do
-                pip install $p --upgrade
-            done
-        fi
-    done
-script:
-  - |
-    for pv in $PYVERSIONS; do
-        if [[ $TEST_SUITE == "behave" && $pv == $EXCLUDE_BEHAVE ]]; then
-            continue
-        fi
-
-        source ~/virtualenv/python${pv}/bin/activate
-
-        if [[ $TEST_SUITE != "behave" ]]; then
-            echo Running unit tests using python${pv}
-            unbuffer $TEST_SUITE test
-            $TEST_SUITE flake8
-        elif [[ $pv != $EXCLUDE_BEHAVE ]]; then
-            echo Running acceptance tests using python${pv}
-            if ! PATH=.:/usr/lib/postgresql/9.6/bin:$PATH unbuffer $TEST_SUITE; then
-                # output all log files when tests are failing
-                grep . features/output/*_failed/*postgres?.*
-                exit 1
-            fi
-        fi
-    done
-
-    set +e
-after_success:
-  # before_cache is executed earlier than after_success, so we need to restore one of virtualenv directories
-  - fpv=$(basename $(readlink $HOME/virtualenv/python3.6)) && mv $HOME/mycache/${fpv} $HOME/virtualenv/${fpv}
-  - coveralls
-  - if [[ $TEST_SUITE != "behave" ]]; then python-codacy-coverage -r coverage.xml; fi
-  - if [[ $DCS == "exhibitor" ]]; then ~/mycache/zookeeper-${ZKVERSION}/bin/zkServer.sh stop; fi
-  - sudo kill $(jobs -p)

Dockerfile

@@ -14,7 +14,7 @@ ARG PGDATA
 ARG LC_ALL
 ARG LANG
 
-ENV ETCDVERSION=2.3.8 CONFDVERSION=0.16.0
+ENV ETCDVERSION=3.3.13 CONFDVERSION=0.16.0
 
 RUN set -ex \
     && export DEBIAN_FRONTEND=noninteractive \
@@ -25,11 +25,12 @@ RUN set -ex \
             | grep -Ev '^python3-(sphinx|etcd|consul|kazoo|kubernetes)' \
             | xargs apt-get install -y vim curl less jq locales haproxy sudo \
                             python3-etcd python3-kazoo python3-pip busybox \
+                            net-tools iputils-ping --fix-missing \
     && pip3 install dumb-init \
 \
     # Cleanup all locales but en_US.UTF-8
     && find /usr/share/i18n/charmaps/ -type f ! -name UTF-8.gz -delete \
-    && find /usr/share/i18n/locales/ -type f ! -name en_US ! -name en_GB ! -name i18n ! -name iso14651_t1 ! -name iso14651_t1_common ! -name 'translit_*' -delete \
+    && find /usr/share/i18n/locales/ -type f ! -name en_US ! -name en_GB ! -name i18n* ! -name iso14651_t1 ! -name iso14651_t1_common ! -name 'translit_*' -delete \
     && echo 'en_US.UTF-8 UTF-8' > /usr/share/i18n/SUPPORTED \
 \
     # Make sure we have a en_US.UTF-8 locale available
@@ -149,6 +150,7 @@ RUN sed -i 's/env python/&3/' /patroni*.py \
     && sed -i 's/^    \(replication\|superuser\|rewind\|unix_socket_directories\|\(\(  \)\{0,1\}\(username\|password\)\)\):/#&/' postgres?.yml \
     && sed -i 's/^      parameters:/      pg_hba:\n      - local all all trust\n      - host replication all all md5\n      - host all all all md5\n&\n        max_connections: 100/'  postgres?.yml \
     && if [ "$COMPRESS" = "true" ]; then chmod u+s /usr/bin/sudo; fi \
+    && chmod +s /bin/ping \
     && chown -R postgres:postgres $PGHOME /run /etc/haproxy
 
 USER postgres

README.rst

@@ -1,4 +1,4 @@
-|Build Status| |Coverage Status|
+|Tests Status| |Coverage Status|
 
 Patroni: A Template for PostgreSQL HA with ZooKeeper, etcd or Consul
 --------------------------------------------------------------------
@@ -12,6 +12,8 @@ Patroni is a template for you to create your own customized, high-availability s
 
 We call Patroni a "template" because it is far from being a one-size-fits-all or plug-and-play replication system. It will have its own caveats. Use wisely.
 
+Currently supported PostgreSQL versions: 9.3 to 13.
+
 **Note to Kubernetes users**: Patroni can run natively on top of Kubernetes. Take a look at the `Kubernetes <https://github.com/zalando/patroni/blob/master/docs/kubernetes.rst>`__ chapter of the Patroni documentation.
 
 .. contents::
@@ -96,7 +98,7 @@ Patroni can be installed with pip:
 
 where dependencies can be either empty, or consist of one or more of the following:
 
-etcd
+etcd or etcd3
     `python-etcd` module in order to use Etcd as DCS
 consul
     `python-consul` module in order to use Consul as DCS
@@ -106,6 +108,8 @@ exhibitor
     `kazoo` module in order to use Exhibitor as DCS (same dependencies as for Zookeeper)
 kubernetes
     `kubernetes` module in order to use Kubernetes as DCS in Patroni
+raft
+    `pysyncobj` module in order to use python Raft implementation as DCS
 aws
     `boto` in order to use AWS callbacks
 
@@ -115,7 +119,7 @@ For example, the command in order to install Patroni together with dependencies
 
     pip install patroni[etcd,aws]
 
-Note that external tools to call in the replica creation or custom bootstap scripts (i.e. WAL-E) should be installed independently of Patroni.
+Note that external tools to call in the replica creation or custom bootstrap scripts (i.e. WAL-E) should be installed independently of Patroni.
 
 =======================
 Running and Configuring
@@ -124,7 +128,7 @@ Running and Configuring
 To get started, do the following from different terminals:
 ::
 
-    > etcd --data-dir=data/etcd
+    > etcd --data-dir=data/etcd --enable-v2=true
     > ./patroni.py postgres0.yml
     > ./patroni.py postgres1.yml
 
@@ -167,7 +171,7 @@ Applications Should Not Use Superusers
 
 When connecting from an application, always use a non-superuser. Patroni requires access to the database to function properly. By using a superuser from an application, you can potentially use the entire connection pool, including the connections reserved for superusers, with the ``superuser_reserved_connections`` setting. If Patroni cannot access the Primary because the connection pool is full, behavior will be undesirable.
 
-.. |Build Status| image:: https://travis-ci.org/zalando/patroni.svg?branch=master
+.. |Tests Status| image:: https://github.com/zalando/patroni/actions/workflows/tests.yaml/badge.svg
-   :target: https://travis-ci.org/zalando/patroni
+   :target: https://github.com/zalando/patroni/actions/workflows/tests.yaml?query=branch%3Amaster
 .. |Coverage Status| image:: https://coveralls.io/repos/zalando/patroni/badge.svg?branch=master
-   :target: https://coveralls.io/r/zalando/patroni?branch=master
+   :target: https://coveralls.io/github/zalando/patroni?branch=master

docker-compose.yml

@@ -6,57 +6,31 @@ networks:
     demo:
 
 services:
-    etcd1:
+    etcd1: &etcd
         image: patroni
         networks: [ demo ]
-        env_file: docker/etcd.env
+        environment:
+            ETCD_LISTEN_PEER_URLS: http://0.0.0.0:2380
+            ETCD_LISTEN_CLIENT_URLS: http://0.0.0.0:2379
+            ETCD_INITIAL_CLUSTER: etcd1=http://etcd1:2380,etcd2=http://etcd2:2380,etcd3=http://etcd3:2380
+            ETCD_INITIAL_CLUSTER_STATE: new
+            ETCD_INITIAL_CLUSTER_TOKEN: tutorial
         container_name: demo-etcd1
         hostname: etcd1
         command: etcd -name etcd1 -initial-advertise-peer-urls http://etcd1:2380
 
     etcd2:
-        image: patroni
+        <<: *etcd
-        networks: [ demo ]
-        env_file: docker/etcd.env
         container_name: demo-etcd2
         hostname: etcd2
         command: etcd -name etcd2 -initial-advertise-peer-urls http://etcd2:2380
 
     etcd3:
-        image: patroni
+        <<: *etcd
-        networks: [ demo ]
-        env_file: docker/etcd.env
         container_name: demo-etcd3
         hostname: etcd3
         command: etcd -name etcd3 -initial-advertise-peer-urls http://etcd3:2380
 
-    patroni1:
-        image: patroni
-        networks: [ demo ]
-        env_file: docker/patroni.env
-        hostname: patroni1
-        container_name: demo-patroni1
-        environment:
-            PATRONI_NAME: patroni1
-
-    patroni2:
-        image: patroni
-        networks: [ demo ]
-        env_file: docker/patroni.env
-        hostname: patroni2
-        container_name: demo-patroni2
-        environment:
-            PATRONI_NAME: patroni2
-
-    patroni3:
-        image: patroni
-        networks: [ demo ]
-        env_file: docker/patroni.env
-        hostname: patroni3
-        container_name: demo-patroni3
-        environment:
-            PATRONI_NAME: patroni3
-
     haproxy:
         image: patroni
         networks: [ demo ]
@@ -67,3 +41,37 @@ services:
             - "5000:5000"
             - "5001:5001"
         command: haproxy
+        environment: &haproxy_env
+            ETCDCTL_ENDPOINTS: http://etcd1:2379,http://etcd2:2379,http://etcd3:2379
+            PATRONI_ETCD3_HOSTS: "'etcd1:2379','etcd2:2379','etcd3:2379'"
+            PATRONI_SCOPE: demo
+
+    patroni1:
+        image: patroni
+        networks: [ demo ]
+        env_file: docker/patroni.env
+        hostname: patroni1
+        container_name: demo-patroni1
+        environment:
+            <<: *haproxy_env
+            PATRONI_NAME: patroni1
+
+    patroni2:
+        image: patroni
+        networks: [ demo ]
+        env_file: docker/patroni.env
+        hostname: patroni2
+        container_name: demo-patroni2
+        environment:
+            <<: *haproxy_env
+            PATRONI_NAME: patroni2
+
+    patroni3:
+        image: patroni
+        networks: [ demo ]
+        env_file: docker/patroni.env
+        hostname: patroni3
+        container_name: demo-patroni3
+        environment:
+            <<: *haproxy_env
+            PATRONI_NAME: patroni3

docker/entrypoint.sh

@@ -25,7 +25,7 @@ case "$1" in
             while ! etcdctl cluster-health 2> /dev/null; do
                 sleep 1
             done
-            exec dumb-init $CONFD etcd -node $(echo $ETCDCTL_ENDPOINTS | sed 's/,/ -node /g')
+            exec dumb-init $CONFD etcdv3 -node $(echo $ETCDCTL_ENDPOINTS | sed 's/,/ -node /g')
         fi
         ;;
     etcd)
@@ -37,7 +37,7 @@ case "$1" in
 esac
 
 ## We start an etcd
-if [ -z "$PATRONI_ETCD_HOSTS" ] && [ -z "$PATRONI_ZOOKEEPER_HOSTS" ]; then
+if [ -z "$PATRONI_ETCD3_HOSTS" ] && [ -z "$PATRONI_ZOOKEEPER_HOSTS" ]; then
     export PATRONI_ETCD_URL="http://127.0.0.1:2379"
     etcd --data-dir /tmp/etcd.data -advertise-client-urls=$PATRONI_ETCD_URL -listen-client-urls=http://0.0.0.0:2379 > /var/log/etcd.log 2> /var/log/etcd.err &
 fi

docker/etcd.env

@@ -1,5 +0,0 @@
-ETCD_LISTEN_PEER_URLS=http://0.0.0.0:2380
-ETCD_LISTEN_CLIENT_URLS=http://0.0.0.0:2379
-ETCD_INITIAL_CLUSTER=etcd1=http://etcd1:2380,etcd2=http://etcd2:2380,etcd3=http://etcd3:2380
-ETCD_INITIAL_CLUSTER_STATE=new
-ETCD_INITIAL_CLUSTER_TOKEN=tutorial

docker/patroni.env

@@ -1,6 +1,3 @@
-PATRONI_SCOPE=demo
-PATRONI_ETCD_HOSTS='etcd1:2379','etcd2:2379','etcd3:2379'
-
 PATRONI_RESTAPI_USERNAME=admin
 PATRONI_RESTAPI_PASSWORD=admin
 PATRONI_SUPERUSER_USERNAME=postgres
@@ -9,6 +6,3 @@ PATRONI_REPLICATION_USERNAME=replicator
 PATRONI_REPLICATION_PASSWORD=replicate
 PATRONI_admin_PASSWORD=admin
 PATRONI_admin_OPTIONS=createdb,createrole
-
-# for etcdctl
-ETCDCTL_ENDPOINTS=http://etcd1:2379,http://etcd2:2379,http://etcd3:2379

docs/CONTRIBUTING.rst

@@ -10,6 +10,39 @@ Chatting
 
 Just want to chat with other Patroni users?  Looking for interactive troubleshooting help? Join us on channel #patroni in the `PostgreSQL Slack <https://postgres-slack.herokuapp.com/>`__.
 
+Running tests
+-------------
+
+Requirements for running behave tests:
+
+1. PostgreSQL packages need to be installed.
+2. PostgreSQL binaries must be available in your `PATH`. You may need to add them to the path with something like `PATH=/usr/lib/postgresql/11/bin:$PATH python -m behave`.
+3. If you'd like to test with external DCSs (e.g., Etcd, Consul, and Zookeeper) you'll need the packages installed and respective services running and accepting unencrypted/unprotected connections on localhost and default port. In the case of Etcd or Consul, the behave test suite could start them up if binaries are available in the `PATH`.
+
+ Install dependencies:
+
+.. code-block:: bash
+
+    # You may want to use Virtualenv or specify pip3.
+    pip install -r requirements.txt
+    pip install -r requirements.dev.txt
+
+After you have all dependencies installed, you can run the various test suites:
+
+.. code-block:: bash
+
+    # You may want to use Virtualenv or specify python3.
+
+    # Run flake8 to check syntax and formatting:
+    python setup.py flake8
+
+    # Run the pytest suite in tests/:
+    python setup.py test
+
+    # Run the behave (https://behave.readthedocs.io/en/latest/) test suite in features/;
+    # modify DCS as desired (raft has no dependencies so is the easiest to start with):
+    DCS=raft python -m behave
+
 Reporting issues
 ----------------
 

docs/ENVIRONMENT.rst

@@ -11,7 +11,11 @@ Global/Universal
 -  **PATRONI\_NAME**: name of the node where the current instance of Patroni is running. Must be unique for the cluster.
 -  **PATRONI\_NAMESPACE**: path within the configuration store where Patroni will keep information about the cluster. Default value: "/service"
 -  **PATRONI\_SCOPE**: cluster name
+
+Log
+---
 -  **PATRONI\_LOG\_LEVEL**: sets the general logging level. Default value is **INFO** (see `the docs for Python logging <https://docs.python.org/3.6/library/logging.html#levels>`_)
+-  **PATRONI\_LOG\_TRACEBACK\_LEVEL**: sets the level where tracebacks will be visible. Default value is **ERROR**. Set it to **DEBUG** if you want to see tracebacks only if you enable **PATRONI\_LOG\_LEVEL=DEBUG**.
 -  **PATRONI\_LOG\_FORMAT**: sets the log formatting string. Default value is **%(asctime)s %(levelname)s: %(message)s** (see `the LogRecord attributes <https://docs.python.org/3.6/library/logging.html#logrecord-attributes>`_)
 -  **PATRONI\_LOG\_DATEFORMAT**: sets the datetime formatting string. (see the `formatTime() documentation <https://docs.python.org/3.6/library/logging.html#logging.Formatter.formatTime>`_)
 -  **PATRONI\_LOG\_MAX\_QUEUE\_SIZE**: Patroni is using two-step logging. Log records are written into the in-memory queue and there is a separate thread which pulls them from the queue and writes to stderr or file. The maximum size of the internal queue is limited by default by **1000** records, which is enough to keep logs for the past 1h20m.
@@ -19,8 +23,6 @@ Global/Universal
 -  **PATRONI\_LOG\_FILE\_NUM**: The number of application logs to retain.
 -  **PATRONI\_LOG\_FILE\_SIZE**: Size of patroni.log file (in bytes) that triggers a log rolling.
 -  **PATRONI\_LOG\_LOGGERS**: Redefine logging level per python module. Example ``PATRONI_LOG_LOGGERS="{patroni.postmaster: WARNING, urllib3: DEBUG}"``
--  **PATRONI\_DEBUG\_MODE**: When set to a non-empty value, makes Patroni run in the special debug mode that enables stepping with a debugger through some execution paths within Patroni `
-Example ``PATRONI_DEBUG_MODE="on"``
 
 Bootstrap configuration
 -----------------------
@@ -33,8 +35,8 @@ Example: defining ``PATRONI_admin_PASSWORD=strongpasswd`` and ``PATRONI_admin_OP
 
 Consul
 ------
--  **PATRONI\_CONSUL\_HOST**: the host:port for the Consul endpoint.
+-  **PATRONI\_CONSUL\_HOST**: the host:port for the Consul local agent.
--  **PATRONI\_CONSUL\_URL**: url for the Consul, in format: http(s)://host:port
+-  **PATRONI\_CONSUL\_URL**: url for the Consul local agent, in format: http(s)://host:port
 -  **PATRONI\_CONSUL\_PORT**: (optional) Consul port
 -  **PATRONI\_CONSUL\_SCHEME**: (optional) **http** or **https**, defaults to **http**
 -  **PATRONI\_CONSUL\_TOKEN**: (optional) ACL token
@@ -44,7 +46,7 @@ Consul
 -  **PATRONI\_CONSUL\_KEY**: (optional) File with the client key. Can be empty if the key is part of certificate.
 -  **PATRONI\_CONSUL\_DC**: (optional) Datacenter to communicate with. By default the datacenter of the host is used.
 -  **PATRONI\_CONSUL\_CONSISTENCY**: (optional) Select consul consistency mode. Possible values are ``default``, ``consistent``, or ``stale`` (more details in `consul API reference <https://www.consul.io/api/features/consistency.html/>`__)
--  **PATRONI\_CONSUL\_CHECKS**: (optional) list of Consul health checks used for the session. If not specified Consul will use "serfHealth" in additional to the TTL based check created by Patroni. Additional checks, in particular the "serfHealth", may cause the leader lock to expire faster than in `ttl` seconds when the leader instance becomes unavailable.
+-  **PATRONI\_CONSUL\_CHECKS**: (optional) list of Consul health checks used for the session. By default an empty list is used.
 -  **PATRONI\_CONSUL\_REGISTER\_SERVICE**: (optional) whether or not to register a service with the name defined by the scope parameter and the tag master, replica or standby-leader depending on the node's role. Defaults to **false**
 -  **PATRONI\_CONSUL\_SERVICE\_CHECK\_INTERVAL**: (optional) how often to perform health check against registered url
 
@@ -64,6 +66,28 @@ Etcd
 -  **PATRONI\_ETCD\_CERT**: File with the client certificate.
 -  **PATRONI\_ETCD\_KEY**: File with the client key. Can be empty if the key is part of certificate.
 
+Etcdv3
+------
+Environment names for Etcdv3 are similar as for Etcd, you just need to use ``ETCD3`` instead of ``ETCD`` in the variable name. Example: ``PATRONI_ETCD3_HOST``, ``PATRONI_ETCD3_CACERT``, and so on.
+
+.. warning::
+    Keys created with protocol version 2 are not visible with protocol version 3 and the other way around, therefore it is not possible to switch from Etcd to Etcdv3 just by updating Patroni configuration.
+
+
+ZooKeeper
+---------
+-  **PATRONI\_ZOOKEEPER\_HOSTS**: Comma separated list of ZooKeeper cluster members: "'host1:port1','host2:port2','etc...'". It is important to quote every single entity!
+-  **PATRONI\_ZOOKEEPER\_USE\_SSL**: (optional) Whether SSL is used or not. Defaults to ``false``. If set to ``false``, all SSL specific parameters are ignored.
+-  **PATRONI\_ZOOKEEPER\_CACERT**: (optional) The CA certificate. If present it will enable validation.
+-  **PATRONI\_ZOOKEEPER\_CERT**: (optional) File with the client certificate.
+-  **PATRONI\_ZOOKEEPER\_KEY**: (optional) File with the client key.
+-  **PATRONI\_ZOOKEEPER\_KEY\_PASSWORD**: (optional) The client key password.
+-  **PATRONI\_ZOOKEEPER\_VERIFY**: (optional) Whether to verify certificate or not. Defaults to ``true``.
+
+.. note::
+    It is required to install ``kazoo>=2.6.0`` to support SSL.
+
+
 Exhibitor
 ---------
 -  **PATRONI\_EXHIBITOR\_HOSTS**: initial list of Exhibitor (ZooKeeper) nodes in format: 'host1,host2,etc...'. This list updates automatically whenever the Exhibitor (ZooKeeper) cluster topology changes.
@@ -73,13 +97,23 @@ Exhibitor
 
 Kubernetes
 ----------
+-  **PATRONI\_KUBERNETES\_BYPASS\_API\_SERVICE**: (optional) When communicating with the Kubernetes API, Patroni is usually relying on the `kubernetes` service, the address of which is exposed in the pods via the `KUBERNETES_SERVICE_HOST` environment variable. If `PATRONI_KUBERNETES_BYPASS_API_SERVICE` is set to ``true``, Patroni will resolve the list of API nodes behind the service and connect directly to them.
 -  **PATRONI\_KUBERNETES\_NAMESPACE**: (optional) Kubernetes namespace where the Patroni pod is running. Default value is `default`.
 -  **PATRONI\_KUBERNETES\_LABELS**: Labels in format ``{label1: value1, label2: value2}``. These labels will be used to find existing objects (Pods and either Endpoints or ConfigMaps) associated with the current cluster. Also Patroni will set them on every object (Endpoint or ConfigMap) it creates.
 -  **PATRONI\_KUBERNETES\_SCOPE\_LABEL**: (optional) name of the label containing cluster name. Default value is `cluster-name`.
 -  **PATRONI\_KUBERNETES\_ROLE\_LABEL**: (optional) name of the label containing Postgres role (`master` or `replica`). Patroni will set this label on the pod it is running in. Default value is `role`.
 -  **PATRONI\_KUBERNETES\_USE\_ENDPOINTS**: (optional) if set to true, Patroni will use Endpoints instead of ConfigMaps to run leader elections and keep cluster state.
 -  **PATRONI\_KUBERNETES\_POD\_IP**: (optional) IP address of the pod Patroni is running in. This value is required when `PATRONI_KUBERNETES_USE_ENDPOINTS` is enabled and is used to populate the leader endpoint subsets when the pod's PostgreSQL is promoted.
--  **PATRONI\_KUBERNETES\_PORTS**: (optional) if the Service object has the name for the port, the same name must appear in the Endpoint object, otherwise service won't work. For example, if your service is defined as ``{Kind: Service, spec: {ports: [{name: postgresql, port: 5432, targetPort: 5432}]}}``, then you have to set ``PATRONI_KUBERNETES_PORTS='{[{"name": "postgresql", "port": 5432}]}'`` and Patroni will use it for updating subsets of the leader Endpoint. This parameter is used only if `PATRONI_KUBERNETES_USE_ENDPOINTS` is set.
+-  **PATRONI\_KUBERNETES\_PORTS**: (optional) if the Service object has the name for the port, the same name must appear in the Endpoint object, otherwise service won't work. For example, if your service is defined as ``{Kind: Service, spec: {ports: [{name: postgresql, port: 5432, targetPort: 5432}]}}``, then you have to set ``PATRONI_KUBERNETES_PORTS='[{"name": "postgresql", "port": 5432}]'`` and Patroni will use it for updating subsets of the leader Endpoint. This parameter is used only if `PATRONI_KUBERNETES_USE_ENDPOINTS` is set.
+
+Raft
+----
+
+-  **PATRONI\_RAFT\_SELF\_ADDR**: ``ip:port`` to listen on for Raft connections. The ``self_addr`` must be accessible from other nodes of the cluster. If not set, the node will not participate in consensus.
+-  **PATRONI\_RAFT\_BIND\_ADDR**: (optional) ``ip:port`` to listen on for Raft connections. If not specified the ``self_addr`` will be used.
+-  **PATRONI\_RAFT\_PARTNER\_ADDRS**: list of other Patroni nodes in the cluster in format ``"'ip1:port1','ip2:port2'"``. It is important to quote every single entity!
+-  **PATRONI\_RAFT\_DATA\_DIR**: directory where to store Raft log and snapshot. If not specified the current working directory is used.
+-  **PATRONI\_RAFT\_PASSWORD**: (optional) Encrypt Raft traffic with a specified password, requires ``cryptography`` python module.
 
 PostgreSQL
 ----------
@@ -91,10 +125,34 @@ PostgreSQL
 -  **PATRONI\_POSTGRESQL\_PGPASS**: path to the `.pgpass <https://www.postgresql.org/docs/current/static/libpq-pgpass.html>`__ password file. Patroni creates this file before executing pg\_basebackup and under some other circumstances. The location must be writable by Patroni.
 -  **PATRONI\_REPLICATION\_USERNAME**: replication username; the user will be created during initialization. Replicas will use this user to access master via streaming replication
 -  **PATRONI\_REPLICATION\_PASSWORD**: replication password; the user will be created during initialization.
+-  **PATRONI\_REPLICATION\_SSLMODE**: (optional) maps to the `sslmode <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLMODE>`__ connection parameter, which allows a client to specify the type of TLS negotiation mode with the server. For more information on how each mode works, please visit the `PostgreSQL documentation <https://www.postgresql.org/docs/current/libpq-ssl.html#LIBPQ-SSL-SSLMODE-STATEMENTS>`__. The default mode is ``prefer``.
+-  **PATRONI\_REPLICATION\_SSLKEY**: (optional) maps to the `sslkey <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLKEY>`__ connection parameter, which specifies the location of the secret key used with the client's certificate.
+-  **PATRONI\_REPLICATION\_SSLPASSWORD**: (optional) maps to the `sslpassword <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLPASSWORD>`__ connection parameter, which specifies the password for the secret key specified in ``PATRONI_REPLICATION_SSLKEY``.
+-  **PATRONI\_REPLICATION\_SSLCERT**: (optional) maps to the `sslcert <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLCERT>`__ connection parameter, which specifies the location of the client certificate.
+-  **PATRONI\_REPLICATION\_SSLROOTCERT**: (optional) maps to the `sslrootcert <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLROOTCERT>`__ connection parameter, which specifies the location of a file containing one ore more certificate authorities (CA) certificates that the client will use to verify a server's certificate.
+-  **PATRONI\_REPLICATION\_SSLCRL**: (optional) maps to the `sslcrl <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLCRL>`__ connection parameter, which specifies the location of a file containing a certificate revocation list. A client will reject connecting to any server that has a certificate present in this list.
+-  **PATRONI\_REPLICATION\_GSSENCMODE**: (optional) maps to the `gssencmode <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-GSSENCMODE>`__ connection parameter, which determines whether or with what priority a secure GSS TCP/IP connection will be negotiated with the server
+-  **PATRONI\_REPLICATION\_CHANNEL\_BINDING**: (optional) maps to the `channel_binding <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-CHANNEL-BINDING>`__ connection parameter, which controls the client's use of channel binding.
 -  **PATRONI\_SUPERUSER\_USERNAME**: name for the superuser, set during initialization (initdb) and later used by Patroni to connect to the postgres. Also this user is used by pg_rewind.
 -  **PATRONI\_SUPERUSER\_PASSWORD**: password for the superuser, set during initialization (initdb).
+-  **PATRONI\_SUPERUSER\_SSLMODE**: (optional) maps to the `sslmode <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLMODE>`__ connection parameter, which allows a client to specify the type of TLS negotiation mode with the server. For more information on how each mode works, please visit the `PostgreSQL documentation <https://www.postgresql.org/docs/current/libpq-ssl.html#LIBPQ-SSL-SSLMODE-STATEMENTS>`__. The default mode is ``prefer``.
+-  **PATRONI\_SUPERUSER\_SSLKEY**: (optional) maps to the `sslkey <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLKEY>`__ connection parameter, which specifies the location of the secret key used with the client's certificate.
+-  **PATRONI\_SUPERUSER\_SSLPASSWORD**: (optional) maps to the `sslpassword <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLPASSWORD>`__ connection parameter, which specifies the password for the secret key specified in ``PATRONI_SUPERUSER_SSLKEY``.
+-  **PATRONI\_SUPERUSER\_SSLCERT**: (optional) maps to the `sslcert <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLCERT>`__ connection parameter, which specifies the location of the client certificate.
+-  **PATRONI\_SUPERUSER\_SSLROOTCERT**: (optional) maps to the `sslrootcert <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLROOTCERT>`__ connection parameter, which specifies the location of a file containing one ore more certificate authorities (CA) certificates that the client will use to verify a server's certificate.
+-  **PATRONI\_SUPERUSER\_SSLCRL**: (optional) maps to the `sslcrl <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLCRL>`__ connection parameter, which specifies the location of a file containing a certificate revocation list. A client will reject connecting to any server that has a certificate present in this list.
+-  **PATRONI\_SUPERUSER\_GSSENCMODE**: (optional) maps to the `gssencmode <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-GSSENCMODE>`__ connection parameter, which determines whether or with what priority a secure GSS TCP/IP connection will be negotiated with the server
+-  **PATRONI\_SUPERUSER\_CHANNEL\_BINDING**: (optional) maps to the `channel_binding <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-CHANNEL-BINDING>`__ connection parameter, which controls the client's use of channel binding.
 -  **PATRONI\_REWIND\_USERNAME**: name for the user for ``pg_rewind``; the user will be created during initialization of postgres 11+ and all necessary `permissions <https://www.postgresql.org/docs/11/app-pgrewind.html#id-1.9.5.8.8>`__ will be granted.
 -  **PATRONI\_REWIND\_PASSWORD**: password for the user for ``pg_rewind``; the user will be created during initialization.
+-  **PATRONI\_REWIND\_SSLMODE**: (optional) maps to the `sslmode <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLMODE>`__ connection parameter, which allows a client to specify the type of TLS negotiation mode with the server. For more information on how each mode works, please visit the `PostgreSQL documentation <https://www.postgresql.org/docs/current/libpq-ssl.html#LIBPQ-SSL-SSLMODE-STATEMENTS>`__. The default mode is ``prefer``.
+-  **PATRONI\_REWIND\_SSLKEY**: (optional) maps to the `sslkey <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLKEY>`__ connection parameter, which specifies the location of the secret key used with the client's certificate.
+-  **PATRONI\_REWIND\_SSLPASSWORD**: (optional) maps to the `sslpassword <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLPASSWORD>`__ connection parameter, which specifies the password for the secret key specified in ``PATRONI_REWIND_SSLKEY``.
+-  **PATRONI\_REWIND\_SSLCERT**: (optional) maps to the `sslcert <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLCERT>`__ connection parameter, which specifies the location of the client certificate.
+-  **PATRONI\_REWIND\_SSLROOTCERT**: (optional) maps to the `sslrootcert <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLROOTCERT>`__ connection parameter, which specifies the location of a file containing one ore more certificate authorities (CA) certificates that the client will use to verify a server's certificate.
+-  **PATRONI\_REWIND\_SSLCRL**: (optional) maps to the `sslcrl <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLCRL>`__ connection parameter, which specifies the location of a file containing a certificate revocation list. A client will reject connecting to any server that has a certificate present in this list.
+-  **PATRONI\_REWIND\_GSSENCMODE**: (optional) maps to the `gssencmode <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-GSSENCMODE>`__ connection parameter, which determines whether or with what priority a secure GSS TCP/IP connection will be negotiated with the server
+-  **PATRONI\_REWIND\_CHANNEL\_BINDING**: (optional) maps to the `channel_binding <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-CHANNEL-BINDING>`__ connection parameter, which controls the client's use of channel binding.
 
 REST API
 --------
@@ -104,7 +162,19 @@ REST API
 -  **PATRONI\_RESTAPI\_PASSWORD**: Basic-auth password to protect unsafe REST API endpoints.
 -  **PATRONI\_RESTAPI\_CERTFILE**: Specifies the file with the certificate in the PEM format. If the certfile is not specified or is left empty, the API server will work without SSL.
 -  **PATRONI\_RESTAPI\_KEYFILE**: Specifies the file with the secret key in the PEM format.
+-  **PATRONI\_RESTAPI\_KEYFILE\_PASSWORD**: Specifies a password for decrypting the keyfile.
+-  **PATRONI\_RESTAPI\_CAFILE**: Specifies the file with the CA_BUNDLE with certificates of trusted CAs to use while verifying client certs.
+-  **PATRONI\_RESTAPI\_CIPHERS**: (optional) Specifies the permitted cipher suites (e.g. "ECDHE-RSA-AES256-GCM-SHA384:DHE-RSA-AES256-GCM-SHA384:ECDHE-RSA-AES128-GCM-SHA256:DHE-RSA-AES128-GCM-SHA256:!SSLv1:!SSLv2:!SSLv3:!TLSv1:!TLSv1.1")
+-  **PATRONI\_RESTAPI\_VERIFY\_CLIENT**: ``none`` (default), ``optional`` or ``required``. When ``none`` REST API will not check client certificates. When ``required`` client certificates are required for all REST API calls. When ``optional`` client certificates are required for all unsafe REST API endpoints. When ``required`` is used, then client authentication succeeds, if the certificate signature verification succeeds. For ``optional`` the client cert will only be checked for ``PUT``, ``POST``, ``PATCH``, and ``DELETE`` requests.
+-  **PATRONI\_RESTAPI\_ALLOWLIST**: (optional): Specifies the set of hosts that are allowed to call unsafe REST API endpoints. The single element could be a host name, an IP address or a network address using CIDR notation. By default ``allow all`` is used. In case if ``allowlist`` or ``allowlist_include_members`` are set, anything that is not included is rejected.
+-  **PATRONI\_RESTAPI\_ALLOWLIST\_INCLUDE\_MEMBERS**: (optional): If set to ``true`` it allows accessing unsafe REST API endpoints from other cluster members registered in DCS (IP address or hostname is taken from the members ``api_url``). Be careful, it might happen that OS will use a different IP for outgoing connections.
+-  **PATRONI\_RESTAPI\_HTTP\_EXTRA\_HEADERS**: (optional) HTTP headers let the REST API server pass additional information with an HTTP response.
+-  **PATRONI\_RESTAPI\_HTTPS\_EXTRA\_HEADERS**: (optional) HTTPS headers let the REST API server pass additional information with an HTTP response when TLS is enabled. This will also pass additional information set in ``http_extra_headers``.
 
-ZooKeeper
+CTL
----------
+---
--  **PATRONI\_ZOOKEEPER\_HOSTS**: comma separated list of ZooKeeper cluster members: "'host1:port1','host2:port2','etc...'". It is important to quote every single entity!
+-  **PATRONICTL\_CONFIG\_FILE**: location of the configuration file.
+-  **PATRONI\_CTL\_INSECURE**: Allow connections to REST API without verifying SSL certs.
+-  **PATRONI\_CTL\_CACERT**: Specifies the file with the CA_BUNDLE file or directory with certificates of trusted CAs to use while verifying REST API SSL certs. If not provided patronictl will use the value provided for REST API "cafile" parameter.
+-  **PATRONI\_CTL\_CERTFILE**: Specifies the file with the client certificate in the PEM format. If not provided patronictl will use the value provided for REST API "certfile" parameter.
+-  **PATRONI\_CTL\_KEYFILE**: Specifies the file with the client secret key in the PEM format. If not provided patronictl will use the value provided for REST API "keyfile" parameter.

docs/README.rst

@@ -72,7 +72,7 @@ Patroni can be installed with pip:
 
 where dependencies can be either empty, or consist of one or more of the following:
 
-etcd
+etcd or etcd3
     `python-etcd` module in order to use Etcd as DCS
 consul
     `python-consul` module in order to use Consul as DCS
@@ -82,6 +82,8 @@ exhibitor
     `kazoo` module in order to use Exhibitor as DCS (same dependencies as for Zookeeper)
 kubernetes
     `kubernetes` module in order to use Kubernetes as DCS in Patroni
+raft
+    `pysyncobj` module in order to use python Raft implementation as DCS
 aws
     `boto` in order to use AWS callbacks
 
@@ -97,6 +99,13 @@ independently of Patroni.
 
 .. _running_configuring:
 
+Planning the Number of PostgreSQL Nodes
+---------------------------------------
+
+Patroni/PostgreSQL nodes are decoupled from DCS nodes (except when Patroni implements RAFT on its own) and therefore
+there is no requirement on the minimal number of nodes. Running a cluster consisting of one master and one standby is 
+perfectly fine. You can add more standby nodes later.
+
 Running and Configuring
 -----------------------
 
@@ -107,7 +116,7 @@ obtain those files from the git repository and replace `./patroni.py` below with
 To get started, do the following from different terminals:
 ::
 
-    > etcd --data-dir=data/etcd
+    > etcd --data-dir=data/etcd --enable-v2=true
     > ./patroni.py postgres0.yml
     > ./patroni.py postgres1.yml
 
@@ -154,3 +163,20 @@ When connecting from an application, always use a non-superuser. Patroni require
    :target: https://travis-ci.org/zalando/patroni
 .. |Coverage Status| image:: https://coveralls.io/repos/zalando/patroni/badge.svg?branch=master
    :target: https://coveralls.io/r/zalando/patroni?branch=master
+
+Testing Your HA Solution
+--------------------------------------
+Testing an HA solution is a time consuming process, with many variables. This is particularly true considering a cross-platform application. You need a trained system administrator or a consultant to do this work. It is not something we can cover in depth in the documentation.
+
+That said, here are some pieces of your infrastructure you should be sure to test:
+
+* Network (the network in front of your system as well as the NICs [physical or virtual] themselves)
+* Disk IO 
+* file limits (nofile in Linux)
+* RAM. Even if you have oomkiller turned off as suggested, the unavailability of RAM could cause issues.
+* CPU
+* Virtualization Contention (overcommitting the hypervisor)
+* Any cgroup limitation (likely to be related to the above)
+* ``kill -9`` of any postgres process (except postmaster!). This is a decent simulation of a segfault.
+
+One thing that you should not do is run ``kill -9`` on a postmaster process. This is because doing so does not mimic any real life scenario. If you are concerned your infrastructure is insecure and an attacker could run ``kill -9``, no amount of HA process is going to fix that. The attacker will simply kill the process again, or cause chaos in another way.

docs/SETTINGS.rst

@@ -4,6 +4,68 @@
 YAML Configuration Settings
 ===========================
 
+.. _dynamic_configuration_settings:
+
+Dynamic configuration settings
+------------------------------
+
+Dynamic configuration is stored in the DCS (Distributed Configuration Store) and applied on all cluster nodes. Some parameters, like **loop_wait**, **ttl**, **postgresql.parameters.max_connections**, **postgresql.parameters.max_worker_processes** and so on could be set only in the dynamic configuration. Some other parameters like **postgresql.listen**, **postgresql.data_dir** could be set only locally, i.e. in the Patroni config file or via :ref:`configuration <environment>` variable. In most cases the local configuration will override the dynamic configuration. In order to change the dynamic configuration you can use either ``patronictl edit-config`` tool or Patroni :ref:`REST API <rest_api>`.
+
+-  **loop\_wait**: the number of seconds the loop will sleep. Default value: 10
+-  **ttl**: the TTL to acquire the leader lock (in seconds). Think of it as the length of time before initiation of the automatic failover process. Default value: 30
+-  **retry\_timeout**: timeout for DCS and PostgreSQL operation retries (in seconds). DCS or network issues shorter than this will not cause Patroni to demote the leader. Default value: 10
+-  **maximum\_lag\_on\_failover**: the maximum bytes a follower may lag to be able to participate in leader election.
+-  **maximum\_lag\_on\_syncnode**: the maximum bytes a synchronous follower may lag before it is considered as an unhealthy candidate and swapped by healthy asynchronous follower. Patroni utilize the max replica lsn if there is more than one follower, otherwise it will use leader's current wal lsn. Default is -1, Patroni will not take action to swap synchronous unhealthy follower when the value is set to 0 or below. Please set the value high enough so Patroni won't swap synchrounous follower fequently during high transaction volume.
+-  **max\_timelines\_history**: maximum number of timeline history items kept in DCS.  Default value: 0. When set to 0, it keeps the full history in DCS.
+-  **master\_start\_timeout**: the amount of time a master is allowed to recover from failures before failover is triggered (in seconds). Default is 300 seconds. When set to 0 failover is done immediately after a crash is detected if possible. When using asynchronous replication a failover can cause lost transactions. Worst case failover time for master failure is: loop\_wait + master\_start\_timeout + loop\_wait, unless master\_start\_timeout is zero, in which case it's just loop\_wait. Set the value according to your durability/availability tradeoff.
+- **master\_stop\_timeout**: The number of seconds Patroni is allowed to wait when stopping Postgres and effective only when synchronous_mode is enabled. When set to > 0 and the synchronous_mode is enabled, Patroni sends SIGKILL to the postmaster if the stop operation is running for more than the value set by master_stop_timeout. Set the value according to your durability/availability tradeoff. If the parameter is not set or set <= 0, master_stop_timeout does not apply.
+-  **synchronous\_mode**: turns on synchronous replication mode. In this mode a replica will be chosen as synchronous and only the latest leader and synchronous replica are able to participate in leader election. Synchronous mode makes sure that successfully committed transactions will not be lost at failover, at the cost of losing availability for writes when Patroni cannot ensure transaction durability. See :ref:`replication modes documentation <replication_modes>` for details.
+-  **synchronous\_mode\_strict**: prevents disabling synchronous replication if no synchronous replicas are available, blocking all client writes to the master. See :ref:`replication modes documentation <replication_modes>` for details.
+-  **postgresql**:
+    -  **use\_pg\_rewind**: whether or not to use pg_rewind. Defaults to `false`.
+    -  **use\_slots**: whether or not to use replication slots. Defaults to `true` on PostgreSQL 9.4+.
+    -  **recovery\_conf**: additional configuration settings written to recovery.conf when configuring follower. There is no recovery.conf anymore in PostgreSQL 12, but you may continue using this section, because Patroni handles it transparently.
+    -  **parameters**: list of configuration settings for Postgres.
+-  **standby\_cluster**: if this section is defined, we want to bootstrap a standby cluster.
+    -  **host**: an address of remote master
+    -  **port**: a port of remote master
+    -  **primary\_slot\_name**: which slot on the remote master to use for replication. This parameter is optional, the default value is derived from the instance name (see function `slot_name_from_member_name`).
+    -  **create\_replica\_methods**: an ordered list of methods that can be used to bootstrap standby leader from the remote master, can be different from the list defined in :ref:`postgresql_settings`
+    -  **restore\_command**: command to restore WAL records from the remote master to standby leader, can be different from the list defined in :ref:`postgresql_settings`
+    -  **archive\_cleanup\_command**: cleanup command for standby leader
+    -  **recovery\_min\_apply\_delay**: how long to wait before actually apply WAL records on a standby leader
+-  **slots**: define permanent replication slots. These slots will be preserved during switchover/failover. The logical slots are copied from the primary to a standby with restart, and after that their position advanced every **loop_wait** seconds (if necessary). Copying logical slot files performed via ``libpq`` connection and using either rewind or superuser credentials (see **postgresql.authentication** section). There is always a chance that the logical slot position on the replica is a bit behind the former primary, therefore application should be prepared that some messages could be received the second time after the failover. The easiest way of doing so - tracking ``confirmed_flush_lsn``. Enabling permanent logical replication slots requires **postgresql.use_slots** to be set and will also automatically enable the ``hot_standby_feedback``. Since the failover of logical replication slots is unsafe on PostgreSQL 9.6 and older and PostgreSQL version 10 is missing some important functions, the feature only works with PostgreSQL 11+.
+    -  **my_slot_name**: the name of replication slot. If the permanent slot name matches with the name of the current primary it will not be created. Everything else is the responsibility of the operator to make sure that there are no clashes in names between replication slots automatically created by Patroni for members and permanent replication slots.
+        -  **type**: slot type. Could be ``physical`` or ``logical``. If the slot is logical, you have to additionally define ``database`` and ``plugin``.
+        -  **database**: the database name where logical slots should be created.
+        -  **plugin**: the plugin name for the logical slot.
+-  **ignore_slots**: list of sets of replication slot properties for which Patroni should ignore matching slots. This configuration/feature/etc. is useful when some replication slots are managed outside of Patroni. Any subset of matching properties will cause a slot to be ignored.
+    -  **name**: the name of the replication slot.
+    -  **type**: slot type. Can be ``physical`` or ``logical``. If the slot is logical, you may additionally define ``database`` and/or ``plugin``.
+    -  **database**: the database name (when matching a ``logical`` slot).
+    -  **plugin**: the logical decoding plugin (when matching a ``logical`` slot).
+
+Note: **slots** is a hashmap while **ignore_slots** is an array. For example:
+
+.. code:: YAML
+
+        slots:
+          permanent_logical_slot_name:
+            type: logical
+            database: my_db
+            plugin: test_decoding
+          permanent_physical_slot_name:
+            type: physical
+          ...
+        ignore_slots:
+          - name: ignored_logical_slot_name
+            type: logical
+            database: my_db
+            plugin: test_decoding
+          - name: ignored_physical_slot_name
+            type: physical
+          ...
+
 Global/Universal
 ----------------
 -  **name**: the name of the host. Must be unique for the cluster.
@@ -13,6 +75,7 @@ Global/Universal
 Log
 ---
 -  **level**: sets the general logging level. Default value is **INFO** (see `the docs for Python logging <https://docs.python.org/3.6/library/logging.html#levels>`_)
+-  **traceback\_level**: sets the level where tracebacks will be visible. Default value is **ERROR**. Set it to **DEBUG** if you want to see tracebacks only if you enable **log.level=DEBUG**.
 -  **format**: sets the log formatting string. Default value is **%(asctime)s %(levelname)s: %(message)s** (see `the LogRecord attributes <https://docs.python.org/3.6/library/logging.html#logrecord-attributes>`_)
 -  **dateformat**: sets the datetime formatting string. (see the `formatTime() documentation <https://docs.python.org/3.6/library/logging.html#logging.Formatter.formatTime>`_)
 -  **max\_queue\_size**: Patroni is using two-step logging. Log records are written into the in-memory queue and there is a separate thread which pulls them from the queue and writes to stderr or file. The maximum size of the internal queue is limited by default by **1000** records, which is enough to keep logs for the past 1h20m.
@@ -27,50 +90,26 @@ Log
 
 Bootstrap configuration
 -----------------------
--  **dcs**: This section will be written into `/<namespace>/<scope>/config` of a given configuration store after initializing of new cluster. This is the global configuration for the cluster. If you want to change some parameters for all cluster nodes - just do it in DCS (or via Patroni API) and all nodes will apply this configuration.
+-  **bootstrap**:
-    -  **loop\_wait**: the number of seconds the loop will sleep. Default value: 10
+    -  **dcs**: This section will be written into `/<namespace>/<scope>/config` of the given configuration store after initializing of new cluster. The global dynamic configuration for the cluster. Under the ``bootstrap.dcs`` you can put any of the parameters described in the :ref:`Dynamic Configuration settings <dynamic_configuration_settings>` and after Patroni initialized (bootstrapped) the new cluster, it will write this section into `/<namespace>/<scope>/config` of the configuration store. All later changes of ``bootstrap.dcs`` will not take any effect! If you want to change them please use either ``patronictl edit-config`` or Patroni :ref:`REST API <rest_api>`.
-    -  **ttl**: the TTL to acquire the leader lock. Think of it as the length of time before initiation of the automatic failover process. Default value: 30
+    -  **method**: custom script to use for bootstrapping this cluster.
-    -  **retry\_timeout**: timeout for DCS and PostgreSQL operation retries. DCS or network issues shorter than this will not cause Patroni to demote the leader. Default value: 10
+       See :ref:`custom bootstrap methods documentation <custom_bootstrap>` for details.
-    -  **maximum\_lag\_on\_failover**: the maximum bytes a follower may lag to be able to participate in leader election.
+       When ``initdb`` is specified revert to the default ``initdb`` command. ``initdb`` is also triggered when no ``method``
-    -  **master\_start\_timeout**: the amount of time a master is allowed to recover from failures before failover is triggered. Default is 300 seconds. When set to 0 failover is done immediately after a crash is detected if possible. When using asynchronous replication a failover can cause lost transactions. Best worst case failover time for master failure is: loop\_wait + master\_start\_timeout + loop\_wait, unless master\_start\_timeout is zero, in which case it's just loop\_wait. Set the value according to your durability/availability tradeoff.
+       parameter is present in the configuration file.
-    -  **synchronous\_mode**: turns on synchronous replication mode. In this mode a replica will be chosen as synchronous and only the latest leader and synchronous replica are able to participate in leader election. Synchronous mode makes sure that successfully committed transactions will not be lost at failover, at the cost of losing availability for writes when Patroni cannot ensure transaction durability. See :ref:`replication modes documentation <replication_modes>` for details.
+    -  **initdb**: List options to be passed on to initdb.
-    -  **synchronous\_mode\_strict**: prevents disabling synchronous replication if no synchronous replicas are available, blocking all client writes to the master. See :ref:`replication modes documentation <replication_modes>` for details.
+            -  **- data-checksums**: Must be enabled when pg_rewind is needed on 9.3.
-    -  **postgresql**:
+            -  **- encoding: UTF8**: default encoding for new databases.
-        -  **use\_pg\_rewind**: whether or not to use pg_rewind
+            -  **- locale: UTF8**: default locale for new databases.
-        -  **use\_slots**: whether or not to use replication_slots. Must be False for PostgreSQL 9.3. You should comment out max_replication_slots before it becomes ineligible for leader status.
+    -  **pg\_hba**: list of lines that you should add to pg\_hba.conf.
-        -  **recovery\_conf**: additional configuration settings written to recovery.conf when configuring follower.
+            -  **- host all all 0.0.0.0/0 md5**.
-        -  **parameters**: list of configuration settings for Postgres. Many of these are required for replication to work.
+            -  **- host replication replicator 127.0.0.1/32 md5**: A line like this is required for replication.
-    -  **standby\_cluster**: if this section is defined, we want to bootstrap a standby cluster.
+    -  **users**: Some additional users which need to be created after initializing new cluster
-        -  **host**: an address of remote master
+        -  **admin**: the name of user
-        -  **port**: a port of remote master
+            -  **password: zalando**:
-        -  **primary\_slot\_name**: which slot on the remote master to use for replication. This parameter is optional, the default value is derived from the instance name (see function `slot_name_from_member_name`).
+            -  **options**: list of options for CREATE USER statement
-        -  **create\_replica\_methods**: an ordered list of methods that can be used to bootstrap standby leader from the remote master, can be different from the list defined in :ref:`postgresql_settings`
+                -  **- createrole**
-        -  **restore\_command**: command to restore WAL records from the remote master to standby leader, can be different from the list defined in :ref:`postgresql_settings`
+                -  **- createdb**
-        -  **archive\_cleanup\_command**: cleanup command for standby leader
+    -  **post\_bootstrap** or **post\_init**: An additional script that will be executed after initializing the cluster. The script receives a connection string URL (with the cluster superuser as a user name). The PGPASSFILE variable is set to the location of pgpass file.
-        -  **recovery\_min\_apply\_delay**: how long to wait before actually apply WAL records on a standby leader
-    -  **slots**: define permanent replication slots. These slots will be preserved during switchover/failover. Patroni will try to create slots before opening connections to the cluster.
-        -  **my_slot_name**: the name of replication slot. It is the responsibility of the operator to make sure that there are no clashes in names between replication slots automatically created by Patroni for members and permanent replication slots.
-            -  **type**: slot type. Could be ``physical`` or ``logical``. If the slot is logical, you have to additionally define ``database`` and ``plugin``.
-            -  **database**: the database name where logical slots should be created.
-            -  **plugin**: the plugin name for the logical slot.
--  **method**: custom script to use for bootstrapping this cluster.
-   See :ref:`custom bootstrap methods documentation <custom_bootstrap>` for details.
-   When ``initdb`` is specified revert to the default ``initdb`` command. ``initdb`` is also triggered when no ``method``
-   parameter is present in the configuration file.
--  **initdb**: List options to be passed on to initdb.
-        -  **- data-checksums**: Must be enabled when pg_rewind is needed on 9.3.
-        -  **- encoding: UTF8**: default encoding for new databases.
-        -  **- locale: UTF8**: default locale for new databases.
--  **pg\_hba**: list of lines that you should add to pg\_hba.conf.
-        -  **- host all all 0.0.0.0/0 md5**.
-        -  **- host replication replicator 127.0.0.1/32 md5**: A line like this is required for replication.
--  **users**: Some additional users which need to be created after initializing new cluster
-    -  **admin**: the name of user
-        -  **password: zalando**:
-        -  **options**: list of options for CREATE USER statement
-            -  **- createrole**
-            -  **- createdb**
--  **post\_bootstrap** or **post\_init**: An additional script that will be executed after initializing the cluster. The script receives a connection string URL (with the cluster superuser as a user name). The PGPASSFILE variable is set to the location of pgpass file.
 
 .. _consul_settings:
 
@@ -78,20 +117,35 @@ Consul
 ------
 Most of the parameters are optional, but you have to specify one of the **host** or **url**
 
--  **host**: the host:port for the Consul endpoint, in format: http(s)://host:port
+-  **host**: the host:port for the Consul local agent.
--  **url**: url for the Consul endpoint
+-  **url**: url for the Consul local agent, in format: http(s)://host:port.
--  **port**: (optional) Consul port
+-  **port**: (optional) Consul port.
--  **scheme**: (optional) **http** or **https**, defaults to **http**
+-  **scheme**: (optional) **http** or **https**, defaults to **http**.
--  **token**: (optional) ACL token
+-  **token**: (optional) ACL token.
--  **verify**: (optional) whether to verify the SSL certificate for HTTPS requests
+-  **verify**: (optional) whether to verify the SSL certificate for HTTPS requests.
 -  **cacert**: (optional) The ca certificate. If present it will enable validation.
--  **cert**: (optional) file with the client certificate
+-  **cert**: (optional) file with the client certificate.
 -  **key**: (optional) file with the client key. Can be empty if the key is part of **cert**.
 -  **dc**: (optional) Datacenter to communicate with. By default the datacenter of the host is used.
 -  **consistency**: (optional) Select consul consistency mode. Possible values are ``default``, ``consistent``, or ``stale`` (more details in `consul API reference <https://www.consul.io/api/features/consistency.html/>`__)
--  **checks**: (optional) list of Consul health checks used for the session. If not specified Consul will use "serfHealth" in additional to the TTL based check created by Patroni. Additional checks, in particular the "serfHealth", may cause the leader lock to expire faster than in `ttl` seconds when the leader instance becomes unavailable
+-  **checks**: (optional) list of Consul health checks used for the session. By default an empty list is used.
--  **register\_service**: (optional) whether or not to register a service with the name defined by the scope parameter and the tag master, replica or standby-leader depending on the node's role. Defaults to **false**
+-  **register\_service**: (optional) whether or not to register a service with the name defined by the scope parameter and the tag master, replica or standby-leader depending on the node's role. Defaults to **false**.
--  **service\_check\_interval**: (optional) how often to perform health check against registered url
+-  **service\_tags**: (optional) additional static tags to add to the Consul service apart from the role (``master``/``replica``/``standby-leader``).  By default an empty list is used.
+-  **service\_check\_interval**: (optional) how often to perform health check against registered url.
+
+The ``token`` needs to have the following ACL permissions:
+
+::
+
+    service_prefix "${scope}" {
+        policy = "write"
+    }
+    key_prefix "${namespace}/${scope}" {
+        policy = "write"
+    }
+    session_prefix "" {
+        policy = "write"
+    }
 
 Etcd
 ----
@@ -100,8 +154,8 @@ Most of the parameters are optional, but you have to specify one of the **host**
 -  **host**: the host:port for the etcd endpoint.
 -  **hosts**: list of etcd endpoint in format host1:port1,host2:port2,etc... Could be a comma separated string or an actual yaml list.
 -  **use\_proxies**: If this parameter is set to true, Patroni will consider **hosts** as a list of proxies and will not perform a topology discovery of etcd cluster.
--  **url**: url for the etcd
+-  **url**: url for the etcd.
--  **proxy**: proxy url for the etcd. If you are connecting to the etcd using proxy, use this parameter instead of **url**
+-  **proxy**: proxy url for the etcd. If you are connecting to the etcd using proxy, use this parameter instead of **url**.
 -  **srv**: Domain to search the SRV record(s) for cluster autodiscovery.
 -  **protocol**: (optional) http or https, if not specified http is used. If the **url** or **proxy** is specified - will take protocol from them.
 -  **username**: (optional) username for etcd authentication.
@@ -110,98 +164,225 @@ Most of the parameters are optional, but you have to specify one of the **host**
 -  **cert**: (optional) file with the client certificate.
 -  **key**: (optional) file with the client key. Can be empty if the key is part of **cert**.
 
+Etcdv3
+------
+If you want that Patroni works with Etcd cluster via protocol version 3, you need to use the ``etcd3`` section in the Patroni configuration file. All configuration parameters are the same as for ``etcd``.
+
+.. warning::
+    Keys created with protocol version 2 are not visible with protocol version 3 and the other way around, therefore it is not possible to switch from ``etcd`` to ``etcd3`` just by updating Patroni config file.
+
+
+ZooKeeper
+----------
+-  **hosts**: List of ZooKeeper cluster members in format: ['host1:port1', 'host2:port2', 'etc...'].
+-  **use_ssl**: (optional) Whether SSL is used or not. Defaults to ``false``. If set to ``false``, all SSL specific parameters are ignored.
+-  **cacert**: (optional) The CA certificate. If present it will enable validation.
+-  **cert**: (optional) File with the client certificate.
+-  **key**: (optional) File with the client key.
+-  **key_password**: (optional) The client key password.
+-  **verify**: (optional) Whether to verify certificate or not. Defaults to ``true``.
+
+.. note::
+    It is required to install ``kazoo>=2.6.0`` to support SSL.
+
+
 Exhibitor
 ---------
 -  **hosts**: initial list of Exhibitor (ZooKeeper) nodes in format: 'host1,host2,etc...'. This list updates automatically whenever the Exhibitor (ZooKeeper) cluster topology changes.
--  **poll\_interval**: how often the list of ZooKeeper and Exhibitor nodes should be updated from Exhibitor
+-  **poll\_interval**: how often the list of ZooKeeper and Exhibitor nodes should be updated from Exhibitor.
 -  **port**: Exhibitor port.
 
 .. _kubernetes_settings:
 
 Kubernetes
 ----------
+-  **bypass\_api\_service**: (optional) When communicating with the Kubernetes API, Patroni is usually relying on the `kubernetes` service, the address of which is exposed in the pods via the `KUBERNETES_SERVICE_HOST` environment variable. If `bypass_api_service` is set to ``true``, Patroni will resolve the list of API nodes behind the service and connect directly to them.
 -  **namespace**: (optional) Kubernetes namespace where Patroni pod is running. Default value is `default`.
 -  **labels**: Labels in format ``{label1: value1, label2: value2}``. These labels will be used to find existing objects (Pods and either Endpoints or ConfigMaps) associated with the current cluster. Also Patroni will set them on every object (Endpoint or ConfigMap) it creates.
 -  **scope\_label**: (optional) name of the label containing cluster name. Default value is `cluster-name`.
 -  **role\_label**: (optional) name of the label containing role (master or replica). Patroni will set this label on the pod it runs in. Default value is ``role``.
 -  **use\_endpoints**: (optional) if set to true, Patroni will use Endpoints instead of ConfigMaps to run leader elections and keep cluster state.
 -  **pod\_ip**: (optional) IP address of the pod Patroni is running in. This value is required when `use_endpoints` is enabled and is used to populate the leader endpoint subsets when the pod's PostgreSQL is promoted.
--  **ports**: (optional) if the Service object has the name for the port, the same name must appear in the Endpoint object, otherwise service won't work. For example, if your service is defined as ``{Kind: Service, spec: {ports: [{name: postgresql, port: 5432, targetPort: 5432}]}}``, then you have to set ``kubernetes.ports: {[{"name": "postgresql", "port": 5432}]}`` and Patroni will use it for updating subsets of the leader Endpoint. This parameter is used only if `kubernetes.use_endpoints` is set.
+-  **ports**: (optional) if the Service object has the name for the port, the same name must appear in the Endpoint object, otherwise service won't work. For example, if your service is defined as ``{Kind: Service, spec: {ports: [{name: postgresql, port: 5432, targetPort: 5432}]}}``, then you have to set ``kubernetes.ports: [{"name": "postgresql", "port": 5432}]`` and Patroni will use it for updating subsets of the leader Endpoint. This parameter is used only if `kubernetes.use_endpoints` is set.
+
+
+.. _raft_settings:
+
+Raft
+----
+-  **self\_addr**: ``ip:port`` to listen on for Raft connections. The ``self_addr`` must be accessible from other nodes of the cluster. If not set, the node will not participate in consensus.
+-  **bind\_addr**: (optional) ``ip:port`` to listen on for Raft connections. If not specified the ``self_addr`` will be used.
+-  **partner\_addrs**: list of other Patroni nodes in the cluster in format: ['ip1:port', 'ip2:port', 'etc...']
+-  **data\_dir**: directory where to store Raft log and snapshot. If not specified the current working directory is used.
+-  **password**: (optional) Encrypt Raft traffic with a specified password, requires ``cryptography`` python module.
+
+  Short FAQ about Raft implementation
+
+  - Q: How to list all the nodes providing consensus?
+
+    A: ``syncobj_admin -conn host:port`` -status where the host:port is the address of one of the cluster nodes
+
+  - Q: Node that was a part of consensus and has gone and I can't reuse the same IP for other node. How to remove this node from the consensus?
+
+    A: ``syncobj_admin -conn host:port -remove host2:port2`` where the ``host2:port2`` is the address of the node you want to remove from consensus.
+
+  - Q: Where to get the ``syncobj_admin`` utility?
+
+    A: It is installed together with ``pysyncobj`` module (python RAFT implementation), which is Patroni dependency.
+
+  - Q: it is possible to run Patroni node without adding in to the consensus?
+
+    A: Yes, just comment out or remove ``raft.self_addr`` from Patroni configuration.
+
+  - Q: It is possible to run Patroni and PostgreSQL only on two nodes?
+
+    A: Yes, on the third node you can run ``patroni_raft_controller`` (without Patroni and PostgreSQL). In such a setup, one can temporarily lose one node without affecting the primary.
+
 
 .. _postgresql_settings:
 
 PostgreSQL
 ----------
--  **authentication**:
+-  **postgresql**:
-    -  **superuser**:
+    -  **authentication**:
-        -  **username**: name for the superuser, set during initialization (initdb) and later used by Patroni to connect to the postgres.
+        -  **superuser**:
-        -  **password**: password for the superuser, set during initialization (initdb).
+            -  **username**: name for the superuser, set during initialization (initdb) and later used by Patroni to connect to the postgres.
-    -  **replication**:
+            -  **password**: password for the superuser, set during initialization (initdb).
-        -  **username**: replication username; the user will be created during initialization. Replicas will use this user to access master via streaming replication
+            -  **sslmode**: (optional) maps to the `sslmode <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLMODE>`__ connection parameter, which allows a client to specify the type of TLS negotiation mode with the server. For more information on how each mode works, please visit the `PostgreSQL documentation <https://www.postgresql.org/docs/current/libpq-ssl.html#LIBPQ-SSL-SSLMODE-STATEMENTS>`__. The default mode is ``prefer``.
-        -  **password**: replication password; the user will be created during initialization.
+            -  **sslkey**: (optional) maps to the `sslkey <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLKEY>`__ connection parameter, which specifies the location of the secret key used with the client's certificate.
-    -  **rewind**:
+            -  **sslpassword**: (optional) maps to the `sslpassword <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLPASSWORD>`__ connection parameter, which specifies the password for the secret key specified in ``sslkey``.
-        -  **username**: name for the user for ``pg_rewind``; the user will be created during initialization of postgres 11+ and all necessary `permissions <https://www.postgresql.org/docs/11/app-pgrewind.html#id-1.9.5.8.8>`__ will be granted.
+            -  **sslcert**: (optional) maps to the `sslcert <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLCERT>`__ connection parameter, which specifies the location of the client certificate.
-        -  **password**: password for the user for ``pg_rewind``; the user will be created during initialization.
+            -  **sslrootcert**: (optional) maps to the `sslrootcert <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLROOTCERT>`__ connection parameter, which specifies the location of a file containing one ore more certificate authorities (CA) certificates that the client will use to verify a server's certificate.
--  **callbacks**: callback scripts to run on certain actions. Patroni will pass the action, role and cluster name. (See scripts/aws.py as an example of how to write them.)
+            -  **sslcrl**: (optional) maps to the `sslcrl <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLCRL>`__ connection parameter, which specifies the location of a file containing a certificate revocation list. A client will reject connecting to any server that has a certificate present in this list.
-        -  **on\_reload**: run this script when configuration reload is triggered.
+            -  **gssencmode**: (optional) maps to the `gssencmode <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-GSSENCMODE>`__ connection parameter, which determines whether or with what priority a secure GSS TCP/IP connection will be negotiated with the server
-        -  **on\_restart**: run this script when the postgres restarts (without changing role).
+            -  **channel_binding**: (optional) maps to the `channel_binding <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-CHANNEL-BINDING>`__ connection parameter, which controls the client's use of channel binding.
-        -  **on\_role\_change**: run this script when the postgres is being promoted or demoted.
+        -  **replication**:
-        -  **on\_start**: run this script when the postgres starts.
+            -  **username**: replication username; the user will be created during initialization. Replicas will use this user to access master via streaming replication
-        -  **on\_stop**: run this script when the postgres stops.
+            -  **password**: replication password; the user will be created during initialization.
--  **connect\_address**: IP address + port through which Postgres is accessible from other nodes and applications.
+            -  **sslmode**: (optional) maps to the `sslmode <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLMODE>`__ connection parameter, which allows a client to specify the type of TLS negotiation mode with the server. For more information on how each mode works, please visit the `PostgreSQL documentation <https://www.postgresql.org/docs/current/libpq-ssl.html#LIBPQ-SSL-SSLMODE-STATEMENTS>`__. The default mode is ``prefer``.
--  **create\_replica\_methods**: an ordered list of the create methods for turning a Patroni node into a new replica.
+            -  **sslkey**: (optional) maps to the `sslkey <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLKEY>`__ connection parameter, which specifies the location of the secret key used with the client's certificate.
-   "basebackup" is the default method; other methods are assumed to refer to scripts, each of which is configured as its
+            -  **sslpassword**: (optional) maps to the `sslpassword <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLPASSWORD>`__ connection parameter, which specifies the password for the secret key specified in ``sslkey``.
-   own config item. See :ref:`custom replica creation methods documentation <custom_replica_creation>` for further explanation.
+            -  **sslcert**: (optional) maps to the `sslcert <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLCERT>`__ connection parameter, which specifies the location of the client certificate.
--  **data\_dir**: The location of the Postgres data directory, either :ref:`existing <existing_data>` or to be initialized by Patroni.
+            -  **sslrootcert**: (optional) maps to the `sslrootcert <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLROOTCERT>`__ connection parameter, which specifies the location of a file containing one ore more certificate authorities (CA) certificates that the client will use to verify a server's certificate.
--  **config\_dir**: The location of the Postgres configuration directory, defaults to the data directory. Must be writable by Patroni.
+            -  **sslcrl**: (optional) maps to the `sslcrl <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLCRL>`__ connection parameter, which specifies the location of a file containing a certificate revocation list. A client will reject connecting to any server that has a certificate present in this list.
--  **bin\_dir**: Path to PostgreSQL binaries (pg_ctl, pg_rewind, pg_basebackup, postgres). The default value is an empty string meaning that PATH environment variable will be used to find the executables.
+            -  **gssencmode**: (optional) maps to the `gssencmode <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-GSSENCMODE>`__ connection parameter, which determines whether or with what priority a secure GSS TCP/IP connection will be negotiated with the server
--  **listen**: IP address + port that Postgres listens to; must be accessible from other nodes in the cluster, if you're using streaming replication. Multiple comma-separated addresses are permitted, as long as the port component is appended after to the last one with a colon, i.e. ``listen: 127.0.0.1,127.0.0.2:5432``. Patroni will use the first address from this list to establish local connections to the PostgreSQL node.
+            -  **channel_binding**: (optional) maps to the `channel_binding <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-CHANNEL-BINDING>`__ connection parameter, which controls the client's use of channel binding.
--  **use\_unix\_socket**: specifies that Patroni should prefer to use unix sockets to connect to the cluster. Default value is ``false``. If ``unix_socket_directories`` is defined, Patroni will use the first suitable value from it to connect to the cluster and fallback to tcp if nothing is suitable. If ``unix_socket_directories`` is not specified in ``postgresql.parameters``, Patroni will assume that the default value should be used and omit ``host`` from the connection parameters.
+        -  **rewind**:
--  **pgpass**: path to the `.pgpass <https://www.postgresql.org/docs/current/static/libpq-pgpass.html>`__ password file. Patroni creates this file before executing pg\_basebackup, the post_init script and under some other circumstances. The location must be writable by Patroni.
+            -  **username**: name for the user for ``pg_rewind``; the user will be created during initialization of postgres 11+ and all necessary `permissions <https://www.postgresql.org/docs/11/app-pgrewind.html#id-1.9.5.8.8>`__ will be granted.
--  **recovery\_conf**: additional configuration settings written to recovery.conf when configuring follower.
+            -  **password**: password for the user for ``pg_rewind``; the user will be created during initialization.
--  **custom\_conf** : path to an optional custom ``postgresql.conf`` file, that will be used in place of ``postgresql.base.conf``. The file must exist on all cluster nodes, be readable by PostgreSQL and will be included from its location on the real ``postgresql.conf``. Note that Patroni will not monitor this file for changes, nor backup it. However, its settings can still be overridden by Patroni's own configuration facilities - see :ref:`dynamic configuration <dynamic_configuration>` for details.
+            -  **sslmode**: (optional) maps to the `sslmode <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLMODE>`__ connection parameter, which allows a client to specify the type of TLS negotiation mode with the server. For more information on how each mode works, please visit the `PostgreSQL documentation <https://www.postgresql.org/docs/current/libpq-ssl.html#LIBPQ-SSL-SSLMODE-STATEMENTS>`__. The default mode is ``prefer``.
--  **parameters**: list of configuration settings for Postgres. Many of these are required for replication to work.
+            -  **sslkey**: (optional) maps to the `sslkey <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLKEY>`__ connection parameter, which specifies the location of the secret key used with the client's certificate.
--  **pg\_hba**: list of lines that Patroni will use to generate ``pg_hba.conf``. This parameter has higher priority than ``bootstrap.pg_hba``. Together with :ref:`dynamic configuration <dynamic_configuration>` it simplifies management of ``pg_hba.conf``.
+            -  **sslpassword**: (optional) maps to the `sslpassword <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLPASSWORD>`__ connection parameter, which specifies the password for the secret key specified in ``sslkey``.
-        -  **- host all all 0.0.0.0/0 md5**.
+            -  **sslcert**: (optional) maps to the `sslcert <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLCERT>`__ connection parameter, which specifies the location of the client certificate.
-        -  **- host replication replicator 127.0.0.1/32 md5**: A line like this is required for replication.
+            -  **sslrootcert**: (optional) maps to the `sslrootcert <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLROOTCERT>`__ connection parameter, which specifies the location of a file containing one ore more certificate authorities (CA) certificates that the client will use to verify a server's certificate.
--  **pg\_ident**: list of lines that Patroni will use to generate ``pg_ident.conf``. Together with :ref:`dynamic configuration <dynamic_configuration>` it simplifies management of ``pg_ident.conf``.
+            -  **sslcrl**: (optional) maps to the `sslcrl <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-SSLCRL>`__ connection parameter, which specifies the location of a file containing a certificate revocation list. A client will reject connecting to any server that has a certificate present in this list.
-        -  **- mapname1 systemname1 pguser1**.
+            -  **gssencmode**: (optional) maps to the `gssencmode <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-GSSENCMODE>`__ connection parameter, which determines whether or with what priority a secure GSS TCP/IP connection will be negotiated with the server
-        -  **- mapname1 systemname2 pguser2**.
+            -  **channel_binding**: (optional) maps to the `channel_binding <https://www.postgresql.org/docs/current/libpq-connect.html#LIBPQ-CONNECT-CHANNEL-BINDING>`__ connection parameter, which controls the client's use of channel binding.
--  **pg\_ctl\_timeout**: How long should pg_ctl wait when doing ``start``, ``stop`` or ``restart``. Default value is 60 seconds.
+    -  **callbacks**: callback scripts to run on certain actions. Patroni will pass the action, role and cluster name. (See scripts/aws.py as an example of how to write them.)
--  **use\_pg\_rewind**: try to use pg\_rewind on the former leader when it joins cluster as a replica.
+            -  **on\_reload**: run this script when configuration reload is triggered.
--  **remove\_data\_directory\_on\_rewind\_failure**: If this option is enabled, Patroni will remove the PostgreSQL data directory and recreate the replica. Otherwise it will try to follow the new leader. Default value is **false**.
+            -  **on\_restart**: run this script when the postgres restarts (without changing role).
--  **remove\_data\_directory\_on\_diverged\_timelines**: Patroni will remove the PostgreSQL data directory and recreate the replica if it notices that timelines are diverging and the former master can not start streaming from the new master. This option is useful when ``pg_rewind`` can not be used. Default value is **false**.
+            -  **on\_role\_change**: run this script when the postgres is being promoted or demoted.
--  **replica\_method**: for each create_replica_methods other than basebackup, you would add a configuration section of the same name. At a minimum, this should include "command" with a full path to the actual script to be executed. Other configuration parameters will be passed along to the script in the form "parameter=value".
+            -  **on\_start**: run this script when the postgres starts.
+            -  **on\_stop**: run this script when the postgres stops.
+    -  **connect\_address**: IP address + port through which Postgres is accessible from other nodes and applications.
+    -  **create\_replica\_methods**: an ordered list of the create methods for turning a Patroni node into a new replica.
+       "basebackup" is the default method; other methods are assumed to refer to scripts, each of which is configured as its
+       own config item. See :ref:`custom replica creation methods documentation <custom_replica_creation>` for further explanation.
+    -  **data\_dir**: The location of the Postgres data directory, either :ref:`existing <existing_data>` or to be initialized by Patroni.
+    -  **config\_dir**: The location of the Postgres configuration directory, defaults to the data directory. Must be writable by Patroni.
+    -  **bin\_dir**: Path to PostgreSQL binaries (pg_ctl, pg_rewind, pg_basebackup, postgres). The default value is an empty string meaning that PATH environment variable will be used to find the executables.
+    -  **listen**: IP address + port that Postgres listens to; must be accessible from other nodes in the cluster, if you're using streaming replication. Multiple comma-separated addresses are permitted, as long as the port component is appended after to the last one with a colon, i.e. ``listen: 127.0.0.1,127.0.0.2:5432``. Patroni will use the first address from this list to establish local connections to the PostgreSQL node.
+    -  **use\_unix\_socket**: specifies that Patroni should prefer to use unix sockets to connect to the cluster. Default value is ``false``. If ``unix_socket_directories`` is defined, Patroni will use the first suitable value from it to connect to the cluster and fallback to tcp if nothing is suitable. If ``unix_socket_directories`` is not specified in ``postgresql.parameters``, Patroni will assume that the default value should be used and omit ``host`` from the connection parameters.
+    -  **use\_unix\_socket\_repl**: specifies that Patroni should prefer to use unix sockets for replication user cluster connection. Default value is ``false``. If ``unix_socket_directories`` is defined, Patroni will use the first suitable value from it to connect to the cluster and fallback to tcp if nothing is suitable. If ``unix_socket_directories`` is not specified in ``postgresql.parameters``, Patroni will assume that the default value should be used and omit ``host`` from the connection parameters.
+    -  **pgpass**: path to the `.pgpass <https://www.postgresql.org/docs/current/static/libpq-pgpass.html>`__ password file. Patroni creates this file before executing pg\_basebackup, the post_init script and under some other circumstances. The location must be writable by Patroni.
+    -  **recovery\_conf**: additional configuration settings written to recovery.conf when configuring follower.
+    -  **custom\_conf** : path to an optional custom ``postgresql.conf`` file, that will be used in place of ``postgresql.base.conf``. The file must exist on all cluster nodes, be readable by PostgreSQL and will be included from its location on the real ``postgresql.conf``. Note that Patroni will not monitor this file for changes, nor backup it. However, its settings can still be overridden by Patroni's own configuration facilities - see :ref:`dynamic configuration <dynamic_configuration>` for details.
+    -  **parameters**: list of configuration settings for Postgres. Many of these are required for replication to work.
+    -  **pg\_hba**: list of lines that Patroni will use to generate ``pg_hba.conf``. This parameter has higher priority than ``bootstrap.pg_hba``. Together with :ref:`dynamic configuration <dynamic_configuration>` it simplifies management of ``pg_hba.conf``.
+            -  **- host all all 0.0.0.0/0 md5**.
+            -  **- host replication replicator 127.0.0.1/32 md5**: A line like this is required for replication.
+    -  **pg\_ident**: list of lines that Patroni will use to generate ``pg_ident.conf``. Together with :ref:`dynamic configuration <dynamic_configuration>` it simplifies management of ``pg_ident.conf``.
+            -  **- mapname1 systemname1 pguser1**.
+            -  **- mapname1 systemname2 pguser2**.
+    -  **pg\_ctl\_timeout**: How long should pg_ctl wait when doing ``start``, ``stop`` or ``restart``. Default value is 60 seconds.
+    -  **use\_pg\_rewind**: try to use pg\_rewind on the former leader when it joins cluster as a replica.
+    -  **remove\_data\_directory\_on\_rewind\_failure**: If this option is enabled, Patroni will remove the PostgreSQL data directory and recreate the replica. Otherwise it will try to follow the new leader. Default value is **false**.
+    -  **remove\_data\_directory\_on\_diverged\_timelines**: Patroni will remove the PostgreSQL data directory and recreate the replica if it notices that timelines are diverging and the former master can not start streaming from the new master. This option is useful when ``pg_rewind`` can not be used. Default value is **false**.
+    -  **replica\_method**: for each create_replica_methods other than basebackup, you would add a configuration section of the same name. At a minimum, this should include "command" with a full path to the actual script to be executed. Other configuration parameters will be passed along to the script in the form "parameter=value".
+    -  **pre\_promote**: a fencing script that executes during a failover after acquiring the leader lock but before promoting the replica. If the script exits with a non-zero code, Patroni does not promote the replica and removes the leader key from DCS.
 
 REST API
 --------
--  **connect\_address**: IP address (or hostname) and port, to access the Patroni's REST API. All the members of the cluster must be able to connect to this address, so unless the Patroni setup is intended for a demo inside the localhost, this address must be a non "localhost" or loopback addres (ie: "localhost" or "127.0.0.1"). It can serve as a endpoint for HTTP health checks (read below about the "listen" REST API parameter), and also for user queries (either directly or via the REST API), as well as for the health checks done by the cluster members during leader elections (for example, to determine whether the master is still running, or if there is a node which has a WAL position that is ahead of the one doing the query; etc.) The connect_address is put in the member key in DCS, making it possible to translate the member name into the address to connect to its REST API.
+-  **restapi**:
+        -  **connect\_address**: IP address (or hostname) and port, to access the Patroni's :ref:`REST API <rest_api>`. All the members of the cluster must be able to connect to this address, so unless the Patroni setup is intended for a demo inside the localhost, this address must be a non "localhost" or loopback address (ie: "localhost" or "127.0.0.1"). It can serve as an endpoint for HTTP health checks (read below about the "listen" REST API parameter), and also for user queries (either directly or via the REST API), as well as for the health checks done by the cluster members during leader elections (for example, to determine whether the master is still running, or if there is a node which has a WAL position that is ahead of the one doing the query; etc.) The connect_address is put in the member key in DCS, making it possible to translate the member name into the address to connect to its REST API.
 
--  **listen**: IP address (or hostname) and port that Patroni will listen to for the REST API - to provide also the same health checks and cluster messaging between the participating nodes, as described above. to provide health-check information for HAProxy (or any other load balancer capable of doing a HTTP "OPTION" or "GET" checks).
+        -  **listen**: IP address (or hostname) and port that Patroni will listen to for the REST API - to provide also the same health checks and cluster messaging between the participating nodes, as described above. to provide health-check information for HAProxy (or any other load balancer capable of doing a HTTP "OPTION" or "GET" checks).
 
--  **Optional**:
+        -  **authentication**: (optional)
-        -  **authentication**:
             -  **username**: Basic-auth username to protect unsafe REST API endpoints.
             -  **password**: Basic-auth password to protect unsafe REST API endpoints.
+        -  **certfile**: (optional): Specifies the file with the certificate in the PEM format. If the certfile is not specified or is left empty, the API server will work without SSL.
+        -  **keyfile**: (optional): Specifies the file with the secret key in the PEM format.
+        -  **keyfile_password**: (optional): Specifies a password for decrypting the keyfile.
+        -  **cafile**: (optional): Specifies the file with the CA_BUNDLE with certificates of trusted CAs to use while verifying client certs.
+        -  **ciphers**: (optional): Specifies the permitted cipher suites (e.g. "ECDHE-RSA-AES256-GCM-SHA384:DHE-RSA-AES256-GCM-SHA384:ECDHE-RSA-AES128-GCM-SHA256:DHE-RSA-AES128-GCM-SHA256:!SSLv1:!SSLv2:!SSLv3:!TLSv1:!TLSv1.1")
+        -  **verify\_client**: (optional): ``none`` (default), ``optional`` or ``required``. When ``none`` REST API will not check client certificates. When ``required`` client certificates are required for all REST API calls. When ``optional`` client certificates are required for all unsafe REST API endpoints. When ``required`` is used, then client authentication succeeds, if the certificate signature verification succeeds.  For ``optional`` the client cert will only be checked for ``PUT``, ``POST``, ``PATCH``, and ``DELETE`` requests.
+        -  **allowlist**: (optional): Specifies the set of hosts that are allowed to call unsafe REST API endpoints. The single element could be a host name, an IP address or a network address using CIDR notation. By default ``allow all`` is used. In case if ``allowlist`` or ``allowlist_include_members`` are set, anything that is not included is rejected.
+        -  **allowlist\_include\_members**: (optional): If set to ``true`` it allows accessing unsafe REST API endpoints from other cluster members registered in DCS (IP address or hostname is taken from the members ``api_url``). Be careful, it might happen that OS will use a different IP for outgoing connections.
+        -  **http\_extra\_headers**: (optional): HTTP headers let the REST API server pass additional information with an HTTP response.
+        -  **https\_extra\_headers**: (optional): HTTPS headers let the REST API server pass additional information with an HTTP response when TLS is enabled. This will also pass additional information set in ``http_extra_headers``.
 
-        -  **certfile**: Specifies the file with the certificate in the PEM format. If the certfile is not specified or is left empty, the API server will work without SSL.
+Here is an example of both **http_extra_headers** and **https_extra_headers**:
-        -  **keyfile**: Specifies the file with the secret key in the PEM format.
+
+.. code:: YAML
+
+        restapi:
+          listen: <listen>
+          connect_address: <connect_address>
+          authentication:
+            username: <username>
+            password: <password>
+          http_extra_headers:
+            'X-Frame-Options': 'SAMEORIGIN'
+            'X-XSS-Protection': '1; mode=block'
+            'X-Content-Type-Options': 'nosniff'
+          cafile: <ca file>
+          certfile: <cert>
+          keyfile: <key>
+          https_extra_headers:
+            'Strict-Transport-Security': 'max-age=31536000; includeSubDomains'
 
 .. _patronictl_settings:
 
 CTL
 ---
--  **Optional**:
+- **ctl**: (optional)
     -  **insecure**: Allow connections to REST API without verifying SSL certs.
-    -  **cacert**: Specifices the file with the CA_BUNDLE file or directory with certificates of trusted CAs to use while verifying REST API SSL certs.
+    -  **cacert**: Specifies the file with the CA_BUNDLE file or directory with certificates of trusted CAs to use while verifying REST API SSL certs. If not provided patronictl will use the value provided for REST API "cafile" parameter.
-    -  **certfile**: Specifies the file with the certificate in the PEM format to use while verifying REST API SSL certs. If not provided patronictl will use the value provided for REST API "certfile" parameter.
+    -  **certfile**: Specifies the file with the client certificate in the PEM format. If not provided patronictl will use the value provided for REST API "certfile" parameter.
-
+    -  **keyfile**: Specifies the file with the client secret key in the PEM format. If not provided patronictl will use the value provided for REST API "keyfile" parameter.
-ZooKeeper
-----------
--  **hosts**: list of ZooKeeper cluster members in format: ['host1:port1', 'host2:port2', 'etc...'].
 
 Watchdog
 --------
 - **mode**: ``off``, ``automatic`` or ``required``. When ``off`` watchdog is disabled. When ``automatic`` watchdog will be used if available, but ignored if it is not. When ``required`` the node will not become a leader unless watchdog can be successfully enabled.
 - **device**: Path to watchdog device. Defaults to ``/dev/watchdog``.
 - **safety_margin**: Number of seconds of safety margin between watchdog triggering and leader key expiration.
+
+.. _tags_settings:
+
+Tags
+----
+- **nofailover**: ``true`` or ``false``, controls whether this node is allowed to participate in the leader race and become a leader. Defaults to ``false``
+- **clonefrom**: ``true`` or ``false``. If set to ``true`` other nodes might prefer to use this node for bootstrap (take ``pg_basebackup`` from). If there are several nodes with ``clonefrom`` tag set to ``true`` the node to bootstrap from will be chosen randomly. The default value is ``false``.
+- **noloadbalance**: ``true`` or ``false``. If set to ``true`` the node will return HTTP Status Code 503 for the ``GET /replica`` REST API health-check and therefore will be excluded from the load-balancing. Defaults to ``false``.
+- **replicatefrom**: The IP address/hostname of another replica. Used to support cascading replication.
+- **nosync**: ``true`` or ``false``. If set to ``true`` the node will never be selected as a synchronous replica.
+
+In addition to these predefined tags, you can also add your own ones:
+
+- **key1**: ``true``
+- **key2**: ``false``
+- **key3**: ``1.4``
+- **key4**: ``"RandomString"``
+
+Tags are visible in the :ref:`REST API <rest_api>` and ``patronictl list`` You can also check for an instance health using these tags. If the tag isn't defined for an instance, or if the respective value doesn't match the querying value, it will return HTTP Status Code 503.

docs/dynamic_configuration.rst

@@ -14,12 +14,14 @@ Patroni configuration is stored in the DCS (Distributed Configuration Store). Th
 
 - Local :ref:`configuration <settings>` (patroni.yml).
 	These options are defined in the configuration file and take precedence over dynamic configuration.
-	patroni.yml could be changed and reload in runtime (without restart of Patroni) by sending SIGHUP to the Patroni process or by performing ``POST /reload`` REST-API request.
+	patroni.yml could be changed and reloaded in runtime (without restart of Patroni) by sending SIGHUP to the Patroni process, performing ``POST /reload`` REST-API request or executing ``patronictl reload``.
 
 - Environment :ref:`configuration <environment>`.
 	It is possible to set/override some of the "Local" configuration parameters with environment variables.
 	Environment configuration is very useful when you are running in a dynamic environment and you don't know some of the parameters in advance (for example it's not possible to know your external IP address when you are running inside ``docker``).
 
+The local configuration can be either a single YAML file or a directory. When it is a directory, all YAML files in that directory are loaded one by one in sorted order. In case a key is defined in multiple files, the occurrence in the last file takes precedence.
+
 Some of the PostgreSQL parameters must hold the same values on the master and the replicas. For those, values set either in the local patroni configuration files or via the environment variables take no effect. To alter or set their values one must change the shared configuration in the DCS. Below is the actual list of such parameters together with the default values:
 
 - max_connections: 100
@@ -35,6 +37,7 @@ For the parameters below, PostgreSQL does not require equal values among the mas
 - max_wal_senders: 5
 - max_replication_slots: 5
 - wal_keep_segments: 8
+- wal_keep_size: 128MB
 
 These parameters are validated to ensure they are sane, or meet a minimum value.
 
@@ -76,6 +79,7 @@ Also, the following Patroni configuration options can be changed only dynamicall
 - loop_wait: 10
 - retry_timeouts: 10
 - maximum_lag_on_failover: 1048576
+- max_timelines_history: 0
 - check_timeline: false
 - postgresql.use_slots: true
 
@@ -83,147 +87,3 @@ Upon changing these options, Patroni will read the relevant section of the confi
 run-time values.
 
 Patroni nodes are dumping the state of the DCS options to disk upon for every change of the configuration into the file ``patroni.dynamic.json`` located in the Postgres data directory. Only the master is allowed to restore these options from the on-disk dump if these are completely absent from the DCS or if they are invalid.
-
-REST API
-========
-
-We provide a REST API endpoint for working with dynamic configuration.
-
-GET /config
------------
-Get current version of dynamic configuration.
-
-.. code-block:: bash
-
-	$ curl -s localhost:8008/config | jq .
-	{
-	  "ttl": 30,
-	  "loop_wait": 10,
-	  "retry_timeout": 10,
-	  "maximum_lag_on_failover": 1048576,
-	  "postgresql": {
-	    "use_slots": true,
-	    "use_pg_rewind": true,
-	    "parameters": {
-	      "hot_standby": "on",
-	      "wal_log_hints": "on",
-	      "wal_keep_segments": 8,
-	      "wal_level": "hot_standby",
-	      "max_wal_senders": 5,
-	      "max_replication_slots": 5,
-	      "max_connections": "100"
-	    }
-	  }
-	}
-
-PATCH /config
--------------
-Change existing configuration.
-
-.. code-block:: bash
-
-	$ curl -s -XPATCH -d \
-		'{"loop_wait":5,"ttl":20,"postgresql":{"parameters":{"max_connections":"101"}}}' \
-		http://localhost:8008/config | jq .
-	{
-	  "ttl": 20,
-	  "loop_wait": 5,
-	  "maximum_lag_on_failover": 1048576,
-	  "retry_timeout": 10,
-	  "postgresql": {
-	    "use_slots": true,
-	    "use_pg_rewind": true,
-	    "parameters": {
-	      "hot_standby": "on",
-	      "wal_log_hints": "on",
-	      "wal_keep_segments": 8,
-	      "wal_level": "hot_standby",
-	      "max_wal_senders": 5,
-	      "max_replication_slots": 5,
-	      "max_connections": "101"
-	    }
-	  }
-	}
-
-The above REST API call patches the existing configuration and returns the new configuration.
-
-Let's check that the node processed this configuration. First of all it should start printing log lines every 5 seconds (loop_wait=5). The change of "max_connections" requires a restart, so the "restart_pending" flag should be exposed:
-
-.. code-block:: bash
-
-	$ curl -s http://localhost:8008/patroni | jq .
-	{
-	  "pending_restart": true,
-	  "database_system_identifier": "6287881213849985952",
-	  "postmaster_start_time": "2016-06-13 13:13:05.211 CEST",
-	  "xlog": {
-	    "location": 2197818976
-	  },
-	  "patroni": {
-	    "scope": "batman",
-	    "version": "1.0"
-	  },
-	  "state": "running",
-	  "role": "master",
-	  "server_version": 90503
-	}
-
-Removing parameters:
-
-If you want to remove (reset) some setting just patch it with ``null``:
-
-.. code-block:: bash
-
-	$ curl -s -XPATCH -d \
-		'{"postgresql":{"parameters":{"max_connections":null}}}' \
-		http://localhost:8008/config | jq .
-	{
-	  "ttl": 20,
-	  "loop_wait": 5,
-	  "retry_timeout": 10,
-	  "maximum_lag_on_failover": 1048576,
-	  "postgresql": {
-	    "use_slots": true,
-	    "use_pg_rewind": true,
-	    "parameters": {
-	      "hot_standby": "on",
-	      "unix_socket_directories": ".",
-	      "wal_keep_segments": 8,
-	      "wal_level": "hot_standby",
-	      "wal_log_hints": "on",
-	      "max_wal_senders": 5,
-	      "max_replication_slots": 5
-	    }
-	  }
-	}
-
-Above call removes ``postgresql.parameters.max_connections`` from the dynamic configuration.
-
-PUT /config
------------
-
-It's also possible to perform the full rewrite of an existing dynamic configuration unconditionally:
-
-.. code-block:: bash
-
-	$ curl -s -XPUT -d \
-		'{"maximum_lag_on_failover":1048576,"retry_timeout":10,"postgresql":{"use_slots":true,"use_pg_rewind":true,"parameters":{"hot_standby":"on","wal_log_hints":"on","wal_keep_segments":8,"wal_level":"hot_standby","unix_socket_directories":".","max_wal_senders":5}},"loop_wait":3,"ttl":20}' \
-		http://localhost:8008/config | jq .
-	{
-	  "ttl": 20,
-	  "maximum_lag_on_failover": 1048576,
-	  "retry_timeout": 10,
-	  "postgresql": {
-	    "use_slots": true,
-	    "parameters": {
-	      "hot_standby": "on",
-	      "unix_socket_directories": ".",
-	      "wal_keep_segments": 8,
-	      "wal_level": "hot_standby",
-	      "wal_log_hints": "on",
-	      "max_wal_senders": 5
-	    },
-	    "use_pg_rewind": true
-	  },
-	  "loop_wait": 3
-	}

docs/existing_data.rst

@@ -23,6 +23,21 @@ A Patroni cluster can be started with a data directory from a single-node Postgr
 3. Start Patroni (e.g. ``patroni /etc/patroni/patroni.yml``). It automatically detects that PostgreSQL daemon is already running but its configuration might be out-of-date.
 4. Ask Patroni to restart the node with ``patronictl restart cluster-name node-name``. This step is only required if PostgreSQL configuration is out-of-date.
 
+Major Upgrade of PostgreSQL Version
+===================================
+
+The only possible way to do a major upgrade currently is:
+
+1. Stop Patroni
+2. Upgrade PostgreSQL binaries and perform `pg_upgrade <https://www.postgresql.org/docs/current/pgupgrade.html>`_ on the master node
+3. Update patroni.yml
+4. Remove the initialize key from DCS or wipe complete cluster state from DCS. The second one could be achieved by running ``patronictl remove <cluster-name>``. It is necessary because pg_upgrade runs initdb which actually creates a new database with a new PostgreSQL system identifier.
+5. If you wiped the cluster state in the previous step, you may wish to copy patroni.dynamic.json from old data dir to the new one.  It will help you to retain some PostgreSQL parameters you had set before.
+6. Start Patroni on the master node.
+7. Upgrade PostgreSQL binaries, update patroni.yml and wipe the data_dir on standby nodes.
+8. Start Patroni on the standby nodes and wait for the replication to complete.
+
+Running pg_upgrade on standby nodes is not supported by PostgreSQL. If you know what you are doing, you can try the rsync procedure described in https://www.postgresql.org/docs/current/pgupgrade.html instead of wiping data_dir on standby nodes. The safest way is however to let Patroni replicate the data for you.
 
 FAQ
 ---

docs/index.rst

@@ -10,6 +10,8 @@ Patroni is a template for you to create your own customized, high-availability s
 
 We call Patroni a "template" because it is far from being a one-size-fits-all or plug-and-play replication system. It will have its own caveats. Use wisely. There are many ways to run high availability with PostgreSQL; for a list, see the `PostgreSQL Documentation <https://wiki.postgresql.org/wiki/Replication,_Clustering,_and_Connection_Pooling>`__.
 
+Currently supported PostgreSQL versions: 9.3 to 13.
+
 **Note to Kubernetes users**: Patroni can run natively on top of Kubernetes. Take a look at the :ref:`Kubernetes <kubernetes>` chapter of the Patroni documentation.
 
 
@@ -19,8 +21,11 @@ We call Patroni a "template" because it is far from being a one-size-fits-all or
 
    README
    dynamic_configuration
+   rest_api
+   existing_data
    ENVIRONMENT
    SETTINGS
+   security
    replica_bootstrap
    replication_modes
    pause

docs/releases.rst

@@ -3,6 +3,840 @@
 Release notes
 =============
 
+Version 2.1.0
+-------------
+
+This version adds compatibility with PostgreSQL v14, makes logical replication slots to survive failover/switchover, implements support of allowlist for REST API, and also reducing the number of logs to one line per heart-beat.
+
+**New features**
+
+- Compatibility with PostgreSQL v14 (Alexander Kukushkin)
+
+  Unpause WAL replay if Patroni is not in a "pause" mode itself. It could be "paused" due to the change of certain parameters like for example ``max_connections`` on the primary.
+
+- Failover logical slots (Alexander)
+
+  Make logical replication slots survive failover/switchover on PostgreSQL v11+. The replication slot if copied from the primary to the replica with restart and later the `pg_replication_slot_advance() <https://www.postgresql.org/docs/11/functions-admin.html#id-1.5.8.31.8.5.2.2.8.1.1>`__ function is used to move it forward. As a result, the slot will already exist before the failover and no events should be lost, but, there is a chance that some events could be delivered more than once.
+
+- Implemented allowlist for Patroni REST API (Alexander)
+
+  If configured, only IP's that matching rules would be allowed to call unsafe endpoints. In addition to that, it is possible to automatically include IP's of members of the cluster to the list.
+
+- Added support of replication connections via unix socket (Mohamad El-Rifai)
+
+  Previously Patroni was always using TCP for replication connection what could cause some issues with SSL verification. Using unix sockets allows exempt replication user from SSL verification.
+
+- Health check on user-defined tags (Arman Jafari Tehrani)
+
+  Along with :ref:`predefined tags: <tags_settings>` it is possible to specify any number of custom tags that become visible in the ``patronictl list`` output and in the REST API. From now on it is possible to use custom tags in health checks.
+
+- Added Prometheus ``/metrics`` endpoint (Mark Mercado, Michael Banck)
+
+  The endpoint exposing the same metrics as ``/patroni``.
+
+- Reduced chattiness of Patroni logs (Alexander)
+
+  When everything goes normal, only one line will be written for every run of HA loop.
+
+
+**Breaking chances**
+
+- The old ``permanent logical replication slots`` feature will no longer work with PostgreSQL v10 and older (Alexander)
+
+  The strategy of creating the logical slots after performing a promotion can't guaranty that no logical events are lost and therefore disabled.
+
+- The ``/leader`` always endpoint returns 200 if the node holds the lock (Alexander)
+
+  Promoting the standby cluster requires updating load-balancer health checks, which is not very convenient and easy to forget. To solve it, we change the behavior of the ``/leader`` health check endpoint. It will return 200 without taking into account whether the cluster is normal or the ``standby_cluster``.
+
+
+**Improvements in Raft support**
+
+- Reliable support of Raft traffic encryption (Alexander)
+
+  Due to the different issues in the ``PySyncObj`` the encryption support was very unstable
+
+- Handle DNS issues in Raft implementation (Alexander)
+
+  If ``self_addr`` and/or ``partner_addrs`` are configured using the DNS name instead of IP's the ``PySyncObj`` was effectively doing resolve only once when the object is created. It was causing problems when the same node was coming back online with a different IP.
+
+
+**Stability improvements**
+
+- Compatibility with ``psycopg2-2.9+`` (Alexander)
+
+  In ``psycopg2`` the ``autocommit = True`` is ignored in the ``with connection`` block, which breaks replication protocol connections.
+
+- Fix excessive HA loop runs with Zookeeper (Alexander)
+
+  Update of member ZNodes was causing a chain reaction and resulted in running the HA loops multiple times in a row.
+
+- Reload if REST API certificate is changed on disk (Michael Todorovic)
+
+  If the REST API certificate file was updated in place Patroni didn't perform a reload.
+
+- Don't create pgpass dir if kerberos auth is used (Kostiantyn Nemchenko)
+
+  Kerberos and password authentication are mutually exclusive.
+
+- Fixed little issues with custom bootstrap (Alexander)
+
+  Start Postgres with ``hot_standby=off`` only when we do a PITR and restart it after PITR is done.
+
+
+**Bugfixes**
+
+- Compatibility with ``kazoo-2.7+`` (Alexander)
+
+  Since Patroni is handling retries on its own, it is relying on the old behavior of ``kazoo`` that requests to a Zookeeper cluster are immediately discarded when there are no connections available.
+
+- Explicitly request the version of Etcd v3 cluster when it is known that we are connecting via proxy (Alexander)
+
+  Patroni is working with Etcd v3 cluster via gPRC-gateway and it depending on the cluster version different endpoints (``/v3``, ``/v3beta``, or ``/v3alpha``) must be used. The version was resolved only together with the cluster topology, but since the latter was never done when connecting via proxy.
+
+
+Version 2.0.2
+-------------
+
+**New features**
+
+- Ability to ignore externally managed replication slots (James Coleman)
+
+  Patroni is trying to remove any replication slot which is unknown to it, but there are certainly cases when replication slots should be managed externally. From now on it is possible to configure slots that should not be removed.
+
+- Added support for cipher suite limitation for REST API (Gunnar "Nick" Bluth)
+
+  It could be configured via ``restapi.ciphers`` or the ``PATRONI_RESTAPI_CIPHERS`` environment variable.
+
+- Added support for encrypted TLS keys for REST API (Jonathan S. Katz)
+
+  It could be configured via ``restapi.keyfile_password`` or the ``PATRONI_RESTAPI_KEYFILE_PASSWORD`` environment variable.
+
+- Constant time comparison of REST API authentication credentials (Alex Brasetvik)
+
+  Use ``hmac.compare_digest()`` instead of ``==``, which is vulnerable to timing attack.
+
+- Choose synchronous nodes based on replication lag (Krishna Sarabu)
+
+  If the replication lag on the synchronous node starts exceeding the configured threshold it could be demoted to asynchronous and/or replaced by the other node. Behaviour is controlled with ``maximum_lag_on_syncnode``.
+
+
+**Stability improvements**
+
+- Start postgres with ``hot_standby = off`` when doing custom bootstrap (Igor Yanchenko)
+
+  During custom bootstrap Patroni is restoring the basebackup, starting Postgres up, and waiting until recovery finishes. Some PostgreSQL parameters on the standby can't be smaller than on the primary and if the new value (restored from WAL) is higher than the configured one, Postgres panics and stops. In order to avoid such behavior we will do custom bootstrap without ``hot_standby`` mode.
+
+- Warn the user if the required watchdog is not healthy (Nicolas Thauvin)
+
+  When the watchdog device is not writable or missing in required mode, the member cannot be promoted. Added a warning to show the user where to search for this misconfiguration.
+
+- Better verbosity for single-user mode recovery (Alexander Kukushkin)
+
+  If Patroni notices that PostgreSQL wasn't shutdown clearly, in certain cases the crash-recovery is executed by starting Postgres in single-user mode. It could happen that the recovery failed (for example due to the lack of space on disk) but errors were swallowed.
+
+- Added compatibility with ``python-consul2`` module (Alexander, Wilfried Roset)
+
+  The good old ``python-consul`` is not maintained since a few years, therefore someone created a fork with new features and bug-fixes.
+
+- Don't use ``bypass_api_service`` when running ``patronictl`` (Alexander)
+
+  When a K8s pod is running in a non-``default`` namespace it does not necessarily have enough permissions to query the ``kubernetes`` endpoint. In this case Patroni shows the warning and ignores the ``bypass_api_service`` setting. In case of ``patronictl`` the warning was a bit annoying.
+
+- Create ``raft.data_dir`` if it doesn't exists or make sure that it is writable (Mark Mercado)
+
+  Improves user-friendliness and usability.
+
+
+**Bugfixes**
+
+- Don't interrupt restart or promote if lost leader lock in pause (Alexander)
+
+  In pause it is allowed to run postgres as primary without lock.
+
+- Fixed issue with ``shutdown_request()`` in the REST API (Nicolas Limage)
+
+  In order to improve handling of SSL connections and delay the handshake until thread is started Patroni overrides a few methods in the ``HTTPServer``. The ``shutdown_request()`` method was forgotten.
+
+- Fixed issue with sleep time when using Zookeeper (Alexander)
+
+  There were chances that Patroni was sleeping up to twice longer between running HA code.
+
+- Fixed invalid ``os.symlink()`` calls when moving data directory after failed bootstrap (Andrew L'Ecuyer)
+
+  If the bootstrap failed Patroni is renaming data directory, pg_wal, and all tablespaces. After that it updates symlinks so filesystem remains consistent. The symlink creation was failing due to the ``src`` and ``dst`` arguments being swapped.
+
+- Fixed bug in the post_bootstrap() method (Alexander)
+
+  If the superuser password wasn't configured Patroni was failing to call the ``post_init`` script and therefore the whole bootstrap was failing.
+
+- Fixed an issues with pg_rewind in the standby cluster (Alexander)
+
+  If the superuser name is different from Postgres, the ``pg_rewind`` in the standby cluster was failing because the connection string didn't contain the database name.
+
+- Exit only if authentication with Etcd v3 explicitly failed (Alexander)
+
+  On start Patroni performs discovery of Etcd cluster topology and authenticates if it is necessarily. It could happen that one of etcd servers is not accessible, Patroni was trying to perform authentication on this server and failing instead of retrying with the next node.
+
+- Handle case with psutil cmdline() returning empty list (Alexander)
+
+  Zombie processes are still postmasters children, but they don't have cmdline()
+
+- Treat ``PATRONI_KUBERNETES_USE_ENDPOINTS`` environment variable as boolean (Alexander)
+
+  Not doing so was making impossible disabling ``kubernetes.use_endpoints`` via environment.
+
+- Improve handling of concurrent endpoint update errors (Alexander)
+
+  Patroni will explicitly query the current endpoint object, verify that the current pod still holds the leader lock and repeat the update.
+
+
+Version 2.0.1
+-------------
+
+**New features**
+
+- Use ``more`` as pager in ``patronictl edit-config`` if ``less`` is not available (Pavel Golub)
+
+  On Windows it would be the ``more.com``. In addition to that, ``cdiff`` was changed to ``ydiff`` in ``requirements.txt``, but ``patronictl`` still supports both for compatibility.
+
+- Added support of ``raft`` ``bind_addr`` and ``password`` (Alexander Kukushkin)
+
+  ``raft.bind_addr`` might be useful when running behind NAT. ``raft.password`` enables traffic encryption (requires the ``cryptography`` module).
+
+- Added ``sslpassword`` connection parameter support (Kostiantyn Nemchenko)
+
+  The connection parameter was introduced in PostgreSQL 13.
+
+**Stability improvements**
+
+- Changed the behavior in pause (Alexander)
+
+  1. Patroni will not call the ``bootstrap`` method if the ``PGDATA`` directory is missing/empty.
+  2. Patroni will not exit on sysid mismatch in pause, only log a warning.
+  3. The node will not try to grab the leader key in pause mode if Postgres is running not in recovery (accepting writes) but the sysid doesn't match with the initialize key.
+
+- Apply ``master_start_timeout`` when executing crash recovery (Alexander)
+
+  If Postgres crashed on the leader node, Patroni does a crash-recovery by starting Postgres in single-user mode. During the crash-recovery the leader lock is being updated. If the crash-recovery didn't finish in ``master_start_timeout`` seconds, Patroni will stop it forcefully and release the leader lock.
+
+- Removed the ``secure`` extra from the ``urllib3`` requirements (Alexander)
+
+  The only reason for adding it there was the ``ipaddress`` dependency for python 2.7.
+
+**Bugfixes**
+
+- Fixed a bug in the ``Kubernetes.update_leader()`` (Alexander)
+
+  An unhandled exception was preventing demoting the primary when the update of the leader object failed.
+
+- Fixed hanging ``patronictl`` when RAFT is being used (Alexander)
+
+  When using ``patronictl`` with Patroni config, ``self_addr`` should be added to the ``partner_addrs``.
+
+- Fixed bug in ``get_guc_value()`` (Alexander)
+
+  Patroni was failing to get the value of ``restore_command`` on PostgreSQL 12, therefore fetching missing WALs for ``pg_rewind`` didn't work.
+
+
+Version 2.0.0
+-------------
+
+This version enhances compatibility with PostgreSQL 13, adds support of multiple synchronous standbys, has significant improvements in handling of ``pg_rewind``, adds support of Etcd v3 and Patroni on pure RAFT (without Etcd, Consul, or Zookeeper), and makes it possible to optionally call the ``pre_promote`` (fencing) script.
+
+**PostgreSQL 13 support**
+
+- Don't fire ``on_reload`` when promoting to ``standby_leader`` on PostgreSQL 13+ (Alexander Kukushkin)
+
+  When promoting to ``standby_leader`` we change ``primary_conninfo``, update the role and reload Postgres. Since ``on_role_change`` and ``on_reload`` effectively duplicate each other, Patroni will call only ``on_role_change``.
+
+- Added support for ``gssencmode`` and ``channel_binding`` connection parameters (Alexander)
+
+  PostgreSQL 12 introduced ``gssencmode`` and 13 ``channel_binding`` connection parameters and now they can be used if defined in the ``postgresql.authentication`` section.
+
+- Handle renaming of ``wal_keep_segments`` to ``wal_keep_size`` (Alexander)
+
+  In case of misconfiguration (``wal_keep_segments`` on 13 and ``wal_keep_size`` on older versions) Patroni will automatically adjust the configuration.
+
+- Use ``pg_rewind`` with ``--restore-target-wal`` on 13 if possible (Alexander)
+
+  On PostgreSQL 13 Patroni checks if ``restore_command`` is configured and tells ``pg_rewind`` to use it.
+
+
+**New features**
+
+- [BETA] Implemented support of Patroni on pure RAFT (Alexander)
+
+  This makes it possible to run Patroni without 3rd party dependencies, like Etcd, Consul, or Zookeeper. For HA you will have to run either three Patroni nodes or two nodes with Patroni and one node with ``patroni_raft_controller``. For more information please check the :ref:`documentation <raft_settings>`.
+
+- [BETA] Implemented support for Etcd v3 protocol via gPRC-gateway (Alexander)
+
+  Etcd 3.0 was released more than four years ago and Etcd 3.4 has v2 disabled by default. There are also chances that v2 will be completely removed from Etcd, therefore we implemented support of Etcd v3 in Patroni. In order to start using it you have to explicitly create the ``etcd3`` section is the Patroni configuration file.
+
+- Supporting multiple synchronous standbys (Krishna Sarabu)
+
+  It allows running a cluster with more than one synchronous replicas. The maximum number of synchronous replicas is controlled by the new parameter ``synchronous_node_count``. It is set to 1 by default and has no effect when the ``synchronous_mode`` is set to ``off``.
+
+- Added possibility to call the ``pre_promote`` script (Sergey Dudoladov)
+
+  Unlike callbacks, the ``pre_promote`` script is called synchronously after acquiring the leader lock, but before promoting Postgres. If the script fails or exits with a non-zero exitcode, the current node will release the leader lock.
+
+- Added support for configuration directories (Floris van Nee)
+
+  YAML files in the directory loaded and applied in alphabetical order.
+
+- Advanced validation of PostgreSQL parameters (Alexander)
+
+  In case the specific parameter is not supported by the current PostgreSQL version or when its value is incorrect, Patroni will remove the parameter completely or try to fix the value.
+
+- Wake up the main thread when the forced checkpoint after promote completed (Alexander)
+
+  Replicas are waiting for checkpoint indication via member key of the leader in DCS. The key is normally updated only once per HA loop. Without waking the main thread up, replicas will have to wait up to ``loop_wait`` seconds longer than necessary.
+
+- Use of ``pg_stat_wal_recevier`` view on 9.6+ (Alexander)
+
+  The view contains up-to-date values of ``primary_conninfo`` and ``primary_slot_name``, while the contents of ``recovery.conf`` could be stale.
+
+- Improved handing of IPv6 addresses in the Patroni config file (Mateusz Kowalski)
+
+  The IPv6 address is supposed to be enclosed into square brackets, but Patroni was expecting to get it plain. Now both formats are supported.
+
+- Added Consul ``service_tags`` configuration parameter (Robert Edström)
+
+  They are useful for dynamic service discovery, for example by load balancers.
+
+- Implemented SSL support for Zookeeper (Kostiantyn Nemchenko)
+
+  It requires ``kazoo>=2.6.0``.
+
+- Implemented ``no_params`` option for custom bootstrap method (Kostiantyn)
+
+  It allows calling ``wal-g``, ``pgBackRest`` and other backup tools without wrapping them into shell scripts.
+
+- Move WAL and tablespaces after a failed init (Feike Steenbergen)
+
+  When doing ``reinit``, Patroni was already removing not only ``PGDATA`` but also the symlinked WAL directory and tablespaces. Now the ``move_data_directory()`` method will do a similar job, i.e. rename WAL directory and tablespaces and update symlinks in PGDATA.
+
+
+**Improved in pg_rewind support**
+
+- Improved timeline divergence check (Alexander)
+
+  We don't need to rewind when the replayed location on the replica is not ahead of the switchpoint or the end of the checkpoint record on the former primary is the same as the switchpoint. In order to get the end of the checkpoint record we use ``pg_waldump`` and parse its output.
+
+- Try to fetch missing WAL if ``pg_rewind`` complains about it (Alexander)
+
+  It could happen that the WAL segment required for ``pg_rewind`` doesn't exist in the ``pg_wal`` directory anymore and therefore ``pg_rewind`` can't find the checkpoint location before the divergence point. Starting from PostgreSQL 13 ``pg_rewind`` could use ``restore_command`` for fetching missing WALs. For older PostgreSQL versions Patroni parses the errors of a failed rewind attempt and tries to fetch the missing WAL by calling the ``restore_command`` on its own.
+
+- Detect a new timeline in the standby cluster and trigger rewind/reinitialize if necessary (Alexander)
+
+  The ``standby_cluster`` is decoupled from the primary cluster and therefore doesn't immediately know about leader elections and timeline switches. In order to detect the fact, the ``standby_leader`` periodically checks for new history files in ``pg_wal``.
+
+- Shorten and beautify history log output (Alexander)
+
+  When Patroni is trying to figure out the necessity of ``pg_rewind``, it could write the content of the history file from the primary into the log. The history file is growing with every failover/switchover and eventually starts taking up too many lines, most of which are not so useful. Instead of showing the raw data, Patroni will show only 3 lines before the current replica timeline and 2 lines after.
+
+
+**Improvements on K8s**
+
+- Get rid of ``kubernetes`` python module (Alexander)
+
+  The official python kubernetes client contains a lot of auto-generated code and therefore very heavy. Patroni uses only a small fraction of K8s API endpoints and implementing support for them wasn't hard.
+
+- Make it possible to bypass the ``kubernetes`` service (Alexander)
+
+  When running on K8s, Patroni is usually communicating with the K8s API via the ``kubernetes`` service, the address of which is exposed in the ``KUBERNETES_SERVICE_HOST`` environment variable. Like any other service, the ``kubernetes`` service is handled by ``kube-proxy``, which in turn, depending on the configuration, is either relying on a userspace program or ``iptables`` for traffic routing. Skipping the intermediate component and connecting directly to the K8s master nodes allows us to implement a better retry strategy and mitigate risks of demoting Postgres when K8s master nodes are upgraded.
+
+- Sync HA loops of all pods of a Patroni cluster (Alexander)
+
+  Not doing so was increasing failure detection time from ``ttl`` to ``ttl + loop_wait``.
+
+- Populate ``references`` and ``nodename`` in the subsets addresses on K8s (Alexander)
+
+  Some load-balancers are relying on this information.
+
+- Fix possible race conditions in the ``update_leader()`` (Alexander)
+
+  The concurrent update of the leader configmap or endpoint happening outside of Patroni might cause the ``update_leader()`` call to fail. In this case Patroni rechecks that the current node is still owning the leader lock and repeats the update.
+
+- Explicitly disallow patching non-existent config (Alexander)
+
+  For DCS other than ``kubernetes`` the PATCH call is failing with an exception due to ``cluster.config`` being ``None``, but on Kubernetes it was happily creating the config annotation and preventing writing bootstrap configuration after the bootstrap finished.
+
+- Fix bug in ``pause`` (Alexander)
+
+  Replicas were removing ``primary_conninfo`` and restarting Postgres when the leader key was absent, but they should do nothing.
+
+
+**Improvements in REST API**
+
+- Defer TLS handshake until worker thread has started (Alexander, Ben Harris)
+
+  If the TLS handshake was done in the API thread and the client-side didn't send any data, the API thread was blocked (risking DoS).
+
+- Check ``basic-auth`` independently from client certificate in REST API (Alexander)
+
+  Previously only the client certificate was validated. Doing two checks independently is an absolutely valid use-case.
+
+- Write double ``CRLF`` after HTTP headers of the ``OPTIONS`` request (Sergey Burladyan)
+
+  HAProxy was happy with a single ``CRLF``, while Consul health-check complained about broken connection and unexpected EOF.
+
+- ``GET /cluster`` was showing stale members info for Zookeeper (Alexander)
+
+  The endpoint was using the Patroni internal cluster view. For Patroni itself it didn't cause any issues, but when exposed to the outside world we need to show up-to-date information, especially replication lag.
+
+- Fixed health-checks for standby cluster (Alexander)
+
+  The ``GET /standby-leader`` for a master and ``GET /master`` for a ``standby_leader`` were incorrectly responding with 200.
+
+- Implemented ``DELETE /switchover`` (Alexander)
+
+  The REST API call deletes the scheduled switchover.
+
+- Created ``/readiness`` and ``/liveness`` endpoints (Alexander)
+
+  They could be useful to eliminate "unhealthy" pods from subsets addresses when the K8s service is used with label selectors.
+
+- Enhanced ``GET /replica`` and ``GET /async`` REST API health-checks (Krishna, Alexander)
+
+  Checks now support optional keyword ``?lag=<max-lag>`` and will respond with 200 only if the lag is smaller than the supplied value. If relying on this feature please keep in mind that information about WAL position on the leader is updated only every ``loop_wait`` seconds!
+
+- Added support for user defined HTTP headers in the REST API response (Yogesh Sharma)
+
+  This feature might be useful if requests are made from a browser.
+
+
+**Improvements in patronictl**
+
+- Don't try to call non-existing leader in ``patronictl pause`` (Alexander)
+
+  While pausing a cluster without a leader on K8s, ``patronictl`` was showing warnings that member "None" could not be accessed.
+
+- Handle the case when member ``conn_url`` is missing (Alexander)
+
+  On K8s it is possible that the pod doesn't have the necessary annotations because Patroni is not yet running. It was making ``patronictl`` to fail.
+
+- Added ability to print ASCII cluster topology (Maxim Fedotov, Alexander)
+
+  It is very useful to get overview of the cluster with cascading replication.
+
+- Implement ``patronictl flush switchover`` (Alexander)
+
+  Before that ``patronictl flush`` only supported cancelling scheduled restarts.
+
+
+**Bugfixes**
+
+- Attribute error during bootstrap of the cluster with existing PGDATA (Krishna)
+
+  When trying to create/update the ``/history`` key, Patroni was accessing the ``ClusterConfig`` object which wasn't created in DCS yet.
+
+- Improved exception handling in Consul (Alexander)
+
+  Unhandled exception in the ``touch_member()`` method caused the whole Patroni process to crash.
+
+- Enforce ``synchronous_commit=local`` for the ``post_init`` script (Alexander)
+
+  Patroni was already doing that when creating users (``replication``, ``rewind``), but missing it in the case of ``post_init`` was an oversight. As a result, if the script wasn't doing it internally on it's own the bootstrap in ``synchronous_mode`` wasn't able to finish.
+
+- Increased ``maxsize`` in the Consul pool manager (ponvenkates)
+
+  With the default ``size=1`` some warnings were generated.
+
+- Patroni was wrongly reporting Postgres as running (Alexander)
+
+  The state wasn't updated when for example Postgres crashed due to an out-of-disk error.
+
+- Put ``*`` into ``pgpass`` instead of missing or empty values (Alexander)
+
+  If for example the ``standby_cluster.port`` is not specified, the ``pgpass`` file was incorrectly generated.
+
+- Skip physical replication slot creation on the leader node with special characters (Krishna)
+
+  Patroni appeared to be creating a dormant slot (when ``slots`` defined) for the leader node when the name contained special chars such as '-'  (for e.g. "abc-us-1").
+
+- Avoid removing non-existent ``pg_hba.conf`` in the custom bootstrap (Krishna)
+
+  Patroni was failing if ``pg_hba.conf`` happened to be located outside of the ``pgdata`` dir after custom bootstrap.
+
+
+Version 1.6.5
+-------------
+
+**New features**
+
+- Master stop timeout (Krishna Sarabu)
+
+  The number of seconds Patroni is allowed to wait when stopping Postgres. Effective only when ``synchronous_mode`` is enabled. When set to value greater than 0 and the ``synchronous_mode`` is enabled, Patroni sends ``SIGKILL`` to the postmaster if the stop operation is running for more than the value set by ``master_stop_timeout``. Set the value according to your durability/availability tradeoff. If the parameter is not set or set to non-positive value, ``master_stop_timeout`` does not have an effect.
+
+- Don't create permanent physical slot with name of the primary (Alexander Kukushkin)
+
+  It is a common problem that the primary recycles WAL segments while the replica is down. Now we have a good solution for static clusters, with a fixed number of nodes and names that never change. You just need to list the names of all nodes in the ``slots`` so the primary will not remove the slot when the node is down (not registered in DCS).
+
+- First draft of Config Validator (Igor Yanchenko)
+
+  Use ``patroni --validate-config patroni.yaml`` in order to validate Patroni configuration.
+
+- Possibility to configure max length of timelines history (Krishna)
+
+  Patroni writes the history of failovers/switchovers into the ``/history`` key in DCS. Over time the size of this key becomes big, but in most cases only the last few lines are interesting. The ``max_timelines_history`` parameter allows to specify the maximum number of timeline history items to be kept in DCS.
+
+- Kazoo 2.7.0 compatibility (Danyal Prout)
+
+  Some non-public methods in Kazoo changed their signatures, but Patroni was relying on them.
+
+
+**Improvements in patronictl**
+
+- Show member tags (Kostiantyn Nemchenko, Alexander)
+
+  Tags are configured individually for every node and there was no easy way to get an overview of them
+
+- Improve members output (Alexander)
+
+  The redundant cluster name won't be shown anymore on every line, only in the table header.
+
+.. code-block:: bash
+
+    $ patronictl list
+    + Cluster: batman (6813309862653668387) +---------+----+-----------+---------------------+
+    |    Member   |      Host      |  Role  |  State  | TL | Lag in MB | Tags                |
+    +-------------+----------------+--------+---------+----+-----------+---------------------+
+    | postgresql0 | 127.0.0.1:5432 | Leader | running |  3 |           | clonefrom: true     |
+    |             |                |        |         |    |           | noloadbalance: true |
+    |             |                |        |         |    |           | nosync: true        |
+    +-------------+----------------+--------+---------+----+-----------+---------------------+
+    | postgresql1 | 127.0.0.1:5433 |        | running |  3 |       0.0 |                     |
+    +-------------+----------------+--------+---------+----+-----------+---------------------+
+
+- Fail if a config file is specified explicitly but not found (Kaarel Moppel)
+
+  Previously ``patronictl`` was only reporting a ``DEBUG`` message.
+
+- Solved the problem of not initialized K8s pod breaking patronictl (Alexander)
+
+  Patroni is relying on certain pod annotations on K8s. When one of the Patroni pods is stopping or starting there is no valid annotation yet and ``patronictl`` was failing with an exception.
+
+
+**Stability improvements**
+
+- Apply 1 second backoff if LIST call to K8s API server failed (Alexander)
+
+  It is mostly necessary to avoid flooding logs, but also helps to prevent starvation of the main thread.
+
+- Retry if the ``retry-after`` HTTP header is returned by K8s API (Alexander)
+
+  If the K8s API server is overwhelmed with requests it might ask to retry.
+
+- Scrub ``KUBERNETES_`` environment from the postmaster (Feike Steenbergen)
+
+  The ``KUBERNETES_`` environment variables are not required for PostgreSQL, yet having them exposed to the postmaster will also expose them to backends and to regular database users (using pl/perl for example).
+
+- Clean up tablespaces on reinitialize (Krishna)
+
+  During reinit, Patroni was removing only ``PGDATA`` and leaving user-defined tablespace directories. This is causing Patroni to loop in reinit. The previous workarond for the problem was implementing the :ref:`custom bootstrap <custom_bootstrap>` script.
+
+- Explicitly execute ``CHECKPOINT`` after promote happened (Alexander)
+
+  It helps to reduce the time before the new primary is usable for ``pg_rewind``.
+
+- Smart refresh of Etcd members (Alexander)
+
+  In case Patroni failed to execute a request on all members of the Etcd cluster, Patroni will re-check ``A`` or ``SRV`` records for changes of IPs/hosts before retrying the next time.
+
+- Skip missing values from ``pg_controldata`` (Feike)
+
+  Values are missing when trying to use binaries of a version that doesn't match PGDATA. Patroni will try to start Postgres anyway, and Postgres will complain that the major version doesn't match and abort with an error.
+
+
+**Bugfixes**
+
+- Disable SSL verification for Consul when required (Julien Riou)
+
+  Starting from a certain version of ``urllib3``, the ``cert_reqs`` must be explicitly set to ``ssl.CERT_NONE`` in order to effectively disable SSL verification.
+
+- Avoid opening replication connection on every cycle of HA loop (Alexander)
+
+  Regression was introduced in 1.6.4.
+
+- Call ``on_role_change`` callback on failed primary (Alexander)
+
+  In certain cases it could lead to the virtual IP remaining attached to the old primary. Regression was introduced in 1.4.5.
+
+- Reset rewind state if postgres started after successful pg_rewind (Alexander)
+
+  As a result of this bug Patroni was starting up manually shut down postgres in the pause mode.
+
+- Convert ``recovery_min_apply_delay`` to ``ms`` when checking ``recovery.conf``
+
+  Patroni was indefinitely restarting replica if ``recovery_min_apply_delay`` was configured on PostgreSQL older than 12.
+
+- PyInstaller compatibility (Alexander)
+
+  PyInstaller freezes (packages) Python applications into stand-alone executables. The compatibility was broken when we switched to the ``spawn`` method instead of ``fork`` for ``multiprocessing``.
+
+
+Version 1.6.4
+-------------
+
+**New features**
+
+- Implemented ``--wait`` option for ``patronictl reinit`` (Igor Yanchenko)
+
+  Patronictl will wait for ``reinit`` to finish is the ``--wait`` option is used.
+
+- Further improvements of Windows support (Igor Yanchenko, Alexander Kukushkin)
+
+  1. All shell scripts which are used for integration testing are rewritten in python
+  2. The ``pg_ctl kill`` will be used to stop postgres on non posix systems
+  3. Don't try to use unix-domain sockets
+
+
+**Stability improvements**
+
+- Make sure ``unix_socket_directories`` and ``stats_temp_directory`` exist (Igor)
+
+  Upon the start of Patroni and Postgres make sure that ``unix_socket_directories`` and ``stats_temp_directory`` exist or try to create them. Patroni will exit if failed to create them.
+
+- Make sure ``postgresql.pgpass`` is located in the place where Patroni has write access (Igor)
+
+  In case if it doesn't have a write access Patroni will exit with exception.
+
+- Disable Consul ``serfHealth`` check by default (Kostiantyn Nemchenko)
+
+  Even in case of little network problems the failing ``serfHealth`` leads to invalidation of all sessions associated with the node. Therefore, the leader key is lost much earlier than ``ttl`` which causes unwanted restarts of replicas and maybe demotion of the primary.
+
+- Configure tcp keepalives for connections to K8s API (Alexander)
+
+  In case if we get nothing from the socket after TTL seconds it can be considered dead.
+
+- Avoid logging of passwords on user creation (Alexander)
+
+  If the password is rejected or logging is configured to verbose or not configured at all it might happen that the password is written into postgres logs. In order to avoid it Patroni will change ``log_statement``, ``log_min_duration_statement``, and ``log_min_error_statement`` to some safe values before doing the attempt to create/update user.
+
+
+**Bugfixes**
+
+- Use ``restore_command`` from the ``standby_cluster`` config on cascading replicas (Alexander)
+
+  The ``standby_leader`` was already doing it from the beginning the feature existed. Not doing the same on replicas might prevent them from catching up with standby leader.
+
+- Update timeline reported by the standby cluster (Alexander)
+
+  In case of timeline switch the standby cluster was correctly replicating from the primary but ``patronictl`` was reporting the old timeline.
+
+- Allow certain recovery parameters be defined in the custom_conf (Alexander)
+
+  When doing validation of recovery parameters on replica Patroni will skip ``archive_cleanup_command``, ``promote_trigger_file``, ``recovery_end_command``, ``recovery_min_apply_delay``, and ``restore_command`` if they are not defined in the patroni config but in files other than ``postgresql.auto.conf`` or ``postgresql.conf``.
+
+- Improve handling of postgresql parameters with period in its name (Alexander)
+
+  Such parameters could be defined by extensions where the unit is not necessarily a string. Changing the value might require a restart (for example ``pg_stat_statements.max``).
+
+- Improve exception handling during shutdown (Alexander)
+
+  During shutdown Patroni is trying to update its status in the DCS. If the DCS is inaccessible an exception might be raised. Lack of exception handling was preventing logger thread from stopping.
+
+
+Version 1.6.3
+-------------
+
+**Bugfixes**
+
+- Don't expose password when running ``pg_rewind`` (Alexander Kukushkin)
+
+  Bug was introduced in the `#1301 <https://github.com/zalando/patroni/pull/1301>`__
+
+- Apply connection parameters specified in the ``postgresql.authentication`` to ``pg_basebackup`` and custom replica creation methods (Alexander)
+
+  They were relying on url-like connection string and therefore parameters never applied.
+
+
+Version 1.6.2
+-------------
+
+**New features**
+
+- Implemented ``patroni --version`` (Igor Yanchenko)
+
+  It prints the current version of Patroni and exits.
+
+- Set the ``user-agent`` http header for all http requests (Alexander Kukushkin)
+
+  Patroni is communicating with Consul, Etcd, and Kubernetes API via the http protocol. Having a specifically crafted ``user-agent`` (example: ``Patroni/1.6.2 Python/3.6.8 Linux``) might be useful for debugging and monitoring.
+
+- Make it possible to configure log level for exception tracebacks (Igor)
+
+  If you set ``log.traceback_level=DEBUG`` the tracebacks will be visible only when ``log.level=DEBUG``. The default behavior remains the same.
+
+
+**Stability improvements**
+
+- Avoid importing all DCS modules when searching for the module required by the config file (Alexander)
+
+  There is no need to import modules for Etcd, Consul, and Kubernetes if we need only e.g. Zookeeper. It helps to reduce memory usage and solves the problem of having INFO messages ``Failed to import smth``.
+
+- Removed python ``requests`` module from explicit requirements (Alexander)
+
+  It wasn't used for anything critical, but causing a lot of problems when the new version of ``urllib3`` is released.
+
+- Improve handling of ``etcd.hosts`` written as a comma-separated string instead of YAML array (Igor)
+
+  Previously it was failing when written in format ``host1:port1, host2:port2`` (the space character after the comma).
+
+
+**Usability improvements**
+
+- Don't force users to choose members from an empty list in ``patronictl`` (Igor)
+
+  If the user provides a wrong cluster name, we will raise an exception rather than ask to choose a member from an empty list.
+
+- Make the error message more helpful if the REST API cannot bind (Igor)
+
+  For an inexperienced user it might be hard to figure out what is wrong from the Python stacktrace.
+
+
+**Bugfixes**
+
+- Fix calculation of ``wal_buffers`` (Alexander)
+
+  The base unit has been changed from 8 kB blocks to bytes in PostgreSQL 11.
+
+- Use ``passfile`` in ``primary_conninfo`` only on PostgreSQL 10+ (Alexander)
+
+  On older versions there is no guarantee that ``passfile`` will work, unless the latest version of ``libpq`` is installed.
+
+
+Version 1.6.1
+-------------
+
+**New features**
+
+- Added ``PATRONICTL_CONFIG_FILE`` environment variable (msvechla)
+
+  It allows configuring the ``--config-file`` argument for ``patronictl`` from the environment.
+
+- Implement ``patronictl history`` (Alexander Kukushkin)
+
+  It shows the history of failovers/switchovers.
+
+- Pass ``-c statement_timeout=0`` in ``PGOPTIONS`` when doing ``pg_rewind`` (Alexander Kukushkin)
+
+  It protects from the case when ``statement_timeout`` on the server is set to some small value and one of the statements executed by pg_rewind is canceled.
+
+- Allow lower values for PostgreSQL configuration (Soulou)
+
+  Patroni didn't allow some of the PostgreSQL configuration parameters be set smaller than some hardcoded values. Now the minimal allowed values are smaller, default values have not been changed.
+
+- Allow for certificate-based authentication (Jonathan S. Katz)
+
+  This feature enables certificate-based authentication for superuser, replication, rewind accounts and allows the user to specify the ``sslmode`` they wish to connect with.
+
+- Use the ``passfile`` in the ``primary_conninfo`` instead of password (Alexander Kukushkin)
+
+  It allows to avoid setting ``600`` permissions on postgresql.conf
+
+- Perform ``pg_ctl reload`` regardless of config changes (Alexander Kukushkin)
+
+  It is possible that some config files are not controlled by Patroni. When somebody is doing a reload via the REST API or by sending SIGHUP to the Patroni process, the usual expectation is that Postgres will also be reloaded. Previously it didn't happen when there were no changes in the ``postgresql`` section of Patroni config.
+
+- Compare all recovery parameters, not only ``primary_conninfo`` (Alexander Kukushkin)
+
+  Previously the ``check_recovery_conf()`` method was only checking whether ``primary_conninfo`` has changed, never taking into account all other recovery parameters.
+
+- Make it possible to apply some recovery parameters without restart (Alexander Kukushkin)
+
+  Starting from PostgreSQL 12 the following recovery parameters could be changed without restart: ``archive_cleanup_command``, ``promote_trigger_file``, ``recovery_end_command``, and ``recovery_min_apply_delay``. In future Postgres releases this list will be extended and Patroni will support it automatically.
+
+- Make it possible to change ``use_slots`` online (Alexander Kukushkin)
+
+  Previously it required restarting Patroni and removing slots manually.
+
+- Remove only ``PATRONI_`` prefixed environment variables when starting up Postgres (Cody Coons)
+
+  It will solve a lot of problems with running different Foreign Data Wrappers.
+
+
+**Stability improvements**
+
+- Use LIST + WATCH when working with K8s API (Alexander Kukushkin)
+
+  It allows to efficiently receive object changes (pods, endpoints/configmaps) and makes less stress on K8s master nodes.
+
+- Improve the workflow when PGDATA is not empty during bootstrap (Alexander Kukushkin)
+
+  According to the ``initdb`` source code it might consider a PGDATA empty when there are only ``lost+found`` and ``.dotfiles`` in it. Now Patroni does the same. If ``PGDATA`` happens to be non-empty, and at the same time not valid from the ``pg_controldata`` point of view, Patroni will complain and exit.
+
+- Avoid calling expensive ``os.listdir()`` on every HA loop (Alexander Kukushkin)
+
+  When the system is under IO stress, ``os.listdir()`` could take a few seconds (or even minutes) to execute, badly affecting the HA loop of Patroni. This could even cause the leader key to disappear from DCS due to the lack of updates. There is a better and less expensive way to check that the PGDATA is not empty. Now we check the presence of the ``global/pg_control`` file in the PGDATA.
+
+- Some improvements in logging infrastructure (Alexander Kukushkin)
+
+  Previously there was a possibility to loose the last few log lines on shutdown because the logging thread was a ``daemon`` thread.
+
+- Use ``spawn`` multiprocessing start method on python 3.4+ (Maciej Kowalczyk)
+
+  It is a known `issue <https://bugs.python.org/issue6721>`__ in Python that threading and multiprocessing do not mix well. Switching from the default method ``fork`` to the ``spawn`` is a recommended workaround. Not doing so might result in the Postmaster starting process hanging and Patroni indefinitely reporting ``INFO: restarting after failure in progress``, while  Postgres is actually up and running.
+
+**Improvements in REST API**
+
+- Make it possible to check client certificates in the REST API (Alexander Kukushkin)
+
+  If the ``verify_client`` is set to ``required``, Patroni will check client certificates for all REST API calls. When it is set to ``optional``, client certificates are checked for all unsafe REST API endpoints.
+
+- Return the response code 503 for the ``GET /replica`` health check request if Postgres is not running (Alexander Anikin)
+
+  Postgres might spend significant time in recovery before it starts accepting client connections.
+
+- Implement ``/history`` and ``/cluster`` endpoints (Alexander Kukushkin)
+
+  The ``/history`` endpoint shows the content of the ``history`` key in DCS. The ``/cluster`` endpoint shows all cluster members and some service info like pending and scheduled restarts or switchovers.
+
+
+**Improvements in Etcd support**
+
+- Retry on Etcd RAFT internal error (Alexander Kukushkin)
+
+  When the Etcd node is being shut down, it sends ``response code=300, data='etcdserver: server stopped'``, which was causing Patroni to demote the primary.
+
+- Don't give up on Etcd request retry too early (Alexander Kukushkin)
+
+  When there were some network problems, Patroni was quickly exhausting the list of Etcd nodes and giving up without using the whole ``retry_timeout``, potentially resulting in demoting the primary.
+
+
+**Bugfixes**
+
+- Disable ``synchronous_commit`` when granting execute permissions to the ``pg_rewind`` user (kremius)
+
+  If the bootstrap is done with ``synchronous_mode_strict: true`` the `GRANT EXECUTE` statement was waiting indefinitely due to the non-synchronous nodes being available.
+
+- Fix memory leak on python 3.7 (Alexander Kukushkin)
+
+  Patroni is using ``ThreadingMixIn`` to process REST API requests and python 3.7 made threads spawn for every request non-daemon by default.
+
+- Fix race conditions in asynchronous actions (Alexander Kukushkin)
+
+  There was a chance that ``patronictl reinit --force`` could be overwritten by the attempt to recover stopped Postgres. This ended up in a situation when Patroni was trying to start Postgres while basebackup was running.
+
+- Fix race condition in ``postmaster_start_time()`` method (Alexander Kukushkin)
+
+  If the method is executed from the REST API thread, it requires a separate cursor object to be created.
+
+- Fix the problem of not promoting the sync standby that had a name containing upper case letters (Alexander Kukushkin)
+
+  We converted the name to the lower case because Postgres was doing the same while comparing the ``application_name`` with the value in ``synchronous_standby_names``.
+
+- Kill all children along with the callback process before starting the new one (Alexander Kukushkin)
+
+  Not doing so makes it hard to implement callbacks in bash and eventually can lead to the situation when two callbacks are running at the same time. 
+
+- Fix 'start failed' issue (Alexander Kukushkin)
+
+  Under certain conditions the Postgres state might be set to 'start failed' despite Postgres being up and running.
+
+
 Version 1.6.0
 -------------
 
@@ -40,7 +874,7 @@ This version adds compatibility with PostgreSQL 12, makes is possible to run pg_
   This functionality works similarly to ``pg_hba.conf``: if the ``postgresql.pg_ident`` is defined in the config file or DCS, Patroni will write its value to ``pg_ident.conf``, however, if ``postgresql.parameters.ident_file`` is defined, Patroni will assume that ``pg_ident`` is managed from outside and not update the file.
 
 
-  **Improvements in REST API**
+**Improvements in REST API**
 
 - Added ``/health`` endpoint (Wilfried Roset)
 
@@ -266,7 +1100,7 @@ Compatibility and bugfix release.
 
 - Fix broken compatibility with postgres 9.3 (Alexander)
 
-  When opening a replication connection we should specify replication=1, beacuse 9.3 does not understand replication='database'
+  When opening a replication connection we should specify replication=1, because 9.3 does not understand replication='database'
 
 - Make sure we refresh Consul session at least once per HA loop and improve handling of consul sessions exceptions (Alexander)
 
@@ -351,7 +1185,7 @@ This version enables Patroni HA cluster to operate in a standby mode, introduces
 
 - Immediately reserve the WAL position upon creation of the replication slot (Alexander Kukushkin)
 
-  Starting from 9.6, `pg_create_physical_replication_slot` function provides an additional boolean parameter `immediately_reserve`. When it is set to `false`, which is also the default, the slot doesn't reserve the WAL position until it receives the first client connection, potentially losing some segments required by the client in a time window between the slot creation and the intiial client connection.
+  Starting from 9.6, `pg_create_physical_replication_slot` function provides an additional boolean parameter `immediately_reserve`. When it is set to `false`, which is also the default, the slot doesn't reserve the WAL position until it receives the first client connection, potentially losing some segments required by the client in a time window between the slot creation and the initial client connection.
 
 - Fix bug in strict synchronous replication (Alexander Kukushkin)
 
@@ -591,7 +1425,7 @@ This version adds support for using Kubernetes as a DCS, allowing to run Patroni
 
 **Upgrade notice**
 
-Installing Patroni via pip will no longer bring in dependencies for (such as libraries for Etcd, Zookeper, Consul or Kubernetes, or support for AWS). In order to enable them one need to list them in pip install command explicitely, for instance `pip install patroni[etcd,kubernetes]`.
+Installing Patroni via pip will no longer bring in dependencies for (such as libraries for Etcd, Zookeper, Consul or Kubernetes, or support for AWS). In order to enable them one need to list them in pip install command explicitly, for instance `pip install patroni[etcd,kubernetes]`.
 
 **Kubernetes support**
 
@@ -610,7 +1444,7 @@ In addition to using Endpoints, Patroni supports ConfigMaps. You can find more i
 
 - Remove leader key on shutdown only when we have the lock (Ants)
 
-  Unconditional removal was generating unnecessary and missleading exceptions.
+  Unconditional removal was generating unnecessary and misleading exceptions.
 
 **Improvements in patronictl**
 
@@ -641,7 +1475,7 @@ In addition to using Endpoints, Patroni supports ConfigMaps. You can find more i
 
 - Alter the behavior of ``patronictl failover`` (Alexander)
 
-  It will work even if there is no leader, but in that case you will have to explicitely specify a node which should become the new leader.
+  It will work even if there is no leader, but in that case you will have to explicitly specify a node which should become the new leader.
 
 **Expose information about timeline and history**
 
@@ -657,7 +1491,7 @@ In addition to using Endpoints, Patroni supports ConfigMaps. You can find more i
 
 - Add new /sync and /async endpoints (Alexander, Oleksii Kliukin)
 
- Those endpoints (also accessible as /synchronous and /asynchronous) return 200 only for synchronous and asynchornous replicas correspondingly (exclusing those marked as `noloadbalance`).
+ Those endpoints (also accessible as /synchronous and /asynchronous) return 200 only for synchronous and asynchronous replicas correspondingly (exclusing those marked as `noloadbalance`).
 
 **Allow multiple hosts for Etcd**
 
@@ -745,7 +1579,7 @@ Version 1.3.4
 
 - Pass the consul token as a header (Andrew Colin Kissa)
 
-  Headers are now the prefered way to pass the token to the consul `API <https://www.consul.io/api/index.html#authentication>`__.
+  Headers are now the preferred way to pass the token to the consul `API <https://www.consul.io/api/index.html#authentication>`__.
 
 
 - Advanced configuration for Consul (Alexander Kukushkin)
@@ -1063,7 +1897,7 @@ In addition, patronictl supports new ``pause`` and ``resume`` commands to toggle
   Originally, ping_timeout and connect_timeout values were calculated from the negotiated session timeout. Patroni loop_wait was not taken into account. As
   a result, a single retry could take more time than the session timeout, forcing Patroni to release the lock and demote.
 
-  This change set ping and connect timeout to half of the value of loop_wait, speeding up detection of connection issues and  leaving enough time to retry the connection attempt before loosing the lock.
+  This change set ping and connect timeout to half of the value of loop_wait, speeding up detection of connection issues and  leaving enough time to retry the connection attempt before losing the lock.
 
 - Update Etcd topology only after original request succeed (Alexander)
 
@@ -1141,7 +1975,7 @@ When upgrading from v0.90 or below, always upgrade all replicas before the maste
 
   See the :ref:`dynamic configuration <dynamic_configuration>`  for the details on which parameters can be changed and the order of processing difference configuration sources.
 
-  The configuration file format *has changed* since the v0.90. Patroni is still compatible with the old configuration files, but in order to take advantage of the bootstrap parameters one needs to change it. Users are encourage to update them by referring to the :ref:`dynamic configuraton documentation page <dynamic_configuration>`.
+  The configuration file format *has changed* since the v0.90. Patroni is still compatible with the old configuration files, but in order to take advantage of the bootstrap parameters one needs to change it. Users are encourage to update them by referring to the :ref:`dynamic configuration documentation page <dynamic_configuration>`.
 
 **More flexible configuration***
 

docs/replica_bootstrap.rst

@@ -24,6 +24,7 @@ arguments to them, i.e. the name of the cluster and the path to the data directo
         <custom_bootstrap_method_name>:
             command: <path_to_custom_bootstrap_script> [param1 [, ...]]
             keep_existing_recovery_conf: False
+            no_params: False
             recovery_conf:
                 recovery_target_action: promote
                 recovery_target_timeline: latest
@@ -40,6 +41,8 @@ in the configuration files, Patroni supplies two cluster-specific ones:
 --datadir
     Path to the data directory of the cluster instance to be bootstrapped
 
+Passing these two additional flags can be disabled by setting a special ``no_params`` parameter to ``True``.
+
 If the bootstrap script returns 0, Patroni tries to configure and start the PostgreSQL instance produced by it. If any
 of the intermediate steps fail, or the script returns a non-zero value, Patroni assumes that the bootstrap has failed,
 cleans up after itself and releases the initialize lock to give another node the opportunity to bootstrap.
@@ -127,7 +130,7 @@ A special ``no_master`` parameter, if defined, allows Patroni to call the replic
 running master or replicas. In that case, an empty string will be passed in a connection string. This is useful for
 restoring the formerly running cluster from the binary backup.
 
-A special ``keep_data`` parameter, if defined, will instuct Patroni to  not clean PGDATA folder before calling restore.
+A special ``keep_data`` parameter, if defined, will instruct Patroni to  not clean PGDATA folder before calling restore.
 
 A special ``no_params`` parameter, if defined, restricts passing parameters to custom command.
 
@@ -140,6 +143,10 @@ possible to specify a ``basebackup`` configuration section. Same rules as with t
 namely, only long (with --) options should be specified there. Not all parameters make sense, if you override a connection
 string or provide an option to created tar-ed or compressed base backups, patroni won't be able to make a replica out
 of it. There is no validation performed on the names or values of the parameters passed to the ``basebackup`` section.
+Also note that in case symlinks are used for the WAL folder it is up to the user to specify the correct ``--waldir``
+path as an option, so that after replica buildup or re-initialization the symlink would persist. This option is supported
+only since v10 though.
+
 You can specify basebackup parameters as either a map (key-value pairs) or a list of elements, where each element
 could be either a key-value pair or a single key (for options that does not receive any values, for instance, ``--verbose``).
 Consider those 2 examples:
@@ -159,6 +166,7 @@ and
         basebackup:
             - verbose
             - max-rate: '100M'
+            - waldir: /pg-wal-mount/external-waldir
 
 If all replica creation methods fail, Patroni will try again all methods in order during the next event loop cycle.
 
@@ -204,4 +212,4 @@ in a patroni configuration:
 Note, that these options will be applied only once during cluster bootstrap,
 and the only way to change them afterwards is through DCS.
 
-If you use replication slots on the standby cluster, you must also create the corresponding replication slot on the primary cluster.  It will not be done automatically by the standby cluster implementation.  You can use Patroni's permenant replication slots feature on the primary cluster to maintain a replication slot with the same name as ``primary_slot_name``, or its default value if ``primary_slot_name`` is not provided.
+If you use replication slots on the standby cluster, you must also create the corresponding replication slot on the primary cluster.  It will not be done automatically by the standby cluster implementation.  You can use Patroni's permanent replication slots feature on the primary cluster to maintain a replication slot with the same name as ``primary_slot_name``, or its default value if ``primary_slot_name`` is not provided.

docs/replication_modes.rst

@@ -46,7 +46,7 @@ When ``synchronous_mode`` is on and a standby crashes, commits will block until
 
 When it is absolutely necessary to guarantee that each write is stored durably
 on at least two nodes, enable ``synchronous_mode_strict`` in addition to the
-``synchronous_node``. This parameter prevents Patroni from switching off the
+``synchronous_mode``. This parameter prevents Patroni from switching off the
 synchronous replication on the primary when no synchronous standby candidates
 are available. As a downside, the primary is not be available for writes
 (unless the Postgres transaction explicitly turns of ``synchronous_mode``),
@@ -57,11 +57,16 @@ You can ensure that a standby never becomes the synchronous standby by setting `
 
 Synchronous mode can be switched on and off via Patroni REST interface. See :ref:`dynamic configuration <dynamic_configuration>` for instructions.
 
+Note: Because of the way synchronous replication is implemented in PostgreSQL it is still possible to lose transactions even when using ``synchronous_mode_strict``. If the PostgreSQL backend is cancelled while waiting to acknowledge replication (as a result of packet cancellation due to client timeout or backend failure) transaction changes become visible for other backends. Such changes are not yet replicated and may be lost in case of standby promotion.
+
+Synchronous Replication Factor
+------------------------------
+The parameter ``synchronous_node_count`` is used by Patroni to manage number of synchronous standby databases. It is set to 1 by default. It has no effect when ``synchronous_mode`` is set to off. When enabled, Patroni manages precise number of synchronous standby databases based on parameter ``synchronous_node_count`` and adjusts the state in DCS & synchronous_standby_names as members join and leave.
 
 Synchronous mode implementation
 -------------------------------
 
-When in synchronous mode Patroni maintains synchronization state in the DCS, containing the latest primary and current synchronous standby. This state is updated with strict ordering constraints to ensure the following invariants:
+When in synchronous mode Patroni maintains synchronization state in the DCS, containing the latest primary and current synchronous standby databases. This state is updated with strict ordering constraints to ensure the following invariants:
 
 - A node must be marked as the latest leader whenever it can accept write transactions. Patroni crashing or PostgreSQL not shutting down can cause violations of this invariant.
 
@@ -69,9 +74,9 @@ When in synchronous mode Patroni maintains synchronization state in the DCS, con
 
 - A node that is not the leader or current synchronous standby is not allowed to promote itself automatically.
 
-Patroni will only ever assign one standby to ``synchronous_standby_names`` because with multiple candidates it is not possible to know which node was acting as synchronous during the failure.
+Patroni will only assign one or more synchronous standby nodes based on ``synchronous_node_count`` parameter to ``synchronous_standby_names``.
 
-On each HA loop iteration Patroni re-evaluates synchronous standby choice. If the current synchronous standby is connected and has not requested its synchronous status to be removed it remains picked. Otherwise the cluster member available for sync that is furthest ahead in replication is picked.
+On each HA loop iteration Patroni re-evaluates synchronous standby nodes choice. If the current list of synchronous standby nodes are connected and has not requested its synchronous status to be removed it remains picked. Otherwise the cluster member available for sync that is furthest ahead in replication is picked.
 
 
 .. [1] The data is still there, but recovering it requires a manual recovery effort by data recovery specialists. When Patroni is allowed to rewind with ``use_pg_rewind`` the forked timeline will be automatically erased to rejoin the failed primary with the cluster.

docs/rest_api.rst

@@ -0,0 +1,391 @@
+.. _rest_api:
+
+Patroni REST API
+================
+
+Patroni has a rich REST API, which is used by Patroni itself during the leader race, by the ``patronictl`` tool in order to perform failovers/switchovers/reinitialize/restarts/reloads, by HAProxy or any other kind of load balancer to perform HTTP health checks, and of course could also be used for monitoring. Below you will find the list of Patroni REST API endpoints.
+
+Health check endpoints
+----------------------
+For all health check ``GET`` requests Patroni returns a JSON document with the status of the node, along with the HTTP status code. If you don't want or don't need the JSON document, you might consider using the ``OPTIONS`` method instead of ``GET``.
+
+- The following requests to Patroni REST API will return HTTP status code **200** only when the Patroni node is running as the primary with leader lock:
+
+  - ``GET /``
+  - ``GET /master``
+  - ``GET /primary``
+  - ``GET /read-write``
+
+- ``GET /standby-leader``: returns HTTP status code **200** only when the Patroni node is running as the leader in a :ref:`standby cluster <standby_cluster>`.
+
+- ``GET /leader``: returns HTTP status code **200** when the Patroni node has the leader lock. The major difference from the two previous endpoints is that it doesn't take into account whether PostgreSQL is running as the ``primary`` or the ``standby_leader``.
+
+- ``GET /replica``: replica health check endpoint. It returns HTTP status code **200** only when the Patroni node is in the state ``running``, the role is ``replica`` and ``noloadbalance`` tag is not set.
+
+- ``GET /replica?lag=<max-lag>``: replica check endpoint. In addition to checks from ``replica``, it also checks replication latency and returns status code **200** only when it is below specified value. The key cluster.last_leader_operation from DCS is used for Leader wal position and compute latency on replica for performance reasons. max-lag can be specified in bytes (integer) or in human readable values, for e.g. 16kB, 64MB, 1GB.
+
+  - ``GET /replica?lag=1048576``
+  - ``GET /replica?lag=1024kB``
+  - ``GET /replica?lag=10MB``
+  - ``GET /replica?lag=1GB``
+
+- ``GET /replica?tag_key1=value1&tag_key2=value2``: replica check endpoint. In addition, It will also check for user defined tags ``key1`` and ``key2`` and their respective values in the **tags** section of the yaml configuration management. If the tag isn't defined for an instance, or if the value in the yaml configuration doesn't match the querying value, it will return HTTP Status Code 503.
+
+ In the following requests, since we are checking for the leader or standby-leader status, Patroni doesn't apply any of the user defined tags and they will be ignored.
+  - ``GET /?tag_key1=value1&tag_key2=value2``
+  - ``GET /master?tag_key1=value1&tag_key2=value2``
+  - ``GET /leader?tag_key1=value1&tag_key2=value2``
+  - ``GET /primary?tag_key1=value1&tag_key2=value2``
+  - ``GET /read-write?tag_key1=value1&tag_key2=value2``
+  - ``GET /standby_leader?tag_key1=value1&tag_key2=value2``
+  - ``GET /standby-leader?tag_key1=value1&tag_key2=value2``
+
+- ``GET /read-only``: like the above endpoint, but also includes the primary.
+
+- ``GET /synchronous`` or ``GET /sync``: returns HTTP status code **200** only when the Patroni node is running as a synchronous standby.
+
+- ``GET /asynchronous`` or ``GET /async``: returns HTTP status code **200** only when the Patroni node is running as an asynchronous standby.
+
+- ``GET /asynchronous?lag=<max-lag>`` or ``GET /async?lag=<max-lag>``: asynchronous standby check endpoint. In addition to checks from ``asynchronous`` or ``async``, it also checks replication latency and returns status code **200** only when it is below specified value. The key cluster.last_leader_operation from DCS is used for Leader wal position and compute latency on replica for performance reasons. max-lag can be specified in bytes (integer) or in human readable values, for e.g. 16kB, 64MB, 1GB.
+
+  - ``GET /async?lag=1048576``
+  - ``GET /async?lag=1024kB``
+  - ``GET /async?lag=10MB``
+  - ``GET /async?lag=1GB``
+
+- ``GET /health``: returns HTTP status code **200** only when PostgreSQL is up and running.
+
+- ``GET /liveness``: always returns HTTP status code **200** what only indicates that Patroni is running. Could be used for ``livenessProbe``.
+
+- ``GET /readiness``: returns HTTP status code **200** when the Patroni node is running as the leader or when PostgreSQL is up and running. The endpoint could be used for ``readinessProbe`` when it is not possible to use Kubernetes endpoints for leader elections (OpenShift).
+
+Both, ``readiness`` and ``liveness`` endpoints are very light-weight and not executing any SQL. Probes should be configured in such a way that they start failing about time when the leader key is expiring. With the default value of ``ttl``, which is ``30s`` example probes would look like:
+
+.. code-block:: yaml
+
+    readinessProbe:
+      httpGet:
+        scheme: HTTP
+        path: /readiness
+        port: 8008
+      initialDelaySeconds: 3
+      periodSeconds: 10
+      timeoutSeconds: 5
+      successThreshold: 1
+      failureThreshold: 3
+    livenessProbe:
+      httpGet:
+        scheme: HTTP
+        path: /liveness
+        port: 8008
+      initialDelaySeconds: 3
+      periodSeconds: 10
+      timeoutSeconds: 5
+      successThreshold: 1
+      failureThreshold: 3
+
+
+Monitoring endpoint
+-------------------
+
+The ``GET /patroni`` is used by Patroni during the leader race. It also could be used by your monitoring system. The JSON document produced by this endpoint has the same structure as the JSON produced by the health check endpoints.
+
+.. code-block:: bash
+
+    $ curl -s http://localhost:8008/patroni | jq .
+    {
+      "state": "running",
+      "postmaster_start_time": "2019-09-24 09:22:32.555 CEST",
+      "role": "master",
+      "server_version": 110005,
+      "cluster_unlocked": false,
+      "xlog": {
+        "location": 25624640
+      },
+      "timeline": 3,
+      "database_system_identifier": "6739877027151648096",
+      "patroni": {
+        "version": "1.6.0",
+        "scope": "batman"
+      }
+    }
+
+Cluster status endpoints
+------------------------
+
+- The ``GET /cluster`` endpoint generates a JSON document describing the current cluster topology and state:
+
+.. code-block:: bash
+
+    $ curl -s http://localhost:8008/cluster | jq .
+    {
+      "members": [
+        {
+          "name": "postgresql0",
+          "host": "127.0.0.1",
+          "port": 5432,
+          "role": "leader",
+          "state": "running",
+          "api_url": "http://127.0.0.1:8008/patroni",
+          "timeline": 5,
+          "tags": {
+            "clonefrom": true
+          }
+        },
+        {
+          "name": "postgresql1",
+          "host": "127.0.0.1",
+          "port": 5433,
+          "role": "replica",
+          "state": "running",
+          "api_url": "http://127.0.0.1:8009/patroni",
+          "timeline": 5,
+          "tags": {
+            "clonefrom": true
+          },
+          "lag": 0
+        }
+      ],
+      "scheduled_switchover": {
+        "at": "2019-09-24T10:36:00+02:00",
+        "from": "postgresql0"
+      }
+    }
+
+
+- The ``GET /history`` endpoint provides a view on the history of cluster switchovers/failovers. The format is very similar to the content of history files in the ``pg_wal`` directory. The only difference is the timestamp field showing when the new timeline was created.
+
+.. code-block:: bash
+
+    $ curl -s http://localhost:8008/history | jq .
+    [
+      [
+        1,
+        25623960,
+        "no recovery target specified",
+        "2019-09-23T16:57:57+02:00"
+      ],
+      [
+        2,
+        25624344,
+        "no recovery target specified",
+        "2019-09-24T09:22:33+02:00"
+      ],
+      [
+        3,
+        25624752,
+        "no recovery target specified",
+        "2019-09-24T09:26:15+02:00"
+      ],
+      [
+        4,
+        50331856,
+        "no recovery target specified",
+        "2019-09-24T09:35:52+02:00"
+      ]
+    ]
+
+
+Config endpoint
+---------------
+
+``GET /config``: Get the current version of the dynamic configuration:
+
+.. code-block:: bash
+
+	$ curl -s localhost:8008/config | jq .
+	{
+	  "ttl": 30,
+	  "loop_wait": 10,
+	  "retry_timeout": 10,
+	  "maximum_lag_on_failover": 1048576,
+	  "postgresql": {
+	    "use_slots": true,
+	    "use_pg_rewind": true,
+	    "parameters": {
+	      "hot_standby": "on",
+	      "wal_log_hints": "on",
+	      "wal_level": "hot_standby",
+	      "max_wal_senders": 5,
+	      "max_replication_slots": 5,
+	      "max_connections": "100"
+	    }
+	  }
+	}
+
+
+``PATCH /config``: Change the existing configuration.
+
+.. code-block:: bash
+
+	$ curl -s -XPATCH -d \
+		'{"loop_wait":5,"ttl":20,"postgresql":{"parameters":{"max_connections":"101"}}}' \
+		http://localhost:8008/config | jq .
+	{
+	  "ttl": 20,
+	  "loop_wait": 5,
+	  "maximum_lag_on_failover": 1048576,
+	  "retry_timeout": 10,
+	  "postgresql": {
+	    "use_slots": true,
+	    "use_pg_rewind": true,
+	    "parameters": {
+	      "hot_standby": "on",
+	      "wal_log_hints": "on",
+	      "wal_level": "hot_standby",
+	      "max_wal_senders": 5,
+	      "max_replication_slots": 5,
+	      "max_connections": "101"
+	    }
+	  }
+	}
+
+The above REST API call patches the existing configuration and returns the new configuration.
+
+Let's check that the node processed this configuration. First of all it should start printing log lines every 5 seconds (loop_wait=5). The change of "max_connections" requires a restart, so the "pending_restart" flag should be exposed:
+
+.. code-block:: bash
+
+	$ curl -s http://localhost:8008/patroni | jq .
+	{
+	  "pending_restart": true,
+	  "database_system_identifier": "6287881213849985952",
+	  "postmaster_start_time": "2016-06-13 13:13:05.211 CEST",
+	  "xlog": {
+	    "location": 2197818976
+	  },
+	  "patroni": {
+	    "scope": "batman",
+	    "version": "1.0"
+	  },
+	  "state": "running",
+	  "role": "master",
+	  "server_version": 90503
+	}
+
+Removing parameters:
+
+If you want to remove (reset) some setting just patch it with ``null``:
+
+.. code-block:: bash
+
+	$ curl -s -XPATCH -d \
+		'{"postgresql":{"parameters":{"max_connections":null}}}' \
+		http://localhost:8008/config | jq .
+	{
+	  "ttl": 20,
+	  "loop_wait": 5,
+	  "retry_timeout": 10,
+	  "maximum_lag_on_failover": 1048576,
+	  "postgresql": {
+	    "use_slots": true,
+	    "use_pg_rewind": true,
+	    "parameters": {
+	      "hot_standby": "on",
+	      "unix_socket_directories": ".",
+	      "wal_level": "hot_standby",
+	      "wal_log_hints": "on",
+	      "max_wal_senders": 5,
+	      "max_replication_slots": 5
+	    }
+	  }
+	}
+
+The above call removes ``postgresql.parameters.max_connections`` from the dynamic configuration.
+
+``PUT /config``: It's also possible to perform the full rewrite of an existing dynamic configuration unconditionally:
+
+.. code-block:: bash
+
+	$ curl -s -XPUT -d \
+		'{"maximum_lag_on_failover":1048576,"retry_timeout":10,"postgresql":{"use_slots":true,"use_pg_rewind":true,"parameters":{"hot_standby":"on","wal_log_hints":"on","wal_level":"hot_standby","unix_socket_directories":".","max_wal_senders":5}},"loop_wait":3,"ttl":20}' \
+		http://localhost:8008/config | jq .
+	{
+	  "ttl": 20,
+	  "maximum_lag_on_failover": 1048576,
+	  "retry_timeout": 10,
+	  "postgresql": {
+	    "use_slots": true,
+	    "parameters": {
+	      "hot_standby": "on",
+	      "unix_socket_directories": ".",
+	      "wal_level": "hot_standby",
+	      "wal_log_hints": "on",
+	      "max_wal_senders": 5
+	    },
+	    "use_pg_rewind": true
+	  },
+	  "loop_wait": 3
+	}
+
+
+Switchover and failover endpoints
+---------------------------------
+
+``POST /switchover`` or ``POST /failover``. These endpoints are very similar to each other. There are a couple of minor differences though:
+
+1. The failover endpoint allows to perform a manual failover when there are no healthy nodes, but at the same time it will not allow you to schedule a switchover.
+
+2. The switchover endpoint is the opposite. It works only when the cluster is healthy (there is a leader) and allows to schedule a switchover at a given time.
+
+
+In the JSON body of the ``POST`` request you must specify at least the ``leader`` or ``candidate`` fields and optionally the ``scheduled_at`` field if you want to schedule a switchover at a specific time.
+
+
+Example: perform a failover to the specific node:
+
+.. code-block:: bash
+
+    $ curl -s http://localhost:8009/failover -XPOST -d '{"candidate":"postgresql1"}'
+    Successfully failed over to "postgresql1"
+
+
+Example: schedule a switchover from the leader to any other healthy replica in the cluster at a specific time:
+
+.. code-block:: bash
+
+    $ curl -s http://localhost:8008/switchover -XPOST -d \
+	    '{"leader":"postgresql0","scheduled_at":"2019-09-24T12:00+00"}'
+    Switchover scheduled
+
+
+Depending on the situation the request might finish with a different HTTP status code and body. The status code **200** is returned when the switchover or failover successfully completed. If the switchover was successfully scheduled, Patroni will return HTTP status code **202**. In case something went wrong, the error status code (one of **400**, **412** or **503**) will be returned with some details in the response body. For more information please check the source code of ``patroni/api.py:do_POST_failover()`` method.
+
+- ``DELETE /switchover``: delete the scheduled switchover
+
+The ``POST /switchover`` and ``POST failover`` endpoints are used by ``patronictl switchover`` and ``patronictl failover``, respectively.
+The ``DELETE /switchover`` is used by ``patronictl flush <cluster-name> switchover``.
+
+
+Restart endpoint
+----------------
+
+- ``POST /restart``: You can restart Postgres on the specific node by performing the ``POST /restart`` call. In the JSON body of ``POST`` request it is possible to optionally specify some restart conditions:
+
+  - **restart_pending**: boolean, if set to ``true`` Patroni will restart PostgreSQL only when restart is pending in order to apply some changes in the PostgreSQL config.
+  - **role**: perform restart only if the current role of the node matches with the role from the POST request.
+  - **postgres_version**: perform restart only if the current version of postgres is smaller than specified in the POST request.
+  - **timeout**: how long we should wait before PostgreSQL starts accepting connections. Overrides ``master_start_timeout``.
+  - **schedule**: timestamp with time zone, schedule the restart somewhere in the future.
+
+- ``DELETE /restart``: delete the scheduled restart
+
+``POST /restart`` and ``DELETE /restart`` endpoints are used by ``patronictl restart`` and ``patronictl flush <cluster-name> restart`` respectively.
+
+
+Reload endpoint
+---------------
+
+The ``POST /reload`` call will order Patroni to re-read and apply the configuration file. This is the equivalent of sending the ``SIGHUP`` signal to the Patroni process. In case you changed some of the Postgres parameters which require a restart (like **shared_buffers**), you still have to explicitly do the restart of Postgres by either calling the ``POST /restart`` endpoint or with the help of ``patronictl restart``.
+
+The reload endpoint is used by ``patronictl reload``.
+
+
+Reinitialize endpoint
+---------------------
+
+``POST /reinitialize``: reinitialize the PostgreSQL data directory on the specified node. It is allowed to be executed only on replicas. Once called, it will remove the data directory and start ``pg_basebackup`` or some alternative :ref:`replica creation method <custom_replica_creation>`.
+
+The call might fail if Patroni is in a loop trying to recover (restart) a failed Postgres. In order to overcome this problem one can specify ``{"force":true}`` in the request body.
+
+The reinitialize endpoint is used by ``patronictl reinit``.

docs/security.rst

@@ -0,0 +1,37 @@
+.. _security:
+
+=======================
+Security Considerations
+=======================
+
+A Patroni cluster has two interfaces to be protected from unauthorized access: the distributed configuration storage (DCS) and the Patroni REST API.
+
+Protecting DCS
+==============
+
+Patroni and patronictl both store and retrieve data to/from the DCS. 
+
+Despite DCS doesn't contain any sensitive information, it allows changing some of Patroni/Postgres configuration. Therefore the very first thing that should be protected is DCS itself.
+
+The details of protection depend on the type of DCS used. The authentication and encryption parameters (tokens/basic-auth/client certificates) for the supported types of DCS are covered in :ref:`SETTINGS <bootstrap_settings>` 
+
+The general recommendation is to enable TLS for all DCS communication.
+
+Protecting the REST API
+=======================
+
+Protecting the REST API is a more complicated task.
+
+The Patroni REST API is used by Patroni itself during the leader race, by the ``patronictl`` tool in order to perform failovers/switchovers/reinitialize/restarts/reloads, by HAProxy or any other kind of load balancer to perform HTTP health checks, and of course could also be used for monitoring. 
+
+From the point of view of security, REST API contains safe (``GET`` requests, only retrieve information) and unsafe (``PUT``, ``POST``, ``PATCH`` and ``DELETE`` requests, change the state of nodes) endpoints.
+
+The unsafe endpoints can be protected with HTTP basic-auth by setting the ``restapi.authentication.username`` and ``restapi.authentication.password`` parameters. There is no way to protect the safe endpoints without enabling TLS.
+
+When TLS for the REST API is enabled and a PKI is established, mutual authentication of the API server and API client is possible for all endpoints.
+
+The ``restapi`` section parameters enable TLS client authentication to the server. Depending on the value of the ``verify_client`` parameter, the API server requires a successful client certificate verification for both safe and unsafe API calls (``verify_client: required``), or only for unsafe API calls (``verify_client: optional``), or for no API calls (``verify_client: none``).
+
+The ``ctl`` section parameters enable TLS server authentication to the client (the ``patronictl`` tool which uses the same config as patroni). Set ``insecure: true`` to disable the server certificate verification by the client. See :ref:`SETTINGS <patronictl_settings>` for a detailed description of the TLS client parameters.
+
+Protecting the PostgreSQL database proper from unauthorized access is beyond the scope of this document and is covered in https://www.postgresql.org/docs/current/client-authentication.html

extras/startup-scripts/patroni.service

@@ -14,7 +14,8 @@ Group=postgres
 # Read in configuration file if it exists, otherwise proceed
 EnvironmentFile=-/etc/patroni_env.conf
 
-WorkingDirectory=~
+# the default is the user's home directory, and if you want to change it, you must provide an absolute path.
+# WorkingDirectory=/home/sameuser
 
 # Where to send early-startup messages from the server
 # This is normally controlled by the global default set by systemd

features/archive-restore.py

@@ -0,0 +1,21 @@
+#!/usr/bin/env python
+import os
+import argparse
+import shutil
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--dirname", required=True)
+    parser.add_argument("--pathname", required=True)
+    parser.add_argument("--filename", required=True)
+    parser.add_argument("--mode", required=True, choices=("archive", "restore"))
+    args, _ = parser.parse_known_args()
+
+    full_filename = os.path.join(args.dirname, args.filename)
+    if args.mode == "archive":
+        if not os.path.isdir(args.dirname):
+            os.makedirs(args.dirname)
+        if not os.path.exists(full_filename):
+            shutil.copy(args.pathname, full_filename)
+    else:
+        shutil.copy(full_filename, args.pathname)

features/backup_create.py

@@ -0,0 +1,14 @@
+#!/usr/bin/env python
+import argparse
+import subprocess
+import sys
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--datadir", required=True)
+    parser.add_argument("--dbname", required=True)
+    parser.add_argument("--walmethod", required=True, choices=("fetch", "stream", "none"))
+    args, _ = parser.parse_known_args()
+
+    walmethod = ["-X", args.walmethod] if args.walmethod != "none" else []
+    sys.exit(subprocess.call(["pg_basebackup", "-D", args.datadir, "-c", "fast", "-d", args.dbname] + walmethod))

features/backup_create.sh

@@ -1,22 +0,0 @@
-#!/bin/bash
-
-while getopts ":-:" optchar; do
-    [[ "${optchar}" == "-" ]] || continue
-    case "${OPTARG}" in
-        datadir=* )
-            PGDATA=${OPTARG#*=}
-            ;;
-        dbname=* )
-            DBNAME=${OPTARG#*=}
-            ;;
-        walmethod=* )
-            WALMETHOD=${OPTARG#*=}
-            ;;
-    esac
-done
-
-[[ -z $PGDATA || -z $DBNAME || -z $WALMETHOD ]] && exit 1
-
-[[ $WALMETHOD != "none" ]] && WALMETHOD="-X $WALMETHOD" || WALMETHOD=""
-
-exec pg_basebackup -D $PGDATA $WALMETHOD -c fast -d $DBNAME

features/backup_restore.py

@@ -0,0 +1,11 @@
+#!/usr/bin/env python
+import argparse
+import shutil
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--datadir", required=True)
+    parser.add_argument("--sourcedir", required=True)
+    args, _ = parser.parse_known_args()
+
+    shutil.copytree(args.sourcedir, args.datadir)

features/backup_restore.sh

@@ -1,21 +0,0 @@
-#!/bin/bash
-
-set -x
-
-while getopts ":-:" optchar; do
-    [[ "${optchar}" == "-" ]] || continue
-    case "${OPTARG}" in
-        datadir=* )
-            PGDATA=${OPTARG#*=}
-            ;;
-        sourcedir=* )
-            SOURCE=${OPTARG#*=}
-            ;;
-    esac
-done
-
-[[ -z $PGDATA || -z $SOURCE ]] && exit 1
-
-mkdir -p $(dirname $PGDATA)
-
-exec cp -af $SOURCE $PGDATA

features/basic_replication.feature

@@ -4,6 +4,7 @@ Feature: basic replication
   Scenario: check replication of a single table
     Given I start postgres0
     Then postgres0 is a leader after 10 seconds
+    And there is a non empty initialize key in DCS after 15 seconds
     When I issue a PATCH request to http://127.0.0.1:8008/config with {"ttl": 20, "loop_wait": 2, "synchronous_mode": true}
     Then I receive a response code 200
     When I start postgres1
@@ -27,6 +28,46 @@ Feature: basic replication
     When I issue a GET request to http://127.0.0.1:8009/async
     Then I receive a response code 200
 
+  Scenario: check stuck sync replica
+    Given I issue a PATCH request to http://127.0.0.1:8008/config with {"pause": true, "maximum_lag_on_syncnode": 15000000, "postgresql": {"parameters": {"synchronous_commit": "remote_apply"}}}
+    Then I receive a response code 200
+    And I create table on postgres0
+    And table mytest is present on postgres1 after 2 seconds
+    And table mytest is present on postgres2 after 2 seconds
+    When I pause wal replay on postgres2
+    And I load data on postgres0
+    Then "sync" key in DCS has sync_standby=postgres1 after 15 seconds
+    And I resume wal replay on postgres2
+    And I sleep for 2 seconds
+    And I issue a GET request to http://127.0.0.1:8009/sync
+    Then I receive a response code 200
+    When I issue a GET request to http://127.0.0.1:8010/async
+    Then I receive a response code 200
+    When I issue a PATCH request to http://127.0.0.1:8008/config with {"pause": null, "maximum_lag_on_syncnode": -1, "postgresql": {"parameters": {"synchronous_commit": "on"}}}
+    Then I receive a response code 200
+    And I drop table on postgres0
+
+  Scenario: check multi sync replication
+    Given I issue a PATCH request to http://127.0.0.1:8008/config with {"synchronous_node_count": 2}
+    Then I receive a response code 200
+    And I sleep for 10 seconds
+    Then "sync" key in DCS has sync_standby=postgres1,postgres2 after 5 seconds
+    When I issue a GET request to http://127.0.0.1:8010/sync
+    Then I receive a response code 200
+    When I issue a GET request to http://127.0.0.1:8009/sync
+    Then I receive a response code 200
+    When I issue a PATCH request to http://127.0.0.1:8008/config with {"synchronous_node_count": 1}
+    Then I receive a response code 200
+    And I shut down postgres1
+    And I sleep for 10 seconds
+    Then "sync" key in DCS has sync_standby=postgres2 after 10 seconds
+    When I start postgres1
+    And "members/postgres1" key in DCS has state=running after 10 seconds
+    When I issue a GET request to http://127.0.0.1:8010/sync
+    Then I receive a response code 200
+    When I issue a GET request to http://127.0.0.1:8009/async
+    Then I receive a response code 200
+
   Scenario: check the basic failover in synchronous mode
     Given I run patronictl.py pause batman
     Then I receive a response returncode 0
@@ -35,10 +76,12 @@ Feature: basic replication
     And I run patronictl.py resume batman
     Then I receive a response returncode 0
     And postgres2 role is the primary after 24 seconds
+    And Response on GET http://127.0.0.1:8010/history contains recovery after 10 seconds
     When I issue a PATCH request to http://127.0.0.1:8010/config with {"synchronous_mode": null, "master_start_timeout": 0}
     Then I receive a response code 200
     When I add the table bar to postgres2
     Then table bar is present on postgres1 after 20 seconds
+    And Response on GET http://127.0.0.1:8010/config contains master_start_timeout after 10 seconds
 
   Scenario: check immediate failover when master_start_timeout=0
     Given I kill postmaster on postgres2

features/callback.sh

@@ -1,5 +0,0 @@
-#!/bin/bash
-
-[[ "$3" == "master" ]] || exit
-
-PGPASSWORD=zalando psql -h localhost -U postgres -p $1 -w -tAc "SELECT slot_name FROM pg_replication_slots WHERE slot_type = 'logical'" >> data/postgres0/label

features/callback2.py

@@ -0,0 +1,5 @@
+#!/usr/bin/env python
+
+import sys
+with open("data/{0}/{0}_cb.log".format(sys.argv[1]), "a+") as log:
+    log.write(" ".join(sys.argv[-3:]) + "\n")

features/custom_bootstrap.feature

@@ -13,5 +13,5 @@ Scenario: make a backup and do a restore into a new cluster
     Given I add the table bar to postgres1
     And I do a backup of postgres1
     When I start postgres2 in a cluster batman2 from backup
-    Then postgres2 is a leader of batman2 after 10 seconds
+    Then postgres2 is a leader of batman2 after 30 seconds
     And table bar is present on postgres2 after 10 seconds

features/environment.py

@@ -1,11 +1,6 @@
 import abc
-import consul
 import datetime
-import etcd
-import kazoo.client
-import kazoo.exceptions
 import os
-import psutil
 import psycopg2
 import json
 import shutil
@@ -18,6 +13,8 @@ import threading
 import time
 import yaml
 
+from six.moves.BaseHTTPServer import BaseHTTPRequestHandler, HTTPServer
+
 
 @six.add_metaclass(abc.ABCMeta)
 class AbstractController(object):
@@ -170,13 +167,21 @@ class PatroniController(AbstractController):
             config = yaml.safe_load(f)
             config.pop('etcd', None)
 
+        raft_port = os.environ.get('RAFT_PORT')
+        if raft_port:
+            os.environ['RAFT_PORT'] = str(int(raft_port) + 1)
+            config['raft'] = {'data_dir': self._output_dir, 'self_addr': 'localhost:' + os.environ['RAFT_PORT']}
+
         host = config['postgresql']['listen'].split(':')[0]
 
         config['postgresql']['listen'] = config['postgresql']['connect_address'] = '{0}:{1}'.format(host, self.__PORT)
 
         config['name'] = name
         config['postgresql']['data_dir'] = self._data_dir
-        config['postgresql']['use_unix_socket'] = True
+        config['postgresql']['basebackup'] = [{'checkpoint': 'fast'}]
+        config['postgresql']['use_unix_socket'] = os.name != 'nt'  # windows doesn't yet support unix-domain sockets
+        config['postgresql']['use_unix_socket_repl'] = os.name != 'nt'  # windows doesn't yet support unix-domain sockets
+        config['postgresql']['pgpass'] = os.path.join(tempfile.gettempdir(), 'pgpass_' + name)
         config['postgresql']['parameters'].update({
             'logging_collector': 'on', 'log_destination': 'csvlog', 'log_directory': self._output_dir,
             'log_filename': name + '.log', 'log_statement': 'all', 'log_min_messages': 'debug1',
@@ -190,6 +195,8 @@ class PatroniController(AbstractController):
         if custom_config is not None:
             self.recursive_update(config, custom_config)
 
+        self.recursive_update(config, {
+            'bootstrap': {'dcs': {'postgresql': {'parameters': {'wal_keep_segments': 100}}}}})
         if config['postgresql'].get('callbacks', {}).get('on_role_change'):
             config['postgresql']['callbacks']['on_role_change'] += ' ' + str(self.__PORT)
 
@@ -249,59 +256,24 @@ class PatroniController(AbstractController):
         except Exception:
             return None
 
-    def database_is_running(self):
-        pid = self._get_pid()
-        if not pid:
-            return False
-        try:
-            os.kill(pid, 0)
-        except OSError:
-            return False
-        return True
-
     def patroni_hang(self, timeout):
         hang = ProcessHang(self._handle.pid, timeout)
         self._closables.append(hang)
         hang.start()
 
-    def checkpoint_hang(self, timeout):
-        pid = self._get_pid()
-        if not pid:
-            return False
-        proc = psutil.Process(pid)
-        for child in proc.children():
-            if 'checkpoint' in child.cmdline()[0]:
-                checkpointer = child
-                break
-        else:
-            return False
-        hang = ProcessHang(checkpointer.pid, timeout)
-        self._closables.append(hang)
-        hang.start()
-        return True
-
     def cancel_background(self):
         for obj in self._closables:
             obj.close()
         self._closables = []
 
-    def terminate_backends(self):
-        pid = self._get_pid()
-        if not pid:
-            return False
-        proc = psutil.Process(pid)
-        for p in proc.children():
-            if 'process' not in p.cmdline()[0]:
-                p.terminate()
-
     @property
     def backup_source(self):
         return 'postgres://{username}:{password}@{host}:{port}/{database}'.format(**self._replication)
 
-    def backup(self, dest='data/basebackup'):
+    def backup(self, dest=os.path.join('data', 'basebackup')):
-        subprocess.call([PatroniPoolController.BACKUP_SCRIPT, '--walmethod=none',
+        subprocess.call(PatroniPoolController.BACKUP_SCRIPT + ['--walmethod=none',
-                         '--datadir=' + os.path.join(self._work_directory, dest),
+                        '--datadir=' + os.path.join(self._work_directory, dest),
-                         '--dbname=' + self.backup_source])
+                        '--dbname=' + self.backup_source])
 
 
 class ProcessHang(object):
@@ -375,9 +347,11 @@ class ConsulController(AbstractDcsController):
         super(ConsulController, self).__init__(context)
         os.environ['PATRONI_CONSUL_HOST'] = 'localhost:8500'
         os.environ['PATRONI_CONSUL_REGISTER_SERVICE'] = 'on'
-        self._client = consul.Consul()
         self._config_file = None
 
+        import consul
+        self._client = consul.Consul()
+
     def _start(self):
         self._config_file = self._work_directory + '.json'
         with open(self._config_file, 'wb') as f:
@@ -410,26 +384,45 @@ class ConsulController(AbstractDcsController):
         super(ConsulController, self).start(max_wait_limit)
 
 
-class EtcdController(AbstractDcsController):
+class AbstractEtcdController(AbstractDcsController):
 
     """ handles all etcd related tasks, used for the tests setup and cleanup """
 
-    def __init__(self, context):
+    def __init__(self, context, client_cls):
-        super(EtcdController, self).__init__(context)
+        super(AbstractEtcdController, self).__init__(context)
-        os.environ['PATRONI_ETCD_HOST'] = 'localhost:2379'
+        self._client_cls = client_cls
-        self._client = etcd.Client(port=2379)
 
     def _start(self):
         return subprocess.Popen(["etcd", "--debug", "--data-dir", self._work_directory],
                                 stdout=self._log, stderr=subprocess.STDOUT)
 
+    def _is_running(self):
+        from patroni.dcs.etcd import DnsCachingResolver
+        # if etcd is running, but we didn't start it
+        try:
+            self._client = self._client_cls({'host': 'localhost', 'port': 2379, 'retry_timeout': 30,
+                                             'patronictl': 1}, DnsCachingResolver())
+            return True
+        except Exception:
+            return False
+
+
+class EtcdController(AbstractEtcdController):
+
+    def __init__(self, context):
+        from patroni.dcs.etcd import EtcdClient
+        super(EtcdController, self).__init__(context, EtcdClient)
+        os.environ['PATRONI_ETCD_HOST'] = 'localhost:2379'
+
     def query(self, key, scope='batman'):
+        import etcd
         try:
             return self._client.get(self.path(key, scope)).value
         except etcd.EtcdKeyNotFound:
             return None
 
     def cleanup_service_tree(self):
+        import etcd
         try:
             self._client.delete(self.path(scope=''), recursive=True)
         except (etcd.EtcdKeyNotFound, etcd.EtcdConnectionFailed):
@@ -437,12 +430,25 @@ class EtcdController(AbstractDcsController):
         except Exception as e:
             assert False, "exception when cleaning up etcd contents: {0}".format(e)
 
-    def _is_running(self):
+
-        # if etcd is running, but we didn't start it
+class Etcd3Controller(AbstractEtcdController):
+
+    def __init__(self, context):
+        from patroni.dcs.etcd3 import Etcd3Client
+        super(Etcd3Controller, self).__init__(context, Etcd3Client)
+        os.environ['PATRONI_ETCD3_HOST'] = 'localhost:2379'
+
+    def query(self, key, scope='batman'):
+        import base64
+        response = self._client.range(self.path(key, scope))
+        for k in response.get('kvs', []):
+            return base64.b64decode(k['value']).decode('utf-8') if 'value' in k else None
+
+    def cleanup_service_tree(self):
         try:
-            return bool(self._client.machines)
+            self._client.deleteprefix(self.path(scope=''))
-        except Exception:
+        except Exception as e:
-            return False
+            assert False, "exception when cleaning up etcd contents: {0}".format(e)
 
 
 class KubernetesController(AbstractDcsController):
@@ -454,8 +460,9 @@ class KubernetesController(AbstractDcsController):
         self._label_selector = ','.join('{0}={1}'.format(k, v) for k, v in self._labels.items())
         os.environ['PATRONI_KUBERNETES_LABELS'] = json.dumps(self._labels)
         os.environ['PATRONI_KUBERNETES_USE_ENDPOINTS'] = 'true'
+        os.environ['PATRONI_KUBERNETES_BYPASS_API_SERVICE'] = 'true'
 
-        from kubernetes import client as k8s_client, config as k8s_config
+        from patroni.dcs.kubernetes import k8s_client, k8s_config
         k8s_config.load_kube_config(context='local')
         self._client = k8s_client
         self._api = self._client.CoreV1Api()
@@ -473,13 +480,13 @@ class KubernetesController(AbstractDcsController):
 
     def delete_pod(self, name):
         try:
-            self._api.delete_namespaced_pod(name, self._namespace, self._client.V1DeleteOptions())
+            self._api.delete_namespaced_pod(name, self._namespace, body=self._client.V1DeleteOptions())
-        except:
+        except Exception:
             pass
         while True:
             try:
                 self._api.read_namespaced_pod(name, self._namespace)
-            except:
+            except Exception:
                 break
 
     def query(self, key, scope='batman'):
@@ -488,22 +495,23 @@ class KubernetesController(AbstractDcsController):
             return (pod.metadata.annotations or {}).get('status', '')
         else:
             try:
-                e = self._api.read_namespaced_endpoints(scope + ('' if key == 'leader' else '-' + key), self._namespace)
+                ep = scope + {'leader': '', 'history': '-config', 'initialize': '-config'}.get(key, '-' + key)
-                if key == 'leader':
+                e = self._api.read_namespaced_endpoints(ep, self._namespace)
+                if key != 'sync':
                     return e.metadata.annotations[key]
                 else:
                     return json.dumps(e.metadata.annotations)
-            except:
+            except Exception:
                 return None
 
     def cleanup_service_tree(self):
         try:
             self._api.delete_collection_namespaced_pod(self._namespace, label_selector=self._label_selector)
-        except:
+        except Exception:
             pass
         try:
             self._api.delete_collection_namespaced_endpoints(self._namespace, label_selector=self._label_selector)
-        except:
+        except Exception:
             pass
 
         while True:
@@ -523,18 +531,22 @@ class ZooKeeperController(AbstractDcsController):
         super(ZooKeeperController, self).__init__(context, False)
         if export_env:
             os.environ['PATRONI_ZOOKEEPER_HOSTS'] = "'localhost:2181'"
+
+        import kazoo.client
         self._client = kazoo.client.KazooClient()
 
     def _start(self):
         pass  # TODO: implement later
 
     def query(self, key, scope='batman'):
+        import kazoo.exceptions
         try:
             return self._client.get(self.path(key, scope))[0].decode('utf-8')
         except kazoo.exceptions.NoNodeError:
             return None
 
     def cleanup_service_tree(self):
+        import kazoo.exceptions
         try:
             self._client.delete(self.path(scope=''), recursive=True)
         except (kazoo.exceptions.NoNodeError):
@@ -552,16 +564,76 @@ class ZooKeeperController(AbstractDcsController):
             return False
 
 
+class MockExhibitor(BaseHTTPRequestHandler):
+
+    def do_GET(self):
+        self.send_response(200)
+        self.end_headers()
+        self.wfile.write(b'{"servers":["127.0.0.1"],"port":2181}')
+
+    def log_message(self, fmt, *args):
+        pass
+
+
 class ExhibitorController(ZooKeeperController):
 
     def __init__(self, context):
         super(ExhibitorController, self).__init__(context, False)
-        os.environ.update({'PATRONI_EXHIBITOR_HOSTS': 'localhost', 'PATRONI_EXHIBITOR_PORT': '8181'})
+        port = 8181
+        exhibitor = HTTPServer(('', port), MockExhibitor)
+        exhibitor.daemon_thread = True
+        exhibitor_thread = threading.Thread(target=exhibitor.serve_forever)
+        exhibitor_thread.daemon = True
+        exhibitor_thread.start()
+        os.environ.update({'PATRONI_EXHIBITOR_HOSTS': 'localhost', 'PATRONI_EXHIBITOR_PORT': str(port)})
+
+
+class RaftController(AbstractDcsController):
+
+    CONTROLLER_ADDR = 'localhost:1234'
+    PASSWORD = '12345'
+
+    def __init__(self, context):
+        super(RaftController, self).__init__(context)
+        os.environ.update(PATRONI_RAFT_PARTNER_ADDRS="'" + self.CONTROLLER_ADDR + "'",
+                          PATRONI_RAFT_PASSWORD=self.PASSWORD, RAFT_PORT='1234')
+        self._raft = None
+
+    def _start(self):
+        env = os.environ.copy()
+        del env['PATRONI_RAFT_PARTNER_ADDRS']
+        env['PATRONI_RAFT_SELF_ADDR'] = self.CONTROLLER_ADDR
+        env['PATRONI_RAFT_DATA_DIR'] = self._work_directory
+        return subprocess.Popen([sys.executable, '-m', 'coverage', 'run',
+                                '--source=patroni', '-p', 'patroni_raft_controller.py'],
+                                stdout=self._log, stderr=subprocess.STDOUT, env=env)
+
+    def query(self, key, scope='batman'):
+        ret = self._raft.get(self.path(key, scope))
+        return ret and ret['value']
+
+    def set(self, key, value):
+        self._raft.set(self.path(key), value)
+
+    def cleanup_service_tree(self):
+        from patroni.dcs.raft import KVStoreTTL
+
+        if self._raft:
+            self._raft.destroy()
+            self.stop()
+            os.makedirs(self._work_directory)
+            self.start()
+
+        ready_event = threading.Event()
+        self._raft = KVStoreTTL(ready_event.set, None, None, partner_addrs=[self.CONTROLLER_ADDR], password=self.PASSWORD)
+        self._raft.startAutoTick()
+        ready_event.wait()
 
 
 class PatroniPoolController(object):
 
-    BACKUP_SCRIPT = 'features/backup_create.sh'
+    BACKUP_SCRIPT = [sys.executable, 'features/backup_create.py']
+    ARCHIVE_RESTORE_SCRIPT = ' '.join((sys.executable, os.path.abspath('features/archive-restore.py')))
 
     def __init__(self, context):
         self._context = context
@@ -594,9 +666,8 @@ class PatroniPoolController(object):
         self._processes[name].start(max_wait_limit)
 
     def __getattr__(self, func):
-        if func not in ['stop', 'query', 'write_label', 'read_label', 'check_role_has_changed_to', 'add_tag_to_config',
+        if func not in ['stop', 'query', 'write_label', 'read_label', 'check_role_has_changed_to',
-                        'get_watchdog', 'database_is_running', 'checkpoint_hang', 'patroni_hang',
+                        'add_tag_to_config', 'get_watchdog', 'patroni_hang', 'backup']:
-                        'terminate_backends', 'backup']:
             raise AttributeError("PatroniPoolController instance has no attribute '{0}'".format(func))
 
         def wrapper(name, *args, **kwargs):
@@ -610,7 +681,7 @@ class PatroniPoolController(object):
         self._processes.clear()
 
     def create_and_set_output_directory(self, feature_name):
-        feature_dir = os.path.join(self.patroni_path, 'features/output', feature_name.replace(' ', '_'))
+        feature_dir = os.path.join(self.patroni_path, 'features', 'output', feature_name.replace(' ', '_'))
         if os.path.exists(feature_dir):
             shutil.rmtree(feature_dir)
         os.makedirs(feature_dir)
@@ -623,7 +694,7 @@ class PatroniPoolController(object):
             'bootstrap': {
                 'method': 'pg_basebackup',
                 'pg_basebackup': {
-                    'command': self.BACKUP_SCRIPT + ' --walmethod=stream --dbname=' + f.backup_source
+                    'command': " ".join(self.BACKUP_SCRIPT) + ' --walmethod=stream --dbname=' + f.backup_source
                 },
                 'dcs': {
                     'postgresql': {
@@ -636,8 +707,9 @@ class PatroniPoolController(object):
             'postgresql': {
                 'parameters': {
                     'archive_mode': 'on',
-                    'archive_command': 'mkdir -p {0} && test ! -f {0}/%f && cp %p {0}/%f'.format(
+                    'archive_command': (self.ARCHIVE_RESTORE_SCRIPT + ' --mode archive ' +
-                            os.path.join(self.patroni_path, 'data/wal_archive'))
+                                        '--dirname {} --filename %f --pathname %p').format(
+                                        os.path.join(self.patroni_path, 'data', 'wal_archive'))
                 },
                 'authentication': {
                     'superuser': {'password': 'zalando1'},
@@ -653,12 +725,14 @@ class PatroniPoolController(object):
             'bootstrap': {
                 'method': 'backup_restore',
                 'backup_restore': {
-                    'command': 'features/backup_restore.sh --sourcedir=' + os.path.join(self.patroni_path,
+                    'command': (sys.executable + ' features/backup_restore.py --sourcedir=' +
-                                                                                        'data/basebackup'),
+                                os.path.join(self.patroni_path, 'data', 'basebackup')),
                     'recovery_conf': {
                         'recovery_target_action': 'promote',
                         'recovery_target_timeline': 'latest',
-                        'restore_command': 'cp {0}/data/wal_archive/%f %p'.format(self.patroni_path)
+                        'restore_command': (self.ARCHIVE_RESTORE_SCRIPT + ' --mode restore ' +
+                                            '--dirname {} --filename %f --pathname %p').format(
+                                            os.path.join(self.patroni_path, 'data', 'wal_archive'))
                     }
                 }
             },
@@ -795,11 +869,11 @@ class WatchdogMonitor(object):
         return triggered
 
 
-# actions to execute on start/stop of the tests and before running invidual features
+# actions to execute on start/stop of the tests and before running individual features
 def before_all(context):
     os.environ.update({'PATRONI_RESTAPI_USERNAME': 'username', 'PATRONI_RESTAPI_PASSWORD': 'password'})
-    context.ci = 'TRAVIS_BUILD_NUMBER' in os.environ or 'BUILD_NUMBER' in os.environ
+    context.ci = any(a in os.environ for a in ('TRAVIS_BUILD_NUMBER', 'BUILD_NUMBER', 'GITHUB_ACTIONS'))
-    context.timeout_multiplier = 2 if context.ci else 1
+    context.timeout_multiplier = 5 if context.ci else 1  # MacOS sometimes is VERY slow
     context.pctl = PatroniPoolController(context)
     context.dcs_ctl = context.pctl.known_dcs[context.pctl.dcs](context)
     context.dcs_ctl.start()

features/ignored_slots.feature

@@ -0,0 +1,61 @@
+Feature: ignored slots
+  Scenario: check ignored slots aren't removed on failover/switchover
+    Given I start postgres1
+    Then postgres1 is a leader after 10 seconds
+    And there is a non empty initialize key in DCS after 15 seconds
+    When I issue a PATCH request to http://127.0.0.1:8009/config with {"loop_wait": 2, "ignore_slots": [{"name": "unmanaged_slot_0", "database": "postgres", "plugin": "test_decoding", "type": "logical"}, {"name": "unmanaged_slot_1", "database": "postgres", "plugin": "test_decoding"}, {"name": "unmanaged_slot_2", "database": "postgres"}, {"name": "unmanaged_slot_3"}], "postgresql": {"parameters": {"wal_level": "logical"}}}
+    Then I receive a response code 200
+    And Response on GET http://127.0.0.1:8009/config contains ignore_slots after 10 seconds
+    # Make sure the wal_level has been changed.
+    When I shut down postgres1
+    And I start postgres1
+    Then postgres1 is a leader after 10 seconds
+    And "members/postgres1" key in DCS has role=master after 3 seconds
+    # Make sure Patroni has finished telling Postgres it should be accepting writes.
+    And postgres1 role is the primary after 20 seconds
+    # 1. Create our test logical replication slot.
+    # Test that ny subset of attributes in the ignore slots matcher is enough to match a slot
+    # by using 3 different slots.
+    When I create a logical replication slot unmanaged_slot_0 on postgres1 with the test_decoding plugin
+    And I create a logical replication slot unmanaged_slot_1 on postgres1 with the test_decoding plugin
+    And I create a logical replication slot unmanaged_slot_2 on postgres1 with the test_decoding plugin
+    And I create a logical replication slot unmanaged_slot_3 on postgres1 with the test_decoding plugin
+    And I create a logical replication slot dummy_slot on postgres1 with the test_decoding plugin
+    # It seems like it'd be obvious that these slots exist since we just created them,
+    # but Patroni can actually end up dropping them almost immediately, so it's helpful
+    # to verify they exist before we begin testing whether they persist through failover
+    # cycles.
+    Then postgres1 has a logical replication slot named unmanaged_slot_0 with the test_decoding plugin
+    And postgres1 has a logical replication slot named unmanaged_slot_1 with the test_decoding plugin
+    And postgres1 has a logical replication slot named unmanaged_slot_2 with the test_decoding plugin
+    And postgres1 has a logical replication slot named unmanaged_slot_3 with the test_decoding plugin
+
+    When I start postgres0
+    Then "members/postgres0" key in DCS has role=replica after 3 seconds
+    And postgres0 role is the secondary after 20 seconds
+    # Verify that the replica has advanced beyond the point in the WAL
+    # where we created the replication slot so that on the next failover
+    # cycle we don't accidentally rewind to before the slot creation.
+    And replication works from postgres1 to postgres0 after 20 seconds
+    When I shut down postgres1
+    Then "members/postgres0" key in DCS has role=master after 3 seconds
+
+    # 2. After a failover the server (now a replica) still has the slot.
+    When I start postgres1
+    Then postgres1 role is the secondary after 20 seconds
+    And "members/postgres1" key in DCS has role=replica after 3 seconds
+    # give Patroni time to sync replication slots
+    And I sleep for 2 seconds
+    And postgres1 has a logical replication slot named unmanaged_slot_0 with the test_decoding plugin
+    And postgres1 has a logical replication slot named unmanaged_slot_1 with the test_decoding plugin
+    And postgres1 has a logical replication slot named unmanaged_slot_2 with the test_decoding plugin
+    And postgres1 has a logical replication slot named unmanaged_slot_3 with the test_decoding plugin
+    And postgres1 does not have a logical replication slot named dummy_slot
+
+    # 3. After a failover the server (now a master) still has the slot.
+    When I shut down postgres0
+    Then "members/postgres1" key in DCS has role=master after 3 seconds
+    And postgres1 has a logical replication slot named unmanaged_slot_0 with the test_decoding plugin
+    And postgres1 has a logical replication slot named unmanaged_slot_1 with the test_decoding plugin
+    And postgres1 has a logical replication slot named unmanaged_slot_2 with the test_decoding plugin
+    And postgres1 has a logical replication slot named unmanaged_slot_3 with the test_decoding plugin

features/patroni_api.feature

@@ -8,6 +8,8 @@ Scenario: check API requests on a stand-alone server
 	Then I receive a response code 200
 	And I receive a response state running
 	And I receive a response role master
+	When I issue a GET request to http://127.0.0.1:8008/standby_leader
+	Then I receive a response code 503
 	When I issue a GET request to http://127.0.0.1:8008/health
 	Then I receive a response code 200
 	When I issue a GET request to http://127.0.0.1:8008/replica
@@ -28,10 +30,7 @@ Scenario: check API requests on a stand-alone server
 	And I receive a response text "Failover could be performed only to a specific candidate"
 
 Scenario: check local configuration reload
-	Given I issue an empty POST request to http://127.0.0.1:8008/reload
+	Given I add tag new_tag new_value to postgres0 config
-	Then I receive a response code 200
-	And I receive a response text nothing changed
-	When I add tag new_tag new_value to postgres0 config
 	And I issue an empty POST request to http://127.0.0.1:8008/reload
 	Then I receive a response code 202
 
@@ -46,11 +45,31 @@ Scenario: check dynamic configuration change via DCS
 	When I issue a GET request to http://127.0.0.1:8008/patroni
 	Then I receive a response code 200
 	And I receive a response tags {'new_tag': 'new_value'}
+	And I sleep for 4 seconds
+
+Scenario: check the scheduled restart
+	Given I issue a PATCH request to http://127.0.0.1:8008/config with {"postgresql": {"parameters": {"superuser_reserved_connections": "6"}}}
+	Then I receive a response code 200
+		And Response on GET http://127.0.0.1:8008/patroni contains pending_restart after 5 seconds
+	Given I issue a scheduled restart at http://127.0.0.1:8008 in 3 seconds with {"role": "replica"}
+	Then I receive a response code 202
+		And I sleep for 4 seconds
+		And Response on GET http://127.0.0.1:8008/patroni contains pending_restart after 10 seconds
+	Given I issue a scheduled restart at http://127.0.0.1:8008 in 3 seconds with {"restart_pending": "True"}
+	Then I receive a response code 202
+		And Response on GET http://127.0.0.1:8008/patroni does not contain pending_restart after 10 seconds
+		And postgres0 role is the primary after 10 seconds
 
 Scenario: check API requests for the primary-replica pair in the pause mode
-	Given I run patronictl.py pause batman
+	Given I start postgres1
+	Then replication works from postgres0 to postgres1 after 20 seconds
+	When I run patronictl.py pause batman
+	Then I receive a response returncode 0
+	When I kill postmaster on postgres1
+	And I issue a GET request to http://127.0.0.1:8009/replica
+	Then I receive a response code 503
+	When I run patronictl.py restart batman postgres1 --force
 	Then I receive a response returncode 0
-	When I start postgres1
 	Then replication works from postgres0 to postgres1 after 20 seconds
 	When I issue a GET request to http://127.0.0.1:8009/replica
 	Then I receive a response code 200
@@ -73,6 +92,7 @@ Scenario: check the switchover via the API in the pause mode
 	And postgres1 role is the primary after 10 seconds
 	And postgres0 role is the secondary after 10 seconds
 	And replication works from postgres1 to postgres0 after 20 seconds
+	And "members/postgres0" key in DCS has state=running after 10 seconds
 	When I issue a GET request to http://127.0.0.1:8008/master
 	Then I receive a response code 503
 	When I issue a GET request to http://127.0.0.1:8008/replica
@@ -94,6 +114,7 @@ Scenario: check the scheduled switchover
 	And postgres0 role is the primary after 10 seconds
 	And postgres1 role is the secondary after 10 seconds
 	And replication works from postgres0 to postgres1 after 25 seconds
+	And "members/postgres1" key in DCS has state=running after 10 seconds
 	When I issue a GET request to http://127.0.0.1:8008/master
 	Then I receive a response code 200
 	When I issue a GET request to http://127.0.0.1:8008/replica
@@ -102,16 +123,3 @@ Scenario: check the scheduled switchover
 	Then I receive a response code 503
 	When I issue a GET request to http://127.0.0.1:8009/replica
 	Then I receive a response code 200
-
-Scenario: check the scheduled restart
-	Given I issue a PATCH request to http://127.0.0.1:8008/config with {"postgresql": {"parameters": {"superuser_reserved_connections": "6"}}}
-	Then I receive a response code 200
-		And Response on GET http://127.0.0.1:8008/patroni contains pending_restart after 5 seconds
-	Given I issue a scheduled restart at http://127.0.0.1:8008 in 3 seconds with {"role": "replica"}
-	Then I receive a response code 202
-		And I sleep for 4 seconds
-		And Response on GET http://127.0.0.1:8008/patroni contains pending_restart after 10 seconds
-	Given I issue a scheduled restart at http://127.0.0.1:8008 in 3 seconds with {"restart_pending": "True"}
-	Then I receive a response code 202
-		And Response on GET http://127.0.0.1:8008/patroni does not contain pending_restart after 10 seconds
-

features/standby_cluster.feature

@@ -1,8 +1,8 @@
 Feature: standby cluster
-  Scenario: check permanent logical slots are preserved on failover/switchover
+  Scenario: prepare the cluster with logical slots
     Given I start postgres1
     Then postgres1 is a leader after 10 seconds
-    And I sleep for 3 seconds
+    And there is a non empty initialize key in DCS after 15 seconds
     When I issue a PATCH request to http://127.0.0.1:8009/config with {"loop_wait": 2, "slots": {"pm_1": {"type": "physical"}}, "postgresql": {"parameters": {"wal_level": "logical"}}}
     Then I receive a response code 200
     And Response on GET http://127.0.0.1:8009/config contains slots after 10 seconds
@@ -10,37 +10,49 @@ Feature: standby cluster
     When I issue a PATCH request to http://127.0.0.1:8009/config with {"slots": {"test_logical": {"type": "logical", "database": "postgres", "plugin": "test_decoding"}}}
     Then I receive a response code 200
     And I do a backup of postgres1
-    When I start postgres0 with callback configured
+    When I start postgres0
     Then "members/postgres0" key in DCS has state=running after 10 seconds
     And replication works from postgres1 to postgres0 after 15 seconds
+
+  @skip
+  Scenario: check permanent logical slots are synced to the replica
+    Given I run patronictl.py restart batman postgres1 --force
+    Then Logical slot test_logical is in sync between postgres0 and postgres1 after 10 seconds
+    When I add the table replicate_me to postgres1
+    And I get all changes from logical slot test_logical on postgres1
+    Then Logical slot test_logical is in sync between postgres0 and postgres1 after 10 seconds
+
+  Scenario: Detach exiting node from the cluster
     When I shut down postgres1
     Then postgres0 is a leader after 10 seconds
     And "members/postgres0" key in DCS has role=master after 3 seconds
     When I issue a GET request to http://127.0.0.1:8008/
     Then I receive a response code 200
-    And there is a label with "test_logical" in postgres0 data directory
 
   Scenario: check replication of a single table in a standby cluster
     Given I start postgres1 in a standby cluster batman1 as a clone of postgres0
     Then postgres1 is a leader of batman1 after 10 seconds
     When I add the table foo to postgres0
     Then table foo is present on postgres1 after 20 seconds
+    And I sleep for 3 seconds
     When I issue a GET request to http://127.0.0.1:8009/master
-    Then I receive a response code 200
+    Then I receive a response code 503
     When I issue a GET request to http://127.0.0.1:8009/standby_leader
     Then I receive a response code 200
     And I receive a response role standby_leader
-    And there is a postgres1_cb.log with "on_start replica batman1\non_role_change standby_leader batman1" in postgres1 data directory
+    And there is a postgres1_cb.log with "on_role_change standby_leader batman1" in postgres1 data directory
     When I start postgres2 in a cluster batman1
     Then postgres2 role is the replica after 24 seconds
     And table foo is present on postgres2 after 20 seconds
+    And postgres1 does not have a logical replication slot named test_logical
 
   Scenario: check failover
     When I kill postgres1
     And I kill postmaster on postgres1
     Then postgres2 is replicating from postgres0 after 32 seconds
     When I issue a GET request to http://127.0.0.1:8010/master
-    Then I receive a response code 200
+    Then I receive a response code 503
+    And I sleep for 3 seconds
     When I issue a GET request to http://127.0.0.1:8010/standby_leader
     Then I receive a response code 200
     And I receive a response role standby_leader

features/steps/basic_replication.py

@@ -33,6 +33,37 @@ def add_table(context, table_name, pg_name):
         assert False, "Error creating table {0} on {1}: {2}".format(table_name, pg_name, e)
 
 
+@step('I {action:w} wal replay on {pg_name:w}')
+def toggle_wal_replay(context, action, pg_name):
+    # pause or resume the wal replay process
+    try:
+       version = context.pctl.query(pg_name, "select pg_catalog.pg_read_file('PG_VERSION', 0, 2)").fetchone()
+       wal = version and version[0] and int(version[0].split('.')[0]) < 10 and "xlog" or "wal"
+       context.pctl.query(pg_name, "SELECT pg_{0}_replay_{1}()".format(wal, action))
+    except pg.Error as e:
+        assert False, "Error during {0} wal recovery on {1}: {2}".format(action, pg_name, e)
+
+
+@step('I {action:w} table on {pg_name:w}')
+def crdr_mytest(context, action, pg_name):
+    try:
+       if (action == "create"):
+           context.pctl.query(pg_name, "create table if not exists mytest(id Numeric)")
+       else:
+           context.pctl.query(pg_name, "drop table if exists mytest")
+    except pg.Error as e:
+        assert False, "Error {0} table mytest on {1}: {2}".format(action, pg_name, e)
+
+
+@step('I load data on {pg_name:w}')
+def initiate_load(context, pg_name):
+    # perform dummy load
+    try:
+       context.pctl.query(pg_name, "begin; insert into mytest select r::numeric from generate_series(1, 350000) r; commit;")
+    except pg.Error as e:
+        assert False, "Error loading test data on {0}: {1}".format(pg_name, e)
+
+
 @then('Table {table_name:w} is present on {pg_name:w} after {max_replication_delay:d} seconds')
 def table_is_present_on(context, table_name, pg_name, max_replication_delay):
     max_replication_delay *= context.timeout_multiplier

features/steps/cascading_replication.py

@@ -12,8 +12,10 @@ def start_patroni_with_a_name_value_tag(context, name, tag_name, tag_value):
 @then('There is a {label} with "{content}" in {name:w} data directory')
 def check_label(context, label, content, name):
     label = context.pctl.read_label(name, label)
+    if label is None:
+        label = ""
     label = label.replace('\n', '\\n')
-    assert label == content, "{0} is not equal to {1}".format(label, content)
+    assert content in label, "\"{0}\" doesn't contain {1}".format(label, content)
 
 
 @step('I create label with "{content:w}" in {name:w} data directory')
@@ -21,16 +23,33 @@ def write_label(context, content, name):
     context.pctl.write_label(name, content)
 
 
-@step('"{name}" key in DCS has {key:w}={value:w} after {time_limit:d} seconds')
+@step('"{name}" key in DCS has {key:w}={value} after {time_limit:d} seconds')
 def check_member(context, name, key, value, time_limit):
     time_limit *= context.timeout_multiplier
     max_time = time.time() + int(time_limit)
+    dcs_value = None
     while time.time() < max_time:
         try:
             response = json.loads(context.dcs_ctl.query(name))
-            if response.get(key) == value:
+            dcs_value = response.get(key)
+            if dcs_value == value:
                 return
         except Exception:
             pass
         time.sleep(1)
-    assert False, "{0} does not have {1}={2} in dcs after {3} seconds".format(name, key, value, time_limit)
+    assert False, "{0} does not have {1}={2} (found {3}) in dcs after {4} seconds".format(name, key, value,
+                                                                                          dcs_value, time_limit)
+
+
+@step('there is a non empty {key:w} key in DCS after {time_limit:d} seconds')
+def check_initialize(context, key, time_limit):
+    time_limit *= context.timeout_multiplier
+    max_time = time.time() + int(time_limit)
+    while time.time() < max_time:
+        try:
+            if context.dcs_ctl.query(key):
+                return
+        except Exception:
+            pass
+        time.sleep(1)
+    assert False, "There is no {0} in dcs after {1} seconds".format(key, time_limit)

features/steps/patroni_api.py

@@ -1,8 +1,6 @@
-import base64
 import json
 import os
 import parse
-import requests
 import shlex
 import subprocess
 import sys
@@ -12,8 +10,10 @@ import yaml
 from behave import register_type, step, then
 from dateutil import tz
 from datetime import datetime, timedelta
+from patroni.request import PatroniRequest
 
 tzutc = tz.tzutc()
+request_executor = PatroniRequest({'ctl': {'auth': 'username:password'}})
 
 
 @parse.with_pattern(r'https?://(?:\w|\.|:|/)+')
@@ -45,9 +45,9 @@ def sleep_for_n_seconds(context, value):
 
 
 def _set_response(context, response):
-    context.status_code = response.status_code
+    context.status_code = response.status
-    data = response.content.decode('utf-8')
+    data = response.data.decode('utf-8')
-    ct = response.headers.get('content-type', '')
+    ct = response.getheader('content-type', '')
     if ct.startswith('application/json') or\
             ct.startswith('text/yaml') or\
             ct.startswith('text/x-yaml') or\
@@ -63,13 +63,7 @@ def _set_response(context, response):
 
 @step('I issue a GET request to {url:url}')
 def do_get(context, url):
-    try:
+    do_request(context, 'GET', url, None)
-        r = requests.get(url)
-    except requests.exceptions.RequestException:
-        context.status_code = None
-        context.response = None
-    else:
-        _set_response(context, r)
 
 
 @step('I issue an empty POST request to {url:url}')
@@ -79,17 +73,11 @@ def do_post_empty(context, url):
 
 @step('I issue a {request_method:w} request to {url:url} with {data}')
 def do_request(context, request_method, url, data):
-    data = data and json.loads(data) or {}
+    data = data and json.loads(data)
-    headers = {'Authorization': 'Basic ' + base64.b64encode('username:password'.encode('utf-8')).decode('utf-8'),
-               'Content-Type': 'application/json'}
     try:
-        if request_method == 'PATCH':
+        r = request_executor.request(request_method, url, data)
-            r = requests.patch(url, headers=headers, json=data)
+    except Exception:
-        else:
+        context.status_code = context.response = None
-            r = requests.post(url, headers=headers, json=data)
-    except requests.exceptions.RequestException:
-        context.status_code = None
-        context.response = None
     else:
         _set_response(context, r)
 
@@ -149,8 +137,8 @@ def add_tag_to_config(context, tag, value, pg_name):
 def check_http_response(context, url, value, timeout, negate=False):
     timeout *= context.timeout_multiplier
     for _ in range(int(timeout)):
-        r = requests.get(url)
+        r = request_executor.request('GET', url)
-        if (value in r.content.decode('utf-8')) != negate:
+        if (value in r.data.decode('utf-8')) != negate:
             break
         time.sleep(1)
     else:

features/steps/slots.py

@@ -0,0 +1,60 @@
+import time
+import psycopg2
+
+from behave import step, then
+
+
+@step('I create a logical replication slot {slot_name} on {pg_name:w} with the {plugin:w} plugin')
+def create_logical_replication_slot(context, slot_name, pg_name, plugin):
+    try:
+        output = context.pctl.query(pg_name, ("SELECT pg_create_logical_replication_slot('{0}', '{1}'),"
+                                              " current_database()").format(slot_name, plugin))
+        print(output.fetchone())
+    except psycopg2.Error as e:
+        print(e)
+        assert False, "Error creating slot {0} on {1} with plugin {2}".format(slot_name, pg_name, plugin)
+
+
+@then('{pg_name:w} has a logical replication slot named {slot_name} with the {plugin:w} plugin')
+def has_logical_replication_slot(context, pg_name, slot_name, plugin):
+    try:
+        row = context.pctl.query(pg_name, ("SELECT slot_type, plugin FROM pg_replication_slots"
+                                           " WHERE slot_name = '{0}'").format(slot_name)).fetchone()
+        assert row, "Couldn't find replication slot named {0}".format(slot_name)
+        assert row[0] == "logical", "Found replication slot named {0} but wasn't a logical slot".format(slot_name)
+        assert row[1] == plugin, ("Found replication slot named {0} but was using plugin "
+                                  "{1} rather than {2}").format(slot_name, row[1], plugin)
+    except psycopg2.Error:
+        assert False, "Error looking for slot {0} on {1} with plugin {2}".format(slot_name, pg_name, plugin)
+
+
+@then('{pg_name:w} does not have a logical replication slot named {slot_name}')
+def does_not_have_logical_replication_slot(context, pg_name, slot_name):
+    try:
+        row = context.pctl.query(pg_name, ("SELECT 1 FROM pg_replication_slots"
+                                           " WHERE slot_name = '{0}'").format(slot_name)).fetchone()
+        assert not row, "Found unexpected replication slot named {0}".format(slot_name)
+    except psycopg2.Error:
+        assert False, "Error looking for slot {0} on {1}".format(slot_name, pg_name)
+
+
+@step('Logical slot {slot_name:w} is in sync between {pg_name1:w} and {pg_name2:w} after {time_limit:d} seconds')
+def logical_slots_in_sync(context, slot_name, pg_name1, pg_name2, time_limit):
+    time_limit *= context.timeout_multiplier
+    max_time = time.time() + int(time_limit)
+    while time.time() < max_time:
+        try:
+            query = "SELECT confirmed_flush_lsn FROM pg_replication_slots WHERE slot_name = '{0}'".format(slot_name)
+            slot1 = context.pctl.query(pg_name1, query).fetchone()
+            slot2 = context.pctl.query(pg_name2, query).fetchone()
+            if slot1[0] == slot2[0]:
+                return
+        except Exception:
+            pass
+        time.sleep(1)
+    assert False, "Logical slot {0} is not in sync between {1} and {2}".format(slot_name, pg_name1, pg_name2)
+
+
+@step('I get all changes from logical slot {slot_name:w} on {pg_name:w}')
+def logical_slot_get_changes(context, slot_name, pg_name):
+    context.pctl.query(pg_name, "SELECT * FROM pg_logical_slot_get_changes('{0}', NULL, NULL)".format(slot_name))

features/steps/standby_cluster.py

@@ -1,4 +1,5 @@
 import os
+import sys
 import time
 
 from behave import step
@@ -9,18 +10,8 @@ SELECT * FROM pg_catalog.pg_stat_replication
 WHERE application_name = '{0}'
 """
 
-callback = "bash -c 'echo \"${*: -3:1} ${*: -2:1} ${*: -1:1}\" >> data/$1/$1_cb.log' -- "
+executable = sys.executable if os.name != 'nt' else sys.executable.replace('\\', '/')
-
+callback = executable + " features/callback2.py "
-
-@step('I start {name:w} with callback configured')
-def start_patroni_with_callbacks(context, name):
-    return context.pctl.start(name, custom_config={
-        "postgresql": {
-            "callbacks": {
-                "on_role_change": "features/callback.sh"
-            }
-        }
-    })
 
 
 @step('I start {name:w} in a cluster {cluster_name:w}')
@@ -30,8 +21,8 @@ def start_patroni(context, name, cluster_name):
         "postgresql": {
             "callbacks": {c: callback + name for c in ('on_start', 'on_stop', 'on_restart', 'on_role_change')},
             "backup_restore": {
-                "command": "features/backup_restore.sh --sourcedir=" + os.path.join(context.pctl.patroni_path,
+                "command": (executable + " features/backup_restore.py --sourcedir=" +
-                                                                                    "data/basebackup")}
+                            os.path.join(context.pctl.patroni_path, 'data', 'basebackup'))}
         }
     })
 
@@ -53,7 +44,8 @@ def start_patroni_standby_cluster(context, name, cluster_name, name2):
                     "port": port,
                     "primary_slot_name": "pm_1",
                     "create_replica_methods": ["backup_restore", "basebackup"]
-                }
+                },
+                "postgresql": {"parameters": {"wal_level": "logical"}}
             }
         },
         "postgresql": {

features/steps/watchdog.py

@@ -44,31 +44,6 @@ def watchdog_was_triggered(context, name, timeout):
     assert False
 
 
-@then('{name:w} watchdog was not triggered')
-def watchdog_was_not_triggered(context, name):
-    assert not context.pctl.get_watchdog(name).was_triggered
-
-
-@step('{name:w} checkpoint takes {timeout:d} seconds')
-def checkpoint_hang(context, name, timeout):
-    assert context.pctl.checkpoint_hang(name, timeout)
-
-
 @step('{name:w} hangs for {timeout:d} seconds')
 def patroni_hang(context, name, timeout):
     return context.pctl.patroni_hang(name, timeout)
-
-
-@step('I terminate {name:w} user processes')
-def terminate_backends(context, name):
-    return context.pctl.terminate_backends(name)
-
-
-@step('Sleep for {timeout:d} seconds')
-def dcs_connection_lost(context, timeout):
-    time.sleep(timeout)
-
-
-@then('{name:w} database is running')
-def database_is_running(context, name):
-    assert context.pctl.database_is_running(name)

kubernetes/entrypoint.sh

@@ -33,7 +33,5 @@ postgresql:
 __EOF__
 
 unset PATRONI_SUPERUSER_PASSWORD PATRONI_REPLICATION_PASSWORD
-export KUBERNETES_NAMESPACE=$PATRONI_KUBERNETES_NAMESPACE
-export POD_NAME=$PATRONI_NAME
 
-exec /usr/bin/python3 /usr/local/bin/patroni /home/postgres/patroni.yml
+exec /usr/bin/python3 /usr/local/bin/patroni /home/postgres/patroni.yml

kubernetes/openshift-example/templates/template_patroni_ephemeral.yml

@@ -118,6 +118,8 @@ objects:
               fieldRef:
                 apiVersion: v1
                 fieldPath: metadata.namespace
+          - name: PATRONI_KUBERNETES_BYPASS_API_SERVICE
+            value: 'true'
           - name: PATRONI_KUBERNETES_LABELS
             value: '{application: ${APPLICATION_NAME}, cluster-name: ${PATRONI_CLUSTER_NAME}}'
           - name: PATRONI_SUPERUSER_USERNAME
@@ -152,6 +154,16 @@ objects:
           image: docker-registry.default.svc:5000/${NAMESPACE}/patroni:latest
           imagePullPolicy: IfNotPresent
           name: ${APPLICATION_NAME}
+          readinessProbe:
+            httpGet:
+              scheme: HTTP
+              path: /readiness
+              port: 8008
+            initialDelaySeconds: 3
+            periodSeconds: 10
+            timeoutSeconds: 5
+            successThreshold: 1
+            failureThreshold: 3
           ports:
           - containerPort: 8008
             protocol: TCP
@@ -240,6 +252,34 @@ objects:
   subjects:
   - kind: ServiceAccount
     name: ${SERVICE_ACCOUNT}
+# Following privileges are only required if deployed not in the "default"
+# namespace and you want Patroni to bypass kubernetes service
+# (PATRONI_KUBERNETES_BYPASS_API_SERVICE=true)
+- apiVersion: rbac.authorization.k8s.io/v1
+  kind: ClusterRole
+  metadata:
+    name: patroni-k8s-ep-access
+  rules:
+  - apiGroups:
+    - ""
+    resources:
+    - endpoints
+    resourceNames:
+    - kubernetes
+    verbs:
+    - get
+- apiVersion: rbac.authorization.k8s.io/v1
+  kind: ClusterRoleBinding
+  metadata:
+    name: ${NAMESPACE}-${SERVICE_ACCOUNT}-k8s-ep-access
+  roleRef:
+    apiGroup: rbac.authorization.k8s.io
+    kind: ClusterRole
+    name: patroni-k8s-ep-access
+  subjects:
+  - kind: ServiceAccount
+    name: ${SERVICE_ACCOUNT}
+    namespace: ${NAMESPACE}
 parameters:
 - description: The name of the application for labelling all artifacts.
   displayName: Application Name

kubernetes/openshift-example/templates/template_patroni_persistent.yaml

@@ -5,11 +5,9 @@ metadata:
   annotations:
     description: |-
       Patroni Postgresql database cluster, with persistent storage.
-
-      WARNING: Any data stored will be lost upon pod destruction. Only use this template for testing.
     iconClass: icon-postgresql
     openshift.io/display-name: Patroni Postgresql (Persistent)
-    openshift.io/long-description: This template deploys a a patroni postgresql HA cluster without persistent storage.
+    openshift.io/long-description: This template deploys a a patroni postgresql HA cluster with persistent storage.
     tags: postgresql
 objects:
 - apiVersion: v1
@@ -106,6 +104,20 @@ objects:
           application: ${APPLICATION_NAME}
           cluster-name: ${PATRONI_CLUSTER_NAME}
       spec:
+        initContainers:
+        - command:
+          - sh
+          - -c 
+          - "mkdir -p /home/postgres/pgdata/pgroot/data && chmod 0700 /home/postgres/pgdata/pgroot/data"
+          image: docker-registry.default.svc:5000/${NAMESPACE}/patroni:latest
+          imagePullPolicy: IfNotPresent
+          name: fix-perms
+          resources: {}
+          terminationMessagePath: /dev/termination-log
+          terminationMessagePolicy: File
+          volumeMounts:
+          - mountPath: /home/postgres/pgdata
+            name: ${APPLICATION_NAME}
         containers:
         - env:
           - name: PATRONI_KUBERNETES_POD_IP
@@ -118,6 +130,8 @@ objects:
               fieldRef:
                 apiVersion: v1
                 fieldPath: metadata.namespace
+          - name: PATRONI_KUBERNETES_BYPASS_API_SERVICE
+            value: 'true'
           - name: PATRONI_KUBERNETES_LABELS
             value: '{application: ${APPLICATION_NAME}, cluster-name: ${PATRONI_CLUSTER_NAME}}'
           - name: PATRONI_SUPERUSER_USERNAME
@@ -152,6 +166,16 @@ objects:
           image: docker-registry.default.svc:5000/${NAMESPACE}/patroni:latest
           imagePullPolicy: IfNotPresent
           name: ${APPLICATION_NAME}
+          readinessProbe:
+            httpGet:
+              scheme: HTTP
+              path: /readiness
+              port: 8008
+            initialDelaySeconds: 3
+            periodSeconds: 10
+            timeoutSeconds: 5
+            successThreshold: 1
+            failureThreshold: 3
           ports:
           - containerPort: 8008
             protocol: TCP
@@ -252,6 +276,34 @@ objects:
   subjects:
   - kind: ServiceAccount
     name: ${SERVICE_ACCOUNT}
+# Following privileges are only required if deployed not in the "default"
+# namespace and you want Patroni to bypass kubernetes service
+# (PATRONI_KUBERNETES_BYPASS_API_SERVICE=true)
+- apiVersion: rbac.authorization.k8s.io/v1
+  kind: ClusterRole
+  metadata:
+    name: patroni-k8s-ep-access
+  rules:
+  - apiGroups:
+    - ""
+    resources:
+    - endpoints
+    resourceNames:
+    - kubernetes
+    verbs:
+    - get
+- apiVersion: rbac.authorization.k8s.io/v1
+  kind: ClusterRoleBinding
+  metadata:
+    name: ${NAMESPACE}-${SERVICE_ACCOUNT}-k8s-ep-access
+  roleRef:
+    apiGroup: rbac.authorization.k8s.io
+    kind: ClusterRole
+    name: patroni-k8s-ep-access
+  subjects:
+  - kind: ServiceAccount
+    name: ${SERVICE_ACCOUNT}
+    namespace: ${NAMESPACE}
 parameters:
 - description: The name of the application for labelling all artifacts.
   displayName: Application Name
@@ -300,4 +352,4 @@ parameters:
 - description: The size of the persistent volume to create. 
   displayName: Persistent Volume Size
   name: PVC_SIZE
-  value: 5Gi
+  value: 5Gi

kubernetes/patroni_k8s.yaml

@@ -10,7 +10,7 @@ spec:
   clusterIP: None
 
 ---
-apiVersion: apps/v1beta1
+apiVersion: apps/v1
 kind: StatefulSet
 metadata:
   name: &cluster_name patronidemo
@@ -31,6 +31,16 @@ spec:
       - name: *cluster_name
         image: patroni  # docker build -t patroni .
         imagePullPolicy: IfNotPresent
+        readinessProbe:
+          httpGet:
+            scheme: HTTP
+            path: /readiness
+            port: 8008
+          initialDelaySeconds: 3
+          periodSeconds: 10
+          timeoutSeconds: 5
+          successThreshold: 1
+          failureThreshold: 3
         ports:
         - containerPort: 8008
           protocol: TCP
@@ -48,6 +58,8 @@ spec:
           valueFrom:
             fieldRef:
               fieldPath: metadata.namespace
+        - name: PATRONI_KUBERNETES_BYPASS_API_SERVICE
+          value: 'true'
         - name: PATRONI_KUBERNETES_USE_ENDPOINTS
           value: 'true'
         - name: PATRONI_KUBERNETES_LABELS
@@ -123,6 +135,25 @@ spec:
   - port: 5432
     targetPort: 5432
 
+---
+apiVersion: v1
+kind: Service
+metadata:
+  name: patronidemo-repl
+  labels:
+    application: patroni
+    cluster-name:  &cluster_name patronidemo
+    role: replica
+spec:
+  type: ClusterIP
+  selector:
+    application: patroni
+    cluster-name: *cluster_name
+    role: replica
+  ports:
+  - port: 5432
+    targetPort: 5432
+
 ---
 apiVersion: v1
 kind: Secret
@@ -210,3 +241,37 @@ roleRef:
 subjects:
 - kind: ServiceAccount
   name: patronidemo
+
+# Following privileges are only required if deployed not in the "default"
+# namespace and you want Patroni to bypass kubernetes service
+# (PATRONI_KUBERNETES_BYPASS_API_SERVICE=true)
+---
+apiVersion: rbac.authorization.k8s.io/v1
+kind: ClusterRole
+metadata:
+  name: patroni-k8s-ep-access
+rules:
+- apiGroups:
+  - ""
+  resources:
+  - endpoints
+  resourceNames:
+  - kubernetes
+  verbs:
+  - get
+
+---
+apiVersion: rbac.authorization.k8s.io/v1
+kind: ClusterRoleBinding
+metadata:
+  name: patroni-k8s-ep-access
+roleRef:
+  apiGroup: rbac.authorization.k8s.io
+  kind: ClusterRole
+  name: patroni-k8s-ep-access
+subjects:
+- kind: ServiceAccount
+  name: patronidemo
+# The namespace must be specified explicitly.
+# If deploying to the different namespace you have to change it.
+  namespace: default

patroni.py

@@ -1,14 +1,6 @@
 #!/usr/bin/env python
 from patroni import main
 
-import os
 
 if __name__ == '__main__':
-
-    if os.getenv("PATRONI_DEBUG_MODE"):
-        # XXX Visual Code specific https://github.com/microsoft/ptvsd/issues/1443
-        # create processes by spawning new Python interpreters instead of forking the current one 
-        import multiprocessing
-        multiprocessing.set_start_method('spawn', True)
-
     main()

patroni/__init__.py

@@ -4,35 +4,36 @@ import signal
 import sys
 import time
 
+from .daemon import AbstractPatroniDaemon, abstract_main
+from .version import __version__
+
 logger = logging.getLogger(__name__)
 
+PATRONI_ENV_PREFIX = 'PATRONI_'
+KUBERNETES_ENV_PREFIX = 'KUBERNETES_'
 
-class Patroni(object):
 
-    def __init__(self):
+class Patroni(AbstractPatroniDaemon):
+
+    def __init__(self, config):
         from patroni.api import RestApiServer
-        from patroni.config import Config
         from patroni.dcs import get_dcs
         from patroni.ha import Ha
-        from patroni.log import PatroniLogger
         from patroni.postgresql import Postgresql
-        from patroni.version import __version__
+        from patroni.request import PatroniRequest
         from patroni.watchdog import Watchdog
 
-        self.setup_signal_handlers()
+        super(Patroni, self).__init__(config)
 
         self.version = __version__
-        self.logger = PatroniLogger()
-        self.config = Config()
-        self.logger.reload_config(self.config.get('log', {}))
         self.dcs = get_dcs(self.config)
         self.watchdog = Watchdog(self.config)
         self.load_dynamic_configuration()
 
         self.postgresql = Postgresql(self.config['postgresql'])
         self.api = RestApiServer(self, self.config['restapi'])
+        self.request = PatroniRequest(self.config, True)
         self.ha = Ha(self)
-        self.is_in_debug_mode = bool(os.getenv("PATRONI_DEBUG_MODE"))
 
         self.tags = self.get_tags()
         self.next_run = time.time()
@@ -67,14 +68,17 @@ class Patroni(object):
     def nosync(self):
         return bool(self.tags.get('nosync', False))
 
-    def reload_config(self):
+    def reload_config(self, sighup=False, local=False):
         try:
-            self.tags = self.get_tags()
+            super(Patroni, self).reload_config(sighup, local)
-            self.logger.reload_config(self.config.get('log', {}))
+            if local:
-            self.dcs.reload_config(self.config)
+                self.tags = self.get_tags()
+                self.request.reload_config(self.config)
+            if local or self.api.reload_local_certificate():
+                self.api.reload_config(self.config['restapi'])
             self.watchdog.reload_config(self.config)
-            self.api.reload_config(self.config['restapi'])
+            self.postgresql.reload_config(self.config['postgresql'], sighup)
-            self.postgresql.reload_config(self.config['postgresql'])
+            self.dcs.reload_config(self.config)
         except Exception:
             logger.exception('Failed to reload config_file=%s', self.config.config_file)
 
@@ -82,15 +86,6 @@ class Patroni(object):
     def replicatefrom(self):
         return self.tags.get('replicatefrom')
 
-    def sighup_handler(self, *args):
-        self._received_sighup = True
-
-    def sigterm_handler(self, *args):
-        with self._sigterm_lock:
-            if not self._received_sigterm:
-                self._received_sigterm = True
-                sys.exit()
-
     @property
     def noloadbalance(self):
         return bool(self.tags.get('noloadbalance', False))
@@ -103,67 +98,45 @@ class Patroni(object):
             self.next_run = current_time
             # Release the GIL so we don't starve anyone waiting on async_executor lock
             time.sleep(0.001)
-            # Warn user that Patroni is not keeping up or runs in debug
+            # Warn user that Patroni is not keeping up
-            msg = "Patroni runs in the debug mode: keys' TTL is infinite, loop wait disabled" if self.is_in_debug_mode else "Loop time exceeded, rescheduling immediately."
+            logger.warning("Loop time exceeded, rescheduling immediately.")
-            logger.warning(msg)
         elif self.ha.watch(nap_time):
             self.next_run = time.time()
 
-    @property
-    def received_sigterm(self):
-        with self._sigterm_lock:
-            return self._received_sigterm
-
     def run(self):
         self.api.start()
         self.next_run = time.time()
+        super(Patroni, self).run()
 
-        while not self.received_sigterm:
+    def _run_cycle(self):
-            if self._received_sighup:
+        logger.info(self.ha.run_cycle())
-                self._received_sighup = False
-                if self.config.reload_local_configuration():
-                    self.reload_config()
 
-            logger.info(self.ha.run_cycle())
+        if self.dcs.cluster and self.dcs.cluster.config and self.dcs.cluster.config.data \
+                and self.config.set_dynamic_configuration(self.dcs.cluster.config):
+            self.reload_config()
 
-            if self.dcs.cluster and self.dcs.cluster.config and self.dcs.cluster.config.data \
+        if self.postgresql.role != 'uninitialized':
-                    and self.config.set_dynamic_configuration(self.dcs.cluster.config):
+            self.config.save_cache()
-                self.reload_config()
 
-            if self.postgresql.role != 'uninitialized':
+        self.schedule_next_run()
-                self.config.save_cache()
 
-            self.schedule_next_run()
+    def _shutdown(self):
-
-    def setup_signal_handlers(self):
-        from threading import Lock
-
-        self._received_sighup = False
-        self._sigterm_lock = Lock()
-        self._received_sigterm = False
-        if os.name != 'nt':
-            signal.signal(signal.SIGHUP, self.sighup_handler)
-        signal.signal(signal.SIGTERM, self.sigterm_handler)
-
-    def shutdown(self):
-        with self._sigterm_lock:
-            self._received_sigterm = True
         try:
             self.api.shutdown()
         except Exception:
             logger.exception('Exception during RestApi.shutdown')
-        self.ha.shutdown()
+        try:
-        self.logger.shutdown()
+            self.ha.shutdown()
+        except Exception:
+            logger.exception('Exception during Ha.shutdown')
 
 
 def patroni_main():
-    patroni = Patroni()
+    from multiprocessing import freeze_support
-    try:
+    from patroni.validator import schema
-        patroni.run()
+
-    except KeyboardInterrupt:
+    freeze_support()
-        pass
+    abstract_main(Patroni, schema)
-    finally:
-        patroni.shutdown()
 
 
 def fatal(string, *args):
@@ -193,8 +166,8 @@ def check_psycopg2():
 
 
 def main():
-    check_psycopg2()
     if os.getpid() != 1:
+        check_psycopg2()
         return patroni_main()
 
     # Patroni started with PID=1, it looks like we are in the container

patroni/api.py

@@ -1,4 +1,5 @@
 import base64
+import hmac
 import json
 import logging
 import psycopg2
@@ -7,35 +8,35 @@ import traceback
 import dateutil.parser
 import datetime
 import os
+import six
 import socket
+import sys
 
-from patroni.postgresql import PostgresConnectionException
+from ipaddress import ip_address, ip_network as _ip_network
-from patroni.postgresql.misc import postgres_version_to_int, PostgresException
-from patroni.utils import deep_compare, parse_bool, patch_config, Retry, \
-    RetryFailedError, parse_int, split_host_port, tzutc, uri
 from six.moves.BaseHTTPServer import BaseHTTPRequestHandler, HTTPServer
 from six.moves.socketserver import ThreadingMixIn
+from six.moves.urllib_parse import urlparse, parse_qs
 from threading import Thread
 
+from .exceptions import PostgresConnectionException, PostgresException
+from .postgresql.misc import postgres_version_to_int
+from .utils import deep_compare, enable_keepalive, parse_bool, patch_config, Retry, \
+    RetryFailedError, parse_int, split_host_port, tzutc, uri, cluster_as_json
+
 logger = logging.getLogger(__name__)
 
 
-def check_auth(func):
+def ip_network(value):
-    """Decorator function to check authorization header.
+    return _ip_network(value.decode('utf-8') if six.PY2 else value, False)
-
-    Usage example:
-    @check_auth
-    def do_PUT_foo():
-        pass
-    """
-    def wrapper(handler, *args, **kwargs):
-        if handler.check_auth_header():
-            return func(handler, *args, **kwargs)
-    return wrapper
 
 
 class RestApiHandler(BaseHTTPRequestHandler):
 
+    def _write_status_code_only(self, status_code):
+        message = self.responses[status_code][0]
+        self.wfile.write('{0} {1} {2}\r\n\r\n'.format(self.protocol_version, status_code, message).encode('utf-8'))
+        self.log_request(status_code)
+
     def _write_response(self, status_code, body, content_type='text/html', headers=None):
         self.send_response(status_code)
         headers = headers or {}
@@ -43,20 +44,28 @@ class RestApiHandler(BaseHTTPRequestHandler):
             headers['Content-Type'] = content_type
         for name, value in headers.items():
             self.send_header(name, value)
+        for name, value in self.server.http_extra_headers.items():
+            self.send_header(name, value)
         self.end_headers()
         self.wfile.write(body.encode('utf-8'))
 
     def _write_json_response(self, status_code, response):
-        self._write_response(status_code, json.dumps(response), content_type='application/json')
+        self._write_response(status_code, json.dumps(response, default=str), content_type='application/json')
 
-    def send_auth_request(self, body):
+    def check_access(func):
-        headers = {'WWW-Authenticate': 'Basic realm="' + self.server.patroni.__class__.__name__ + '"'}
+        """Decorator function to check the source ip, authorization header. or client certificates
-        self._write_response(401, body, headers=headers)
 
-    def check_auth_header(self):
+        Usage example:
-        auth_header = self.headers.get('Authorization')
+        @check_access
-        status = self.server.check_auth_header(auth_header)
+        def do_PUT_foo():
-        return not status or self.send_auth_request(status)
+            pass
+        """
+
+        def wrapper(self, *args, **kwargs):
+            if self.server.check_access(self):
+                return func(self, *args, **kwargs)
+
+        return wrapper
 
     def _write_status_response(self, status_code, response):
         patroni = self.server.patroni
@@ -87,59 +96,115 @@ class RestApiHandler(BaseHTTPRequestHandler):
     def do_GET(self, write_status_code_only=False):
         """Default method for processing all GET requests which can not be routed to other methods"""
 
-        time_start = time.time()
-        request_type = 'OPTIONS' if write_status_code_only else 'GET'
-
         path = '/master' if self.path == '/' else self.path
         response = self.get_postgresql_status()
 
         patroni = self.server.patroni
         cluster = patroni.dcs.cluster
 
-        if not cluster and patroni.ha.is_paused():
+        leader_optime = cluster and cluster.last_lsn or 0
-            primary_status_code = 200 if response['role'] == 'master' else 503
+        replayed_location = response.get('xlog', {}).get('replayed_location', 0)
-        else:
+        max_replica_lag = parse_int(self.path_query.get('lag', [sys.maxsize])[0], 'B')
-            primary_status_code = 200 if patroni.ha.is_leader() else 503
+        if max_replica_lag is None:
+            max_replica_lag = sys.maxsize
+        is_lagging = leader_optime and leader_optime > replayed_location + max_replica_lag
 
-        replica_status_code = 200 if not patroni.noloadbalance and \
+        replica_status_code = 200 if not patroni.noloadbalance and not is_lagging and \
             response.get('role') == 'replica' and response.get('state') == 'running' else 503
+
+        if not cluster and patroni.ha.is_paused():
+            leader_status_code = 200 if response.get('role') in ('master', 'standby_leader') else 503
+            primary_status_code = 200 if response.get('role') == 'master' else 503
+            standby_leader_status_code = 200 if response.get('role') == 'standby_leader' else 503
+        elif patroni.ha.is_leader():
+            leader_status_code = 200
+            if patroni.ha.is_standby_cluster():
+                primary_status_code = replica_status_code = 503
+                standby_leader_status_code = 200 if response.get('role') in ('replica', 'standby_leader') else 503
+            else:
+                primary_status_code = 200
+                standby_leader_status_code = 503
+        else:
+            leader_status_code = primary_status_code = standby_leader_status_code = 503
+
         status_code = 503
 
-        if patroni.ha.is_standby_cluster() and ('standby_leader' in path or 'standby-leader' in path):
+        ignore_tags = False
-            status_code = 200 if patroni.ha.is_leader() else 503
+        if 'standby_leader' in path or 'standby-leader' in path:
-        elif 'master' in path or 'leader' in path or 'primary' in path or 'read-write' in path:
+            status_code = standby_leader_status_code
+            ignore_tags = True
+        elif 'leader' in path:
+            status_code = leader_status_code
+            ignore_tags = True
+        elif 'master' in path or 'primary' in path or 'read-write' in path:
             status_code = primary_status_code
+            ignore_tags = True
         elif 'replica' in path:
             status_code = replica_status_code
         elif 'read-only' in path:
-            status_code = 200 if primary_status_code == 200 else replica_status_code
+            status_code = 200 if 200 in (primary_status_code, standby_leader_status_code) else replica_status_code
         elif 'health' in path:
             status_code = 200 if response.get('state') == 'running' else 503
         elif cluster:  # dcs is available
             is_synchronous = cluster.is_synchronous_mode() and cluster.sync \
-                    and cluster.sync.sync_standby == patroni.postgresql.name
+                    and patroni.postgresql.name in cluster.sync.members
             if path in ('/sync', '/synchronous') and is_synchronous:
                 status_code = replica_status_code
             elif path in ('/async', '/asynchronous') and not is_synchronous:
                 status_code = replica_status_code
 
+        # check for user defined tags in query params
+        if not ignore_tags and status_code == 200:
+            qs_tag_prefix = "tag_"
+            for qs_key, qs_value in self.path_query.items():
+                if not qs_key.startswith(qs_tag_prefix):
+                    continue
+                qs_key = qs_key[len(qs_tag_prefix):]
+                qs_value = qs_value[0]
+                instance_tag_value = patroni.tags.get(qs_key)
+                # tag not registered for instance
+                if instance_tag_value is None:
+                    status_code = 503
+                    break
+                if not isinstance(instance_tag_value, six.string_types):
+                    instance_tag_value = str(instance_tag_value).lower()
+                if instance_tag_value != qs_value:
+                    status_code = 503
+                    break
+
         if write_status_code_only:  # when haproxy sends OPTIONS request it reads only status code and nothing more
-            message = self.responses[status_code][0]
+            self._write_status_code_only(status_code)
-            self.wfile.write('{0} {1} {2}\r\n'.format(self.protocol_version, status_code, message).encode('utf-8'))
         else:
             self._write_status_response(status_code, response)
 
-        time_end = time.time()
-        self.log_message('%s %s %s latency: %s ms', request_type, path,
-                         status_code, (time_end - time_start) * 1000)
-
     def do_OPTIONS(self):
         self.do_GET(write_status_code_only=True)
 
+    def do_GET_liveness(self):
+        self._write_status_code_only(200)
+
+    def do_GET_readiness(self):
+        patroni = self.server.patroni
+        if patroni.ha.is_leader():
+            status_code = 200
+        elif patroni.postgresql.state == 'running':
+            status_code = 200 if patroni.dcs.cluster else 503
+        else:
+            status_code = 503
+        self._write_status_code_only(status_code)
+
     def do_GET_patroni(self):
         response = self.get_postgresql_status(True)
         self._write_status_response(200, response)
 
+    def do_GET_cluster(self):
+        cluster = self.server.patroni.dcs.get_cluster(True)
+        self._write_json_response(200, cluster_as_json(cluster))
+
+    def do_GET_history(self):
+        cluster = self.server.patroni.dcs.cluster or self.server.patroni.dcs.get_cluster()
+        self._write_json_response(200, cluster.history and cluster.history.lines or [])
+
     def do_GET_config(self):
         cluster = self.server.patroni.dcs.cluster or self.server.patroni.dcs.get_cluster()
         if cluster.config:
@@ -147,6 +212,84 @@ class RestApiHandler(BaseHTTPRequestHandler):
         else:
             self.send_error(502)
 
+    def do_GET_metrics(self):
+        postgres = self.get_postgresql_status(True)
+        patroni = self.server.patroni
+        epoch = datetime.datetime(1970, 1, 1, tzinfo=tzutc)
+
+        metrics = []
+
+        scope_label = '{{scope="{0}"}}'.format(patroni.postgresql.scope)
+        metrics.append("# HELP patroni_version Patroni semver without periods.")
+        metrics.append("# TYPE patroni_version gauge")
+        padded_semver = ''.join([x.zfill(2) for x in patroni.version.split('.')])  # 2.0.2 => 020002
+        metrics.append("patroni_version{0} {1}".format(scope_label, padded_semver))
+
+        metrics.append("# HELP patroni_postgres_running Value is 1 if Postgres is running, 0 otherwise.")
+        metrics.append("# TYPE patroni_postgres_running gauge")
+        metrics.append("patroni_postgres_running{0} {1}".format(scope_label, int(postgres['state'] == 'running')))
+
+        metrics.append("# HELP patroni_postmaster_start_time Epoch seconds since Postgres started.")
+        metrics.append("# TYPE patroni_postmaster_start_time gauge")
+        postmaster_start_time = postgres.get('postmaster_start_time')
+        postmaster_start_time = (postmaster_start_time - epoch).total_seconds() if postmaster_start_time else 0
+        metrics.append("patroni_postmaster_start_time{0} {1}".format(scope_label, postmaster_start_time))
+
+        metrics.append("# HELP patroni_master Value is 1 if this node is the leader, 0 otherwise.")
+        metrics.append("# TYPE patroni_master gauge")
+        metrics.append("patroni_master{0} {1}".format(scope_label, int(postgres['role'] == 'master')))
+
+        metrics.append("# HELP patroni_xlog_location Current location of the Postgres"
+                       " transaction log, 0 if this node is not the leader.")
+        metrics.append("# TYPE patroni_xlog_location counter")
+        metrics.append("patroni_xlog_location{0} {1}".format(scope_label, postgres.get('xlog', {}).get('location', 0)))
+
+        metrics.append("# HELP patroni_standby_leader Value is 1 if this node is the standby_leader, 0 otherwise.")
+        metrics.append("# TYPE patroni_standby_leader gauge")
+        metrics.append("patroni_standby_leader{0} {1}".format(scope_label, int(postgres['role'] == 'standby_leader')))
+
+        metrics.append("# HELP patroni_replica Value is 1 if this node is a replica, 0 otherwise.")
+        metrics.append("# TYPE patroni_replica gauge")
+        metrics.append("patroni_replica{0} {1}".format(scope_label, int(postgres['role'] == 'replica')))
+
+        metrics.append("# HELP patroni_xlog_received_location Current location of the received"
+                       " Postgres transaction log, 0 if this node is not a replica.")
+        metrics.append("# TYPE patroni_xlog_received_location counter")
+        metrics.append("patroni_xlog_received_location{0} {1}"
+                       .format(scope_label, postgres.get('xlog', {}).get('received_location', 0)))
+
+        metrics.append("# HELP patroni_xlog_replayed_location Current location of the replayed"
+                       " Postgres transaction log, 0 if this node is not a replica.")
+        metrics.append("# TYPE patroni_xlog_replayed_location counter")
+        metrics.append("patroni_xlog_replayed_location{0} {1}"
+                       .format(scope_label, postgres.get('xlog', {}).get('replayed_location', 0)))
+
+        metrics.append("# HELP patroni_xlog_replayed_timestamp Current timestamp of the replayed"
+                       " Postgres transaction log, 0 if null.")
+        metrics.append("# TYPE patroni_xlog_replayed_timestamp gauge")
+        replayed_timestamp = postgres.get('xlog', {}).get('replayed_timestamp')
+        replayed_timestamp = (replayed_timestamp - epoch).total_seconds() if replayed_timestamp else 0
+        metrics.append("patroni_xlog_replayed_timestamp{0} {1}".format(scope_label, replayed_timestamp))
+
+        metrics.append("# HELP patroni_xlog_paused Value is 1 if the Postgres xlog is paused, 0 otherwise.")
+        metrics.append("# TYPE patroni_xlog_paused gauge")
+        metrics.append("patroni_xlog_paused{0} {1}"
+                       .format(scope_label, int(postgres.get('xlog', {}).get('paused', False) is True)))
+
+        metrics.append("# HELP patroni_postgres_server_version Version of Postgres (if running), 0 otherwise.")
+        metrics.append("# TYPE patroni_postgres_server_version gauge")
+        metrics.append("patroni_postgres_server_version {0} {1}".format(scope_label, postgres.get('server_version', 0)))
+
+        metrics.append("# HELP patroni_cluster_unlocked Value is 1 if the cluster is unlocked, 0 if locked.")
+        metrics.append("# TYPE patroni_cluster_unlocked gauge")
+        metrics.append("patroni_cluster_unlocked{0} {1}".format(scope_label, int(postgres['cluster_unlocked'])))
+
+        metrics.append("# HELP patroni_postgres_timeline Postgres timeline of this node (if running), 0 otherwise.")
+        metrics.append("# TYPE patroni_postgres_timeline counter")
+        metrics.append("patroni_postgres_timeline{0} {1}".format(scope_label, postgres.get('timeline', 0)))
+
+        self._write_response(200, '\n'.join(metrics)+'\n', content_type='text/plain')
+
     def _read_json_content(self, body_is_optional=False):
         if 'content-length' not in self.headers:
             return self.send_error(411) if not body_is_optional else {}
@@ -161,11 +304,13 @@ class RestApiHandler(BaseHTTPRequestHandler):
             logger.exception('Bad request')
         self.send_error(400)
 
-    @check_auth
+    @check_access
     def do_PATCH_config(self):
         request = self._read_json_content()
         if request:
-            cluster = self.server.patroni.dcs.get_cluster()
+            cluster = self.server.patroni.dcs.get_cluster(True)
+            if not (cluster.config and cluster.config.modify_index):
+                return self.send_error(503)
             data = cluster.config.data.copy()
             if patch_config(data, request):
                 value = json.dumps(data, separators=(',', ':'))
@@ -174,7 +319,7 @@ class RestApiHandler(BaseHTTPRequestHandler):
             self.server.patroni.ha.wakeup()
             self._write_json_response(200, data)
 
-    @check_auth
+    @check_access
     def do_PUT_config(self):
         request = self._read_json_content()
         if request:
@@ -185,20 +330,10 @@ class RestApiHandler(BaseHTTPRequestHandler):
                     return self.send_error(502)
             self._write_json_response(200, request)
 
-    @check_auth
+    @check_access
     def do_POST_reload(self):
-        try:
+        self.server.patroni.sighup_handler()
-            if self.server.patroni.config.reload_local_configuration(True):
+        self._write_response(202, 'reload scheduled')
-                status_code = 202
-                response = 'reload scheduled'
-                self.server.patroni.sighup_handler()
-            else:
-                status_code = 200
-                response = 'nothing changed'
-        except Exception as e:
-            status_code = 500
-            response = str(e)
-        self._write_response(status_code, response)
 
     @staticmethod
     def parse_schedule(schedule, action):
@@ -221,7 +356,7 @@ class RestApiHandler(BaseHTTPRequestHandler):
             status_code = 422
         return (status_code, error, scheduled_at)
 
-    @check_auth
+    @check_access
     def do_POST_restart(self):
         status_code = 500
         data = 'restart failed'
@@ -282,7 +417,7 @@ class RestApiHandler(BaseHTTPRequestHandler):
                     status_code = 409
         self._write_response(status_code, data)
 
-    @check_auth
+    @check_access
     def do_DELETE_restart(self):
         if self.server.patroni.ha.delete_future_restart():
             data = "scheduled restart deleted"
@@ -292,7 +427,21 @@ class RestApiHandler(BaseHTTPRequestHandler):
             code = 404
         self._write_response(code, data)
 
-    @check_auth
+    @check_access
+    def do_DELETE_switchover(self):
+        failover = self.server.patroni.dcs.get_cluster().failover
+        if failover and failover.scheduled_at:
+            if not self.server.patroni.dcs.manual_failover('', '', index=failover.index):
+                return self.send_error(409)
+            else:
+                data = "scheduled switchover deleted"
+                code = 200
+        else:
+            data = "no switchover is scheduled"
+            code = 404
+        self._write_response(code, data)
+
+    @check_access
     def do_POST_reinitialize(self):
         request = self._read_json_content(body_is_optional=True)
 
@@ -324,20 +473,20 @@ class RestApiHandler(BaseHTTPRequestHandler):
                 if not cluster.failover:
                     return 503, action.title() + ' failed'
             except Exception as e:
-                logger.debug('Exception occured during polling %s result: %s', action, e)
+                logger.debug('Exception occurred during polling %s result: %s', action, e)
         return 503, action.title() + ' status unknown'
 
     def is_failover_possible(self, cluster, leader, candidate, action):
         if leader and (not cluster.leader or cluster.leader.name != leader):
             return 'leader name does not match'
         if candidate:
-            if action == 'switchover' and cluster.is_synchronous_mode() and cluster.sync.sync_standby != candidate:
+            if action == 'switchover' and cluster.is_synchronous_mode() and candidate not in cluster.sync.members:
                 return 'candidate name does not match with sync_standby'
             members = [m for m in cluster.members if m.name == candidate]
             if not members:
                 return 'candidate does not exists'
         elif cluster.is_synchronous_mode():
-            members = [m for m in cluster.members if m.name == cluster.sync.sync_standby]
+            members = [m for m in cluster.members if m.name in cluster.sync.members]
             if not members:
                 return action + ' is not possible: can not find sync_standby'
         else:
@@ -349,7 +498,7 @@ class RestApiHandler(BaseHTTPRequestHandler):
                 return None
         return action + ' is not possible: no good candidates have been found'
 
-    @check_auth
+    @check_access
     def do_POST_failover(self, action='failover'):
         request = self._read_json_content()
         (status_code, data) = (400, '')
@@ -414,6 +563,9 @@ class RestApiHandler(BaseHTTPRequestHandler):
 
         ret = BaseHTTPRequestHandler.parse_request(self)
         if ret:
+            urlpath = urlparse(self.path)
+            self.path = urlpath.path
+            self.path_query = parse_qs(urlpath.query) or {}
             mname = self.path.lstrip('/').split('/')[0]
             mname = self.command + ('_' + mname if mname else '')
             if hasattr(self, 'do_' + mname):
@@ -427,78 +579,77 @@ class RestApiHandler(BaseHTTPRequestHandler):
         return retry(self.server.query, sql, *params)
 
     def get_postgresql_status(self, retry=False):
+        postgresql = self.server.patroni.postgresql
         try:
             cluster = self.server.patroni.dcs.cluster
 
-            if self.server.patroni.postgresql.state not in ('running', 'restarting', 'starting'):
+            if postgresql.state not in ('running', 'restarting', 'starting'):
                 raise RetryFailedError('')
-            stmt = ("WITH replication_info AS ("
+            stmt = ("SELECT " + postgresql.POSTMASTER_START_TIME + ", " + postgresql.TL_LSN + ","
-                    "SELECT usename, application_name, client_addr, state, sync_state, sync_priority"
+                    " pg_catalog.pg_last_xact_replay_timestamp(),"
-                    " FROM pg_catalog.pg_stat_replication) SELECT"
+                    " pg_catalog.array_to_json(pg_catalog.array_agg(pg_catalog.row_to_json(ri))) "
-                    " pg_catalog.to_char(pg_catalog.pg_postmaster_start_time(), 'YYYY-MM-DD HH24:MI:SS.MS TZ'),"
+                    "FROM (SELECT (SELECT rolname FROM pg_authid WHERE oid = usesysid) AS usename,"
-                    " CASE WHEN pg_catalog.pg_is_in_recovery() THEN 0"
+                    " application_name, client_addr, w.state, sync_state, sync_priority"
-                    " ELSE ('x' || pg_catalog.substr(pg_catalog.pg_{0}file_name("
+                    " FROM pg_catalog.pg_stat_get_wal_senders() w, pg_catalog.pg_stat_get_activity(pid)) AS ri")
-                    "pg_catalog.pg_current_{0}_{1}()), 1, 8))::bit(32)::int END,"
-                    " CASE WHEN pg_catalog.pg_is_in_recovery() THEN 0"
-                    " ELSE pg_catalog.pg_{0}_{1}_diff(pg_catalog.pg_current_{0}_{1}(), '0/0')::bigint END,"
-                    " pg_catalog.pg_{0}_{1}_diff(COALESCE(pg_catalog.pg_last_{0}_receive_{1}(),"
-                    " pg_catalog.pg_last_{0}_replay_{1}()), '0/0')::bigint,"
-                    " pg_catalog.pg_{0}_{1}_diff(pg_catalog.pg_last_{0}_replay_{1}(), '0/0')::bigint,"
-                    " pg_catalog.to_char(pg_catalog.pg_last_xact_replay_timestamp(), 'YYYY-MM-DD HH24:MI:SS.MS TZ'),"
-                    " pg_catalog.pg_is_in_recovery() AND pg_catalog.pg_is_{0}_replay_paused(), "
-                    "(SELECT pg_catalog.array_to_json(pg_catalog.array_agg("
-                    "pg_catalog.row_to_json(ri))) FROM replication_info ri)")
 
-            row = self.query(stmt.format(self.server.patroni.postgresql.wal_name,
+            row = self.query(stmt.format(postgresql.wal_name, postgresql.lsn_name), retry=retry)[0]
-                                         self.server.patroni.postgresql.lsn_name), retry=retry)[0]
 
             result = {
-                'state': self.server.patroni.postgresql.state,
+                'state': postgresql.state,
                 'postmaster_start_time': row[0],
                 'role': 'replica' if row[1] == 0 else 'master',
-                'server_version': self.server.patroni.postgresql.server_version,
+                'server_version': postgresql.server_version,
                 'cluster_unlocked': bool(not cluster or cluster.is_unlocked()),
                 'xlog': ({
-                    'received_location': row[3],
+                    'received_location': row[4] or row[3],
-                    'replayed_location': row[4],
+                    'replayed_location': row[3],
-                    'replayed_timestamp': row[5],
+                    'replayed_timestamp': row[6],
-                    'paused': row[6]} if row[1] == 0 else {
+                    'paused': row[5]} if row[1] == 0 else {
                     'location': row[2]
                 })
             }
 
             if result['role'] == 'replica' and self.server.patroni.ha.is_standby_cluster():
-                result['role'] = self.server.patroni.postgresql.role
+                result['role'] = postgresql.role
 
             if row[1] > 0:
                 result['timeline'] = row[1]
             else:
                 leader_timeline = None if not cluster or cluster.is_unlocked() else cluster.leader.timeline
-                result['timeline'] = self.server.patroni.postgresql.replica_cached_timeline(leader_timeline)
+                result['timeline'] = postgresql.replica_cached_timeline(leader_timeline)
 
             if row[7]:
                 result['replication'] = row[7]
 
             return result
         except (psycopg2.Error, RetryFailedError, PostgresConnectionException):
-            state = self.server.patroni.postgresql.state
+            state = postgresql.state
             if state == 'running':
                 logger.exception('get_postgresql_status')
                 state = 'unknown'
-            return {'state': state, 'role': self.server.patroni.postgresql.role}
+            return {'state': state, 'role': postgresql.role}
+
+    def handle_one_request(self):
+        self.__start_time = time.time()
+        BaseHTTPRequestHandler.handle_one_request(self)
 
     def log_message(self, fmt, *args):
-        logger.debug("API thread: %s - - [%s] %s", self.client_address[0], self.log_date_time_string(), fmt % args)
+        latency = 1000.0 * (time.time() - self.__start_time)
+        logger.debug("API thread: %s - - %s latency: %0.3f ms", self.client_address[0], fmt % args, latency)
 
 
 class RestApiServer(ThreadingMixIn, HTTPServer, Thread):
+    # On 3.7+ the `ThreadingMixIn` gathers all non-daemon worker threads in order to join on them at server close.
+    daemon_threads = True  # Make worker threads "fire and forget" to prevent a memory leak.
 
     def __init__(self, patroni, config):
         self.patroni = patroni
         self.__listen = None
-        self.__initialize(config)
+        self.__ssl_options = None
-        self.__set_config_parameters(config)
+        self.reload_config(config)
         self.daemon = True
+        self.__ssl_serial_number = None
+        self._received_new_cert = False
 
     def query(self, sql, *params):
         cursor = None
@@ -519,7 +670,7 @@ class RestApiServer(ThreadingMixIn, HTTPServer, Thread):
             fcntl.fcntl(fd, fcntl.F_SETFD, flags | fcntl.FD_CLOEXEC)
 
     def check_basic_auth_key(self, key):
-        return self.__auth_key == key
+        return hmac.compare_digest(self.__auth_key, key.encode('utf-8'))
 
     def check_auth_header(self, auth_header):
         if self.__auth_key:
@@ -529,12 +680,43 @@ class RestApiServer(ThreadingMixIn, HTTPServer, Thread):
                 return 'not authenticated'
 
     @staticmethod
-    def __get_ssl_options(config):
+    def __resolve_ips(host, port):
-        return {option: config[option] for option in ['certfile', 'keyfile'] if option in config}
+        try:
+            for _, _, _, _, sa in socket.getaddrinfo(host, port, 0, socket.SOCK_STREAM, socket.IPPROTO_TCP):
+                yield ip_network(sa[0])
+        except Exception as e:
+            logger.error('Failed to resolve %s: %r', host, e)
 
-    def __set_config_parameters(self, config):
+    def __members_ips(self):
-        self.__auth_key = base64.b64encode(config['auth'].encode('utf-8')).decode('utf-8') if 'auth' in config else None
+        cluster = self.patroni.dcs.cluster
-        self.connection_string = uri(self.__protocol, config.get('connect_address') or self.__listen, 'patroni')
+        if self.__allowlist_include_members and cluster:
+            for member in cluster.members:
+                if member.api_url:
+                    try:
+                        r = urlparse(member.api_url)
+                        host = r.hostname
+                        port = r.port or (443 if r.scheme == 'https' else 80)
+                        for ip in self.__resolve_ips(host, port):
+                            yield ip
+                    except Exception as e:
+                        logger.debug('Failed to parse url %s: %r', member.api_url, e)
+
+    def check_access(self, rh):
+        if self.__allowlist or self.__allowlist_include_members:
+            incoming_ip = rh.client_address[0]
+            incoming_ip = ip_address(incoming_ip.decode('utf-8') if six.PY2 else incoming_ip)
+            if not any(incoming_ip in net for net in self.__allowlist + tuple(self.__members_ips())):
+                return rh._write_response(403, 'Access is denied')
+
+        if not hasattr(rh.request, 'getpeercert') or not rh.request.getpeercert():  # valid client cert isn't present
+            if self.__protocol == 'https' and self.__ssl_options.get('verify_client') in ('required', 'optional'):
+                return rh._write_response(403, 'client certificate required')
+
+        reason = self.check_auth_header(rh.headers.get('Authorization'))
+        if reason:
+            headers = {'WWW-Authenticate': 'Basic realm="' + self.patroni.__class__.__name__ + '"'}
+            return rh._write_response(401, reason, headers=headers)
+        return True
 
     @staticmethod
     def __has_dual_stack():
@@ -553,7 +735,7 @@ class RestApiServer(ThreadingMixIn, HTTPServer, Thread):
 
     def __httpserver_init(self, host, port):
         dual_stack = self.__has_dual_stack()
-        if host == '':
+        if host in ('', '*'):
             host = None
 
         info = socket.getaddrinfo(host, port, socket.AF_UNSPEC, socket.SOCK_STREAM, 0, socket.AI_PASSIVE)
@@ -561,45 +743,127 @@ class RestApiServer(ThreadingMixIn, HTTPServer, Thread):
         info.sort(key=lambda x: x[0] == socket.AF_INET, reverse=not dual_stack)
 
         self.address_family = info[0][0]
-        HTTPServer.__init__(self, info[0][-1][:2], RestApiHandler)
-
-    def __initialize(self, config):
         try:
-            host, port = split_host_port(config['listen'], None)
+            HTTPServer.__init__(self, info[0][-1][:2], RestApiHandler)
+        except socket.error:
+            logger.error(
+                    "Couldn't start a service on '%s:%s', please check your `restapi.listen` configuration", host, port)
+            raise
+
+    def __initialize(self, listen, ssl_options):
+        try:
+            host, port = split_host_port(listen, None)
         except Exception:
             raise ValueError('Invalid "restapi" config: expected <HOST>:<PORT> for "listen", but got "{0}"'
-                             .format(config['listen']))
+                             .format(listen))
 
-        if self.__listen is not None:  # changing config in runtime
+        reloading_config = self.__listen is not None  # changing config in runtime
+        if reloading_config:
             self.shutdown()
 
-        self.__listen = config['listen']
+        self.__listen = listen
-        self.__ssl_options = self.__get_ssl_options(config)
+        self.__ssl_options = ssl_options
+        self._received_new_cert = False  # reset to False after reload_config()
 
         self.__httpserver_init(host, port)
         Thread.__init__(self, target=self.serve_forever)
         self._set_fd_cloexec(self.socket)
 
-        self.__protocol = 'http'
-
         # wrap socket with ssl if 'certfile' is defined in a config.yaml
         # Sometime it's also needed to pass reference to a 'keyfile'.
+        self.__protocol = 'https' if ssl_options.get('certfile') else 'http'
+        if self.__protocol == 'https':
+            import ssl
+            ctx = ssl.create_default_context(ssl.Purpose.CLIENT_AUTH, cafile=ssl_options.get('cafile'))
+            if ssl_options.get('ciphers'):
+                ctx.set_ciphers(ssl_options['ciphers'])
+            ctx.load_cert_chain(certfile=ssl_options['certfile'], keyfile=ssl_options.get('keyfile'),
+                                password=ssl_options.get('keyfile_password'))
+            verify_client = ssl_options.get('verify_client')
+            if verify_client:
+                modes = {'none': ssl.CERT_NONE, 'optional': ssl.CERT_OPTIONAL, 'required': ssl.CERT_REQUIRED}
+                if verify_client in modes:
+                    ctx.verify_mode = modes[verify_client]
+                else:
+                    logger.error('Bad value in the "restapi.verify_client": %s', verify_client)
+            self.__ssl_serial_number = self.get_certificate_serial_number()
+            self.socket = ctx.wrap_socket(self.socket, server_side=True)
+        if reloading_config:
+            self.start()
+
+    def process_request_thread(self, request, client_address):
+        if isinstance(request, tuple):
+            sock, newsock = request
+            try:
+                request = sock.context.wrap_socket(newsock, do_handshake_on_connect=sock.do_handshake_on_connect,
+                                                   suppress_ragged_eofs=sock.suppress_ragged_eofs, server_side=True)
+            except socket.error:
+                return
+        super(RestApiServer, self).process_request_thread(request, client_address)
+
+    def get_request(self):
+        sock = self.socket
+        newsock, addr = socket.socket.accept(sock)
+        enable_keepalive(newsock, 10, 3)
+        if hasattr(sock, 'context'):  # SSLSocket, we want to do the deferred handshake from a thread
+            newsock = (sock, newsock)
+        return newsock, addr
+
+    def shutdown_request(self, request):
+        if isinstance(request, tuple):
+            _, request = request  # SSLSocket
+        return super(RestApiServer, self).shutdown_request(request)
+
+    def get_certificate_serial_number(self):
         if self.__ssl_options.get('certfile'):
             import ssl
-            ctx = ssl.create_default_context(ssl.Purpose.CLIENT_AUTH)
+            try:
-            ctx.load_cert_chain(**self.__ssl_options)
+                crt = ssl._ssl._test_decode_cert(self.__ssl_options['certfile'])
-            self.socket = ctx.wrap_socket(self.socket, server_side=True)
+                return crt.get('serialNumber')
-            self.__protocol = 'https'
+            except ssl.SSLError as e:
-        return True
+                logger.error('Failed to get serial number from certificate %s: %r', self.__ssl_options['certfile'], e)
+
+    def reload_local_certificate(self):
+        if self.__protocol == 'https':
+            on_disk_cert_serial_number = self.get_certificate_serial_number()
+            if on_disk_cert_serial_number != self.__ssl_serial_number:
+                self._received_new_cert = True
+                self.__ssl_serial_number = on_disk_cert_serial_number
+                return True
+
+    def _build_allowlist(self, value):
+        if isinstance(value, list):
+            for v in value:
+                if '/' in v:  # netmask
+                    try:
+                        yield ip_network(v)
+                    except Exception as e:
+                        logger.error('Invalid value "%s" in the allowlist: %r', v, e)
+                else:  # ip or hostname, try to resolve it
+                    for ip in self.__resolve_ips(v, 8080):
+                        yield ip
 
     def reload_config(self, config):
         if 'listen' not in config:  # changing config in runtime
             raise ValueError('Can not find "restapi.listen" config')
 
-        elif (self.__listen != config['listen'] or self.__ssl_options != self.__get_ssl_options(config)) \
+        self.__allowlist = tuple(self._build_allowlist(config.get('allowlist')))
-                and self.__initialize(config):
+        self.__allowlist_include_members = config.get('allowlist_include_members')
-            self.start()
+
-        self.__set_config_parameters(config)
+        ssl_options = {n: config[n] for n in ('certfile', 'keyfile', 'keyfile_password',
+                                              'cafile', 'ciphers') if n in config}
+
+        self.http_extra_headers = config.get('http_extra_headers') or {}
+        self.http_extra_headers.update((config.get('https_extra_headers') or {}) if ssl_options.get('certfile') else {})
+
+        if isinstance(config.get('verify_client'), six.string_types):
+            ssl_options['verify_client'] = config['verify_client'].lower()
+
+        if self.__listen != config['listen'] or self.__ssl_options != ssl_options or self._received_new_cert:
+            self.__initialize(config['listen'], ssl_options)
+
+        self.__auth_key = base64.b64encode(config['auth'].encode('utf-8')) if 'auth' in config else None
+        self.connection_string = uri(self.__protocol, config.get('connect_address') or self.__listen, 'patroni')
 
     @staticmethod
     def handle_error(request, client_address):

patroni/async_executor.py

@@ -110,6 +110,12 @@ class AsyncExecutor(object):
     def run_async(self, func, args=()):
         Thread(target=self.run, args=(func, args)).start()
 
+    def try_run_async(self, action, func, args=()):
+        prev = self.schedule(action)
+        if prev is None:
+            return self.run_async(func, args)
+        return 'Failed to run {0}, {1} is already in progress'.format(action, prev)
+
     def cancel(self):
         with self:
             with self._scheduled_action_lock:

patroni/config.py

@@ -2,19 +2,37 @@ import json
 import logging
 import os
 import shutil
-import sys
 import tempfile
 import yaml
 
 from collections import defaultdict
 from copy import deepcopy
+from patroni import PATRONI_ENV_PREFIX
+from patroni.exceptions import ConfigParseError
 from patroni.dcs import ClusterConfig
-from patroni.postgresql.config import ConfigHandler
+from patroni.postgresql.config import CaseInsensitiveDict, ConfigHandler
 from patroni.utils import deep_compare, parse_bool, parse_int, patch_config
-from requests.structures import CaseInsensitiveDict
 
 logger = logging.getLogger(__name__)
 
+_AUTH_ALLOWED_PARAMETERS = (
+    'username',
+    'password',
+    'sslmode',
+    'sslcert',
+    'sslkey',
+    'sslpassword',
+    'sslrootcert',
+    'sslcrl',
+    'gssencmode',
+    'channel_binding'
+)
+
+
+def default_validator(conf):
+    if not conf:
+        return "Config is empty."
+
 
 class Config(object):
     """
@@ -36,17 +54,19 @@ class Config(object):
          to work with it as with the old `config` object.
     """
 
-    PATRONI_ENV_PREFIX = 'PATRONI_'
     PATRONI_CONFIG_VARIABLE = PATRONI_ENV_PREFIX + 'CONFIGURATION'
 
     __CACHE_FILENAME = 'patroni.dynamic.json'
     __DEFAULT_CONFIG = {
         'ttl': 30, 'loop_wait': 10, 'retry_timeout': 10,
         'maximum_lag_on_failover': 1048576,
+        'maximum_lag_on_syncnode': -1,
         'check_timeline': False,
         'master_start_timeout': 300,
+        'master_stop_timeout': 0,
         'synchronous_mode': False,
         'synchronous_mode_strict': False,
+        'synchronous_node_count': 1,
         'standby_cluster': {
             'create_replica_methods': '',
             'host': '',
@@ -59,34 +79,34 @@ class Config(object):
         'postgresql': {
             'bin_dir': '',
             'use_slots': True,
-            'parameters': CaseInsensitiveDict({p: v[0] for p, v in ConfigHandler.CMDLINE_OPTIONS.items()})
+            'parameters': CaseInsensitiveDict({p: v[0] for p, v in ConfigHandler.CMDLINE_OPTIONS.items()
+                                               if p not in ('wal_keep_segments', 'wal_keep_size')})
         },
         'watchdog': {
             'mode': 'automatic',
         }
     }
 
-    def __init__(self):
+    def __init__(self, configfile, validator=default_validator):
         self._modify_index = -1
         self._dynamic_configuration = {}
 
         self.__environment_configuration = self._build_environment_configuration()
 
         # Patroni reads the configuration from the command-line argument if it exists, otherwise from the environment
-        self._config_file = len(sys.argv) >= 2 and os.path.isfile(sys.argv[1]) and sys.argv[1]
+        self._config_file = configfile and os.path.exists(configfile) and configfile
         if self._config_file:
             self._local_configuration = self._load_config_file()
         else:
             config_env = os.environ.pop(self.PATRONI_CONFIG_VARIABLE, None)
             self._local_configuration = config_env and yaml.safe_load(config_env) or self.__environment_configuration
-            if not self._local_configuration:
+        if validator:
-                print('Usage: {0} config.yml'.format(sys.argv[0]))
+            error = validator(self._local_configuration)
-                print('\tPatroni may also read the configuration from the {0} environment variable'.
+            if error:
-                      format(self.PATRONI_CONFIG_VARIABLE))
+                raise ConfigParseError(error)
-                sys.exit(1)
 
         self.__effective_configuration = self._build_effective_configuration({}, self._local_configuration)
-        self._data_dir = self.__effective_configuration['postgresql']['data_dir']
+        self._data_dir = self.__effective_configuration.get('postgresql', {}).get('data_dir', "")
         self._cache_file = os.path.join(self._data_dir, self.__CACHE_FILENAME)
         self._load_cache()
         self._cache_needs_saving = False
@@ -102,12 +122,32 @@ class Config(object):
     def check_mode(self, mode):
         return bool(parse_bool(self._dynamic_configuration.get(mode)))
 
+    def _load_config_path(self, path):
+        """
+        If path is a file, loads the yml file pointed to by path.
+        If path is a directory, loads all yml files in that directory in alphabetical order
+        """
+        if os.path.isfile(path):
+            files = [path]
+        elif os.path.isdir(path):
+            files = [os.path.join(path, f) for f in sorted(os.listdir(path))
+                     if (f.endswith('.yml') or f.endswith('.yaml')) and os.path.isfile(os.path.join(path, f))]
+        else:
+            logger.error('config path %s is neither directory nor file', path)
+            raise ConfigParseError('invalid config path')
+
+        overall_config = {}
+        for fname in files:
+            with open(fname) as f:
+                config = yaml.safe_load(f)
+                patch_config(overall_config, config)
+        return overall_config
+
     def _load_config_file(self):
         """Loads config.yaml from filesystem and applies some values which were set via ENV"""
-        with open(self._config_file) as f:
+        config = self._load_config_path(self._config_file)
-            config = yaml.safe_load(f)
+        patch_config(config, self.__environment_configuration)
-            patch_config(config, self.__environment_configuration)
+        return config
-            return config
 
     def _load_cache(self):
         if os.path.isfile(self._cache_file):
@@ -158,23 +198,19 @@ class Config(object):
             except Exception:
                 logger.exception('Exception when setting dynamic_configuration')
 
-    def reload_local_configuration(self, dry_run=False):
+    def reload_local_configuration(self):
         if self.config_file:
             try:
                 configuration = self._load_config_file()
                 if not deep_compare(self._local_configuration, configuration):
                     new_configuration = self._build_effective_configuration(self._dynamic_configuration, configuration)
-                    if dry_run:
-                        return not deep_compare(new_configuration, self.__effective_configuration)
                     self._local_configuration = configuration
                     self.__effective_configuration = new_configuration
                     return True
                 else:
-                    logger.info('No configuration items changed, nothing to reload.')
+                    logger.info('No local configuration items changed.')
             except Exception:
                 logger.exception('Exception when reloading local configuration from %s', self.config_file)
-                if dry_run:
-                    raise
 
     @staticmethod
     def _process_postgresql_parameters(parameters, is_local=False):
@@ -196,7 +232,7 @@ class Config(object):
                 for name, value in (value or {}).items():
                     if name in self.__DEFAULT_CONFIG['standby_cluster']:
                         config['standby_cluster'][name] = deepcopy(value)
-            elif name in config:  # only variables present in __DEFAULT_CONFIG allowed to be overriden from DCS
+            elif name in config:  # only variables present in __DEFAULT_CONFIG allowed to be overridden from DCS
                 if name in ('synchronous_mode', 'synchronous_mode_strict'):
                     config[name] = value
                 else:
@@ -208,7 +244,7 @@ class Config(object):
         ret = defaultdict(dict)
 
         def _popenv(name):
-            return os.environ.pop(Config.PATRONI_ENV_PREFIX + name.upper(), None)
+            return os.environ.pop(PATRONI_ENV_PREFIX + name.upper(), None)
 
         for param in ('name', 'namespace', 'scope'):
             value = _popenv(param)
@@ -217,7 +253,7 @@ class Config(object):
 
         def _fix_log_env(name, oldname):
             value = _popenv(oldname)
-            name = Config.PATRONI_ENV_PREFIX + 'LOG_' + name.upper()
+            name = PATRONI_ENV_PREFIX + 'LOG_' + name.upper()
             if value and name not in os.environ:
                 os.environ[name] = value
 
@@ -230,10 +266,44 @@ class Config(object):
                 if value:
                     ret[section][param] = value
 
-        _set_section_values('restapi', ['listen', 'connect_address', 'certfile', 'keyfile'])
+        _set_section_values('restapi', ['listen', 'connect_address', 'certfile', 'keyfile', 'keyfile_password',
+                                        'cafile', 'ciphers', 'verify_client', 'http_extra_headers',
+                                        'https_extra_headers', 'allowlist', 'allowlist_include_members'])
+        _set_section_values('ctl', ['insecure', 'cacert', 'certfile', 'keyfile'])
         _set_section_values('postgresql', ['listen', 'connect_address', 'config_dir', 'data_dir', 'pgpass', 'bin_dir'])
-        _set_section_values('log', ['level', 'format', 'dateformat', 'max_queue_size',
+        _set_section_values('log', ['level', 'traceback_level', 'format', 'dateformat', 'max_queue_size',
                                     'dir', 'file_size', 'file_num', 'loggers'])
+        _set_section_values('raft', ['data_dir', 'self_addr', 'partner_addrs', 'password', 'bind_addr'])
+
+        for first, second in (('restapi', 'allowlist_include_members'), ('ctl', 'insecure')):
+            value = ret.get(first, {}).pop(second, None)
+            if value:
+                value = parse_bool(value)
+                if value is not None:
+                    ret[first][second] = value
+
+        for second in ('max_queue_size', 'file_size', 'file_num'):
+            value = ret.get('log', {}).pop(second, None)
+            if value:
+                value = parse_int(value)
+                if value is not None:
+                    ret['log'][second] = value
+
+        def _parse_list(value):
+            if not (value.strip().startswith('-') or '[' in value):
+                value = '[{0}]'.format(value)
+            try:
+                return yaml.safe_load(value)
+            except Exception:
+                logger.exception('Exception when parsing list %s', value)
+                return None
+
+        for first, second in (('raft', 'partner_addrs'), ('restapi', 'allowlist')):
+            value = ret.get(first, {}).pop(second, None)
+            if value:
+                value = _parse_list(value)
+                if value:
+                    ret[first][second] = value
 
         def _parse_dict(value):
             if not value.strip().startswith('{'):
@@ -244,15 +314,17 @@ class Config(object):
                 logger.exception('Exception when parsing dict %s', value)
                 return None
 
-        value = ret.get('log', {}).pop('loggers', None)
+        for first, params in (('restapi', ('http_extra_headers', 'https_extra_headers')), ('log', ('loggers',))):
-        if value:
+            for second in params:
-            value = _parse_dict(value)
+                value = ret.get(first, {}).pop(second, None)
-            if value:
+                if value:
-                ret['log']['loggers'] = value
+                    value = _parse_dict(value)
+                    if value:
+                        ret[first][second] = value
 
-        def _get_auth(name):
+        def _get_auth(name, params=None):
             ret = {}
-            for param in ('username', 'password'):
+            for param in params or _AUTH_ALLOWED_PARAMETERS[:2]:
                 value = _popenv(name + '_' + param)
                 if value:
                     ret[param] = value
@@ -264,30 +336,22 @@ class Config(object):
 
         authentication = {}
         for user_type in ('replication', 'superuser', 'rewind'):
-            entry = _get_auth(user_type)
+            entry = _get_auth(user_type, _AUTH_ALLOWED_PARAMETERS)
             if entry:
                 authentication[user_type] = entry
 
         if authentication:
             ret['postgresql']['authentication'] = authentication
 
-        def _parse_list(value):
-            if not (value.strip().startswith('-') or '[' in value):
-                value = '[{0}]'.format(value)
-            try:
-                return yaml.safe_load(value)
-            except Exception:
-                logger.exception('Exception when parsing list %s', value)
-                return None
-
         for param in list(os.environ.keys()):
-            if param.startswith(Config.PATRONI_ENV_PREFIX):
+            if param.startswith(PATRONI_ENV_PREFIX):
                 # PATRONI_(ETCD|CONSUL|ZOOKEEPER|EXHIBITOR|...)_(HOSTS?|PORT|..)
                 name, suffix = (param[8:].split('_', 1) + [''])[:2]
                 if suffix in ('HOST', 'HOSTS', 'PORT', 'USE_PROXIES', 'PROTOCOL', 'SRV', 'URL', 'PROXY',
                               'CACERT', 'CERT', 'KEY', 'VERIFY', 'TOKEN', 'CHECKS', 'DC', 'CONSISTENCY',
                               'REGISTER_SERVICE', 'SERVICE_CHECK_INTERVAL', 'NAMESPACE', 'CONTEXT',
-                              'USE_ENDPOINTS', 'SCOPE_LABEL', 'ROLE_LABEL', 'POD_IP', 'PORTS', 'LABELS') and name:
+                              'USE_ENDPOINTS', 'SCOPE_LABEL', 'ROLE_LABEL', 'POD_IP', 'PORTS', 'LABELS',
+                              'BYPASS_API_SERVICE', 'KEY_PASSWORD', 'USE_SSL') and name:
                     value = os.environ.pop(param)
                     if suffix == 'PORT':
                         value = value and parse_int(value)
@@ -295,16 +359,17 @@ class Config(object):
                         value = value and _parse_list(value)
                     elif suffix == 'LABELS':
                         value = _parse_dict(value)
-                    elif suffix in ('USE_PROXIES', 'REGISTER_SERVICE'):
+                    elif suffix in ('USE_PROXIES', 'REGISTER_SERVICE', 'USE_ENDPOINTS', 'BYPASS_API_SERVICE', 'VERIFY'):
                         value = parse_bool(value)
                     if value:
                         ret[name.lower()][suffix.lower()] = value
-        if 'etcd' in ret:
+        for dcs in ('etcd', 'etcd3'):
-            ret['etcd'].update(_get_auth('etcd'))
+            if dcs in ret:
+                ret[dcs].update(_get_auth(dcs))
 
         users = {}
         for param in list(os.environ.keys()):
-            if param.startswith(Config.PATRONI_ENV_PREFIX):
+            if param.startswith(PATRONI_ENV_PREFIX):
                 name, suffix = (param[8:].rsplit('_', 1) + [''])[:2]
                 # PATRONI_<username>_PASSWORD=<password>, PATRONI_<username>_OPTIONS=<option1,option2,...>
                 # CREATE USER "<username>" WITH <OPTIONS> PASSWORD '<password>'
@@ -332,14 +397,9 @@ class Config(object):
                         config['postgresql'][name] = deepcopy(value)
             elif name not in config or name in ['watchdog']:
                 config[name] = deepcopy(value) if value else {}
-        
-        if os.getenv("PATRONI_DEBUG_MODE"):
-            config['ttl'] = 24 * 60 * 60 # practical infinity for a debugging session
-            config['loop_wait'] = -1
-
 
         # restapi server expects to get restapi.auth = 'username:password'
-        if 'authentication' in config['restapi']:
+        if 'restapi' in config and 'authentication' in config['restapi']:
             config['restapi']['auth'] = '{username}:{password}'.format(**config['restapi']['authentication'])
 
         # special treatment for old config
@@ -358,6 +418,11 @@ class Config(object):
         if 'superuser' not in pg_config['authentication'] and 'pg_rewind' in pg_config:
             pg_config['authentication']['superuser'] = pg_config['pg_rewind']
 
+        # handle setting additional connection parameters that may be available
+        # in the configuration file, such as SSL connection parameters
+        for name, value in pg_config['authentication'].items():
+            pg_config['authentication'][name] = {n: v for n, v in value.items() if n in _AUTH_ALLOWED_PARAMETERS}
+
         # no 'name' in config
         if 'name' not in config and 'name' in pg_config:
             config['name'] = pg_config['name']
@@ -368,6 +433,8 @@ class Config(object):
             'retry_timeout',
             'synchronous_mode',
             'synchronous_mode_strict',
+            'synchronous_node_count',
+            'maximum_lag_on_syncnode'
         )
 
         pg_config.update({p: config[p] for p in updated_fields if p in config})

patroni/ctl.py

@@ -2,12 +2,11 @@
 Patroni Control
 '''
 
-import base64
 import click
 import codecs
 import datetime
 import dateutil.parser
-import cdiff
+import dateutil.tz
 import copy
 import difflib
 import io
@@ -15,26 +14,31 @@ import json
 import logging
 import os
 import random
-import requests
+import six
 import subprocess
 import sys
 import tempfile
 import time
-import tzlocal
 import yaml
 
 from click import ClickException
+from collections import defaultdict
 from contextlib import contextmanager
-from patroni.config import Config
+from prettytable import ALL, FRAME, PrettyTable
-from patroni.dcs import get_dcs as _get_dcs
-from patroni.exceptions import PatroniException
-from patroni.postgresql import Postgresql
-from patroni.postgresql.misc import postgres_version_to_int
-from patroni.utils import patch_config, polling_loop
-from patroni.version import __version__
-from prettytable import PrettyTable
 from six.moves.urllib_parse import urlparse
-from six import text_type
+
+try:
+    from ydiff import markup_to_pager, PatchStream
+except ImportError:  # pragma: no cover
+    from cdiff import markup_to_pager, PatchStream
+
+from .dcs import get_dcs as _get_dcs
+from .exceptions import PatroniException
+from .postgresql import Postgresql
+from .postgresql.misc import postgres_version_to_int
+from .utils import cluster_as_json, find_executable, patch_config, polling_loop
+from .request import PatroniRequest
+from .version import __version__
 
 CONFIG_DIR_PATH = click.get_app_dir('patroni')
 CONFIG_FILE_PATH = os.path.join(CONFIG_DIR_PATH, 'patronictl.yaml')
@@ -48,14 +52,42 @@ class PatroniCtlException(ClickException):
     pass
 
 
+class PatronictlPrettyTable(PrettyTable):
+
+    def __init__(self, header, *args, **kwargs):
+        PrettyTable.__init__(self, *args, **kwargs)
+        self.__table_header = header
+        self.__hline_num = 0
+        self.__hline = None
+
+    def _is_first_hline(self):
+        return self.__hline_num == 0
+
+    def _set_hline(self, value):
+        self.__hline = value
+
+    def _get_hline(self):
+        ret = self.__hline
+
+        # Inject nice table header
+        if self._is_first_hline() and self.__table_header:
+            header = self.__table_header[:len(ret) - 2]
+            ret = "".join([ret[0], header, ret[1 + len(header):]])
+
+        self.__hline_num += 1
+        return ret
+
+    _hrule = property(_get_hline, _set_hline)
+
+
 def parse_dcs(dcs):
     if dcs is None:
         return None
+    elif '//' not in dcs:
+        dcs = '//' + dcs
 
     parsed = urlparse(dcs)
     scheme = parsed.scheme
-    if scheme == '' and parsed.netloc == '':
-        parsed = urlparse('//' + dcs)
     port = int(parsed.port) if parsed.port else None
 
     if scheme == '':
@@ -68,21 +100,17 @@ def parse_dcs(dcs):
 
 
 def load_config(path, dcs):
+    from patroni.config import Config
+
     if not (os.path.exists(path) and os.access(path, os.R_OK)):
-        logging.debug('Ignoring configuration file "%s". It does not exists or is not readable.', path)
+        if path != CONFIG_FILE_PATH:    # bail if non-default config location specified but file not found / readable
+            raise PatroniCtlException('Provided config file {0} not existing or no read rights.'
+                                      ' Check the -c/--config-file parameter'.format(path))
+        else:
+            logging.debug('Ignoring configuration file "%s". It does not exists or is not readable.', path)
     else:
         logging.debug('Loading configuration from file %s', path)
-    config = {}
+    config = Config(path, validator=None).copy()
-    old_argv = list(sys.argv)
-    try:
-        sys.argv[1] = path
-        if Config.PATRONI_CONFIG_VARIABLE not in os.environ:
-            for p in ('PATRONI_RESTAPI_LISTEN', 'PATRONI_POSTGRESQL_DATA_DIR'):
-                if p not in os.environ:
-                    os.environ[p] = '.'
-        config = Config().copy()
-    finally:
-        sys.argv = old_argv
 
     dcs = parse_dcs(dcs) or parse_dcs(config.get('dcs_api')) or {}
     if dcs:
@@ -100,7 +128,7 @@ def store_config(config, path):
         yaml.dump(config, fd)
 
 
-option_format = click.option('--format', '-f', 'fmt', help='Output format (pretty, json, yaml)', default='pretty')
+option_format = click.option('--format', '-f', 'fmt', help='Output format (pretty, tsv, json, yaml)', default='pretty')
 option_watchrefresh = click.option('-w', '--watch', type=float, help='Auto update the screen every X seconds')
 option_watch = click.option('-W', is_flag=True, help='Auto update the screen every 2 seconds')
 option_force = click.option('--force', is_flag=True, help='Do not ask for confirmation at any point')
@@ -135,63 +163,42 @@ def get_dcs(config, scope):
         raise PatroniCtlException(str(e))
 
 
-def auth_header(config):
+def request_patroni(member, method='GET', endpoint=None, data=None):
-    if config.get('restapi', {}).get('auth', ''):
-        return {'Authorization': 'Basic ' + base64.b64encode(config['restapi']['auth'].encode('utf-8')).decode('utf-8')}
-
-
-def request_patroni(member, request_type, endpoint, content=None, headers=None):
     ctx = click.get_current_context()  # the current click context
-    headers = headers or {}
+    request_executor = ctx.obj.get('__request_patroni')
-    url_parts = urlparse(member.api_url)
+    if not request_executor:
-    logging.debug(url_parts)
+        request_executor = ctx.obj['__request_patroni'] = PatroniRequest(ctx.obj)
-    if 'Content-Type' not in headers:
+    return request_executor(member, method, endpoint, data)
-        headers['Content-Type'] = 'application/json'
-
-    url = '{0}://{1}/{2}'.format(url_parts.scheme, url_parts.netloc, endpoint)
-
-    insecure = ctx.obj.get('ctl', {}).get('insecure', False)
-    # Get certfile if any from several configuration namespace
-    cert = ctx.obj.get('ctl', {}).get('cacert') or \
-        ctx.obj.get('restapi', {}).get('cacert') or \
-        ctx.obj.get('restapi', {}).get('certfile')
-    # In the case we specificaly disable SSL cert verification we don't want to have the warning
-    if insecure:
-        verify = False
-    elif cert:
-        verify = cert
-    else:
-        verify = True
-    return getattr(requests, request_type)(url, headers=headers,
-                                           data=json.dumps(content) if content else None, timeout=60,
-                                           verify=verify)
 
 
-def print_output(columns, rows=None, alignment=None, fmt='pretty', header=True, delimiter='\t'):
+def print_output(columns, rows, alignment=None, fmt='pretty', header=None, delimiter='\t'):
-    rows = rows or []
+    if fmt in {'json', 'yaml', 'yml'}:
-    if fmt == 'pretty':
+        elements = [{k: v for k, v in zip(columns, r) if not header or str(v)} for r in rows]
-        t = PrettyTable(columns)
+        func = json.dumps if fmt == 'json' else format_config_for_editing
-        for k, v in (alignment or {}).items():
+        click.echo(func(elements))
-            t.align[k] = v
+    elif fmt in {'pretty', 'tsv', 'topology'}:
-        for r in rows:
+        list_cluster = bool(header and columns and columns[0] == 'Cluster')
-            t.add_row(r)
+        if list_cluster and 'Tags' in columns:  # we want to format member tags as YAML
-        click.echo(t)
+            i = columns.index('Tags')
-        return
+            for row in rows:
+                if row[i]:
+                    row[i] = format_config_for_editing(row[i], fmt != 'pretty').strip()
+        if list_cluster and fmt != 'tsv':  # skip cluster name if pretty-printing
+            columns = columns[1:] if columns else []
+            rows = [row[1:] for row in rows]
 
-    if fmt in ['json', 'yaml', 'yml']:
+        if fmt == 'tsv':
-        elements = [dict(zip(columns, r)) for r in rows]
+            for r in ([columns] if columns else []) + rows:
-        if fmt == 'json':
+                click.echo(delimiter.join(map(str, r)))
-            click.echo(json.dumps(elements))
+        else:
-        elif fmt in ('yaml', 'yml'):
+            hrules = ALL if any(any(isinstance(c, six.string_types) and '\n' in c for c in r) for r in rows) else FRAME
-            click.echo(yaml.safe_dump(elements, encoding=None, default_flow_style=False, allow_unicode=True, width=200))
+            table = PatronictlPrettyTable(header, columns, hrules=hrules)
-
+            table.align = 'l'
-    if fmt == 'tsv':
+            for k, v in (alignment or {}).items():
-        if columns is not None and header:
+                table.align[k] = v
-            click.echo(delimiter.join(columns))
+            for r in rows:
-
+                table.add_row(r)
-        for r in rows:
+            click.echo(table)
-            c = [str(c) for c in r]
-            click.echo(delimiter.join(c))
 
 
 def watching(w, watch, max_count=None, clear=True):
@@ -224,7 +231,7 @@ def watching(w, watch, max_count=None, clear=True):
 
 def get_all_members(cluster, role='master'):
     if role == 'master':
-        if cluster.leader is not None:
+        if cluster.leader is not None and cluster.leader.name:
             yield cluster.leader
         return
 
@@ -241,6 +248,15 @@ def get_any_member(cluster, role='master', member=None):
             return m
 
 
+def get_all_members_leader_first(cluster):
+    leader_name = cluster.leader.member.name if cluster.leader and cluster.leader.member.api_url else None
+    if leader_name:
+        yield cluster.leader.member
+    for member in cluster.members:
+        if member.api_url and member.name != leader_name:
+            yield member
+
+
 def get_cursor(cluster, connect_parameters, role='master', member=None):
     member = get_any_member(cluster, role=role, member=member)
     if member is None:
@@ -271,10 +287,12 @@ def get_cursor(cluster, connect_parameters, role='master', member=None):
     return None
 
 
-def get_members(cluster, cluster_name, member_names, role, force, action, scheduled_at=None):
+def get_members(cluster, cluster_name, member_names, role, force, action, ask_confirmation=True):
     candidates = {m.name: m for m in cluster.members}
 
     if not force or role:
+        if not member_names and not candidates:
+            raise PatroniCtlException('{0} cluster doesn\'t have any members'.format(cluster_name))
         output_members(cluster, cluster_name)
 
     if role:
@@ -294,21 +312,26 @@ def get_members(cluster, cluster_name, member_names, role, force, action, schedu
         if member_name not in candidates:
             raise PatroniCtlException('{0} is not a member of cluster'.format(member_name))
 
+    members = [candidates[n] for n in member_names]
+    if ask_confirmation:
+        confirm_members_action(members, force, action)
+    return members
+
+
+def confirm_members_action(members, force, action, scheduled_at=None):
     if scheduled_at:
         if not force:
             confirm = click.confirm('Are you sure you want to schedule {0} of members {1} at {2}?'
-                                    .format(action, ', '.join(member_names), scheduled_at))
+                                    .format(action, ', '.join([m.name for m in members]), scheduled_at))
             if not confirm:
                 raise PatroniCtlException('Aborted scheduled {0}'.format(action))
     else:
         if not force:
             confirm = click.confirm('Are you sure you want to {0} members {1}?'
-                                    .format(action, ', '.join(member_names)))
+                                    .format(action, ', '.join([m.name for m in members])))
             if not confirm:
                 raise PatroniCtlException('Aborted {0}'.format(action))
 
-    return [candidates[n] for n in member_names]
-
 
 @ctl.command('dsn', help='Generate a dsn for the provided member, defaults to a dsn of the master')
 @click.option('--role', '-r', help='Give a dsn of any member with this role', type=click.Choice(['master', 'replica',
@@ -333,8 +356,7 @@ def dsn(obj, cluster_name, role, member):
 
 @ctl.command('query', help='Query a Patroni PostgreSQL member')
 @arg_cluster_name
-@option_format
+@click.option('--format', 'fmt', help='Output format (pretty, tsv, json, yaml)', default='tsv')
-@click.option('--format', 'fmt', help='Output format (pretty, json)', default='tsv')
 @click.option('--file', '-f', 'p_file', help='Execute the SQL commands from this file', type=click.File('rb'))
 @click.option('--password', help='force password prompt', is_flag=True)
 @click.option('-U', '--username', help='database user name', type=str)
@@ -391,8 +413,8 @@ def query(
         if cursor is None:
             cluster = dcs.get_cluster()
 
-        output, cursor = query_member(cluster, cursor, member, role, command, connect_parameters)
+        output, header = query_member(cluster, cursor, member, role, command, connect_parameters)
-        print_output(None, output, fmt=fmt, delimiter=delimiter)
+        print_output(header, output, fmt=fmt, delimiter=delimiter)
 
 
 def query_member(cluster, cursor, member, role, command, connect_parameters):
@@ -409,15 +431,8 @@ def query_member(cluster, cursor, member, role, command, connect_parameters):
             logging.debug(message)
             return [[timestamp(0), message]], None
 
-        cursor.execute('SELECT pg_catalog.pg_is_in_recovery()')
-        in_recovery = cursor.fetchone()[0]
-
-        if in_recovery and role == 'master' or not in_recovery and role == 'replica':
-            cursor.connection.close()
-            return None, None
-
         cursor.execute(command)
-        return cursor.fetchall(), cursor
+        return cursor.fetchall(), [d.name for d in cursor.description]
     except (psycopg2.OperationalError, psycopg2.DatabaseError) as oe:
         logging.debug(oe)
         if cursor is not None and not cursor.connection.closed:
@@ -457,9 +472,9 @@ def remove(obj, cluster_name, fmt):
 
 
 def check_response(response, member_name, action_name, silent_success=False):
-    if response.status_code >= 400:
+    if response.status >= 400:
         click.echo('Failed: {0} for member {1}, status code={2}, ({3})'.format(
-            action_name, member_name, response.status_code, response.text
+            action_name, member_name, response.status, response.data.decode('utf-8')
         ))
         return False
     elif not silent_success:
@@ -472,7 +487,7 @@ def parse_scheduled(scheduled):
         try:
             scheduled_at = dateutil.parser.parse(scheduled)
             if scheduled_at.tzinfo is None:
-                scheduled_at = tzlocal.get_localzone().localize(scheduled_at)
+                scheduled_at = scheduled_at.replace(tzinfo=dateutil.tz.tzlocal())
         except (ValueError, TypeError):
             message = 'Unable to parse scheduled timestamp ({0}). It should be in an unambiguous format (e.g. ISO 8601)'
             raise PatroniCtlException(message.format(scheduled))
@@ -493,18 +508,17 @@ def reload(obj, cluster_name, member_names, force, role):
 
     members = get_members(cluster, cluster_name, member_names, role, force, 'reload')
 
-    content = {}
     for member in members:
-        r = request_patroni(member, 'post', 'reload', content, auth_header(obj))
+        r = request_patroni(member, 'post', 'reload')
-        if r.status_code == 200:
+        if r.status == 200:
             click.echo('No changes to apply on member {0}'.format(member.name))
-        elif r.status_code == 202:
+        elif r.status == 202:
             click.echo('Reload request received for member {0} and will be processed within {1} seconds'.format(
                 member.name, cluster.config.data.get('loop_wait'))
             )
         else:
             click.echo('Failed: reload for member {0}, status code={1}, ({2})'.format(
-                member.name, r.status_code, r.text)
+                member.name, r.status, r.data.decode('utf-8'))
             )
 
 
@@ -526,14 +540,15 @@ def reload(obj, cluster_name, member_names, force, role):
 def restart(obj, cluster_name, member_names, force, role, p_any, scheduled, version, pending, timeout):
     cluster = get_dcs(obj, cluster_name).get_cluster()
 
+    members = get_members(cluster, cluster_name, member_names, role, force, 'restart', False)
     if scheduled is None and not force:
         next_hour = (datetime.datetime.now() + datetime.timedelta(hours=1)).strftime('%Y-%m-%dT%H:%M')
         scheduled = click.prompt('When should the restart take place (e.g. ' + next_hour + ') ',
                                  type=str, default='now')
 
     scheduled_at = parse_scheduled(scheduled)
+    confirm_members_action(members, force, 'restart', scheduled_at)
 
-    members = get_members(cluster, cluster_name, member_names, role, force, 'restart', scheduled_at)
     if p_any:
         random.shuffle(members)
         members = members[:1]
@@ -565,19 +580,19 @@ def restart(obj, cluster_name, member_names, force, role, p_any, scheduled, vers
     for member in members:
         if 'schedule' in content:
             if force and member.data.get('scheduled_restart'):
-                r = request_patroni(member, 'delete', 'restart', headers=auth_header(obj))
+                r = request_patroni(member, 'delete', 'restart')
                 check_response(r, member.name, 'flush scheduled restart', True)
 
-        r = request_patroni(member, 'post', 'restart', content, auth_header(obj))
+        r = request_patroni(member, 'post', 'restart', content)
-        if r.status_code == 200:
+        if r.status == 200:
             click.echo('Success: restart on member {0}'.format(member.name))
-        elif r.status_code == 202:
+        elif r.status == 202:
             click.echo('Success: restart scheduled on member {0}'.format(member.name))
-        elif r.status_code == 409:
+        elif r.status == 409:
             click.echo('Failed: another restart is already scheduled on member {0}'.format(member.name))
         else:
             click.echo('Failed: restart for member {0}, status code={1}, ({2})'.format(
-                member.name, r.status_code, r.text)
+                member.name, r.status, r.data.decode('utf-8'))
             )
 
 
@@ -585,20 +600,39 @@ def restart(obj, cluster_name, member_names, force, role, p_any, scheduled, vers
 @click.argument('cluster_name')
 @click.argument('member_names', nargs=-1)
 @option_force
+@click.option('--wait', help='Wait until reinitialization completes', is_flag=True)
 @click.pass_obj
-def reinit(obj, cluster_name, member_names, force):
+def reinit(obj, cluster_name, member_names, force, wait):
     cluster = get_dcs(obj, cluster_name).get_cluster()
     members = get_members(cluster, cluster_name, member_names, None, force, 'reinitialize')
 
+    wait_on_members = []
     for member in members:
         body = {'force': force}
         while True:
-            r = request_patroni(member, 'post', 'reinitialize', body, auth_header(obj))
+            r = request_patroni(member, 'post', 'reinitialize', body)
-            if not check_response(r, member.name, 'reinitialize') and r.text.endswith(' already in progress') \
+            started = check_response(r, member.name, 'reinitialize')
+            if not started and r.data.endswith(b' already in progress') \
                     and not force and click.confirm('Do you want to cancel it and reinitialize anyway?'):
                 body['force'] = True
                 continue
             break
+        if started and wait:
+            wait_on_members.append(member)
+
+    last_display = []
+    while wait_on_members:
+        if wait_on_members != last_display:
+            click.echo('Waiting for reinitialize to complete on: {0}'.format(
+                ", ".join(member.name for member in wait_on_members))
+            )
+            last_display[:] = wait_on_members
+        time.sleep(2)
+        for member in wait_on_members:
+            data = json.loads(request_patroni(member, 'get', 'patroni').data.decode('utf-8'))
+            if data.get('state') != 'creating replica':
+                click.echo('Reinitialize is completed on: {0}'.format(member.name))
+                wait_on_members.remove(member)
 
 
 def _do_failover_or_switchover(obj, action, cluster_name, master, candidate, force, scheduled=None):
@@ -612,7 +646,7 @@ def _do_failover_or_switchover(obj, action, cluster_name, master, candidate, for
     dcs = get_dcs(obj, cluster_name)
     cluster = dcs.get_cluster()
 
-    if action == 'switchover' and cluster.leader is None:
+    if action == 'switchover' and (cluster.leader is None or not cluster.leader.name):
         raise PatroniCtlException('This cluster has no master')
 
     if master is None:
@@ -682,19 +716,19 @@ def _do_failover_or_switchover(obj, action, cluster_name, master, candidate, for
     try:
         member = cluster.leader.member if cluster.leader else cluster.get_member(candidate, False)
 
-        r = request_patroni(member, 'post', action, failover_value, auth_header(obj))
+        r = request_patroni(member, 'post', action, failover_value)
 
         # probably old patroni, which doesn't support switchover yet
-        if r.status_code == 501 and action == 'switchover' and 'Server does not support this operation' in r.text:
+        if r.status == 501 and action == 'switchover' and b'Server does not support this operation' in r.data:
-            r = request_patroni(member, 'post', 'failover', failover_value, auth_header(obj))
+            r = request_patroni(member, 'post', 'failover', failover_value)
 
-        if r.status_code in (200, 202):
+        if r.status in (200, 202):
             logging.debug(r)
             cluster = dcs.get_cluster()
             logging.debug(cluster)
-            click.echo('{0} {1}'.format(timestamp(), r.text))
+            click.echo('{0} {1}'.format(timestamp(), r.data.decode('utf-8')))
         else:
-            click.echo('{0} failed, details: {1}, {2}'.format(action.title(), r.status_code, r.text))
+            click.echo('{0} failed, details: {1}, {2}'.format(action.title(), r.status, r.data.decode('utf-8')))
             return
     except Exception:
         logging.exception(r)
@@ -728,83 +762,84 @@ def switchover(obj, cluster_name, master, candidate, force, scheduled):
     _do_failover_or_switchover(obj, 'switchover', cluster_name, master, candidate, force, scheduled)
 
 
+def generate_topology(level, member, topology):
+    members = topology.get(member['name'], [])
+
+    if level > 0:
+        member['name'] = '{0}+ {1}'.format((' ' * (level - 1) * 2), member['name'])
+
+    if member['name']:
+        yield member
+
+    for member in members:
+        for member in generate_topology(level + 1, member, topology):
+            yield member
+
+
+def topology_sort(members):
+    topology = defaultdict(list)
+    leader = next((m for m in members if m['role'].endswith('leader')), {'name': None})
+    replicas = set(member['name'] for member in members if not member['role'].endswith('leader'))
+    for member in members:
+        if not member['role'].endswith('leader'):
+            parent = member.get('tags', {}).get('replicatefrom')
+            parent = parent if parent and parent != member['name'] and parent in replicas else leader['name']
+            topology[parent].append(member)
+    for member in generate_topology(0, leader, topology):
+        yield member
+
+
 def output_members(cluster, name, extended=False, fmt='pretty'):
     rows = []
     logging.debug(cluster)
-    leader_name = None
+    initialize = {None: 'uninitialized', '': 'initializing'}.get(cluster.initialize, cluster.initialize)
-    if cluster.leader:
+    cluster = cluster_as_json(cluster)
-        leader_name = cluster.leader.name
-
-    xlog_location_cluster = cluster.last_leader_operation or 0
-
-    # Mainly for consistent pretty printing and watching we sort the output
-    cluster.members.sort(key=lambda x: x.name)
-
-    has_scheduled_restarts = any(m.data.get('scheduled_restart') for m in cluster.members)
-    has_pending_restarts = any(m.data.get('pending_restart') for m in cluster.members)
-
-    # Show Host as 'host:port' if somebody is running on non-standard port or two nodes are running on the same host
-    append_port = any(str(m.conn_kwargs()['port']) != '5432' for m in cluster.members) or\
-        len(set(m.conn_kwargs()['host'] for m in cluster.members)) < len(cluster.members)
-
-    for m in cluster.members:
-        logging.debug(m)
-
-        role = ''
-        if m.name == leader_name:
-            role = 'Leader'
-        elif m.name == cluster.sync.sync_standby:
-            role = 'Sync standby'
-
-        xlog_location = m.data.get('xlog_location')
-        lag = ''
-        if xlog_location is None:
-            lag = 'unknown'
-        elif xlog_location_cluster >= xlog_location:
-            lag = round((xlog_location_cluster - xlog_location)/1024/1024)
-
-        host = m.conn_kwargs()['host']
-        if append_port:
-            host += ':{0}'.format(m.conn_kwargs()['port'])
-
-        row = [name, m.name, host, role, m.data.get('state', ''), m.data.get('timeline', ''), lag]
-
-        if extended or has_pending_restarts:
-            row.append('*' if m.data.get('pending_restart') else '')
-
-        if extended or has_scheduled_restarts:
-            value = ''
-            scheduled_restart = m.data.get('scheduled_restart')
-            if scheduled_restart:
-                value = scheduled_restart['schedule']
-                if 'postgres_version' in scheduled_restart:
-                    value += ' if version < {0}'.format(scheduled_restart['postgres_version'])
-
-            row.append(value)
-
-        rows.append(row)
 
     columns = ['Cluster', 'Member', 'Host', 'Role', 'State', 'TL', 'Lag in MB']
-    alignment = {'Lag in MB': 'r', 'TL': 'r'}
+    for c in ('Pending restart', 'Scheduled restart', 'Tags'):
+        if extended or any(m.get(c.lower().replace(' ', '_')) for m in cluster['members']):
+            columns.append(c)
 
-    if extended or has_pending_restarts:
+    # Show Host as 'host:port' if somebody is running on non-standard port or two nodes are running on the same host
-        columns.append('Pending restart')
+    members = [m for m in cluster['members'] if 'host' in m]
+    append_port = any('port' in m and m['port'] != 5432 for m in members) or\
+        len(set(m['host'] for m in members)) < len(members)
 
-    if extended or has_scheduled_restarts:
+    sort = topology_sort if fmt == 'topology' else iter
-        columns.append('Scheduled restart')
+    for m in sort(cluster['members']):
+        logging.debug(m)
 
-    print_output(columns, rows, alignment, fmt)
+        lag = m.get('lag', '')
+        m.update(cluster=name, member=m['name'], host=m.get('host', ''), tl=m.get('timeline', ''),
+                 role=m['role'].replace('_', ' ').title(),
+                 lag_in_mb=round(lag/1024/1024) if isinstance(lag, six.integer_types) else lag,
+                 pending_restart='*' if m.get('pending_restart') else '')
+
+        if append_port and m['host'] and m.get('port'):
+            m['host'] = ':'.join([m['host'], str(m['port'])])
+
+        if 'scheduled_restart' in m:
+            value = m['scheduled_restart']['schedule']
+            if 'postgres_version' in m['scheduled_restart']:
+                value += ' if version < {0}'.format(m['scheduled_restart']['postgres_version'])
+            m['scheduled_restart'] = value
+
+        rows.append([m.get(n.lower().replace(' ', '_'), '') for n in columns])
+
+    print_output(columns, rows, {'Lag in MB': 'r', 'TL': 'r'}, fmt, ' Cluster: {0} ({1}) '.format(name, initialize))
+
+    if fmt not in ('pretty', 'topology'):  # Omit service info when using machine-readable formats
+        return
 
     service_info = []
-    if cluster.is_paused():
+    if cluster.get('pause'):
         service_info.append('Maintenance mode: on')
 
-    if cluster.failover and cluster.failover.scheduled_at:
+    if 'scheduled_switchover' in cluster:
-        info = 'Switchover scheduled at: ' + cluster.failover.scheduled_at.isoformat()
+        info = 'Switchover scheduled at: ' + cluster['scheduled_switchover']['at']
-        if cluster.failover.leader:
+        for name in ('from', 'to'):
-            info += '\n                    from: ' + cluster.failover.leader
+            if name in cluster['scheduled_switchover']:
-        if cluster.failover.candidate:
+                info += '\n{0:>24}: {1}'.format(name, cluster['scheduled_switchover'][name])
-            info += '\n                      to: ' + cluster.failover.candidate
         service_info.append(info)
 
     if service_info:
@@ -837,6 +872,16 @@ def members(obj, cluster_names, fmt, watch, w, extended, ts):
             output_members(cluster, cluster_name, extended, fmt)
 
 
+@ctl.command('topology', help='Prints ASCII topology for given cluster')
+@click.argument('cluster_names', nargs=-1)
+@option_watch
+@option_watchrefresh
+@click.pass_obj
+@click.pass_context
+def topology(ctx, obj, cluster_names, watch, w):
+    ctx.forward(members, fmt='topology')
+
+
 def timestamp(precision=6):
     return datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S.%f')[:precision - 7]
 
@@ -904,25 +949,45 @@ def scaffold(obj, cluster_name, sysid):
     click.echo("Cluster {0} has been created successfully".format(cluster_name))
 
 
-@ctl.command('flush', help='Flush scheduled events')
+@ctl.command('flush', help='Discard scheduled events')
 @click.argument('cluster_name')
 @click.argument('member_names', nargs=-1)
-@click.argument('target', type=click.Choice(['restart']))
+@click.argument('target', type=click.Choice(['restart', 'switchover']))
 @click.option('--role', '-r', help='Flush only members with this role', default='any',
               type=click.Choice(['master', 'replica', 'any']))
 @option_force
 @click.pass_obj
 def flush(obj, cluster_name, member_names, force, role, target):
-    cluster = get_dcs(obj, cluster_name).get_cluster()
+    dcs = get_dcs(obj, cluster_name)
+    cluster = dcs.get_cluster()
 
-    members = get_members(cluster, cluster_name, member_names, role, force, 'flush')
+    if target == 'restart':
-    for member in members:
+        for member in get_members(cluster, cluster_name, member_names, role, force, 'flush'):
-        if target == 'restart':
             if member.data.get('scheduled_restart'):
-                r = request_patroni(member, 'delete', 'restart', None, auth_header(obj))
+                r = request_patroni(member, 'delete', 'restart')
                 check_response(r, member.name, 'flush scheduled restart')
             else:
                 click.echo('No scheduled restart for member {0}'.format(member.name))
+    elif target == 'switchover':
+        failover = cluster.failover
+        if not failover or not failover.scheduled_at:
+            return click.echo('No pending scheduled switchover')
+        for member in get_all_members_leader_first(cluster):
+            try:
+                r = request_patroni(member, 'delete', 'switchover')
+                if r.status in (200, 404):
+                    prefix = 'Success' if r.status == 200 else 'Failed'
+                    return click.echo('{0}: {1}'.format(prefix, r.data.decode('utf-8')))
+            except Exception as err:
+                logging.warning(str(err))
+                logging.warning('Member %s is not accessible', member.name)
+
+            click.echo('Failed: member={0}, status_code={1}, ({2})'.format(
+                member.name, r.status, r.data.decode('utf-8')))
+
+        logging.warning('Failing over to DCS')
+        click.echo('{0} Could not find any accessible member of cluster {1}'.format(timestamp(), cluster_name))
+        dcs.manual_failover('', '', index=failover.index)
 
 
 def wait_until_pause_is_applied(dcs, paused, old_cluster):
@@ -949,27 +1014,22 @@ def toggle_pause(config, cluster_name, paused, wait):
     if cluster.is_paused() == paused:
         raise PatroniCtlException('Cluster is {0} paused'.format(paused and 'already' or 'not'))
 
-    members = []
+    for member in get_all_members_leader_first(cluster):
-    if cluster.leader:
-        members.append(cluster.leader.member)
-    members.extend([m for m in cluster.members if m.api_url and (not members or members[0].name != m.name)])
-
-    for member in members:
         try:
-            r = request_patroni(member, 'patch', 'config', {'pause': paused or None}, auth_header(config))
+            r = request_patroni(member, 'patch', 'config', {'pause': paused or None})
         except Exception as err:
             logging.warning(str(err))
             logging.warning('Member %s is not accessible', member.name)
             continue
 
-        if r.status_code == 200:
+        if r.status == 200:
             if wait:
                 wait_until_pause_is_applied(dcs, paused, cluster)
             else:
                 click.echo('Success: cluster management is {0}'.format(paused and 'paused' or 'resumed'))
         else:
             click.echo('Failed: {0} cluster management status code={1}, ({2})'.format(
-                       paused and 'pause' or 'resume', r.status_code, r.text))
+                       paused and 'pause' or 'resume', r.status, r.data.decode('utf-8')))
         break
     else:
         raise PatroniCtlException('Can not find accessible cluster member')
@@ -1016,7 +1076,7 @@ def show_diff(before_editing, after_editing):
     If the output is to a tty the diff will be colored. Inputs are expected to be unicode strings.
     """
     def listify(string):
-        return [l+'\n' for l in string.rstrip('\n').split('\n')]
+        return [line + '\n' for line in string.rstrip('\n').split('\n')]
 
     unified_diff = difflib.unified_diff(listify(before_editing), listify(after_editing))
 
@@ -1024,25 +1084,32 @@ def show_diff(before_editing, after_editing):
         buf = io.StringIO()
         for line in unified_diff:
             # Force cast to unicode as difflib on Python 2.7 returns a mix of unicode and str.
-            buf.write(text_type(line))
+            buf.write(six.text_type(line))
         buf.seek(0)
 
         class opts:
             side_by_side = False
             width = 80
             tab_width = 8
-        cdiff.markup_to_pager(cdiff.PatchStream(buf), opts)
+            wrap = True
+            if find_executable('less'):
+                pager = None
+            else:
+                pager = 'more.com' if sys.platform == 'win32' else 'more'
+            pager_options = None
+
+        markup_to_pager(PatchStream(buf), opts)
     else:
         for line in unified_diff:
             click.echo(line.rstrip('\n'))
 
 
-def format_config_for_editing(data):
+def format_config_for_editing(data, default_flow_style=False):
     """Formats configuration as YAML for human consumption.
 
     :param data: configuration as nested dictionaries
     :returns unicode YAML of the configuration"""
-    return yaml.safe_dump(data, default_flow_style=False, encoding=None, allow_unicode=True)
+    return yaml.safe_dump(data, default_flow_style=default_flow_style, encoding=None, allow_unicode=True, width=200)
 
 
 def apply_config_changes(before_editing, data, kvpairs):
@@ -1106,24 +1173,6 @@ def apply_yaml_file(data, filename):
     return format_config_for_editing(changed_data), changed_data
 
 
-def find_executable(executable, path=None):
-    _, ext = os.path.splitext(executable)
-
-    if (sys.platform == 'win32') and (ext != '.exe'):
-        executable = executable + '.exe'
-
-    if os.path.isfile(executable):
-        return executable
-
-    if path is None:
-        path = os.environ.get('PATH', os.defpath)
-
-    for p in path.split(os.pathsep):
-        f = os.path.join(p, executable)
-        if os.path.isfile(f):
-            return f
-
-
 def invoke_editor(before_editing, cluster_name):
     """Starts editor command to edit configuration in human readable format
 
@@ -1233,8 +1282,8 @@ def version(obj, cluster_name, member_names):
         if m.api_url:
             if not member_names or m.name in member_names:
                 try:
-                    response = request_patroni(m, 'get', 'patroni')
+                    response = request_patroni(m)
-                    data = response.json()
+                    data = json.loads(response.data.decode('utf-8'))
                     version = data.get('patroni', {}).get('version')
                     pg_version = data.get('server_version')
                     pg_version_str = " PostgreSQL {0}".format(format_pg_version(pg_version)) if pg_version else ""
@@ -1243,6 +1292,19 @@ def version(obj, cluster_name, member_names):
                     click.echo("{0}: failed to get version: {1}".format(m.name, e))
 
 
+@ctl.command('history', help="Show the history of failovers/switchovers")
+@arg_cluster_name
+@option_format
+@click.pass_obj
+def history(obj, cluster_name, fmt):
+    cluster = get_dcs(obj, cluster_name).get_cluster()
+    history = cluster.history and cluster.history.lines or []
+    for line in history:
+        if len(line) < 4:
+            line.append('')
+    print_output(['TL', 'LSN', 'Reason', 'Timestamp'], history, {'TL': 'r', 'LSN': 'r'}, fmt)
+
+
 def format_pg_version(version):
     if version < 100000:
         return "{0}.{1}.{2}".format(version // 10000, version // 100 % 100, version % 100)

patroni/daemon.py

@@ -0,0 +1,104 @@
+import abc
+import os
+import signal
+import six
+import sys
+
+from threading import Lock
+
+
+@six.add_metaclass(abc.ABCMeta)
+class AbstractPatroniDaemon(object):
+
+    def __init__(self, config):
+        from patroni.log import PatroniLogger
+
+        self.setup_signal_handlers()
+
+        self.logger = PatroniLogger()
+        self.config = config
+        AbstractPatroniDaemon.reload_config(self, local=True)
+
+    def sighup_handler(self, *args):
+        self._received_sighup = True
+
+    def sigterm_handler(self, *args):
+        with self._sigterm_lock:
+            if not self._received_sigterm:
+                self._received_sigterm = True
+                sys.exit()
+
+    def setup_signal_handlers(self):
+        self._received_sighup = False
+        self._sigterm_lock = Lock()
+        self._received_sigterm = False
+        if os.name != 'nt':
+            signal.signal(signal.SIGHUP, self.sighup_handler)
+        signal.signal(signal.SIGTERM, self.sigterm_handler)
+
+    @property
+    def received_sigterm(self):
+        with self._sigterm_lock:
+            return self._received_sigterm
+
+    def reload_config(self, sighup=False, local=False):
+        if local:
+            self.logger.reload_config(self.config.get('log', {}))
+
+    @abc.abstractmethod
+    def _run_cycle(self):
+        """_run_cycle"""
+
+    def run(self):
+        self.logger.start()
+        while not self.received_sigterm:
+            if self._received_sighup:
+                self._received_sighup = False
+                self.reload_config(True, self.config.reload_local_configuration())
+
+            self._run_cycle()
+
+    @abc.abstractmethod
+    def _shutdown(self):
+        """_shutdown"""
+
+    def shutdown(self):
+        with self._sigterm_lock:
+            self._received_sigterm = True
+        self._shutdown()
+        self.logger.shutdown()
+
+
+def abstract_main(cls, validator=None):
+    import argparse
+
+    from .config import Config, ConfigParseError
+    from .version import __version__
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--version', action='version', version='%(prog)s {0}'.format(__version__))
+    if validator:
+        parser.add_argument('--validate-config', action='store_true', help='Run config validator and exit')
+    parser.add_argument('configfile', nargs='?', default='',
+                        help='Patroni may also read the configuration from the {0} environment variable'
+                        .format(Config.PATRONI_CONFIG_VARIABLE))
+    args = parser.parse_args()
+    try:
+        if validator and args.validate_config:
+            Config(args.configfile, validator=validator)
+            sys.exit()
+
+        config = Config(args.configfile)
+    except ConfigParseError as e:
+        if e.value:
+            print(e.value)
+        parser.print_help()
+        sys.exit(1)
+
+    controller = cls(config)
+    try:
+        controller.run()
+    except KeyboardInterrupt:
+        pass
+    finally:
+        controller.shutdown()

patroni/dcs/__init__.py

@@ -13,12 +13,13 @@ import time
 
 from collections import defaultdict, namedtuple
 from copy import deepcopy
-from patroni.exceptions import PatroniException
-from patroni.utils import parse_bool, uri
 from random import randint
 from six.moves.urllib_parse import urlparse, urlunparse, parse_qsl
 from threading import Event, Lock
 
+from ..exceptions import PatroniFatalException
+from ..utils import deep_compare, parse_bool, uri
+
 slot_name_re = re.compile('^[a-z0-9_]{1,63}$')
 logger = logging.getLogger(__name__)
 
@@ -66,33 +67,48 @@ def dcs_modules():
     module_prefix = __package__ + '.'
 
     if getattr(sys, 'frozen', False):
-        importer = pkgutil.get_importer(dcs_dirname)
+        toc = set()
-        return [module for module in list(importer.toc) if module.startswith(module_prefix) and module.count('.') == 2]
+        # dcs_dirname may contain a dot, which causes pkgutil.iter_importers()
+        # to misinterpret the path as a package name. This can be avoided
+        # altogether by not passing a path at all, because PyInstaller's
+        # FrozenImporter is a singleton and registered as top-level finder.
+        for importer in pkgutil.iter_importers():
+            if hasattr(importer, 'toc'):
+                toc |= importer.toc
+        return [module for module in toc if module.startswith(module_prefix) and module.count('.') == 2]
     else:
         return [module_prefix + name for _, name, is_pkg in pkgutil.iter_modules([dcs_dirname]) if not is_pkg]
 
 
 def get_dcs(config):
-    available_implementations = set()
+    modules = dcs_modules()
-    for module_name in dcs_modules():
+
-        try:
+    for module_name in modules:
-            module = importlib.import_module(module_name)
+        name = module_name.split('.')[-1]
-            for name in filter(lambda name: not name.startswith('__'), dir(module)):  # iterate through module content
+        if name in config:  # we will try to import only modules which have configuration section in the config file
-                item = getattr(module, name)
+            try:
-                name = name.lower()
+                module = importlib.import_module(module_name)
-                # try to find implementation of AbstractDCS interface, class name must match with module_name
+                for key, item in module.__dict__.items():  # iterate through the module content
-                if inspect.isclass(item) and issubclass(item, AbstractDCS) and __package__ + '.' + name == module_name:
+                    # try to find implementation of AbstractDCS interface, class name must match with module_name
-                    available_implementations.add(name)
+                    if key.lower() == name and inspect.isclass(item) and issubclass(item, AbstractDCS):
-                    if name in config:  # which has configuration section in the config file
                         # propagate some parameters
                         config[name].update({p: config[p] for p in ('namespace', 'name', 'scope', 'loop_wait',
                                              'patronictl', 'ttl', 'retry_timeout') if p in config})
                         return item(config[name])
+            except ImportError:
+                logger.debug('Failed to import %s', module_name)
+
+    available_implementations = []
+    for module_name in modules:
+        name = module_name.split('.')[-1]
+        try:
+            module = importlib.import_module(module_name)
+            available_implementations.extend(name for key, item in module.__dict__.items() if key.lower() == name
+                                             and inspect.isclass(item) and issubclass(item, AbstractDCS))
         except ImportError:
-            if not config.get('patronictl'):
+            logger.info('Failed to import %s', module_name)
-                logger.info('Failed to import %s', module_name)
+    raise PatroniFatalException("""Can not find suitable configuration of distributed configuration store
-    raise PatroniException("""Can not find suitable configuration of distributed configuration store
+Available implementations: """ + ', '.join(sorted(set(available_implementations))))
-Available implementations: """ + ', '.join(available_implementations))
 
 
 class Member(namedtuple('Member', 'index,name,session,data')):
@@ -122,6 +138,8 @@ class Member(namedtuple('Member', 'index,name,session,data')):
         else:
             try:
                 data = json.loads(data)
+                if not isinstance(data, dict):
+                    data = {}
             except (TypeError, ValueError):
                 data = {}
         return Member(index, name, session, data)
@@ -129,10 +147,10 @@ class Member(namedtuple('Member', 'index,name,session,data')):
     @property
     def conn_url(self):
         conn_url = self.data.get('conn_url')
-        conn_kwargs = self.data.get('conn_kwargs')
         if conn_url:
             return conn_url
 
+        conn_kwargs = self.data.get('conn_kwargs')
         if conn_kwargs:
             conn_url = uri('postgresql', (conn_kwargs.get('host'), conn_kwargs.get('port', 5432)))
             self.data['conn_url'] = conn_url
@@ -140,16 +158,19 @@ class Member(namedtuple('Member', 'index,name,session,data')):
 
     def conn_kwargs(self, auth=None):
         defaults = {
-            "host": "",
+            "host": None,
-            "port": "",
+            "port": None,
-            "database": ""
+            "database": None
         }
         ret = self.data.get('conn_kwargs')
         if ret:
             defaults.update(ret)
             ret = defaults
         else:
-            r = urlparse(self.conn_url)
+            conn_url = self.conn_url
+            if not conn_url:
+                return {}  # due to the invalid conn_url we don't care about authentication parameters
+            r = urlparse(conn_url)
             ret = {
                 'host': r.hostname,
                 'port': r.port or 5432,
@@ -157,11 +178,11 @@ class Member(namedtuple('Member', 'index,name,session,data')):
             }
             self.data['conn_kwargs'] = ret.copy()
 
+        # apply any remaining authentication parameters
         if auth and isinstance(auth, dict):
+            ret.update({k: v for k, v in auth.items() if v is not None})
             if 'username' in auth:
-                ret['user'] = auth['username']
+                ret['user'] = ret.pop('username')
-            if 'password' in auth:
-                ret['password'] = auth['password']
         return ret
 
     @property
@@ -192,6 +213,15 @@ class Member(namedtuple('Member', 'index,name,session,data')):
     def is_running(self):
         return self.state == 'running'
 
+    @property
+    def version(self):
+        version = self.data.get('version')
+        if version:
+            try:
+                return tuple(map(int, version.split('.')))
+            except Exception:
+                logger.debug('Failed to parse Patroni version %s', version)
+
 
 class RemoteMember(Member):
     """ Represents a remote master for a standby cluster
@@ -232,23 +262,23 @@ class Leader(namedtuple('Leader', 'index,session,member')):
     def conn_url(self):
         return self.member.conn_url
 
+    @property
+    def data(self):
+        return self.member.data
+
     @property
     def timeline(self):
-        return self.member.data.get('timeline')
+        return self.data.get('timeline')
 
     @property
     def checkpoint_after_promote(self):
         """
         >>> Leader(1, '', Member.from_node(1, '', '', '{"version":"z"}')).checkpoint_after_promote
         """
-        version = self.member.data.get('version')
+        version = self.member.version
-        if version:
+        # 1.5.6 is the last version which doesn't expose checkpoint_after_promote: false
-            try:
+        if version and version > (1, 5, 6):
-                # 1.5.6 is the last version which doesn't expose checkpoint_after_promote: false
+            return self.data.get('role') == 'master' and 'checkpoint_after_promote' not in self.data
-                if tuple(map(int, version.split('.'))) > (1, 5, 6):
-                    return self.member.data['role'] == 'master' and 'checkpoint_after_promote' not in self.member.data
-            except Exception:
-                logger.debug('Failed to parse Patroni version %s', version)
 
 
 class Failover(namedtuple('Failover', 'index,leader,candidate,scheduled_at')):
@@ -323,13 +353,21 @@ class ClusterConfig(namedtuple('ClusterConfig', 'index,data,modify_index')):
                 self.data.get('permanent_slots') or self.data.get('slots')
         ) or {}
 
+    @property
+    def ignore_slots_matchers(self):
+        return isinstance(self.data, dict) and self.data.get('ignore_slots') or []
+
+    @property
+    def max_timelines_history(self):
+        return self.data.get('max_timelines_history', 0)
+
 
 class SyncState(namedtuple('SyncState', 'index,leader,sync_standby')):
     """Immutable object (namedtuple) which represents last observed synhcronous replication state
 
     :param index: modification index of a synchronization key in a Configuration Store
     :param leader: reference to member that was leader
-    :param sync_standby: standby that was last synchronized to leader
+    :param sync_standby: synchronous standby list (comma delimited) which are last synchronized to leader
     """
 
     @staticmethod
@@ -361,15 +399,22 @@ class SyncState(namedtuple('SyncState', 'index,leader,sync_standby')):
             data = {}
         return SyncState(index, data.get('leader'), data.get('sync_standby'))
 
+    @property
+    def members(self):
+        """ Returns sync_standby in list """
+        return self.sync_standby and self.sync_standby.split(',') or []
+
     def matches(self, name):
         """
         Returns if a node name matches one of the nodes in the sync state
 
-        >>> s = SyncState(1, 'foo', 'bar')
+        >>> s = SyncState(1, 'foo', 'bar,zoo')
         >>> s.matches('foo')
         True
         >>> s.matches('bar')
         True
+        >>> s.matches('zoo')
+        True
         >>> s.matches('baz')
         False
         >>> s.matches(None)
@@ -377,7 +422,7 @@ class SyncState(namedtuple('SyncState', 'index,leader,sync_standby')):
         >>> SyncState(1, None, None).matches('foo')
         False
         """
-        return name is not None and name in (self.leader, self.sync_standby)
+        return name is not None and name in [self.leader] + self.members
 
 
 class TimelineHistory(namedtuple('TimelineHistory', 'index,value,lines')):
@@ -399,19 +444,20 @@ class TimelineHistory(namedtuple('TimelineHistory', 'index,value,lines')):
         return TimelineHistory(index, value, lines)
 
 
-class Cluster(namedtuple('Cluster', 'initialize,config,leader,last_leader_operation,members,failover,sync,history')):
+class Cluster(namedtuple('Cluster', 'initialize,config,leader,last_lsn,members,failover,sync,history,slots')):
 
     """Immutable object (namedtuple) which represents PostgreSQL cluster.
     Consists of the following fields:
     :param initialize: shows whether this cluster has initialization key stored in DC or not.
     :param config: global dynamic configuration, reference to `ClusterConfig` object
     :param leader: `Leader` object which represents current leader of the cluster
-    :param last_leader_operation: int or long object containing position of last known leader operation.
+    :param last_lsn: int or long object containing position of last known leader LSN.
-        This value is stored in `/optime/leader` key
+        This value is stored in the `/status` key or `/optime/leader` (legacy) key
     :param members: list of Member object, all PostgreSQL cluster members including leader
     :param failover: reference to `Failover` object
     :param sync: reference to `SyncState` object, last observed synchronous replication state.
     :param history: reference to `TimelineHistory` object
+    :param slots: state of permanent logical replication slots on the primary in the format: {"slot_name": int}
     """
 
     def is_unlocked(self):
@@ -437,20 +483,41 @@ class Cluster(namedtuple('Cluster', 'initialize,config,leader,last_leader_operat
     def is_synchronous_mode(self):
         return self.check_mode('synchronous_mode')
 
-    def get_replication_slots(self, name, role):
+    @property
+    def __permanent_slots(self):
+        return self.config and self.config.permanent_slots or {}
+
+    @property
+    def __permanent_physical_slots(self):
+        return {name: value for name, value in self.__permanent_slots.items()
+                if not value or isinstance(value, dict) and value.get('type', 'physical') == 'physical'}
+
+    @property
+    def __permanent_logical_slots(self):
+        return {name: value for name, value in self.__permanent_slots.items() if isinstance(value, dict)
+                and value.get('type', 'logical') == 'logical' and value.get('database') and value.get('plugin')}
+
+    @property
+    def use_slots(self):
+        return self.config and self.config.data.get('postgresql', {}).get('use_slots', True)
+
+    def get_replication_slots(self, my_name, role, nofailover, major_version, show_error=False):
         # if the replicatefrom tag is set on the member - we should not create the replication slot for it on
         # the current master, because that member would replicate from elsewhere. We still create the slot if
         # the replicatefrom destination member is currently not a member of the cluster (fallback to the
         # master), or if replicatefrom destination member happens to be the current master
+        use_slots = self.use_slots
         if role in ('master', 'standby_leader'):
-            slot_members = [m.name for m in self.members if m.name != name and
+            slot_members = [m.name for m in self.members if use_slots and m.name != my_name and
-                            (m.replicatefrom is None or m.replicatefrom == name or
+                            (m.replicatefrom is None or m.replicatefrom == my_name or
                              not self.has_member(m.replicatefrom))]
-            permanent_slots = (self.config and self.config.permanent_slots or {}).copy()
+            permanent_slots = self.__permanent_slots if use_slots and \
+                role == 'master' else self.__permanent_physical_slots
         else:
             # only manage slots for replicas that replicate from this one, except for the leader among them
-            slot_members = [m.name for m in self.members if m.replicatefrom == name and m.name != self.leader.name]
+            slot_members = [m.name for m in self.members if use_slots and
-            permanent_slots = {}
+                            m.replicatefrom == my_name and m.name != self.leader.name]
+            permanent_slots = self.__permanent_logical_slots if use_slots and not nofailover else {}
 
         slots = {slot_name_from_member_name(name): {'type': 'physical'} for name in slot_members}
 
@@ -464,46 +531,82 @@ class Cluster(namedtuple('Cluster', 'initialize,config,leader,last_leader_operat
                                    for k, v in slot_conflicts.items() if len(v) > 1))
 
         # "merge" replication slots for members with permanent_replication_slots
+        disabled_permanent_logical_slots = []
         for name, value in permanent_slots.items():
             if not slot_name_re.match(name):
                 logger.error("Invalid permanent replication slot name '%s'", name)
                 logger.error("Slot name may only contain lower case letters, numbers, and the underscore chars")
                 continue
 
-            if name in slots:
+            value = deepcopy(value) if value else {'type': 'physical'}
-                logger.error("Permanent replication slot {'%s': %s} is conflicting with" +
-                             " physical replication slot for cluster member", name, value)
-                continue
-
-            value = deepcopy(value)
-            if not value:
-                value = {'type': 'physical'}
-
             if isinstance(value, dict):
                 if 'type' not in value:
                     value['type'] = 'logical' if value.get('database') and value.get('plugin') else 'physical'
 
-                if value['type'] == 'physical' or value['type'] == 'logical' \
+                if value['type'] == 'physical':
-                        and value.get('database') and value.get('plugin'):
+                    # Don't try to create permanent physical replication slot for yourself
-                    slots[name] = value
+                    if name != slot_name_from_member_name(my_name):
+                        slots[name] = value
+                    continue
+                elif value['type'] == 'logical' and value.get('database') and value.get('plugin'):
+                    if major_version < 110000:
+                        disabled_permanent_logical_slots.append(name)
+                    elif name in slots:
+                        logger.error("Permanent logical replication slot {'%s': %s} is conflicting with" +
+                                     " physical replication slot for cluster member", name, value)
+                    else:
+                        slots[name] = value
                     continue
 
             logger.error("Bad value for slot '%s' in permanent_slots: %s", name, permanent_slots[name])
 
+        if disabled_permanent_logical_slots and show_error:
+            logger.error("Permanent logical replication slots supported by Patroni only starting from PostgreSQL 11. "
+                         "Following slots will not be created: %s.", disabled_permanent_logical_slots)
+
         return slots
 
-    def has_permanent_logical_slots(self, name):
+    def has_permanent_logical_slots(self, my_name, nofailover, major_version=110000):
-        slots = self.get_replication_slots(name, 'master').values()
+        if major_version < 110000:
+            return False
+        slots = self.get_replication_slots(my_name, 'replica', nofailover, major_version).values()
         return any(v for v in slots if v.get("type") == "logical")
 
+    def should_enforce_hot_standby_feedback(self, my_name, nofailover, major_version):
+        """
+        The hot_standby_feedback must be enabled if the current replica has logical slots
+        or it is working as a cascading replica for the other node that has logical slots.
+        """
+
+        if major_version < 110000:
+            return False
+
+        if self.has_permanent_logical_slots(my_name, nofailover, major_version):
+            return True
+
+        if self.use_slots:
+            members = [m for m in self.members if m.replicatefrom == my_name and m.name != self.leader.name]
+            return any(self.should_enforce_hot_standby_feedback(m.name, m.nofailover, major_version) for m in members)
+        return False
+
+    def get_my_slot_name_on_primary(self, my_name, replicatefrom):
+        """
+        P <-- I <-- L
+        In case of cascading replication we have to check not our physical slot,
+        but slot of the replica that connects us to the primary.
+        """
+
+        m = self.get_member(replicatefrom, False) if replicatefrom else None
+        return self.get_my_slot_name_on_primary(m.name, m.replicatefrom) if m else slot_name_from_member_name(my_name)
+
     @property
     def timeline(self):
         """
-        >>> Cluster(0, 0, 0, 0, 0, 0, 0, 0).timeline
+        >>> Cluster(0, 0, 0, 0, 0, 0, 0, 0, 0).timeline
         0
-        >>> Cluster(0, 0, 0, 0, 0, 0, 0, TimelineHistory.from_node(1, '[]')).timeline
+        >>> Cluster(0, 0, 0, 0, 0, 0, 0, TimelineHistory.from_node(1, '[]'), 0).timeline
         1
-        >>> Cluster(0, 0, 0, 0, 0, 0, 0, TimelineHistory.from_node(1, '[["a"]]')).timeline
+        >>> Cluster(0, 0, 0, 0, 0, 0, 0, TimelineHistory.from_node(1, '[["a"]]'), 0).timeline
         0
         """
         if self.history:
@@ -516,6 +619,10 @@ class Cluster(namedtuple('Cluster', 'initialize,config,leader,last_leader_operat
                 return 1
         return 0
 
+    @property
+    def min_version(self):
+        return next(iter(sorted(filter(lambda v: v, [m.version for m in self.members])) + [None]))
+
 
 @six.add_metaclass(abc.ABCMeta)
 class AbstractDCS(object):
@@ -527,7 +634,8 @@ class AbstractDCS(object):
     _HISTORY = 'history'
     _MEMBERS = 'members/'
     _OPTIME = 'optime'
-    _LEADER_OPTIME = _OPTIME + '/' + _LEADER
+    _STATUS = 'status'  # JSON, contains "leader_lsn" and confirmed_flush_lsn of logical "slots" on the leader
+    _LEADER_OPTIME = _OPTIME + '/' + _LEADER  # legacy
     _SYNC = 'sync'
 
     def __init__(self, config):
@@ -543,7 +651,8 @@ class AbstractDCS(object):
         self._cluster = None
         self._cluster_valid_till = 0
         self._cluster_thread_lock = Lock()
-        self._last_leader_operation = ''
+        self._last_lsn = ''
+        self._last_status = {}
         self.event = Event()
 
     def client_path(self, path):
@@ -577,6 +686,10 @@ class AbstractDCS(object):
     def history_path(self):
         return self.client_path(self._HISTORY)
 
+    @property
+    def status_path(self):
+        return self.client_path(self._STATUS)
+
     @property
     def leader_optime_path(self):
         return self.client_path(self._LEADER_OPTIME)
@@ -619,7 +732,12 @@ class AbstractDCS(object):
            If the current node was running as a master and exception raised,
            instance would be demoted."""
 
-    def get_cluster(self):
+    def _bypass_caches(self):
+        """Used only in zookeeper"""
+
+    def get_cluster(self, force=False):
+        if force:
+            self._bypass_caches()
         try:
             cluster = self._load_cluster()
         except Exception:
@@ -642,14 +760,30 @@ class AbstractDCS(object):
             self._cluster_valid_till = 0
 
     @abc.abstractmethod
-    def _write_leader_optime(self, last_operation):
+    def _write_leader_optime(self, last_lsn):
-        """write current xlog location into `/optime/leader` key in DCS
+        """write current WAL LSN into `/optime/leader` key in DCS
-        :param last_operation: absolute xlog location in bytes
+
+        :param last_lsn: absolute WAL LSN in bytes
         :returns: `!True` on success."""
 
-    def write_leader_optime(self, last_operation):
+    def write_leader_optime(self, last_lsn):
-        if self._last_leader_operation != last_operation and self._write_leader_optime(last_operation):
+        if self._last_lsn != last_lsn and self._write_leader_optime(last_lsn):
-            self._last_leader_operation = last_operation
+            self._last_lsn = last_lsn
+
+    @abc.abstractmethod
+    def _write_status(self, value):
+        """write current WAL LSN and confirmed_flush_lsn of permanent slots into the `/status` key in DCS
+
+        :param value: status serialized in JSON forman
+        :returns: `!True` on success."""
+
+    def write_status(self, value):
+        if not deep_compare(self._last_status, value) and self._write_status(json.dumps(value, separators=(',', ':'))):
+            self._last_status = value
+        cluster = self.cluster
+        min_version = cluster and cluster.min_version
+        if min_version and min_version < (2, 1, 0):
+            self._write_leader_optime(str(value[self._OPTIME]))
 
     @abc.abstractmethod
     def _update_leader(self):
@@ -661,16 +795,20 @@ class AbstractDCS(object):
         You have to use CAS (Compare And Swap) operation in order to update leader key,
         for example for etcd `prevValue` parameter must be used."""
 
-    def update_leader(self, last_operation, access_is_restricted=False):
+    def update_leader(self, last_lsn, slots=None):
         """Update leader key (or session) ttl and optime/leader
 
-        :param last_operation: absolute xlog location in bytes
+        :param last_lsn: absolute WAL LSN in bytes
+        :param slots: dict with permanent slots confirmed_flush_lsn
         :returns: `!True` if leader key (or session) has been updated successfully.
             If not, `!False` must be returned and current instance would be demoted."""
 
         ret = self._update_leader()
-        if ret and last_operation:
+        if ret and last_lsn:
-            self.write_leader_optime(last_operation)
+            status = {self._OPTIME: last_lsn}
+            if slots:
+                status['slots'] = slots
+            self.write_status(status)
         return ret
 
     @abc.abstractmethod
@@ -735,10 +873,19 @@ class AbstractDCS(object):
         otherwise it should return `!False`"""
 
     @abc.abstractmethod
-    def delete_leader(self):
+    def _delete_leader(self):
-        """Voluntarily remove leader key from DCS
+        """Remove leader key from DCS.
         This method should remove leader key if current instance is the leader"""
 
+    def delete_leader(self, last_lsn=None):
+        """Update optime/leader and voluntarily remove leader key from DCS.
+        This method should remove leader key if current instance is the leader.
+        :param last_lsn: latest checkpoint location in bytes"""
+
+        if last_lsn:
+            self.write_leader_optime(last_lsn)
+        return self._delete_leader()
+
     @abc.abstractmethod
     def cancel_initialization(self):
         """ Removes the initialize key for a cluster """
@@ -749,8 +896,10 @@ class AbstractDCS(object):
 
     @staticmethod
     def sync_state(leader, sync_standby):
-        """Build sync_state dict"""
+        """Build sync_state dict
-        return {'leader': leader, 'sync_standby': sync_standby}
+           sync_standby dictionary key being kept for backward compatibility
+        """
+        return {'leader': leader, 'sync_standby': sync_standby and ','.join(sorted(sync_standby)) or None}
 
     def write_sync_state(self, leader, sync_standby, index=None):
         sync_value = self.sync_state(leader, sync_standby)

patroni/dcs/consul.py

@@ -8,14 +8,16 @@ import ssl
 import time
 import urllib3
 
+from collections import namedtuple
 from consul import ConsulException, NotFound, base
-from patroni.dcs import AbstractDCS, ClusterConfig, Cluster, Failover, Leader, Member, SyncState, TimelineHistory
-from patroni.exceptions import DCSError
-from patroni.utils import deep_compare, parse_bool, Retry, RetryFailedError, split_host_port, uri
 from urllib3.exceptions import HTTPError
 from six.moves.urllib.parse import urlencode, urlparse, quote
 from six.moves.http_client import HTTPException
 
+from . import AbstractDCS, Cluster, ClusterConfig, Failover, Leader, Member, SyncState, TimelineHistory
+from ..exceptions import DCSError
+from ..utils import deep_compare, parse_bool, Retry, RetryFailedError, split_host_port, uri, USER_AGENT
+
 logger = logging.getLogger(__name__)
 
 
@@ -35,6 +37,9 @@ class InvalidSession(ConsulException):
     """invalid session"""
 
 
+Response = namedtuple('Response', 'code,headers,body,content')
+
+
 class HTTPClient(object):
 
     def __init__(self, host='127.0.0.1', port=8500, token=None, scheme='http', verify=True, cert=None, ca_cert=None):
@@ -52,9 +57,8 @@ class HTTPClient(object):
                 kwargs['cert_file'] = cert
         if ca_cert:
             kwargs['ca_certs'] = ca_cert
-        if verify or ca_cert:
+        kwargs['cert_reqs'] = ssl.CERT_REQUIRED if verify or ca_cert else ssl.CERT_NONE
-            kwargs['cert_reqs'] = ssl.CERT_REQUIRED
+        self.http = urllib3.PoolManager(num_pools=10, maxsize=10, **kwargs)
-        self.http = urllib3.PoolManager(num_pools=10, **kwargs)
         self._ttl = None
 
     def set_read_timeout(self, timeout):
@@ -71,16 +75,17 @@ class HTTPClient(object):
 
     @staticmethod
     def response(response):
-        data = response.data.decode('utf-8')
+        content = response.data
+        body = content.decode('utf-8')
         if response.status == 500:
-            msg = '{0} {1}'.format(response.status, data)
+            msg = '{0} {1}'.format(response.status, body)
-            if data.startswith('Invalid Session TTL'):
+            if body.startswith('Invalid Session TTL'):
                 raise InvalidSessionTTL(msg)
-            elif data.startswith('invalid session'):
+            elif body.startswith('invalid session'):
                 raise InvalidSession(msg)
             else:
                 raise ConsulInternalError(msg)
-        return base.Response(response.status, response.headers, data)
+        return Response(response.status, response.headers, body, content)
 
     def uri(self, path, params=None):
         return '{0}{1}{2}'.format(self.base_uri, path, params and '?' + urlencode(params) or '')
@@ -89,7 +94,7 @@ class HTTPClient(object):
         if method not in ('get', 'post', 'put', 'delete'):
             raise AttributeError("HTTPClient instance has no attribute '{0}'".format(method))
 
-        def wrapper(callback, path, params=None, data=''):
+        def wrapper(callback, path, params=None, data='', headers=None):
             # python-consul doesn't allow to specify ttl smaller then 10 seconds
             # because session_ttl_min defaults to 10s, so we have to do this ugly dirty hack...
             if method == 'put' and path == '/v1/session/create':
@@ -106,13 +111,15 @@ class HTTPClient(object):
                 # According to the documentation a small random amount of additional wait time is added to the
                 # supplied maximum wait time to spread out the wake up time of any concurrent requests. This adds
                 # up to wait / 16 additional time to the maximum duration. Since our goal is actually getting a
-                # response rather read timeout we will add to the timeout a sligtly bigger value.
+                # response rather read timeout we will add to the timeout a slightly bigger value.
                 kwargs['timeout'] = timeout + max(timeout/15.0, 1)
             else:
                 kwargs['timeout'] = self._read_timeout
+            kwargs['headers'] = (headers or {}).copy()
+            kwargs['headers'].update(urllib3.make_headers(user_agent=USER_AGENT))
             token = params.pop('token', self.token) if isinstance(params, dict) else self.token
             if token:
-                kwargs['headers'] = {'X-Consul-Token': token}
+                kwargs['headers']['X-Consul-Token'] = token
             return callback(self.response(self.http.request(method.upper(), self.uri(path, params), **kwargs)))
         return wrapper
 
@@ -125,7 +132,7 @@ class ConsulClient(base.Consul):
         self.token = kwargs.get('token')
         super(ConsulClient, self).__init__(*args, **kwargs)
 
-    def connect(self, *args, **kwargs):
+    def http_connect(self, *args, **kwargs):
         kwargs.update(dict(zip(['host', 'port', 'scheme', 'verify'], args)))
         if self._cert:
             kwargs['cert'] = self._cert
@@ -135,6 +142,9 @@ class ConsulClient(base.Consul):
             kwargs['token'] = self.token
         return HTTPClient(**kwargs)
 
+    def connect(self, *args, **kwargs):
+        return self.http_connect(*args, **kwargs)
+
     def reload_config(self, config):
         self.http.token = self.token = config.get('token')
         self.consistency = config.get('consistency', 'default')
@@ -215,9 +225,10 @@ class Consul(AbstractDCS):
         self.set_retry_timeout(config['retry_timeout'])
         self.set_ttl(config.get('ttl') or 30)
         self._last_session_refresh = 0
-        self.__session_checks = config.get('checks')
+        self.__session_checks = config.get('checks', [])
         self._register_service = config.get('register_service', False)
         if self._register_service:
+            self._service_tags = config.get('service_tags', [])
             self._service_name = service_name_from_scope_name(self._scope)
             if self._scope != self._service_name:
                 logger.warning('Using %s as consul service name instead of scope name %s', self._service_name,
@@ -326,9 +337,24 @@ class Consul(AbstractDCS):
             history = nodes.get(self._HISTORY)
             history = history and TimelineHistory.from_node(history['ModifyIndex'], history['Value'])
 
-            # get last leader operation
+            # get last known leader lsn and slots
-            last_leader_operation = nodes.get(self._LEADER_OPTIME)
+            status = nodes.get(self._STATUS)
-            last_leader_operation = 0 if last_leader_operation is None else int(last_leader_operation['Value'])
+            if status:
+                try:
+                    status = json.loads(status['Value'])
+                    last_lsn = status.get(self._OPTIME)
+                    slots = status.get('slots')
+                except Exception:
+                    slots = last_lsn = None
+            else:
+                last_lsn = nodes.get(self._LEADER_OPTIME)
+                last_lsn = last_lsn and last_lsn['Value']
+                slots = None
+
+            try:
+                last_lsn = int(last_lsn)
+            except Exception:
+                last_lsn = 0
 
             # get list of members
             members = [self.member(n) for k, n in nodes.items() if k.startswith(self._MEMBERS) and k.count('/') == 1]
@@ -355,9 +381,9 @@ class Consul(AbstractDCS):
             sync = nodes.get(self._SYNC)
             sync = SyncState.from_node(sync and sync['ModifyIndex'], sync and sync['Value'])
 
-            return Cluster(initialize, config, leader, last_leader_operation, members, failover, sync, history)
+            return Cluster(initialize, config, leader, last_lsn, members, failover, sync, history, slots)
         except NotFound:
-            return Cluster(None, None, None, None, [], None, None, None)
+            return Cluster(None, None, None, None, [], None, None, None, None)
         except Exception:
             logger.exception('get_cluster')
             raise ConsulError('Consul is not responding properly')
@@ -366,7 +392,11 @@ class Consul(AbstractDCS):
     def touch_member(self, data, permanent=False):
         cluster = self.cluster
         member = cluster and cluster.get_member(self._name, fallback_to_leader=False)
-        create_member = not permanent and self.refresh_session()
+
+        try:
+            create_member = not permanent and self.refresh_session()
+        except DCSError:
+            return False
 
         if member and (create_member or member.session != self._session):
             self._client.kv.delete(self.member_path)
@@ -409,12 +439,14 @@ class Consul(AbstractDCS):
         conn_parts = urlparse(data['conn_url'])
         check = base.Check.http(api_parts.geturl(), self._service_check_interval,
                                 deregister='{0}s'.format(self._client.http.ttl * 10))
+        tags = self._service_tags[:]
+        tags.append(role)
         params = {
             'service_id': '{0}/{1}'.format(self._scope, self._name),
             'address': conn_parts.hostname,
             'port': conn_parts.port,
             'check': check,
-            'tags': [role]
+            'tags': tags
         }
 
         if state == 'stopped':
@@ -474,8 +506,12 @@ class Consul(AbstractDCS):
         return self._client.kv.put(self.config_path, value, cas=index)
 
     @catch_consul_errors
-    def _write_leader_optime(self, last_operation):
+    def _write_leader_optime(self, last_lsn):
-        return self._client.kv.put(self.leader_optime_path, last_operation)
+        return self._client.kv.put(self.leader_optime_path, last_lsn)
+
+    @catch_consul_errors
+    def _write_status(self, value):
+        return self._client.kv.put(self.status_path, value)
 
     @catch_consul_errors
     def _update_leader(self):
@@ -502,7 +538,7 @@ class Consul(AbstractDCS):
         return self._client.kv.put(self.history_path, value)
 
     @catch_consul_errors
-    def delete_leader(self):
+    def _delete_leader(self):
         cluster = self.cluster
         if cluster and isinstance(cluster.leader, Leader) and cluster.leader.name == self._name:
             return self._client.kv.delete(self.leader_path, cas=cluster.leader.index)

patroni/dcs/etcd.py

@@ -1,30 +1,36 @@
 from __future__ import absolute_import
+import abc
 import etcd
 import json
 import logging
 import os
 import urllib3.util.connection
 import random
-import requests
 import six
 import socket
 import time
 
 from dns.exception import DNSException
 from dns import resolver
-from patroni.dcs import AbstractDCS, ClusterConfig, Cluster, Failover, Leader, Member, SyncState, TimelineHistory
+from urllib3 import Timeout
-from patroni.exceptions import DCSError
+from urllib3.exceptions import HTTPError, ReadTimeoutError, ProtocolError
-from patroni.utils import Retry, RetryFailedError, split_host_port, uri
-from urllib3.exceptions import HTTPError, ReadTimeoutError
-from requests.exceptions import RequestException
 from six.moves.queue import Queue
 from six.moves.http_client import HTTPException
 from six.moves.urllib_parse import urlparse
 from threading import Thread
 
+from . import AbstractDCS, Cluster, ClusterConfig, Failover, Leader, Member, SyncState, TimelineHistory
+from ..exceptions import DCSError
+from ..request import get as requests_get
+from ..utils import Retry, RetryFailedError, split_host_port, uri, USER_AGENT
+
 logger = logging.getLogger(__name__)
 
 
+class EtcdRaftInternal(etcd.EtcdException):
+    """Raft Internal Error"""
+
+
 class EtcdError(DCSError):
     pass
 
@@ -65,16 +71,20 @@ class DnsCachingResolver(Thread):
     def resolve_async(self, host, port, attempt=0):
         self._resolve_queue.put(((host, port), attempt))
 
+    def remove(self, host, port):
+        self._cache.pop((host, port), None)
+
     @staticmethod
     def _do_resolve(host, port):
         try:
             return socket.getaddrinfo(host, port, 0, socket.SOCK_STREAM, socket.IPPROTO_TCP)
-        except socket.gaierror:
+        except Exception as e:
-            logger.warning('failed to resolve host %s', host)
+            logger.warning('failed to resolve host %s: %s', host, e)
             return []
 
 
-class Client(etcd.Client):
+@six.add_metaclass(abc.ABCMeta)
+class AbstractEtcdClientWithFailover(etcd.Client):
 
     def __init__(self, config, dns_resolver, cache_ttl=300):
         self._dns_resolver = dns_resolver
@@ -82,34 +92,81 @@ class Client(etcd.Client):
         self._machines_cache_updated = 0
         args = {p: config.get(p) for p in ('host', 'port', 'protocol', 'use_proxies', 'username', 'password',
                                            'cert', 'ca_cert') if config.get(p)}
-        super(Client, self).__init__(read_timeout=config['retry_timeout'], **args)
+        super(AbstractEtcdClientWithFailover, self).__init__(read_timeout=config['retry_timeout'], **args)
         # For some reason python3-etcd on debian and ubuntu are not based on the latest version
         # Workaround for the case when https://github.com/jplana/python-etcd/pull/196 is not applied
         self.http.connection_pool_kw.pop('ssl_version', None)
         self._config = config
+        self._initial_machines_cache = []
         self._load_machines_cache()
         self._allow_reconnect = True
+        # allow passing retry argument to api_execute in params
+        self._comparison_conditions.add('retry')
+        self._read_options.add('retry')
+        self._del_conditions.add('retry')
 
-    def _build_request_parameters(self):
+    def _calculate_timeouts(self, etcd_nodes, timeout=None):
-        kwargs = {'headers': self._get_headers(), 'redirect': self.allow_redirect}
+        """Calculate a request timeout and number of retries per single etcd node.
+        In case if the timeout per node is too small (less than one second) we will reduce the number of nodes.
+        For the cluster with only one node we will try to do 2 retries.
+        For clusters with 2 nodes we will try to do 1 retry for every node.
+        No retries for clusters with 3 or more nodes. We better rely on switching to a different node."""
 
-        # calculate the number of retries and timeout *per node*
+        per_node_timeout = timeout = float(timeout or self.read_timeout)
-        # actual number of retries depends on the number of nodes
-        etcd_nodes = len(self._machines_cache) + 1
-        kwargs['retries'] = 0 if etcd_nodes > 3 else (1 if etcd_nodes > 1 else 2)
 
-        # if etcd_nodes > 3:
+        max_retries = 4 - min(etcd_nodes, 3)
-        #     kwargs.update({'retries': 0, 'timeout': float(self.read_timeout)/etcd_nodes})
+        per_node_retries = 1
-        # elif etcd_nodes > 1:
+        min_timeout = 1.0
-        #     kwargs.update({'retries': 1, 'timeout': self.read_timeout/2.0/etcd_nodes})
+
-        # else:
+        while etcd_nodes > 0:
-        #     kwargs.update({'retries': 2, 'timeout': self.read_timeout/3.0})
+            per_node_timeout = float(timeout) / etcd_nodes
-        kwargs['timeout'] = self.read_timeout/float(kwargs['retries'] + 1)/etcd_nodes
+            if per_node_timeout >= min_timeout:
+                # for small clusters we will try to do more than on try on every node
+                while per_node_retries < max_retries and per_node_timeout / (per_node_retries + 1) >= min_timeout:
+                    per_node_retries += 1
+                per_node_timeout /= per_node_retries
+                break
+            # if the timeout per one node is to small try to reduce number of nodes
+            etcd_nodes -= 1
+            max_retries = 1
+
+        return etcd_nodes, per_node_timeout, per_node_retries - 1
+
+    def reload_config(self, config):
+        self.username = config.get('username')
+        self.password = config.get('password')
+
+    def _get_headers(self):
+        basic_auth = ':'.join((self.username, self.password)) if self.username and self.password else None
+        return urllib3.make_headers(basic_auth=basic_auth, user_agent=USER_AGENT)
+
+    def _prepare_common_parameters(self, etcd_nodes, timeout=None):
+        kwargs = {'headers': self._get_headers(), 'redirect': self.allow_redirect, 'preload_content': False}
+
+        if timeout is not None:
+            kwargs.update(retries=0, timeout=timeout)
+        else:
+            _, per_node_timeout, per_node_retries = self._calculate_timeouts(etcd_nodes)
+            connect_timeout = max(1, per_node_timeout/2)
+            kwargs.update(timeout=Timeout(connect=connect_timeout, total=per_node_timeout), retries=per_node_retries)
         return kwargs
 
     def set_machines_cache_ttl(self, cache_ttl):
         self._machines_cache_ttl = cache_ttl
 
+    @abc.abstractmethod
+    def _prepare_get_members(self, etcd_nodes):
+        """returns: request parameters"""
+
+    @abc.abstractmethod
+    def _get_members(self, base_uri, **kwargs):
+        """returns: list of clientURLs"""
+
+    @property
+    def machines_cache(self):
+        base_uri, cache = self._base_uri, self._machines_cache
+        return ([base_uri] if base_uri in cache else []) + [machine for machine in cache if machine != base_uri]
+
     @property
     def machines(self):
         """Original `machines` method(property) of `etcd.Client` class raise exception
@@ -122,100 +179,101 @@ class Client(etcd.Client):
         Also this method implements the same timeout-retry logic as `api_execute`, because
         the original method was retrying 2 times with the `read_timeout` on each node."""
 
-        kwargs = self._build_request_parameters()
+        machines_cache = self.machines_cache
+        kwargs = self._prepare_get_members(len(machines_cache))
 
-        while True:
+        for base_uri in machines_cache:
             try:
-                response = self.http.request(self._MGET, self._base_uri + self.version_prefix + '/machines', **kwargs)
+                machines = list(self._get_members(base_uri, **kwargs))
-                data = self._handle_server_response(response).data.decode('utf-8')
-                machines = [m.strip() for m in data.split(',') if m.strip()]
                 logger.debug("Retrieved list of machines: %s", machines)
-                if not machines:
+                if machines:
-                    raise etcd.EtcdException
+                    random.shuffle(machines)
-                random.shuffle(machines)
+                    self._update_dns_cache(self._dns_resolver.resolve_async, machines)
-                for url in machines:
+                    return machines
-                    r = urlparse(url)
-                    port = r.port or (443 if r.scheme == 'https' else 80)
-                    self._dns_resolver.resolve_async(r.hostname, port)
-                return machines
             except Exception as e:
-                # We can't get the list of machines, if one server is in the
+                self.http.clear()
-                # machines cache, try on it
+                logger.error("Failed to get list of machines from %s%s: %r", base_uri, self.version_prefix, e)
-                logger.error("Failed to get list of machines from %s%s: %r", self._base_uri, self.version_prefix, e)
+
-                if self._machines_cache:
+        raise etcd.EtcdConnectionFailed('No more machines in the cluster')
-                    self._base_uri = self._machines_cache.pop(0)
-                    logger.info("Retrying on %s", self._base_uri)
-                elif self._update_machines_cache:
-                    raise etcd.EtcdException("Could not get the list of servers, "
-                                             "maybe you provided the wrong "
-                                             "host(s) to connect to?")
-                else:
-                    return []
 
     def set_read_timeout(self, timeout):
         self._read_timeout = timeout
 
-    def _do_http_request(self, request_executor, method, url, fields=None, **kwargs):
+    def _do_http_request(self, retry, machines_cache, request_executor, method, path, fields=None, **kwargs):
-        try:
+        if fields is not None:
-            response = request_executor(method, url, fields=fields, **kwargs)
+            kwargs['fields'] = fields
-            response.data.decode('utf-8')
+        some_request_failed = False
-            self._check_cluster_id(response)
+        for i, base_uri in enumerate(machines_cache):
-        except (HTTPError, HTTPException, socket.error, socket.timeout) as e:
+            if i > 0:
-            if (isinstance(fields, dict) and fields.get("wait") == "true" and
+                logger.info("Retrying on %s", base_uri)
-                    isinstance(e, ReadTimeoutError)):
+            try:
-                logger.debug("Watch timed out.")
+                response = request_executor(method, base_uri + path, **kwargs)
-                raise etcd.EtcdWatchTimedOut("Watch timed out: {0}".format(e), cause=e)
+                response.data.decode('utf-8')
-            logger.error("Request to server %s failed: %r", self._base_uri, e)
+                if some_request_failed:
-            logger.info("Reconnection allowed, looking for another server.")
+                    self.set_base_uri(base_uri)
-            self._base_uri = self._next_server(cause=e)
+                    self._refresh_machines_cache()
-            response = False
+                return response
-        return response
+            except (HTTPError, HTTPException, socket.error, socket.timeout) as e:
+                self.http.clear()
+                # switch to the next etcd node because we don't know exactly what happened,
+                # whether the key didn't received an update or there is a network problem.
+                if not retry and i + 1 < len(machines_cache):
+                    self.set_base_uri(machines_cache[i + 1])
+                if (isinstance(fields, dict) and fields.get("wait") == "true" and
+                        isinstance(e, (ReadTimeoutError, ProtocolError))):
+                    logger.debug("Watch timed out.")
+                    raise etcd.EtcdWatchTimedOut("Watch timed out: {0}".format(e), cause=e)
+                logger.error("Request to server %s failed: %r", base_uri, e)
+                logger.info("Reconnection allowed, looking for another server.")
+                if not retry:
+                    raise etcd.EtcdException('{0} {1} request failed'.format(method, path))
+                some_request_failed = True
+
+        raise etcd.EtcdConnectionFailed('No more machines in the cluster')
+
+    @abc.abstractmethod
+    def _prepare_request(self, kwargs, params=None, method=None):
+        """returns: request_executor"""
 
     def api_execute(self, path, method, params=None, timeout=None):
-        if not path.startswith('/'):
+        retry = params.pop('retry', None) if isinstance(params, dict) else None
-            raise ValueError('Path does not start with /')
-
-        kwargs = {'fields': params, 'preload_content': False}
-
-        if method in [self._MGET, self._MDELETE]:
-            request_executor = self.http.request
-        elif method in [self._MPUT, self._MPOST]:
-            request_executor = self.http.request_encode_body
-            kwargs['encode_multipart'] = False
-        else:
-            raise etcd.EtcdException('HTTP method {0} not supported'.format(method))
 
         # Update machines_cache if previous attempt of update has failed
         if self._update_machines_cache:
             self._load_machines_cache()
         elif not self._use_proxies and time.time() - self._machines_cache_updated > self._machines_cache_ttl:
             self._refresh_machines_cache()
-            self._machines_cache_updated = time.time()
 
-        kwargs.update(self._build_request_parameters())
+        machines_cache = self.machines_cache
+        etcd_nodes = len(machines_cache)
 
-        if timeout is not None:
+        kwargs = self._prepare_common_parameters(etcd_nodes, timeout)
-            kwargs.update({'retries': 0, 'timeout': timeout})
+        request_executor = self._prepare_request(kwargs, params, method)
 
-        response = False
+        while True:
-
+            try:
-        try:
+                response = self._do_http_request(retry, machines_cache, request_executor, method, path, **kwargs)
-            some_request_failed = False
+                return self._handle_server_response(response)
-            while not response:
+            except etcd.EtcdWatchTimedOut:
-                response = self._do_http_request(request_executor, method, self._base_uri + path, **kwargs)
-
-                if response is False:
-                    some_request_failed = True
-            if some_request_failed:
-                self._refresh_machines_cache()
-        except etcd.EtcdConnectionFailed as e:
-            if isinstance(e, etcd.EtcdWatchTimedOut) and self._machines_cache:
-                self._base_uri = self._next_server()
-            else:
-                self._update_machines_cache = True
-            if not response:
                 raise
-        return self._handle_server_response(response)
+            except etcd.EtcdConnectionFailed as ex:
+                try:
+                    if self._load_machines_cache():
+                        machines_cache = self.machines_cache
+                        etcd_nodes = len(machines_cache)
+                except Exception as e:
+                    logger.debug('Failed to update list of etcd nodes: %r', e)
+                sleeptime = retry.sleeptime
+                remaining_time = retry.stoptime - sleeptime - time.time()
+                nodes, timeout, retries = self._calculate_timeouts(etcd_nodes, remaining_time)
+                if nodes == 0:
+                    self._update_machines_cache = True
+                    raise ex
+                retry.sleep_func(sleeptime)
+                retry.update_delay()
+                # We still have some time left. Partially reduce `machines_cache` and retry request
+                kwargs.update(timeout=Timeout(connect=max(1, timeout/2), total=timeout), retries=retries)
+                machines_cache = machines_cache[:nodes]
 
     @staticmethod
     def get_srv_record(host):
@@ -237,12 +295,12 @@ class Client(etcd.Client):
                 url = uri(protocol, (host, port), endpoint)
                 if endpoint:
                     try:
-                        response = requests.get(url, timeout=self.read_timeout, verify=False)
+                        response = requests_get(url, timeout=self.read_timeout, verify=False)
-                        if response.ok:
+                        if response.status < 400:
-                            for member in response.json():
+                            for member in json.loads(response.data.decode('utf-8')):
                                 ret.extend(member['clientURLs'])
                             break
-                    except RequestException:
+                    except Exception:
                         logger.exception('GET %s', url)
                 else:
                     ret.append(url)
@@ -276,6 +334,13 @@ class Client(etcd.Client):
             machines_cache = self._get_machines_cache_from_dns(self._config['host'], self._config['port'])
         return machines_cache
 
+    @staticmethod
+    def _update_dns_cache(func, machines):
+        for url in machines:
+            r = urlparse(url)
+            port = r.port or (443 if r.scheme == 'https' else 80)
+            func(r.hostname, port)
+
     def _load_machines_cache(self):
         """This method should fill up `_machines_cache` from scratch.
         It could happen only in two cases:
@@ -287,40 +352,95 @@ class Client(etcd.Client):
         if 'srv' not in self._config and 'host' not in self._config and 'hosts' not in self._config:
             raise Exception('Neither srv, hosts, host nor url are defined in etcd section of config')
 
-        self._machines_cache = self._get_machines_cache_from_config()
+        machines_cache = self._get_machines_cache_from_config()
-
         # Can not bootstrap list of etcd-cluster members, giving up
-        if not self._machines_cache:
+        if not machines_cache:
             raise etcd.EtcdException
 
-        # After filling up initial list of machines_cache we should ask etcd-cluster about actual list
+        # enforce resolving dns name,they might get new ips
-        self._base_uri = self._next_server()
+        self._update_dns_cache(self._dns_resolver.remove, machines_cache)
-        self._refresh_machines_cache()
+
+        # The etcd cluster could change its topology over time and depending on how we resolve the initial
+        # topology (list of hosts in the Patroni config or DNS records, A or SRV) we might get into the situation
+        # the the real topology doesn't match anymore with the topology resolved from the configuration file.
+        # In case if the "initial" topology is the same as before we will not override the `_machines_cache`.
+        ret = set(machines_cache) != set(self._initial_machines_cache)
+        if ret:
+            self._initial_machines_cache = self._machines_cache = machines_cache
+
+            # After filling up the initial list of machines_cache we should ask etcd-cluster about actual list
+            self._refresh_machines_cache(True)
 
         self._update_machines_cache = False
+        return ret
+
+    def _refresh_machines_cache(self, updating_cache=False):
+        if self._use_proxies:
+            self._machines_cache = self._get_machines_cache_from_config()
+        else:
+            try:
+                self._machines_cache = self.machines
+            except etcd.EtcdConnectionFailed:
+                if updating_cache:
+                    raise etcd.EtcdException("Could not get the list of servers, "
+                                             "maybe you provided the wrong "
+                                             "host(s) to connect to?")
+                return
+
+        if self._base_uri not in self._machines_cache:
+            self.set_base_uri(self._machines_cache[0])
         self._machines_cache_updated = time.time()
 
-    def _refresh_machines_cache(self):
+    def set_base_uri(self, value):
-        self._machines_cache = self._get_machines_cache_from_config() if self._use_proxies else self.machines
+        logger.info('Selected new etcd server %s', value)
-        if self._base_uri in self._machines_cache:
+        self._base_uri = value
-            self._machines_cache.remove(self._base_uri)
 
 
-class Etcd(AbstractDCS):
+class EtcdClient(AbstractEtcdClientWithFailover):
 
-    def __init__(self, config):
+    ERROR_CLS = EtcdError
-        super(Etcd, self).__init__(config)
+
-        self._ttl = int(config.get('ttl') or 30)
+    def __del__(self):
+        if self.http is not None:
+            try:
+                self.http.clear()
+            except (ReferenceError, TypeError, AttributeError):
+                pass
+
+    def _prepare_get_members(self, etcd_nodes):
+        return self._prepare_common_parameters(etcd_nodes)
+
+    def _get_members(self, base_uri, **kwargs):
+        response = self.http.request(self._MGET, base_uri + self.version_prefix + '/machines', **kwargs)
+        data = self._handle_server_response(response).data.decode('utf-8')
+        return [m.strip() for m in data.split(',') if m.strip()]
+
+    def _prepare_request(self, kwargs, params=None, method=None):
+        kwargs['fields'] = params
+        if method in (self._MPOST, self._MPUT):
+            kwargs['encode_multipart'] = False
+        return self.http.request
+
+
+class AbstractEtcd(AbstractDCS):
+
+    def __init__(self, config, client_cls, retry_errors_cls):
+        super(AbstractEtcd, self).__init__(config)
         self._retry = Retry(deadline=config['retry_timeout'], max_delay=1, max_tries=-1,
-                            retry_exceptions=(etcd.EtcdLeaderElectionInProgress,
+                            retry_exceptions=retry_errors_cls)
-                                              etcd.EtcdWatcherCleared,
+        self._ttl = int(config.get('ttl') or 30)
-                                              etcd.EtcdEventIndexCleared))
+        self._client = self.get_etcd_client(config, client_cls)
-        self._client = self.get_etcd_client(config)
         self.__do_not_watch = False
         self._has_failed = False
 
+    def reload_config(self, config):
+        super(AbstractEtcd, self).reload_config(config)
+        self._client.reload_config(config.get(self.__class__.__name__.lower(), {}))
+
     def retry(self, *args, **kwargs):
-        return self._retry.copy()(*args, **kwargs)
+        retry = self._retry.copy()
+        kwargs['retry'] = retry
+        return retry(*args, **kwargs)
 
     def _handle_exception(self, e, name='', do_sleep=False, raise_ex=None):
         if not self._has_failed:
@@ -333,22 +453,13 @@ class Etcd(AbstractDCS):
         if isinstance(raise_ex, Exception):
             raise raise_ex
 
-    def catch_etcd_errors(func):
-        def wrapper(self, *args, **kwargs):
-            try:
-                retval = func(self, *args, **kwargs) is not None
-                self._has_failed = False
-                return retval
-            except (RetryFailedError, etcd.EtcdException) as e:
-                self._handle_exception(e)
-                return False
-            except Exception as e:
-                self._handle_exception(e, raise_ex=EtcdError('unexpected error'))
-
-        return wrapper
-
     @staticmethod
-    def get_etcd_client(config):
+    def set_socket_options(sock, socket_options):
+        if socket_options:
+            for opt in socket_options:
+                sock.setsockopt(*opt)
+
+    def get_etcd_client(self, config, client_cls):
         if 'proxy' in config:
             config['use_proxies'] = True
             config['url'] = config['proxy']
@@ -368,7 +479,7 @@ class Etcd(AbstractDCS):
             config['hosts'] = []
             for value in hosts:
                 if isinstance(value, six.string_types):
-                    config['hosts'].append(uri(protocol, split_host_port(value, default_port)))
+                    config['hosts'].append(uri(protocol, split_host_port(value.strip(), default_port)))
         elif 'host' in config:
             host, port = split_host_port(config['host'], 2379)
             config['host'] = host
@@ -397,9 +508,7 @@ class Etcd(AbstractDCS):
                 sock = None
                 try:
                     sock = socket.socket(af, socktype, proto)
-                    if socket_options:
+                    self.set_socket_options(sock, socket_options)
-                        for opt in socket_options:
-                            sock.setsockopt(*opt)
                     if timeout is not socket._GLOBAL_DEFAULT_TIMEOUT:
                         sock.settimeout(timeout)
                     if source_address:
@@ -423,7 +532,7 @@ class Etcd(AbstractDCS):
         client = None
         while not client:
             try:
-                client = Client(config, dns_resolver)
+                client = client_cls(config, dns_resolver)
                 if 'use_proxies' in config and not client.machines:
                     raise etcd.EtcdException
             except etcd.EtcdException:
@@ -433,9 +542,10 @@ class Etcd(AbstractDCS):
 
     def set_ttl(self, ttl):
         ttl = int(ttl)
-        self.__do_not_watch = self._ttl != ttl
+        ret = self._ttl != ttl
         self._ttl = ttl
         self._client.set_machines_cache_ttl(ttl*10)
+        return ret
 
     @property
     def ttl(self):
@@ -445,6 +555,31 @@ class Etcd(AbstractDCS):
         self._retry.deadline = retry_timeout
         self._client.set_read_timeout(retry_timeout)
 
+
+def catch_etcd_errors(func):
+    def wrapper(self, *args, **kwargs):
+        try:
+            retval = func(self, *args, **kwargs) is not None
+            self._has_failed = False
+            return retval
+        except (RetryFailedError, etcd.EtcdException) as e:
+            self._handle_exception(e)
+            return False
+        except Exception as e:
+            self._handle_exception(e, raise_ex=self._client.ERROR_CLS('unexpected error'))
+
+    return wrapper
+
+
+class Etcd(AbstractEtcd):
+
+    def __init__(self, config):
+        super(Etcd, self).__init__(config, EtcdClient, (etcd.EtcdLeaderElectionInProgress, EtcdRaftInternal))
+        self.__do_not_watch = False
+
+    def set_ttl(self, ttl):
+        self.__do_not_watch = super(Etcd, self).set_ttl(ttl)
+
     @staticmethod
     def member(node):
         return Member.from_node(node.modifiedIndex, os.path.basename(node.key), node.ttl, node.value)
@@ -467,9 +602,24 @@ class Etcd(AbstractDCS):
             history = nodes.get(self._HISTORY)
             history = history and TimelineHistory.from_node(history.modifiedIndex, history.value)
 
-            # get last leader operation
+            # get last know leader lsn and slots
-            last_leader_operation = nodes.get(self._LEADER_OPTIME)
+            status = nodes.get(self._STATUS)
-            last_leader_operation = 0 if last_leader_operation is None else int(last_leader_operation.value)
+            if status:
+                try:
+                    status = json.loads(status.value)
+                    last_lsn = status.get(self._OPTIME)
+                    slots = status.get('slots')
+                except Exception:
+                    slots = last_lsn = None
+            else:
+                last_lsn = nodes.get(self._LEADER_OPTIME)
+                last_lsn = last_lsn and last_lsn.value
+                slots = None
+
+            try:
+                last_lsn = int(last_lsn)
+            except Exception:
+                last_lsn = 0
 
             # get list of members
             members = [self.member(n) for k, n in nodes.items() if k.startswith(self._MEMBERS) and k.count('/') == 1]
@@ -491,9 +641,9 @@ class Etcd(AbstractDCS):
             sync = nodes.get(self._SYNC)
             sync = SyncState.from_node(sync and sync.modifiedIndex, sync and sync.value)
 
-            cluster = Cluster(initialize, config, leader, last_leader_operation, members, failover, sync, history)
+            cluster = Cluster(initialize, config, leader, last_lsn, members, failover, sync, history, slots)
         except etcd.EtcdKeyNotFound:
-            cluster = Cluster(None, None, None, None, [], None, None, None)
+            cluster = Cluster(None, None, None, None, [], None, None, None, None)
         except Exception as e:
             self._handle_exception(e, 'get_cluster', raise_ex=EtcdError('Etcd is not responding properly'))
         self._has_failed = False
@@ -506,7 +656,7 @@ class Etcd(AbstractDCS):
 
     @catch_etcd_errors
     def take_leader(self):
-        return self.retry(self._client.set, self.leader_path, self._name, self._ttl)
+        return self.retry(self._client.write, self.leader_path, self._name, ttl=self._ttl)
 
     def attempt_to_acquire_leader(self, permanent=False):
         try:
@@ -530,19 +680,23 @@ class Etcd(AbstractDCS):
         return self._client.write(self.config_path, value, prevIndex=index or 0)
 
     @catch_etcd_errors
-    def _write_leader_optime(self, last_operation):
+    def _write_leader_optime(self, last_lsn):
-        return self._client.set(self.leader_optime_path, last_operation)
+        return self._client.set(self.leader_optime_path, last_lsn)
+
+    @catch_etcd_errors
+    def _write_status(self, value):
+        return self._client.set(self.status_path, value)
 
     @catch_etcd_errors
     def _update_leader(self):
-        return self.retry(self._client.test_and_set, self.leader_path, self._name, self._name, self._ttl)
+        return self.retry(self._client.write, self.leader_path, self._name, prevValue=self._name, ttl=self._ttl)
 
     @catch_etcd_errors
     def initialize(self, create_new=True, sysid=""):
         return self.retry(self._client.write, self.initialize_path, sysid, prevExist=(not create_new))
 
     @catch_etcd_errors
-    def delete_leader(self):
+    def _delete_leader(self):
         return self._client.delete(self.leader_path, prevValue=self._name)
 
     @catch_etcd_errors
@@ -575,13 +729,14 @@ class Etcd(AbstractDCS):
 
             while timeout >= 1:  # when timeout is too small urllib3 doesn't have enough time to connect
                 try:
-                    self._client.watch(self.leader_path, index=leader_index, timeout=timeout + 0.5)
+                    result = self._client.watch(self.leader_path, index=leader_index, timeout=timeout + 0.5)
                     self._has_failed = False
+                    if result.action == 'compareAndSwap':
+                        time.sleep(0.01)
                     # Synchronous work of all cluster members with etcd is less expensive
                     # than reestablishing http connection every time from every replica.
                     return True
                 except etcd.EtcdWatchTimedOut:
-                    self._client.http.clear()
                     self._has_failed = False
                     return False
                 except (etcd.EtcdEventIndexCleared, etcd.EtcdWatcherCleared):  # Watch failed
@@ -596,3 +751,6 @@ class Etcd(AbstractDCS):
             return super(Etcd, self).watch(None, timeout)
         finally:
             self.event.clear()
+
+
+etcd.EtcdError.error_exceptions[300] = EtcdRaftInternal

patroni/dcs/etcd3.py

@@ -0,0 +1,823 @@
+from __future__ import absolute_import
+import base64
+import etcd
+import json
+import logging
+import os
+import six
+import socket
+import sys
+import time
+import urllib3
+
+from threading import Condition, Lock, Thread
+
+from . import ClusterConfig, Cluster, Failover, Leader, Member, SyncState, TimelineHistory
+from .etcd import AbstractEtcdClientWithFailover, AbstractEtcd, catch_etcd_errors
+from ..exceptions import DCSError, PatroniException
+from ..utils import deep_compare, enable_keepalive, iter_response_objects, RetryFailedError, USER_AGENT
+
+logger = logging.getLogger(__name__)
+
+
+class Etcd3Error(DCSError):
+    pass
+
+
+class UnsupportedEtcdVersion(PatroniException):
+    pass
+
+
+# google.golang.org/grpc/codes
+GRPCCode = type('Enum', (), {'OK': 0, 'Canceled': 1, 'Unknown': 2, 'InvalidArgument': 3, 'DeadlineExceeded': 4,
+                             'NotFound': 5, 'AlreadyExists': 6, 'PermissionDenied': 7, 'ResourceExhausted': 8,
+                             'FailedPrecondition': 9, 'Aborted': 10, 'OutOfRange': 11, 'Unimplemented': 12,
+                             'Internal': 13, 'Unavailable': 14, 'DataLoss': 15, 'Unauthenticated': 16})
+GRPCcodeToText = {v: k for k, v in GRPCCode.__dict__.items() if not k.startswith('__') and isinstance(v, int)}
+
+
+class Etcd3Exception(etcd.EtcdException):
+    pass
+
+
+class Etcd3ClientError(Etcd3Exception):
+
+    def __init__(self, code=None, error=None, status=None):
+        if not hasattr(self, 'error'):
+            self.error = error and error.strip()
+        self.codeText = GRPCcodeToText.get(code)
+        self.status = status
+
+    def __repr__(self):
+        return "<{0} error: '{1}', code: {2}>".format(self.__class__.__name__, self.error, self.code)
+
+    __str__ = __repr__
+
+    def as_dict(self):
+        return {'error': self.error, 'code': self.code, 'codeText': self.codeText, 'status': self.status}
+
+    @classmethod
+    def get_subclasses(cls):
+        for subclass in cls.__subclasses__():
+            for subsubclass in subclass.get_subclasses():
+                yield subsubclass
+            yield subclass
+
+
+class Unknown(Etcd3ClientError):
+    code = GRPCCode.Unknown
+
+
+class InvalidArgument(Etcd3ClientError):
+    code = GRPCCode.InvalidArgument
+
+
+class DeadlineExceeded(Etcd3ClientError):
+    code = GRPCCode.DeadlineExceeded
+    error = "context deadline exceeded"
+
+
+class NotFound(Etcd3ClientError):
+    code = GRPCCode.NotFound
+
+
+class FailedPrecondition(Etcd3ClientError):
+    code = GRPCCode.FailedPrecondition
+
+
+class Unavailable(Etcd3ClientError):
+    code = GRPCCode.Unavailable
+
+
+# https://github.com/etcd-io/etcd/blob/master/etcdserver/api/v3rpc/rpctypes/error.go
+class LeaseNotFound(NotFound):
+    error = "etcdserver: requested lease not found"
+
+
+class UserEmpty(InvalidArgument):
+    error = "etcdserver: user name is empty"
+
+
+class AuthFailed(InvalidArgument):
+    error = "etcdserver: authentication failed, invalid user ID or password"
+
+
+class PermissionDenied(Etcd3ClientError):
+    code = GRPCCode.PermissionDenied
+    error = "etcdserver: permission denied"
+
+
+class AuthNotEnabled(FailedPrecondition):
+    error = "etcdserver: authentication is not enabled"
+
+
+class InvalidAuthToken(Etcd3ClientError):
+    code = GRPCCode.Unauthenticated
+    error = "etcdserver: invalid auth token"
+
+
+errStringToClientError = {s.error: s for s in Etcd3ClientError.get_subclasses() if hasattr(s, 'error')}
+errCodeToClientError = {s.code: s for s in Etcd3ClientError.__subclasses__()}
+
+
+def _raise_for_data(data, status_code=None):
+    try:
+        error = data.get('error') or data.get('Error')
+        if isinstance(error, dict):  # streaming response
+            status_code = error.get('http_code')
+            code = error['grpc_code']
+            error = error['message']
+        else:
+            code = data.get('code') or data.get('Code')
+    except Exception:
+        error = str(data)
+        code = GRPCCode.Unknown
+    err = errStringToClientError.get(error) or errCodeToClientError.get(code) or Unknown
+    raise err(code, error, status_code)
+
+
+def to_bytes(v):
+    return v if isinstance(v, bytes) else v.encode('utf-8')
+
+
+def prefix_range_end(v):
+    v = bytearray(to_bytes(v))
+    for i in range(len(v) - 1, -1, -1):
+        if v[i] < 0xff:
+            v[i] += 1
+            break
+    return bytes(v)
+
+
+def base64_encode(v):
+    return base64.b64encode(to_bytes(v)).decode('utf-8')
+
+
+def base64_decode(v):
+    return base64.b64decode(v).decode('utf-8')
+
+
+def build_range_request(key, range_end=None):
+    fields = {'key': base64_encode(key)}
+    if range_end:
+        fields['range_end'] = base64_encode(range_end)
+    return fields
+
+
+class Etcd3Client(AbstractEtcdClientWithFailover):
+
+    ERROR_CLS = Etcd3Error
+
+    def __init__(self, config, dns_resolver, cache_ttl=300):
+        self._token = None
+        self._cluster_version = None
+        self.version_prefix = '/v3beta'
+        super(Etcd3Client, self).__init__(config, dns_resolver, cache_ttl)
+
+        if six.PY2:  # pragma: no cover
+            # Old grpc-gateway sometimes sends double 'transfer-encoding: chunked' headers,
+            # what breaks the old (python2.7) httplib.HTTPConnection (it closes the socket).
+            def dedup_addheader(httpm, key, value):
+                prev = httpm.dict.get(key)
+                if prev is None:
+                    httpm.dict[key] = value
+                elif key != 'transfer-encoding' or prev != value:
+                    combined = ", ".join((prev, value))
+                    httpm.dict[key] = combined
+
+            import httplib
+            httplib.HTTPMessage.addheader = dedup_addheader
+
+        try:
+            self.authenticate()
+        except AuthFailed as e:
+            logger.fatal('Etcd3 authentication failed: %r', e)
+            sys.exit(1)
+
+    def _get_headers(self):
+        headers = urllib3.make_headers(user_agent=USER_AGENT)
+        if self._token and self._cluster_version >= (3, 3, 0):
+            headers['authorization'] = self._token
+        return headers
+
+    def _prepare_request(self, kwargs, params=None, method=None):
+        if params is not None:
+            kwargs['body'] = json.dumps(params)
+            kwargs['headers']['Content-Type'] = 'application/json'
+        return self.http.urlopen
+
+    @staticmethod
+    def _handle_server_response(response):
+        data = response.data
+        try:
+            data = data.decode('utf-8')
+            data = json.loads(data)
+        except (TypeError, ValueError, UnicodeError) as e:
+            if response.status < 400:
+                raise etcd.EtcdException('Server response was not valid JSON: %r' % e)
+        if response.status < 400:
+            return data
+        _raise_for_data(data, response.status)
+
+    def _ensure_version_prefix(self, base_uri, **kwargs):
+        if self.version_prefix != '/v3':
+            response = self.http.urlopen(self._MGET, base_uri + '/version', **kwargs)
+            response = self._handle_server_response(response)
+
+            server_version_str = response['etcdserver']
+            server_version = tuple(int(x) for x in server_version_str.split('.'))
+            cluster_version_str = response['etcdcluster']
+            self._cluster_version = tuple(int(x) for x in cluster_version_str.split('.'))
+
+            if self._cluster_version < (3, 0) or server_version < (3, 0, 4):
+                raise UnsupportedEtcdVersion('Detected Etcd version {0} is lower than 3.0.4'.format(server_version_str))
+
+            if self._cluster_version < (3, 3):
+                if self.version_prefix != '/v3alpha':
+                    if self._cluster_version < (3, 1):
+                        logger.warning('Detected Etcd version %s is lower than 3.1.0, watches are not supported',
+                                       cluster_version_str)
+                    if self.username and self.password:
+                        logger.warning('Detected Etcd version %s is lower than 3.3.0, authentication is not supported',
+                                       cluster_version_str)
+                    self.version_prefix = '/v3alpha'
+            elif self._cluster_version < (3, 4):
+                self.version_prefix = '/v3beta'
+            else:
+                self.version_prefix = '/v3'
+
+    def _prepare_get_members(self, etcd_nodes):
+        kwargs = self._prepare_common_parameters(etcd_nodes)
+        self._prepare_request(kwargs, {})
+        return kwargs
+
+    def _get_members(self, base_uri, **kwargs):
+        self._ensure_version_prefix(base_uri, **kwargs)
+        resp = self.http.urlopen(self._MPOST, base_uri + self.version_prefix + '/cluster/member/list', **kwargs)
+        members = self._handle_server_response(resp)['members']
+        return set(url for member in members for url in member.get('clientURLs', []))
+
+    def call_rpc(self, method, fields, retry=None):
+        fields['retry'] = retry
+        return self.api_execute(self.version_prefix + method, self._MPOST, fields)
+
+    def authenticate(self):
+        if self._use_proxies and self._cluster_version is None:
+            kwargs = self._prepare_common_parameters(1)
+            self._ensure_version_prefix(self._base_uri, **kwargs)
+        if self._cluster_version >= (3, 3) and self.username and self.password:
+            logger.info('Trying to authenticate on Etcd...')
+            old_token, self._token = self._token, None
+            try:
+                response = self.call_rpc('/auth/authenticate', {'name': self.username, 'password': self.password})
+            except AuthNotEnabled:
+                logger.info('Etcd authentication is not enabled')
+                self._token = None
+            except Exception:
+                self._token = old_token
+                raise
+            else:
+                self._token = response.get('token')
+            return old_token != self._token
+
+    def _handle_auth_errors(func):
+        def wrapper(self, *args, **kwargs):
+            def retry(ex):
+                if self.username and self.password:
+                    self.authenticate()
+                    return func(self, *args, **kwargs)
+                else:
+                    logger.fatal('Username or password not set, authentication is not possible')
+                    raise ex
+
+            try:
+                return func(self, *args, **kwargs)
+            except (UserEmpty, PermissionDenied) as e:  # no token provided
+                # PermissionDenied is raised on 3.0 and 3.1
+                if self._cluster_version < (3, 3) and (not isinstance(e, PermissionDenied)
+                                                       or self._cluster_version < (3, 2)):
+                    raise UnsupportedEtcdVersion('Authentication is required by Etcd cluster but not '
+                                                 'supported on version lower than 3.3.0. Cluster version: '
+                                                 '{0}'.format('.'.join(map(str, self._cluster_version))))
+                return retry(e)
+            except InvalidAuthToken as e:
+                logger.error('Invalid auth token: %s', self._token)
+                return retry(e)
+
+        return wrapper
+
+    @_handle_auth_errors
+    def range(self, key, range_end=None, retry=None):
+        params = build_range_request(key, range_end)
+        params['serializable'] = True  # For better performance. We can tolerate stale reads.
+        return self.call_rpc('/kv/range', params, retry)
+
+    def prefix(self, key, retry=None):
+        return self.range(key, prefix_range_end(key), retry)
+
+    def lease_grant(self, ttl, retry=None):
+        return self.call_rpc('/lease/grant', {'TTL': ttl}, retry)['ID']
+
+    def lease_keepalive(self, ID, retry=None):
+        return self.call_rpc('/lease/keepalive', {'ID': ID}, retry).get('result', {}).get('TTL')
+
+    def txn(self, compare, success, retry=None):
+        return self.call_rpc('/kv/txn', {'compare': [compare], 'success': [success]}, retry).get('succeeded')
+
+    @_handle_auth_errors
+    def put(self, key, value, lease=None, create_revision=None, mod_revision=None, retry=None):
+        fields = {'key': base64_encode(key), 'value': base64_encode(value)}
+        if lease:
+            fields['lease'] = lease
+        if create_revision is not None:
+            compare = {'target': 'CREATE', 'create_revision': create_revision}
+        elif mod_revision is not None:
+            compare = {'target': 'MOD', 'mod_revision': mod_revision}
+        else:
+            return self.call_rpc('/kv/put', fields, retry)
+        compare['key'] = fields['key']
+        return self.txn(compare, {'request_put': fields}, retry)
+
+    @_handle_auth_errors
+    def deleterange(self, key, range_end=None, mod_revision=None, retry=None):
+        fields = build_range_request(key, range_end)
+        if mod_revision is None:
+            return self.call_rpc('/kv/deleterange', fields, retry)
+        compare = {'target': 'MOD', 'mod_revision': mod_revision, 'key': fields['key']}
+        return self.txn(compare, {'request_delete_range': fields}, retry)
+
+    def deleteprefix(self, key, retry=None):
+        return self.deleterange(key, prefix_range_end(key), retry=retry)
+
+    def watchrange(self, key, range_end=None, start_revision=None, filters=None):
+        """returns: response object"""
+        params = build_range_request(key, range_end)
+        if start_revision is not None:
+            params['start_revision'] = start_revision
+        params['filters'] = filters or []
+        kwargs = self._prepare_common_parameters(1, self.read_timeout)
+        request_executor = self._prepare_request(kwargs, {'create_request': params})
+        kwargs.update(timeout=urllib3.Timeout(connect=kwargs['timeout']), retries=0)
+        return request_executor(self._MPOST, self._base_uri + self.version_prefix + '/watch', **kwargs)
+
+    def watchprefix(self, key, start_revision=None, filters=None):
+        return self.watchrange(key, prefix_range_end(key), start_revision, filters)
+
+
+class KVCache(Thread):
+
+    def __init__(self, dcs, client):
+        Thread.__init__(self)
+        self.daemon = True
+        self._dcs = dcs
+        self._client = client
+        self.condition = Condition()
+        self._config_key = base64_encode(dcs.config_path)
+        self._leader_key = base64_encode(dcs.leader_path)
+        self._optime_key = base64_encode(dcs.leader_optime_path)
+        self._status_key = base64_encode(dcs.status_path)
+        self._name = base64_encode(dcs._name)
+        self._is_ready = False
+        self._response = None
+        self._response_lock = Lock()
+        self._object_cache = {}
+        self._object_cache_lock = Lock()
+        self.start()
+
+    def set(self, value, overwrite=False):
+        with self._object_cache_lock:
+            name = value['key']
+            old_value = self._object_cache.get(name)
+            ret = not old_value or int(old_value['mod_revision']) < int(value['mod_revision'])
+            if ret or overwrite and old_value['mod_revision'] == value['mod_revision']:
+                self._object_cache[name] = value
+        return ret, old_value
+
+    def delete(self, name, mod_revision):
+        with self._object_cache_lock:
+            old_value = self._object_cache.get(name)
+            ret = old_value and int(old_value['mod_revision']) < int(mod_revision)
+            if ret:
+                del self._object_cache[name]
+        return not old_value or ret, old_value
+
+    def copy(self):
+        with self._object_cache_lock:
+            return [v.copy() for v in self._object_cache.values()]
+
+    def get(self, name):
+        with self._object_cache_lock:
+            return self._object_cache.get(name)
+
+    def _process_event(self, event):
+        kv = event['kv']
+        key = kv['key']
+        if event.get('type') == 'DELETE':
+            success, old_value = self.delete(key, kv['mod_revision'])
+        else:
+            success, old_value = self.set(kv, True)
+
+        if success:
+            old_value = old_value and old_value.get('value')
+            new_value = kv.get('value')
+
+            value_changed = old_value != new_value and \
+                (key == self._leader_key or key in (self._optime_key, self._status_key) and new_value is not None or
+                 key == self._config_key and old_value is not None and new_value is not None)
+
+            if value_changed:
+                logger.debug('%s changed from %s to %s', key, old_value, new_value)
+
+            # We also want to wake up HA loop on replicas if leader optime (or status key) was updated
+            if value_changed and (key not in (self._optime_key, self._status_key) or
+                                  (self.get(self._leader_key) or {}).get('value') != self._name):
+                self._dcs.event.set()
+
+    def _process_message(self, message):
+        logger.debug('Received message: %s', message)
+        if 'error' in message:
+            _raise_for_data(message)
+        for event in message.get('result', {}).get('events', []):
+            self._process_event(event)
+
+    @staticmethod
+    def _finish_response(response):
+        try:
+            response.close()
+        finally:
+            response.release_conn()
+
+    def _do_watch(self, revision):
+        with self._response_lock:
+            self._response = None
+        response = self._client.watchprefix(self._dcs.cluster_prefix, revision)
+        with self._response_lock:
+            if self._response is None:
+                self._response = response
+
+        if not self._response:
+            return self._finish_response(response)
+
+        for message in iter_response_objects(response):
+            self._process_message(message)
+
+    def _build_cache(self):
+        result = self._dcs.retry(self._client.prefix, self._dcs.cluster_prefix)
+        with self._object_cache_lock:
+            self._object_cache = {node['key']: node for node in result.get('kvs', [])}
+        with self.condition:
+            self._is_ready = True
+            self.condition.notify()
+
+        try:
+            self._do_watch(result['header']['revision'])
+        except Exception as e:
+            logger.error('watchprefix failed: %r', e)
+        finally:
+            with self.condition:
+                self._is_ready = False
+            with self._response_lock:
+                response, self._response = self._response, None
+            if response:
+                self._finish_response(response)
+
+    def run(self):
+        while True:
+            try:
+                self._build_cache()
+            except Exception as e:
+                logger.error('KVCache.run %r', e)
+                time.sleep(1)
+
+    def kill_stream(self):
+        sock = None
+        with self._response_lock:
+            if self._response:
+                try:
+                    sock = self._response.connection.sock
+                except Exception:
+                    sock = None
+            else:
+                self._response = False
+        if sock:
+            try:
+                sock.shutdown(socket.SHUT_RDWR)
+                sock.close()
+            except Exception as e:
+                logger.debug('Error on socket.shutdown: %r', e)
+
+    def is_ready(self):
+        """Must be called only when holding the lock on `condition`"""
+        return self._is_ready
+
+
+class PatroniEtcd3Client(Etcd3Client):
+
+    def __init__(self, *args, **kwargs):
+        self._kv_cache = None
+        super(PatroniEtcd3Client, self).__init__(*args, **kwargs)
+
+    def configure(self, etcd3):
+        self._etcd3 = etcd3
+
+    def start_watcher(self):
+        if self._cluster_version >= (3, 1):
+            self._kv_cache = KVCache(self._etcd3, self)
+
+    def _restart_watcher(self):
+        if self._kv_cache:
+            self._kv_cache.kill_stream()
+
+    def set_base_uri(self, value):
+        super(PatroniEtcd3Client, self).set_base_uri(value)
+        self._restart_watcher()
+
+    def authenticate(self):
+        ret = super(PatroniEtcd3Client, self).authenticate()
+        if ret:
+            self._restart_watcher()
+        return ret
+
+    def _wait_cache(self, timeout):
+        stop_time = time.time() + timeout
+        while not self._kv_cache.is_ready():
+            timeout = stop_time - time.time()
+            if timeout <= 0:
+                raise RetryFailedError('Exceeded retry deadline')
+            self._kv_cache.condition.wait(timeout)
+
+    def get_cluster(self):
+        if self._kv_cache:
+            with self._kv_cache.condition:
+                self._wait_cache(self._etcd3._retry.deadline)
+                return self._kv_cache.copy()
+        else:
+            return self._etcd3.retry(self.prefix, self._etcd3.cluster_prefix).get('kvs', [])
+
+    def call_rpc(self, method, fields, retry=None):
+        ret = super(PatroniEtcd3Client, self).call_rpc(method, fields, retry)
+
+        if self._kv_cache:
+            value = delete = None
+            if method == '/kv/txn' and ret.get('succeeded'):
+                on_success = fields['success'][0]
+                value = on_success.get('request_put')
+                delete = on_success.get('request_delete_range')
+            elif method == '/kv/put' and ret:
+                value = fields
+            elif method == '/kv/deleterange' and ret:
+                delete = fields
+
+            if value:
+                value['mod_revision'] = ret['header']['revision']
+                self._kv_cache.set(value)
+            elif delete and 'range_end' not in delete:
+                self._kv_cache.delete(delete['key'], ret['header']['revision'])
+
+        return ret
+
+
+class Etcd3(AbstractEtcd):
+
+    def __init__(self, config):
+        super(Etcd3, self).__init__(config, PatroniEtcd3Client, (DeadlineExceeded, Unavailable, FailedPrecondition))
+        self.__do_not_watch = False
+        self._lease = None
+        self._last_lease_refresh = 0
+
+        self._client.configure(self)
+        if not self._ctl:
+            self._client.start_watcher()
+            self.create_lease()
+
+    def set_socket_options(self, sock, socket_options):
+        enable_keepalive(sock, self.ttl, int(self.loop_wait + self._retry.deadline))
+
+    def set_ttl(self, ttl):
+        self.__do_not_watch = super(Etcd3, self).set_ttl(ttl)
+        if self.__do_not_watch:
+            self._lease = None
+
+    def _do_refresh_lease(self, retry=None):
+        if self._lease and self._last_lease_refresh + self._loop_wait > time.time():
+            return False
+
+        if self._lease and not self._client.lease_keepalive(self._lease, retry):
+            self._lease = None
+
+        ret = not self._lease
+        if ret:
+            self._lease = self._client.lease_grant(self._ttl, retry)
+
+        self._last_lease_refresh = time.time()
+        return ret
+
+    def refresh_lease(self):
+        try:
+            return self.retry(self._do_refresh_lease)
+        except (Etcd3ClientError, RetryFailedError):
+            logger.exception('refresh_lease')
+        raise Etcd3Error('Failed ro keepalive/grant lease')
+
+    def create_lease(self):
+        while not self._lease:
+            try:
+                self.refresh_lease()
+            except Etcd3Error:
+                logger.info('waiting on etcd')
+                time.sleep(5)
+
+    @property
+    def cluster_prefix(self):
+        return self.client_path('')
+
+    @staticmethod
+    def member(node):
+        return Member.from_node(node['mod_revision'], os.path.basename(node['key']), node['lease'], node['value'])
+
+    def _load_cluster(self):
+        cluster = None
+        try:
+            path_len = len(self.cluster_prefix)
+
+            nodes = {}
+            for node in self._client.get_cluster():
+                node['key'] = base64_decode(node['key'])
+                node['value'] = base64_decode(node.get('value', ''))
+                node['lease'] = node.get('lease')
+                nodes[node['key'][path_len:].lstrip('/')] = node
+
+            # get initialize flag
+            initialize = nodes.get(self._INITIALIZE)
+            initialize = initialize and initialize['value']
+
+            # get global dynamic configuration
+            config = nodes.get(self._CONFIG)
+            config = config and ClusterConfig.from_node(config['mod_revision'], config['value'])
+
+            # get timeline history
+            history = nodes.get(self._HISTORY)
+            history = history and TimelineHistory.from_node(history['mod_revision'], history['value'])
+
+            # get last know leader lsn and slots
+            status = nodes.get(self._STATUS)
+            if status:
+                try:
+                    status = json.loads(status['value'])
+                    last_lsn = status.get(self._OPTIME)
+                    slots = status.get('slots')
+                except Exception:
+                    slots = last_lsn = None
+            else:
+                last_lsn = nodes.get(self._LEADER_OPTIME)
+                last_lsn = last_lsn and last_lsn['value']
+                slots = None
+
+            try:
+                last_lsn = int(last_lsn)
+            except Exception:
+                last_lsn = 0
+
+            # get list of members
+            members = [self.member(n) for k, n in nodes.items() if k.startswith(self._MEMBERS) and k.count('/') == 1]
+
+            # get leader
+            leader = nodes.get(self._LEADER)
+            if not self._ctl and leader and leader['value'] == self._name and self._lease != leader.get('lease'):
+                logger.warning('I am the leader but not owner of the lease')
+
+            if leader:
+                member = Member(-1, leader['value'], None, {})
+                member = ([m for m in members if m.name == leader['value']] or [member])[0]
+                leader = Leader(leader['mod_revision'], leader['lease'], member)
+
+            # failover key
+            failover = nodes.get(self._FAILOVER)
+            if failover:
+                failover = Failover.from_node(failover['mod_revision'], failover['value'])
+
+            # get synchronization state
+            sync = nodes.get(self._SYNC)
+            sync = SyncState.from_node(sync and sync['mod_revision'], sync and sync['value'])
+
+            cluster = Cluster(initialize, config, leader, last_lsn, members, failover, sync, history, slots)
+        except UnsupportedEtcdVersion:
+            raise
+        except Exception as e:
+            self._handle_exception(e, 'get_cluster', raise_ex=Etcd3Error('Etcd is not responding properly'))
+        self._has_failed = False
+        return cluster
+
+    @catch_etcd_errors
+    def touch_member(self, data, permanent=False):
+        if not permanent:
+            try:
+                self.refresh_lease()
+            except Etcd3Error:
+                return False
+
+        cluster = self.cluster
+        member = cluster and cluster.get_member(self._name, fallback_to_leader=False)
+
+        if member and member.session == self._lease and deep_compare(data, member.data):
+            return True
+
+        data = json.dumps(data, separators=(',', ':'))
+        try:
+            return self._client.put(self.member_path, data, None if permanent else self._lease)
+        except LeaseNotFound:
+            self._lease = None
+            logger.error('Our lease disappeared from Etcd, can not "touch_member"')
+
+    @catch_etcd_errors
+    def take_leader(self):
+        return self.retry(self._client.put, self.leader_path, self._name, self._lease)
+
+    @catch_etcd_errors
+    def _do_attempt_to_acquire_leader(self, permanent):
+        try:
+            return self.retry(self._client.put, self.leader_path, self._name, None if permanent else self._lease, 0)
+        except LeaseNotFound:
+            self._lease = None
+            logger.error('Our lease disappeared from Etcd. Will try to get a new one and retry attempt')
+            self.refresh_lease()
+            return self.retry(self._client.put, self.leader_path, self._name, None if permanent else self._lease, 0)
+
+    def attempt_to_acquire_leader(self, permanent=False):
+        if not self._lease and not permanent:
+            self.refresh_lease()
+
+        ret = self._do_attempt_to_acquire_leader(permanent)
+        if not ret:
+            logger.info('Could not take out TTL lock')
+        return ret
+
+    @catch_etcd_errors
+    def set_failover_value(self, value, index=None):
+        return self._client.put(self.failover_path, value, mod_revision=index)
+
+    @catch_etcd_errors
+    def set_config_value(self, value, index=None):
+        return self._client.put(self.config_path, value, mod_revision=index)
+
+    @catch_etcd_errors
+    def _write_leader_optime(self, last_lsn):
+        return self._client.put(self.leader_optime_path, last_lsn)
+
+    @catch_etcd_errors
+    def _write_status(self, value):
+        return self._client.put(self.status_path, value)
+
+    @catch_etcd_errors
+    def _update_leader(self):
+        if not self._lease:
+            self.refresh_lease()
+        elif self.retry(self._client.lease_keepalive, self._lease):
+            self._last_lease_refresh = time.time()
+
+        if self._lease:
+            cluster = self.cluster
+            leader_lease = cluster and isinstance(cluster.leader, Leader) and cluster.leader.session
+            if leader_lease != self._lease:
+                self.take_leader()
+        return bool(self._lease)
+
+    @catch_etcd_errors
+    def initialize(self, create_new=True, sysid=""):
+        return self.retry(self._client.put, self.initialize_path, sysid, None, 0 if create_new else None)
+
+    @catch_etcd_errors
+    def _delete_leader(self):
+        cluster = self.cluster
+        if cluster and isinstance(cluster.leader, Leader) and cluster.leader.name == self._name:
+            return self._client.deleterange(self.leader_path, mod_revision=cluster.leader.index)
+
+    @catch_etcd_errors
+    def cancel_initialization(self):
+        return self.retry(self._client.deleterange, self.initialize_path)
+
+    @catch_etcd_errors
+    def delete_cluster(self):
+        return self.retry(self._client.deleteprefix, self.cluster_prefix)
+
+    @catch_etcd_errors
+    def set_history_value(self, value):
+        return self._client.put(self.history_path, value)
+
+    @catch_etcd_errors
+    def set_sync_state_value(self, value, index=None):
+        return self.retry(self._client.put, self.sync_path, value, mod_revision=index)
+
+    @catch_etcd_errors
+    def delete_sync_state(self, index=None):
+        return self.retry(self._client.deleterange, self.sync_path, mod_revision=index)
+
+    def watch(self, leader_index, timeout):
+        if self.__do_not_watch:
+            self.__do_not_watch = False
+            return True
+
+        try:
+            return super(Etcd3, self).watch(None, timeout)
+        finally:
+            self.event.clear()

patroni/dcs/exhibitor.py

@@ -1,11 +1,11 @@
+import json
 import logging
 import random
-import requests
 import time
 
 from patroni.dcs.zookeeper import ZooKeeper
+from patroni.request import get as requests_get
 from patroni.utils import uri
-from requests.exceptions import RequestException
 
 logger = logging.getLogger(__name__)
 
@@ -48,10 +48,10 @@ class ExhibitorEnsembleProvider(object):
         random.shuffle(exhibitors)
         for host in exhibitors:
             try:
-                response = requests.get(uri('http', (host, self._exhibitor_port), self._uri_path), timeout=self.TIMEOUT)
+                response = requests_get(uri('http', (host, self._exhibitor_port), self._uri_path), timeout=self.TIMEOUT)
-                return response.json()
+                return json.loads(response.data.decode('utf-8'))
-            except RequestException:
+            except Exception:
-                pass
+                logging.debug('Request to %s failed', host)
         return None
 
     @property

patroni/dcs/raft.py

@@ -0,0 +1,423 @@
+import json
+import logging
+import os
+import threading
+import time
+
+from pysyncobj import SyncObj, SyncObjConf, replicated, FAIL_REASON
+from pysyncobj.dns_resolver import globalDnsResolver
+from pysyncobj.node import TCPNode
+from pysyncobj.transport import TCPTransport, CONNECTION_STATE
+from pysyncobj.utility import TcpUtility
+
+from . import AbstractDCS, ClusterConfig, Cluster, Failover, Leader, Member, SyncState, TimelineHistory
+from ..utils import validate_directory
+
+logger = logging.getLogger(__name__)
+
+
+class _TCPTransport(TCPTransport):
+
+    def __init__(self, syncObj, selfNode, otherNodes):
+        super(_TCPTransport, self).__init__(syncObj, selfNode, otherNodes)
+        self.setOnUtilityMessageCallback('members', syncObj.getMembers)
+
+    def _connectIfNecessarySingle(self, node):
+        try:
+            return super(_TCPTransport, self)._connectIfNecessarySingle(node)
+        except Exception as e:
+            logger.debug('Connection to %s failed: %r', node, e)
+            return False
+
+
+def resolve_host(self):
+    return globalDnsResolver().resolve(self.host)
+
+
+setattr(TCPNode, 'ip', property(resolve_host))
+
+
+class SyncObjUtility(object):
+
+    def __init__(self, otherNodes, conf):
+        self._nodes = otherNodes
+        self._utility = TcpUtility(conf.password)
+
+    def executeCommand(self, command):
+        try:
+            return self._utility.executeCommand(self.__node, command)
+        except Exception:
+            return None
+
+    def getMembers(self):
+        for self.__node in self._nodes:
+            response = self.executeCommand(['members'])
+            if response:
+                return [member['addr'] for member in response]
+
+
+class DynMemberSyncObj(SyncObj):
+
+    def __init__(self, selfAddress, partnerAddrs, conf):
+        self.__early_apply_local_log = selfAddress is not None
+        self.applied_local_log = False
+
+        utility = SyncObjUtility(partnerAddrs, conf)
+        members = utility.getMembers()
+        add_self = members and selfAddress not in members
+
+        partnerAddrs = [member for member in (members or partnerAddrs) if member != selfAddress]
+
+        super(DynMemberSyncObj, self).__init__(selfAddress, partnerAddrs, conf, transportClass=_TCPTransport)
+
+        if add_self:
+            thread = threading.Thread(target=utility.executeCommand, args=(['add', selfAddress],))
+            thread.daemon = True
+            thread.start()
+
+    def getMembers(self, args, callback):
+        callback([{'addr': node.id, 'leader': node == self._getLeader(), 'status': CONNECTION_STATE.CONNECTED
+                   if self.isNodeConnected(node) else CONNECTION_STATE.DISCONNECTED} for node in self.otherNodes] +
+                 [{'addr': self.selfNode.id, 'leader': self._isLeader(), 'status': CONNECTION_STATE.CONNECTED}], None)
+
+    def _onTick(self, timeToWait=0.0):
+        super(DynMemberSyncObj, self)._onTick(timeToWait)
+
+        # The SyncObj calls onReady callback only when cluster got the leader and is ready for writes.
+        # In some cases for us it is safe to "signal" the Raft object when the local log is fully applied.
+        # We are using the `applied_local_log` property for that, but not calling the callback function.
+        if self.__early_apply_local_log and not self.applied_local_log and self.raftLastApplied == self.raftCommitIndex:
+            self.applied_local_log = True
+
+
+class KVStoreTTL(DynMemberSyncObj):
+
+    def __init__(self, on_ready, on_set, on_delete, **config):
+        self.__thread = None
+        self.__on_set = on_set
+        self.__on_delete = on_delete
+        self.__limb = {}
+        self.__retry_timeout = None
+
+        self_addr = config.get('self_addr')
+        partner_addrs = set(config.get('partner_addrs', []))
+        if config.get('patronictl'):
+            if self_addr:
+                partner_addrs.add(self_addr)
+            self_addr = None
+
+        # Create raft data_dir if necessary
+        raft_data_dir = config.get('data_dir', '')
+        if raft_data_dir != '':
+            validate_directory(raft_data_dir)
+
+        file_template = (self_addr or '')
+        file_template = file_template.replace(':', '_') if os.name == 'nt' else file_template
+        file_template = os.path.join(raft_data_dir, file_template)
+        conf = SyncObjConf(password=config.get('password'), autoTick=False, appendEntriesUseBatch=False,
+                           bindAddress=config.get('bind_addr'), dnsFailCacheTime=(config.get('loop_wait') or 10),
+                           dnsCacheTime=(config.get('ttl') or 30), commandsWaitLeader=config.get('commandsWaitLeader'),
+                           fullDumpFile=(file_template + '.dump' if self_addr else None),
+                           journalFile=(file_template + '.journal' if self_addr else None),
+                           onReady=on_ready, dynamicMembershipChange=True)
+
+        super(KVStoreTTL, self).__init__(self_addr, partner_addrs, conf)
+        self.__data = {}
+
+    @staticmethod
+    def __check_requirements(old_value, **kwargs):
+        return ('prevExist' not in kwargs or bool(kwargs['prevExist']) == bool(old_value)) and \
+            ('prevValue' not in kwargs or old_value and old_value['value'] == kwargs['prevValue']) and \
+            (not kwargs.get('prevIndex') or old_value and old_value['index'] == kwargs['prevIndex'])
+
+    def set_retry_timeout(self, retry_timeout):
+        self.__retry_timeout = retry_timeout
+
+    def retry(self, func, *args, **kwargs):
+        event = threading.Event()
+        ret = {'result': None, 'error': -1}
+
+        def callback(result, error):
+            ret.update(result=result, error=error)
+            event.set()
+
+        kwargs['callback'] = callback
+        timeout = kwargs.pop('timeout', None) or self.__retry_timeout
+        deadline = timeout and time.time() + timeout
+
+        while True:
+            event.clear()
+            func(*args, **kwargs)
+            event.wait(timeout)
+            if ret['error'] == FAIL_REASON.SUCCESS:
+                return ret['result']
+            elif ret['error'] == FAIL_REASON.REQUEST_DENIED:
+                break
+            elif deadline:
+                timeout = deadline - time.time()
+                if timeout <= 0:
+                    break
+            time.sleep(1)
+        return False
+
+    @replicated
+    def _set(self, key, value, **kwargs):
+        old_value = self.__data.get(key, {})
+        if not self.__check_requirements(old_value, **kwargs):
+            return False
+
+        if old_value and old_value['created'] != value['created']:
+            value['created'] = value['updated']
+        value['index'] = self.raftLastApplied + 1
+
+        self.__data[key] = value
+        if self.__on_set:
+            self.__on_set(key, value)
+        return True
+
+    def set(self, key, value, ttl=None, **kwargs):
+        old_value = self.__data.get(key, {})
+        if not self.__check_requirements(old_value, **kwargs):
+            return False
+
+        value = {'value': value, 'updated': time.time()}
+        value['created'] = old_value.get('created', value['updated'])
+        if ttl:
+            value['expire'] = value['updated'] + ttl
+        return self.retry(self._set, key, value, **kwargs)
+
+    def __pop(self, key):
+        self.__data.pop(key)
+        if self.__on_delete:
+            self.__on_delete(key)
+
+    @replicated
+    def _delete(self, key, recursive=False, **kwargs):
+        if recursive:
+            for k in list(self.__data.keys()):
+                if k.startswith(key):
+                    self.__pop(k)
+        elif not self.__check_requirements(self.__data.get(key, {}), **kwargs):
+            return False
+        else:
+            self.__pop(key)
+        return True
+
+    def delete(self, key, recursive=False, **kwargs):
+        if not recursive and not self.__check_requirements(self.__data.get(key, {}), **kwargs):
+            return False
+        return self.retry(self._delete, key, recursive=recursive, **kwargs)
+
+    @staticmethod
+    def __values_match(old, new):
+        return all(old.get(n) == new.get(n) for n in ('created', 'updated', 'expire', 'value'))
+
+    @replicated
+    def _expire(self, key, value, callback=None):
+        current = self.__data.get(key)
+        if current and self.__values_match(current, value):
+            self.__pop(key)
+
+    def __expire_keys(self):
+        for key, value in self.__data.items():
+            if value and 'expire' in value and value['expire'] <= time.time() and \
+                    not (key in self.__limb and self.__values_match(self.__limb[key], value)):
+                self.__limb[key] = value
+
+                def callback(*args):
+                    if key in self.__limb and self.__values_match(self.__limb[key], value):
+                        self.__limb.pop(key)
+                self._expire(key, value, callback=callback)
+
+    def get(self, key, recursive=False):
+        if not recursive:
+            return self.__data.get(key)
+        return {k: v for k, v in self.__data.items() if k.startswith(key)}
+
+    def _onTick(self, timeToWait=0.0):
+        super(KVStoreTTL, self)._onTick(timeToWait)
+
+        if self._isLeader():
+            self.__expire_keys()
+        else:
+            self.__limb.clear()
+
+    def _autoTickThread(self):
+        self.__destroying = False
+        while not self.__destroying:
+            self.doTick(self.conf.autoTickPeriod)
+
+    def startAutoTick(self):
+        self.__thread = threading.Thread(target=self._autoTickThread)
+        self.__thread.daemon = True
+        self.__thread.start()
+
+    def destroy(self):
+        if self.__thread:
+            self.__destroying = True
+            self.__thread.join()
+        super(KVStoreTTL, self).destroy()
+
+
+class Raft(AbstractDCS):
+
+    def __init__(self, config):
+        super(Raft, self).__init__(config)
+        self._ttl = int(config.get('ttl') or 30)
+
+        ready_event = threading.Event()
+        self._sync_obj = KVStoreTTL(ready_event.set, self._on_set, self._on_delete, commandsWaitLeader=False, **config)
+        self._sync_obj.startAutoTick()
+
+        while True:
+            ready_event.wait(5)
+            if ready_event.isSet() or self._sync_obj.applied_local_log:
+                break
+            else:
+                logger.info('waiting on raft')
+        self.set_retry_timeout(int(config.get('retry_timeout') or 10))
+
+    def _on_set(self, key, value):
+        leader = (self._sync_obj.get(self.leader_path) or {}).get('value')
+        if key == value['created'] == value['updated'] and \
+                (key.startswith(self.members_path) or key == self.leader_path and leader != self._name) or \
+                key in (self.leader_optime_path, self.status_path) and leader != self._name or \
+                key in (self.config_path, self.sync_path):
+            self.event.set()
+
+    def _on_delete(self, key):
+        if key == self.leader_path:
+            self.event.set()
+
+    def set_ttl(self, ttl):
+        self._ttl = ttl
+
+    @property
+    def ttl(self):
+        return self._ttl
+
+    def set_retry_timeout(self, retry_timeout):
+        self._sync_obj.set_retry_timeout(retry_timeout)
+
+    def reload_config(self, config):
+        super(Raft, self).reload_config(config)
+        globalDnsResolver().setTimeouts(self.ttl, self.loop_wait)
+
+    @staticmethod
+    def member(key, value):
+        return Member.from_node(value['index'], os.path.basename(key), None, value['value'])
+
+    def _load_cluster(self):
+        prefix = self.client_path('')
+        response = self._sync_obj.get(prefix, recursive=True)
+        if not response:
+            return Cluster(None, None, None, None, [], None, None, None, None)
+        nodes = {os.path.relpath(key, prefix).replace('\\', '/'): value for key, value in response.items()}
+
+        # get initialize flag
+        initialize = nodes.get(self._INITIALIZE)
+        initialize = initialize and initialize['value']
+
+        # get global dynamic configuration
+        config = nodes.get(self._CONFIG)
+        config = config and ClusterConfig.from_node(config['index'], config['value'])
+
+        # get timeline history
+        history = nodes.get(self._HISTORY)
+        history = history and TimelineHistory.from_node(history['index'], history['value'])
+
+        # get last know leader lsn and slots
+        status = nodes.get(self._STATUS)
+        if status:
+            try:
+                status = json.loads(status['value'])
+                last_lsn = status.get(self._OPTIME)
+                slots = status.get('slots')
+            except Exception:
+                slots = last_lsn = None
+        else:
+            last_lsn = nodes.get(self._LEADER_OPTIME)
+            last_lsn = last_lsn and last_lsn['value']
+            slots = None
+
+        try:
+            last_lsn = int(last_lsn)
+        except Exception:
+            last_lsn = 0
+
+        # get list of members
+        members = [self.member(k, n) for k, n in nodes.items() if k.startswith(self._MEMBERS) and k.count('/') == 1]
+
+        # get leader
+        leader = nodes.get(self._LEADER)
+        if leader:
+            member = Member(-1, leader['value'], None, {})
+            member = ([m for m in members if m.name == leader['value']] or [member])[0]
+            leader = Leader(leader['index'], None, member)
+
+        # failover key
+        failover = nodes.get(self._FAILOVER)
+        if failover:
+            failover = Failover.from_node(failover['index'], failover['value'])
+
+        # get synchronization state
+        sync = nodes.get(self._SYNC)
+        sync = SyncState.from_node(sync and sync['index'], sync and sync['value'])
+
+        return Cluster(initialize, config, leader, last_lsn, members, failover, sync, history, slots)
+
+    def _write_leader_optime(self, last_lsn):
+        return self._sync_obj.set(self.leader_optime_path, last_lsn, timeout=1)
+
+    def _write_status(self, value):
+        return self._sync_obj.set(self.status_path, value, timeout=1)
+
+    def _update_leader(self):
+        ret = self._sync_obj.set(self.leader_path, self._name, ttl=self._ttl, prevValue=self._name)
+        if not ret and self._sync_obj.get(self.leader_path) is None:
+            ret = self.attempt_to_acquire_leader()
+        return ret
+
+    def attempt_to_acquire_leader(self, permanent=False):
+        return self._sync_obj.set(self.leader_path, self._name, prevExist=False,
+                                  ttl=None if permanent else self._ttl)
+
+    def set_failover_value(self, value, index=None):
+        return self._sync_obj.set(self.failover_path, value, prevIndex=index)
+
+    def set_config_value(self, value, index=None):
+        return self._sync_obj.set(self.config_path, value, prevIndex=index)
+
+    def touch_member(self, data, permanent=False):
+        data = json.dumps(data, separators=(',', ':'))
+        return self._sync_obj.set(self.member_path, data, None if permanent else self._ttl, timeout=2)
+
+    def take_leader(self):
+        return self._sync_obj.set(self.leader_path, self._name, ttl=self._ttl)
+
+    def initialize(self, create_new=True, sysid=''):
+        return self._sync_obj.set(self.initialize_path, sysid, prevExist=(not create_new))
+
+    def _delete_leader(self):
+        return self._sync_obj.delete(self.leader_path, prevValue=self._name, timeout=1)
+
+    def cancel_initialization(self):
+        return self._sync_obj.delete(self.initialize_path)
+
+    def delete_cluster(self):
+        return self._sync_obj.delete(self.client_path(''), recursive=True)
+
+    def set_history_value(self, value):
+        return self._sync_obj.set(self.history_path, value)
+
+    def set_sync_state_value(self, value, index=None):
+        return self._sync_obj.set(self.sync_path, value, prevIndex=index)
+
+    def delete_sync_state(self, index=None):
+        return self._sync_obj.delete(self.sync_path, prevIndex=index)
+
+    def watch(self, leader_index, timeout):
+        try:
+            return super(Raft, self).watch(leader_index, timeout)
+        finally:
+            self.event.clear()

patroni/dcs/zookeeper.py

@@ -4,11 +4,13 @@ import select
 import time
 
 from kazoo.client import KazooClient, KazooState, KazooRetry
-from kazoo.exceptions import NoNodeError, NodeExistsError
+from kazoo.exceptions import NoNodeError, NodeExistsError, SessionExpiredError
 from kazoo.handlers.threading import SequentialThreadingHandler
-from patroni.dcs import AbstractDCS, ClusterConfig, Cluster, Failover, Leader, Member, SyncState, TimelineHistory
+from kazoo.protocol.states import KeeperState
-from patroni.exceptions import DCSError
+
-from patroni.utils import deep_compare
+from . import AbstractDCS, ClusterConfig, Cluster, Failover, Leader, Member, SyncState, TimelineHistory
+from ..exceptions import DCSError
+from ..utils import deep_compare
 
 logger = logging.getLogger(__name__)
 
@@ -54,6 +56,20 @@ class PatroniSequentialThreadingHandler(SequentialThreadingHandler):
             raise select.error(9, str(e))
 
 
+class PatroniKazooClient(KazooClient):
+
+    def _call(self, request, async_object):
+        # Before kazoo==2.7.0 it wasn't possible to send requests to zookeeper if
+        # the connection is in the SUSPENDED state and Patroni was strongly relying on it.
+        # The https://github.com/python-zk/kazoo/pull/588 changed it, and now such requests are queued.
+        # We override the `_call()` method in order to keep the old behavior.
+
+        if self._state == KeeperState.CONNECTING:
+            async_object.set_exception(SessionExpiredError())
+            return False
+        return super(PatroniKazooClient, self)._call(request, async_object)
+
+
 class ZooKeeper(AbstractDCS):
 
     def __init__(self, config):
@@ -63,20 +79,25 @@ class ZooKeeper(AbstractDCS):
         if isinstance(hosts, list):
             hosts = ','.join(hosts)
 
-        self._client = KazooClient(hosts, handler=PatroniSequentialThreadingHandler(config['retry_timeout']),
+        mapping = {'use_ssl': 'use_ssl', 'verify': 'verify_certs', 'cacert': 'ca',
-                                   timeout=config['ttl'], connection_retry=KazooRetry(max_delay=1, max_tries=-1,
+                   'cert': 'certfile', 'key': 'keyfile', 'key_password': 'keyfile_password'}
-                                   sleep_func=time.sleep), command_retry=KazooRetry(deadline=config['retry_timeout'],
+        kwargs = {v: config[k] for k, v in mapping.items() if k in config}
-                                   max_delay=1, max_tries=-1, sleep_func=time.sleep))
+
+        self._client = PatroniKazooClient(hosts, handler=PatroniSequentialThreadingHandler(config['retry_timeout']),
+                                          timeout=config['ttl'], connection_retry=KazooRetry(max_delay=1, max_tries=-1,
+                                          sleep_func=time.sleep), command_retry=KazooRetry(max_delay=1, max_tries=-1,
+                                          deadline=config['retry_timeout'], sleep_func=time.sleep), **kwargs)
         self._client.add_listener(self.session_listener)
 
         self._fetch_cluster = True
+        self._fetch_status = True
 
         self._orig_kazoo_connect = self._client._connection._connect
         self._client._connection._connect = self._kazoo_connect
 
         self._client.start()
 
-    def _kazoo_connect(self, host, port):
+    def _kazoo_connect(self, *args):
         """Kazoo is using Ping's to determine health of connection to zookeeper. If there is no
         response on Ping after Ping interval (1/2 from read_timeout) it will consider current
         connection dead and try to connect to another node. Without this "magic" it was taking
@@ -88,16 +109,20 @@ class ZooKeeper(AbstractDCS):
         than loop_wait, because we can spend up to 2 seconds when calling `touch_member()` and
         `write_leader_optime()` methods, which also may hang..."""
 
-        ret = self._orig_kazoo_connect(host, port)
+        ret = self._orig_kazoo_connect(*args)
         return max(self.loop_wait - 2, 2)*1000, ret[1]
 
     def session_listener(self, state):
         if state in [KazooState.SUSPENDED, KazooState.LOST]:
             self.cluster_watcher(None)
 
+    def status_watcher(self, event):
+        self._fetch_status = True
+        self.event.set()
+
     def cluster_watcher(self, event):
         self._fetch_cluster = True
-        self.event.set()
+        self.status_watcher(event)
 
     def reload_config(self, config):
         self.set_retry_timeout(config['retry_timeout'])
@@ -114,13 +139,14 @@ class ZooKeeper(AbstractDCS):
         # the same time, set_ttl method will reestablish connection and return
         # `!True`, otherwise we will close existing connection and let kazoo
         # open the new one.
-        if not self.set_ttl(int(config['ttl'] * 1000)) and loop_wait_changed:
+        if not self.set_ttl(config['ttl']) and loop_wait_changed:
             self._client._connection._socket.close()
 
     def set_ttl(self, ttl):
         """It is not possible to change ttl (session_timeout) in zookeeper without
         destroying old session and creating the new one. This method returns `!True`
         if session_timeout has been changed (`restart()` has been called)."""
+        ttl = int(ttl * 1000)
         if self._client._session_timeout != ttl:
             self._client._session_timeout = ttl
             self._client.restart()
@@ -128,7 +154,7 @@ class ZooKeeper(AbstractDCS):
 
     @property
     def ttl(self):
-        return self._client._session_timeout
+        return self._client._session_timeout / 1000.0
 
     def set_retry_timeout(self, retry_timeout):
         retry = self._client.retry if isinstance(self._client.retry, KazooRetry) else self._client._retry
@@ -141,6 +167,30 @@ class ZooKeeper(AbstractDCS):
         except NoNodeError:
             return None
 
+    def get_status(self, leader):
+        watch = self.status_watcher if not leader or leader.name != self._name else None
+
+        status = self.get_node(self.status_path, watch)
+        if status:
+            try:
+                status = json.loads(status[0])
+                last_lsn = status.get(self._OPTIME)
+                slots = status.get('slots')
+            except Exception:
+                slots = last_lsn = None
+        else:
+            last_lsn = self.get_node(self.leader_optime_path, watch)
+            last_lsn = last_lsn and last_lsn[0]
+            slots = None
+
+        try:
+            last_lsn = int(last_lsn)
+        except Exception:
+            last_lsn = 0
+
+        self._fetch_status = False
+        return last_lsn, slots
+
     @staticmethod
     def member(name, value, znode):
         return Member.from_node(znode.version, name, znode.ephemeralOwner, value)
@@ -151,11 +201,10 @@ class ZooKeeper(AbstractDCS):
         except NoNodeError:
             return []
 
-    def load_members(self, sync_standby):
+    def load_members(self):
         members = []
         for member in self.get_children(self.members_path, self.cluster_watcher):
-            watch = member == sync_standby and self.cluster_watcher or None
+            data = self.get_node(self.members_path + member)
-            data = self.get_node(self.members_path + member, watch)
             if data is not None:
                 members.append(self.member(member, *data))
         return members
@@ -178,17 +227,12 @@ class ZooKeeper(AbstractDCS):
         history = self.get_node(self.history_path, watch=self.cluster_watcher) if self._HISTORY in nodes else None
         history = history and TimelineHistory.from_node(history[1].mzxid, history[0])
 
-        # get last leader operation
-        last_leader_operation = self._OPTIME in nodes and self._fetch_cluster and self.get_node(self.leader_optime_path)
-        last_leader_operation = last_leader_operation and int(last_leader_operation[0]) or 0
-
         # get synchronization state
         sync = self.get_node(self.sync_path, watch=self.cluster_watcher) if self._SYNC in nodes else None
         sync = SyncState.from_node(sync and sync[1].version, sync and sync[0])
 
         # get list of members
-        sync_standby = sync.leader == self._name and sync.sync_standby or None
+        members = self.load_members() if self._MEMBERS[:-1] in nodes else []
-        members = self.load_members(sync_standby) if self._MEMBERS[:-1] in nodes else []
 
         # get leader
         leader = self.get_node(self.leader_path) if self._LEADER in nodes else None
@@ -206,11 +250,14 @@ class ZooKeeper(AbstractDCS):
                 leader = Leader(leader[1].version, leader[1].ephemeralOwner, member)
                 self._fetch_cluster = member.index == -1
 
+        # get last known leader lsn and slots
+        last_lsn, slots = self.get_status(leader)
+
         # failover key
         failover = self.get_node(self.failover_path, watch=self.cluster_watcher) if self._FAILOVER in nodes else None
         failover = failover and Failover.from_node(failover[1].version, failover[0])
 
-        return Cluster(initialize, config, leader, last_leader_operation, members, failover, sync, history)
+        return Cluster(initialize, config, leader, last_lsn, members, failover, sync, history, slots)
 
     def _load_cluster(self):
         cluster = self.cluster
@@ -221,8 +268,22 @@ class ZooKeeper(AbstractDCS):
                 logger.exception('get_cluster')
                 self.cluster_watcher(None)
                 raise ZooKeeperError('ZooKeeper in not responding properly')
+        # The /status ZNode was updated or doesn't exist and we are not leader
+        elif (self._fetch_status and not self._fetch_cluster or not cluster.last_lsn
+              or cluster.has_permanent_logical_slots(self._name, False) and not cluster.slots) and\
+                not (cluster.leader and cluster.leader.name == self._name):
+            try:
+                last_lsn, slots = self.get_status(cluster.leader)
+                self.event.clear()
+                cluster = Cluster(cluster.initialize, cluster.config, cluster.leader, last_lsn,
+                                  cluster.members, cluster.failover, cluster.sync, cluster.history, slots)
+            except Exception:
+                pass
         return cluster
 
+    def _bypass_caches(self):
+        self._fetch_cluster = True
+
     def _create(self, path, value, retry=False, ephemeral=False):
         try:
             if retry:
@@ -309,13 +370,16 @@ class ZooKeeper(AbstractDCS):
     def take_leader(self):
         return self.attempt_to_acquire_leader()
 
-    def _write_leader_optime(self, last_operation):
+    def _write_leader_optime(self, last_lsn):
-        return self._set_or_create(self.leader_optime_path, last_operation)
+        return self._set_or_create(self.leader_optime_path, last_lsn)
+
+    def _write_status(self, value):
+        return self._set_or_create(self.status_path, value)
 
     def _update_leader(self):
         return True
 
-    def delete_leader(self):
+    def _delete_leader(self):
         self._client.restart()
         return True
 
@@ -346,6 +410,7 @@ class ZooKeeper(AbstractDCS):
         return self.set_sync_state_value("{}", index)
 
     def watch(self, leader_index, timeout):
-        if super(ZooKeeper, self).watch(leader_index, timeout):
+        ret = super(ZooKeeper, self).watch(leader_index, timeout + 0.5)
+        if ret and not self._fetch_status:
             self._fetch_cluster = True
-        return self._fetch_cluster
+        return ret or self._fetch_cluster

patroni/exceptions.py

@@ -13,6 +13,10 @@ class PatroniException(Exception):
         return repr(self.value)
 
 
+class PatroniFatalException(PatroniException):
+    pass
+
+
 class PostgresException(PatroniException):
     pass
 
@@ -27,3 +31,7 @@ class PostgresConnectionException(PostgresException):
 
 class WatchdogError(PatroniException):
     pass
+
+
+class ConfigParseError(PatroniException):
+    pass

patroni/ha.py

@@ -3,7 +3,6 @@ import functools
 import json
 import logging
 import psycopg2
-import requests
 import sys
 import time
 import uuid
@@ -11,11 +10,11 @@ import uuid
 from collections import namedtuple
 from multiprocessing.pool import ThreadPool
 from patroni.async_executor import AsyncExecutor, CriticalTask
-from patroni.exceptions import DCSError, PostgresConnectionException, PatroniException
+from patroni.exceptions import DCSError, PostgresConnectionException, PatroniFatalException
 from patroni.postgresql import ACTION_ON_START, ACTION_ON_ROLE_CHANGE
 from patroni.postgresql.misc import postgres_version_to_int
 from patroni.postgresql.rewind import Rewind
-from patroni.utils import polling_loop, tzutc
+from patroni.utils import polling_loop, tzutc, is_standby_cluster as _is_standby_cluster, parse_int
 from patroni.dcs import RemoteMember
 from threading import RLock
 
@@ -67,12 +66,12 @@ class Ha(object):
         self.old_cluster = None
         self._is_leader = False
         self._is_leader_lock = RLock()
-        self._leader_access_is_restricted = False
         self._was_paused = False
         self._leader_timeline = None
         self.recovering = False
-        self._post_bootstrap_task = None
+        self._async_response = CriticalTask()
         self._crash_recovery_executed = False
+        self._crash_recovery_started = None
         self._start_timeout = None
         self._async_executor = AsyncExecutor(self.state_handler.cancellable, self.wakeup)
         self.watchdog = patroni.watchdog
@@ -85,9 +84,12 @@ class Ha(object):
         self._disable_sync = 0
 
         # We need following property to avoid shutdown of postgres when join of Patroni to the postgres
-        # already running as replica was aborted due to cluster not beeing initialized in DCS.
+        # already running as replica was aborted due to cluster not being initialized in DCS.
         self._join_aborted = False
 
+        # used only in backoff after failing a pre_promote script
+        self._released_leader_key_timestamp = 0
+
     def check_mode(self, mode):
         # Try to protect from the case when DCS was wiped out during pause
         if self.cluster and self.cluster.config and self.cluster.config.modify_index:
@@ -95,6 +97,11 @@ class Ha(object):
         else:
             return self.patroni.config.check_mode(mode)
 
+    def master_stop_timeout(self):
+        """ Master stop timeout """
+        ret = parse_int(self.patroni.config['master_stop_timeout'])
+        return ret if ret and ret > 0 and self.is_synchronous_mode() else None
+
     def is_paused(self):
         return self.check_mode('pause')
 
@@ -109,22 +116,16 @@ class Ha(object):
         return config.get('standby_cluster')
 
     def is_standby_cluster(self):
-        config = self.get_standby_cluster_config()
+        return _is_standby_cluster(self.get_standby_cluster_config())
-        # Check whether or not provided configuration describes a standby cluster
-        return isinstance(config, dict) and (config.get('host') or config.get('port') or config.get('restore_command'))
 
     def is_leader(self):
         with self._is_leader_lock:
-            return self._is_leader > time.time() and not self._leader_access_is_restricted
+            return self._is_leader > time.time()
 
     def set_is_leader(self, value):
         with self._is_leader_lock:
             self._is_leader = time.time() + self.dcs.ttl if value else 0
 
-    def set_leader_access_is_restricted(self, value):
-        with self._is_leader_lock:
-            self._leader_access_is_restricted = value
-
     def load_cluster_from_dcs(self):
         cluster = self.dcs.get_cluster()
 
@@ -139,19 +140,23 @@ class Ha(object):
         self._leader_timeline = None if cluster.is_unlocked() else cluster.leader.timeline
 
     def acquire_lock(self):
-        self.set_leader_access_is_restricted(self.cluster.has_permanent_logical_slots(self.state_handler.name))
         ret = self.dcs.attempt_to_acquire_leader()
         self.set_is_leader(ret)
         return ret
 
     def update_lock(self, write_leader_optime=False):
-        last_operation = None
+        last_lsn = slots = None
         if write_leader_optime:
             try:
-                last_operation = self.state_handler.last_operation()
+                last_lsn = self.state_handler.last_operation()
+                slots = self.state_handler.slots()
             except Exception:
                 logger.exception('Exception when called state_handler.last_operation()')
-        ret = self.dcs.update_leader(last_operation, self._leader_access_is_restricted)
+        try:
+            ret = self.dcs.update_leader(last_lsn, slots)
+        except Exception:
+            logger.exception('Unexpected exception raised from update_leader, please report it as a BUG')
+            ret = False
         self.set_is_leader(ret)
         if ret:
             self.watchdog.keepalive()
@@ -194,10 +199,22 @@ class Ha(object):
             if self._async_executor.scheduled_action in (None, 'promote') \
                     and data['state'] in ['running', 'restarting', 'starting']:
                 try:
-                    timeline, wal_position = self.state_handler.timeline_wal_position()
+                    timeline, wal_position, pg_control_timeline = self.state_handler.timeline_wal_position()
                     data['xlog_location'] = wal_position
+                    if not timeline:  # try pg_stat_wal_receiver to get the timeline
+                        timeline = self.state_handler.received_timeline()
                     if not timeline:
-                        timeline = self.state_handler.replica_cached_timeline(self._leader_timeline)
+                        # So far the only way to get the current timeline on the standby is from
+                        # the replication connection. In order to avoid opening the replication
+                        # connection on every iteration of HA loop we will do it only when noticed
+                        # that the timeline on the primary has changed.
+                        # Unfortunately such optimization isn't possible on the standby_leader,
+                        # therefore we will get the timeline from pg_control, either by calling
+                        # pg_control_checkpoint() on 9.6+ or by parsing the output of pg_controldata.
+                        if self.state_handler.role == 'standby_leader':
+                            timeline = pg_control_timeline or self.state_handler.pg_control_timeline()
+                        else:
+                            timeline = self.state_handler.replica_cached_timeline(self._leader_timeline)
                     if timeline:
                         data['timeline'] = timeline
                 except Exception:
@@ -231,27 +248,23 @@ class Ha(object):
             clone_member = self.cluster.get_clone_member(self.state_handler.name)
             member_role = 'leader' if clone_member == self.cluster.leader else 'replica'
             msg = "from {0} '{1}'".format(member_role, clone_member.name)
-            self._async_executor.schedule('bootstrap {0}'.format(msg))
+            ret = self._async_executor.try_run_async('bootstrap {0}'.format(msg), self.clone, args=(clone_member, msg))
-            self._async_executor.run_async(self.clone, args=(clone_member, msg))
+            return ret or 'trying to bootstrap {0}'.format(msg)
-            return 'trying to bootstrap {0}'.format(msg)
 
         # no initialize key and node is allowed to be master and has 'bootstrap' section in a configuration file
         elif self.cluster.initialize is None and not self.patroni.nofailover and 'bootstrap' in self.patroni.config:
             if self.dcs.initialize(create_new=True):  # race for initialization
                 self.state_handler.bootstrapping = True
-                self._post_bootstrap_task = CriticalTask()
+                with self._async_response:
+                    self._async_response.reset()
 
                 if self.is_standby_cluster():
-                    self._async_executor.schedule('bootstrap_standby_leader')
+                    ret = self._async_executor.try_run_async('bootstrap_standby_leader', self.bootstrap_standby_leader)
-                    self._async_executor.run_async(self.bootstrap_standby_leader)
+                    return ret or 'trying to bootstrap a new standby leader'
-                    return 'trying to bootstrap a new standby leader'
                 else:
-                    self._async_executor.schedule('bootstrap')
+                    ret = self._async_executor.try_run_async('bootstrap', self.state_handler.bootstrap.bootstrap,
-                    self._async_executor.run_async(
+                                                             args=(self.patroni.config['bootstrap'],))
-                        self.state_handler.bootstrap.bootstrap,
+                    return ret or 'trying to bootstrap a new cluster'
-                        args=(self.patroni.config['bootstrap'],)
-                    )
-                    return 'trying to bootstrap a new cluster'
             else:
                 return 'failed to acquire initialize lock'
         else:
@@ -259,9 +272,7 @@ class Ha(object):
                                      if self.is_standby_cluster() else None
             if self.state_handler.can_create_replica_without_replication_connection(create_replica_methods):
                 msg = 'bootstrap (without leader)'
-                self._async_executor.schedule(msg)
+                return self._async_executor.try_run_async(msg, self.clone) or 'trying to ' + msg
-                self._async_executor.run_async(self.clone)
-                return 'trying to ' + msg
             return 'waiting for {0}leader to bootstrap'.format('standby_' if self.is_standby_cluster() else '')
 
     def bootstrap_standby_leader(self):
@@ -272,7 +283,8 @@ class Ha(object):
         clone_source = self.get_remote_master()
         msg = 'clone from remote master {0}'.format(clone_source.conn_url)
         result = self.clone(clone_source, msg)
-        self._post_bootstrap_task.complete(result)
+        with self._async_response:  # pretend that post_bootstrap was already executed
+            self._async_response.complete(result)
         if result:
             self.state_handler.set_role('standby_leader')
 
@@ -284,15 +296,13 @@ class Ha(object):
             return None
 
         if self._rewind.can_rewind:
-            self._async_executor.schedule('running pg_rewind from ' + leader.name)
+            msg = 'running pg_rewind from ' + leader.name
-            self._async_executor.run_async(self._rewind.execute, (leader,))
+            return self._async_executor.try_run_async(msg, self._rewind.execute, args=(leader,)) or msg
-            return True
 
         # remove_data_directory_on_diverged_timelines is set
         if not self.is_standby_cluster():
-            self._async_executor.schedule('reinitializing due to diverged timelines')
+            msg = 'reinitializing due to diverged timelines'
-            self._async_executor.run_async(self._do_reinitialize, args=(self.cluster, ))
+            return self._async_executor.try_run_async(msg, self._do_reinitialize, args=(self.cluster,)) or msg
-            return True
 
     def recover(self):
         # Postgres is not running and we will restart in standby mode. Watchdog is not needed until we promote.
@@ -319,9 +329,9 @@ class Ha(object):
                 and not self._crash_recovery_executed and \
                 (self.cluster.is_unlocked() or self._rewind.can_rewind):
             self._crash_recovery_executed = True
-            self._async_executor.schedule('doing crash recovery in a single user mode')
+            self._crash_recovery_started = time.time()
-            self._async_executor.run_async(self.state_handler.fix_cluster_state)
+            msg = 'doing crash recovery in a single user mode'
-            return self._async_executor.scheduled_action
+            return self._async_executor.try_run_async(msg, self._rewind.ensure_clean_shutdown) or msg
 
         self.load_cluster_from_dcs()
 
@@ -329,8 +339,9 @@ class Ha(object):
         if self.is_standby_cluster() or not self.has_lock():
             if not self._rewind.executed:
                 self._rewind.trigger_check_diverged_lsn()
-            if self._handle_rewind_or_reinitialize():
+            msg = self._handle_rewind_or_reinitialize()
-                return self._async_executor.scheduled_action
+            if msg:
+                return msg
 
             if self.has_lock():  # in standby cluster
                 msg = "starting as a standby leader because i had the session lock"
@@ -346,23 +357,34 @@ class Ha(object):
             msg = "starting as readonly because i had the session lock"
             node_to_follow = None
 
-        self.recovering = True
+        if self._async_executor.try_run_async('restarting after failure', self.state_handler.follow,
-
+                                              args=(node_to_follow, role, timeout)) is None:
-        self._async_executor.schedule('restarting after failure')
+            self.recovering = True
-        self._async_executor.run_async(self.state_handler.follow, (node_to_follow, role, timeout))
         return msg
 
     def _get_node_to_follow(self, cluster):
         # determine the node to follow. If replicatefrom tag is set,
         # try to follow the node mentioned there, otherwise, follow the leader.
-        if self.is_standby_cluster() and (self.cluster.is_unlocked() or self.has_lock(False)):
+        standby_config = self.get_standby_cluster_config()
+        is_standby_cluster = _is_standby_cluster(standby_config)
+        if is_standby_cluster and (self.cluster.is_unlocked() or self.has_lock(False)):
             node_to_follow = self.get_remote_master()
         elif self.patroni.replicatefrom and self.patroni.replicatefrom != self.state_handler.name:
             node_to_follow = cluster.get_member(self.patroni.replicatefrom)
         else:
-            node_to_follow = cluster.leader
+            node_to_follow = cluster.leader if cluster.leader and cluster.leader.name else None
 
-        return node_to_follow if node_to_follow and node_to_follow.name != self.state_handler.name else None
+        node_to_follow = node_to_follow if node_to_follow and node_to_follow.name != self.state_handler.name else None
+
+        if node_to_follow and not isinstance(node_to_follow, RemoteMember):
+            # we are going to abuse Member.data to pass following parameters
+            params = ('restore_command', 'archive_cleanup_command')
+            for param in params:  # It is highly unlikely to happen, but we want to protect from the case
+                node_to_follow.data.pop(param, None)  # when above-mentioned params came from outside.
+            if is_standby_cluster:
+                node_to_follow.data.update({p: standby_config[p] for p in params if standby_config.get(p)})
+
+        return node_to_follow
 
     def follow(self, demote_reason, follow_reason, refresh=True):
         if refresh:
@@ -375,29 +397,46 @@ class Ha(object):
         if self.is_paused():
             if not (self._rewind.is_needed and self._rewind.can_rewind_or_reinitialize_allowed)\
                     or self.cluster.is_unlocked():
-                self.state_handler.set_role('master' if is_leader else 'replica')
                 if is_leader:
+                    self.state_handler.set_role('master')
                     return 'continue to run as master without lock'
-                elif not node_to_follow:
+                elif self.state_handler.role != 'standby_leader':
-                    return 'no action'
+                    self.state_handler.set_role('replica')
+
+                if not node_to_follow:
+                    return 'no action. I am ({0})'.format(self.state_handler.name)
         elif is_leader:
             self.demote('immediate-nolock')
             return demote_reason
 
-        if self._handle_rewind_or_reinitialize():
+        if self.is_standby_cluster() and self._leader_timeline and \
-            return self._async_executor.scheduled_action
+                self.state_handler.get_history(self._leader_timeline + 1):
+            self._rewind.trigger_check_diverged_lsn()
+
+        msg = self._handle_rewind_or_reinitialize()
+        if msg:
+            return msg
+
+        if not self.is_paused():
+            self.state_handler.handle_parameter_change()
 
         role = 'standby_leader' if isinstance(node_to_follow, RemoteMember) and self.has_lock(False) else 'replica'
         # It might happen that leader key in the standby cluster references non-exiting member.
         # In this case it is safe to continue running without changing recovery.conf
         if self.is_standby_cluster() and role == 'replica' and not (node_to_follow and node_to_follow.conn_url):
             return 'continue following the old known standby leader'
-        elif not self.state_handler.config.check_recovery_conf(node_to_follow):
+        else:
-            self._async_executor.schedule('changing primary_conninfo and restarting')
+            change_required, restart_required = self.state_handler.config.check_recovery_conf(node_to_follow)
-            self._async_executor.run_async(self.state_handler.follow, (node_to_follow, role))
+            if change_required:
-        elif role == 'standby_leader' and self.state_handler.role != role:
+                if restart_required:
-            self.state_handler.set_role(role)
+                    self._async_executor.try_run_async('changing primary_conninfo and restarting',
-            self.state_handler.call_nowait(ACTION_ON_ROLE_CHANGE)
+                                                       self.state_handler.follow, args=(node_to_follow, role))
+                else:
+                    self.state_handler.follow(node_to_follow, role, do_reload=True)
+                self._rewind.trigger_check_diverged_lsn()
+            elif role == 'standby_leader' and self.state_handler.role != role:
+                self.state_handler.set_role(role)
+                self.state_handler.call_nowait(ACTION_ON_ROLE_CHANGE)
 
         return follow_reason
 
@@ -417,28 +456,38 @@ class Ha(object):
         promoting standbys that were guaranteed to be replicating synchronously.
         """
         if self.is_synchronous_mode():
-            current = self.cluster.sync.leader and self.cluster.sync.sync_standby
+            sync_node_count = self.patroni.config['synchronous_node_count']
-            picked, allow_promote = self.state_handler.pick_synchronous_standby(self.cluster)
+            current = self.cluster.sync.leader and self.cluster.sync.members or []
-            if picked != current:
+            picked, allow_promote = self.state_handler.pick_synchronous_standby(self.cluster, sync_node_count,
-                # We need to revoke privilege from current before replacing it in the config
+                                                                                self.patroni.config[
-                if current:
+                                                                                    'maximum_lag_on_syncnode'])
-                    logger.info("Removing synchronous privilege from %s", current)
+            if set(picked) != set(current):
-                    if not self.dcs.write_sync_state(self.state_handler.name, None, index=self.cluster.sync.index):
+                # update synchronous standby list in dcs temporarily to point to common nodes in current and picked
+                sync_common = list(set(current).intersection(set(allow_promote)))
+                if set(sync_common) != set(current):
+                    logger.info("Updating synchronous privilege temporarily from %s to %s", current, sync_common)
+                    if not self.dcs.write_sync_state(self.state_handler.name,
+                                                     sync_common or None,
+                                                     index=self.cluster.sync.index):
                         logger.info('Synchronous replication key updated by someone else.')
                         return
 
-                if self.is_synchronous_mode_strict() and picked is None:
+                # Update  db param and wait for x secs
-                    picked = '*'
+                if self.is_synchronous_mode_strict() and not picked:
+                    picked = ['*']
                     logger.warning("No standbys available!")
 
                 logger.info("Assigning synchronous standby status to %s", picked)
                 self.state_handler.config.set_synchronous_standby(picked)
 
-                if picked and picked != '*' and not allow_promote:
+                if picked and picked[0] != '*' and set(allow_promote) != set(picked) and not allow_promote:
                     # Wait for PostgreSQL to enable synchronous mode and see if we can immediately set sync_standby
                     time.sleep(2)
-                    picked, allow_promote = self.state_handler.pick_synchronous_standby(self.cluster)
+                    _, allow_promote = self.state_handler.pick_synchronous_standby(self.cluster,
-                if allow_promote:
+                                                                                   sync_node_count,
+                                                                                   self.patroni.config[
+                                                                                       'maximum_lag_on_syncnode'])
+                if allow_promote and set(allow_promote) != set(sync_common):
                     try:
                         cluster = self.dcs.get_cluster()
                     except DCSError:
@@ -446,18 +495,18 @@ class Ha(object):
                     if cluster.sync.leader and cluster.sync.leader != self.state_handler.name:
                         logger.info("Synchronous replication key updated by someone else")
                         return
-                    if not self.dcs.write_sync_state(self.state_handler.name, picked, index=cluster.sync.index):
+                    if not self.dcs.write_sync_state(self.state_handler.name, allow_promote, index=cluster.sync.index):
                         logger.info("Synchronous replication key updated by someone else")
                         return
-                    logger.info("Synchronous standby status assigned to %s", picked)
+                    logger.info("Synchronous standby status assigned to %s", allow_promote)
         else:
             if self.cluster.sync.leader and self.dcs.delete_sync_state(index=self.cluster.sync.index):
                 logger.info("Disabled synchronous replication")
-            self.state_handler.config.set_synchronous_standby(None)
+            self.state_handler.config.set_synchronous_standby([])
 
     def is_sync_standby(self, cluster):
         return cluster.leader and cluster.sync.leader == cluster.leader.name \
-            and cluster.sync.sync_standby == self.state_handler.name
+            and self.state_handler.name in cluster.sync.members
 
     def while_not_sync_standby(self, func):
         """Runs specified action while trying to make sure that the node is not assigned synchronous standby status.
@@ -503,9 +552,10 @@ class Ha(object):
             if cluster_history:
                 self.dcs.set_history_value('[]')
         elif not cluster_history or cluster_history[-1][0] != master_timeline - 1 or len(cluster_history[-1]) != 4:
-            cluster_history = {l[0]: l for l in cluster_history or []}
+            cluster_history = {line[0]: line for line in cluster_history or []}
             history = self.state_handler.get_history(master_timeline)
-            if history:
+            if history and self.cluster.config:
+                history = history[-self.cluster.config.max_timelines_history:]
                 for line in history:
                     # enrich current history with promotion timestamps stored in DCS
                     if len(line) == 3 and line[0] in cluster_history \
@@ -519,13 +569,27 @@ class Ha(object):
         return self.follow(demote_reason, message)
 
     def enforce_master_role(self, message, promote_message):
-        if not self.is_paused() and not self.watchdog.is_running and not self.watchdog.activate():
+        """
-            if self.state_handler.is_leader():
+        Ensure the node that has won the race for the leader key meets criteria
-                self.demote('immediate')
+        for promoting its PG server to the 'master' role.
-                return 'Demoting self because watchdog could not be activated'
+        """
-            else:
+        if not self.is_paused():
-                self.release_leader_key_voluntarily()
+            if not self.watchdog.is_running and not self.watchdog.activate():
-                return 'Not promoting self because watchdog could not be activated'
+                if self.state_handler.is_leader():
+                    self.demote('immediate')
+                    return 'Demoting self because watchdog could not be activated'
+                else:
+                    self.release_leader_key_voluntarily()
+                    return 'Not promoting self because watchdog could not be activated'
+
+            with self._async_response:
+                if self._async_response.result is False:
+                    logger.warning("Releasing the leader key voluntarily because the pre-promote script failed")
+                    self._released_leader_key_timestamp = time.time()
+                    self.release_leader_key_voluntarily()
+                    # discard the result of the failed pre-promote script to be able to re-try promote
+                    self._async_response.reset()
+                    return 'Promotion cancelled because the pre-promote script failed'
 
         if self.state_handler.is_leader():
             # Inform the state handler about its master role.
@@ -545,29 +609,28 @@ class Ha(object):
                     # Somebody else updated sync state, it may be due to us losing the lock. To be safe, postpone
                     # promotion until next cycle. TODO: trigger immediate retry of run_cycle
                     return 'Postponing promotion because synchronous replication state was updated by somebody else'
-                self.state_handler.config.set_synchronous_standby('*' if self.is_synchronous_mode_strict() else None)
+                self.state_handler.config.set_synchronous_standby(['*'] if self.is_synchronous_mode_strict() else [])
             if self.state_handler.role != 'master':
-                self.set_leader_access_is_restricted(self.cluster.has_permanent_logical_slots(self.state_handler.name))
-
                 def on_success():
                     self._rewind.reset_state()
                     logger.info("cleared rewind state after becoming the leader")
 
-                self._async_executor.schedule('promote')
+                with self._async_response:
-                self._async_executor.run_async(self.state_handler.promote,
+                    self._async_response.reset()
-                                               args=(self.dcs.loop_wait, on_success, self._leader_access_is_restricted))
+                self._async_executor.try_run_async('promote', self.state_handler.promote,
+                                                   args=(self.dcs.loop_wait, self._async_response, on_success))
             return promote_message
 
-    @staticmethod
+    def fetch_node_status(self, member):
-    def fetch_node_status(member):
         """This function perform http get request on member.api_url and fetches its status
         :returns: `_MemberStatus` object
         """
 
         try:
-            response = requests.get(member.api_url, timeout=2, verify=False)
+            response = self.patroni.request(member, timeout=2, retries=0)
-            logger.info('Got response from %s %s: %s', member.name, member.api_url, response.content)
+            data = response.data.decode('utf-8')
-            return _MemberStatus.from_api_response(member, response.json())
+            logger.info('Got response from %s %s: %s', member.name, member.api_url, data)
+            return _MemberStatus.from_api_response(member, json.loads(data))
         except Exception as e:
             logger.warning("Request failed to %s: GET %s (%s)", member.name, member.api_url, e)
         return _MemberStatus.unknown(member)
@@ -585,13 +648,13 @@ class Ha(object):
         :param wal_position: Current wal position.
         :returns True when node is lagging
         """
-        lag = (self.cluster.last_leader_operation or 0) - wal_position
+        lag = (self.cluster.last_lsn or 0) - wal_position
         return lag > self.patroni.config.get('maximum_lag_on_failover', 0)
 
     def _is_healthiest_node(self, members, check_replication_lag=True):
         """This method tries to determine whether I am healthy enough to became a new leader candidate or not."""
 
-        _, my_wal_position = self.state_handler.timeline_wal_position()
+        my_wal_position = self.state_handler.last_operation()
         if check_replication_lag and self.is_lagging(my_wal_position):
             logger.info('My wal position exceeds maximum replication lag')
             return False  # Too far behind last reported wal position on master
@@ -689,6 +752,10 @@ class Ha(object):
         return self._is_healthiest_node(members, check_replication_lag=False)
 
     def is_healthiest_node(self):
+        if time.time() - self._released_leader_key_timestamp < self.dcs.ttl:
+            logger.info('backoff: skip leader race after pre_promote script failure and releasing the lock voluntarily')
+            return False
+
         if self.is_paused() and not self.patroni.nofailover and \
                 self.cluster.failover and not self.cluster.failover.scheduled_at:
             ret = self.manual_failover_process_no_leader()
@@ -698,8 +765,10 @@ class Ha(object):
         if self.state_handler.is_starting():  # postgresql still starting up is unhealthy
             return False
 
-        if self.state_handler.is_leader():  # leader is always the healthiest
+        if self.state_handler.is_leader():
-            return True
+            # in pause leader is the healthiest only when no initialize or sysid matches with initialize!
+            return not self.is_paused() or not self.cluster.initialize\
+                    or self.state_handler.sysid == self.cluster.initialize
 
         if self.is_paused():
             return False
@@ -711,6 +780,7 @@ class Ha(object):
             return self.manual_failover_process_no_leader()
 
         if not self.watchdog.is_healthy:
+            logger.warning('Watchdog device is not usable')
             return False
 
         # When in sync mode, only last known master and sync standby are allowed to promote automatically.
@@ -726,13 +796,13 @@ class Ha(object):
 
         return self._is_healthiest_node(members.values())
 
-    def _delete_leader(self):
+    def _delete_leader(self, last_lsn=None):
         self.set_is_leader(False)
-        self.dcs.delete_leader()
+        self.dcs.delete_leader(last_lsn)
         self.dcs.reset_cluster()
 
-    def release_leader_key_voluntarily(self):
+    def release_leader_key_voluntarily(self, last_lsn=None):
-        self._delete_leader()
+        self._delete_leader(last_lsn)
         self.touch_member()
         logger.info("Leader key released")
 
@@ -756,13 +826,15 @@ class Ha(object):
 
         self._rewind.trigger_check_diverged_lsn()
         self.state_handler.stop(mode_control['stop'], checkpoint=mode_control['checkpoint'],
-                                on_safepoint=self.watchdog.disable if self.watchdog.is_running else None)
+                                on_safepoint=self.watchdog.disable if self.watchdog.is_running else None,
+                                stop_timeout=self.master_stop_timeout())
         self.state_handler.set_role('demoted')
         self.set_is_leader(False)
 
         if mode_control['release']:
+            checkpoint_location = self.state_handler.latest_checkpoint_location() if mode == 'graceful' else None
             with self._async_executor:
-                self.release_leader_key_voluntarily()
+                self.release_leader_key_voluntarily(checkpoint_location)
             time.sleep(2)  # Give a time to somebody to take the leader lock
         if mode_control['offline']:
             node_to_follow, leader = None, None
@@ -774,11 +846,10 @@ class Ha(object):
         # there could be an async action already running, calling follow from here will lead
         # to racy state handler state updates.
         if mode_control['async_req']:
-            self._async_executor.schedule('starting after demotion')
+            self._async_executor.try_run_async('starting after demotion', self.state_handler.follow, (node_to_follow,))
-            self._async_executor.run_async(self.state_handler.follow, (node_to_follow,))
         else:
             if self.is_synchronous_mode():
-                self.state_handler.config.set_synchronous_standby(None)
+                self.state_handler.config.set_synchronous_standby([])
             if self._rewind.rewind_or_reinitialize_needed_and_possible(leader):
                 return False  # do not start postgres, but run pg_rewind on the next iteration
             self.state_handler.follow(node_to_follow)
@@ -839,7 +910,7 @@ class Ha(object):
                 else:
                     if self.is_synchronous_mode():
                         if failover.candidate and not self.cluster.sync.matches(failover.candidate):
-                            logger.warning('Failover candidate=%s does not match with sync_standby=%s',
+                            logger.warning('Failover candidate=%s does not match with sync_standbys=%s',
                                            failover.candidate, self.cluster.sync.sync_standby)
                             members = []
                         else:
@@ -848,9 +919,8 @@ class Ha(object):
                         members = [m for m in self.cluster.members
                                    if not failover.candidate or m.name == failover.candidate]
                     if self.is_failover_possible(members):  # check that there are healthy members
-                        self._async_executor.schedule('manual failover: demote')
+                        ret = self._async_executor.try_run_async('manual failover: demote', self.demote, ('graceful',))
-                        self._async_executor.run_async(self.demote, ('graceful',))
+                        return ret or 'manual failover: demoting myself'
-                        return 'manual failover: demoting myself'
                     else:
                         logger.warning('manual failover: no healthy members found, failover is not possible')
             else:
@@ -878,9 +948,9 @@ class Ha(object):
                 self.load_cluster_from_dcs()
 
                 if self.is_standby_cluster():
-                    # standby leader disappeared, and this is a healthiest
+                    # standby leader disappeared, and this is the healthiest
                     # replica, so it should become a new standby leader.
-                    # This imply that we need to start following a remote master
+                    # This implies we need to start following a remote master
                     msg = 'promoted self to a standby leader by acquiring session lock'
                     return self.enforce_follow_remote_master(msg)
                 else:
@@ -913,31 +983,26 @@ class Ha(object):
                 self._delete_leader()
                 return 'removed leader lock because postgres is not running as master'
 
-            if self.state_handler.is_leader() and self._leader_access_is_restricted:
-                self.state_handler.slots_handler.sync_replication_slots(self.cluster)
-                self.state_handler.call_nowait(ACTION_ON_ROLE_CHANGE)
-                self.set_leader_access_is_restricted(False)
-
             if self.update_lock(True):
                 msg = self.process_manual_failover_from_leader()
                 if msg is not None:
                     return msg
 
                 # check if the node is ready to be used by pg_rewind
-                self._rewind.check_for_checkpoint_after_promote()
+                self._rewind.ensure_checkpoint_after_promote(self.wakeup)
 
                 if self.is_standby_cluster():
                     # in case of standby cluster we don't really need to
                     # enforce anything, since the leader is not a master.
                     # So just remind the role.
-                    msg = 'no action.  i am the standby leader with the lock' \
+                    msg = 'no action. I am ({0}) the standby leader with the lock'.format(self.state_handler.name) \
                           if self.state_handler.role == 'standby_leader' else \
                           'promoted self to a standby leader because i had the session lock'
                     return self.enforce_follow_remote_master(msg)
                 else:
                     return self.enforce_master_role(
-                        'no action.  i am the leader with the lock',
+                        'no action. I am ({0}) the leader with the lock'.format(self.state_handler.name),
-                        'promoted self to leader because i had the session lock'
+                        'promoted self to leader because I had the session lock'
                     )
             else:
                 # Either there is no connection to DCS or someone else acquired the lock
@@ -950,12 +1015,15 @@ class Ha(object):
                 else:
                     return 'not promoting because failed to update leader lock in DCS'
         else:
-            logger.info('does not have lock')
+            logger.debug('does not have lock')
+        lock_owner = self.cluster.leader and self.cluster.leader.name
         if self.is_standby_cluster():
-            return self.follow('cannot be a real master in standby cluster',
+            return self.follow('cannot be a real primary in a standby cluster',
-                               'no action.  i am a secondary and i am following a standby leader', refresh=False)
+                               'no action. I am a secondary ({0}) and following a standby leader ({1})'.format(
-        return self.follow('demoting self because i do not have the lock and i was a leader',
+                                    self.state_handler.name, lock_owner), refresh=False)
-                           'no action.  i am a secondary and i am following a leader', refresh=False)
+        return self.follow('demoting self because I do not have the lock and I was a leader',
+                           'no action. I am a secondary ({0}) and following a leader ({1})'.format(
+                                self.state_handler.name, lock_owner), refresh=False)
 
     def evaluate_scheduled_restart(self):
         if self._async_executor.busy:  # Restart already in progress
@@ -1069,7 +1137,7 @@ class Ha(object):
                 return (False, 'restart failed')
 
     def _do_reinitialize(self, cluster):
-        self.state_handler.stop('immediate')
+        self.state_handler.stop('immediate', stop_timeout=self.patroni.config['retry_timeout'])
         # Commented redundant data directory cleanup here
         # self.state_handler.remove_data_directory()
 
@@ -1098,10 +1166,28 @@ class Ha(object):
         self._async_executor.run_async(self._do_reinitialize, args=(self.cluster, ))
 
     def handle_long_action_in_progress(self):
+        """
+        Figure out what to do with the task AsyncExecutor is performing.
+        """
         if self.has_lock() and self.update_lock():
+            if self._async_executor.scheduled_action == 'doing crash recovery in a single user mode':
+                time_left = self.patroni.config['master_start_timeout'] - (time.time() - self._crash_recovery_started)
+                if time_left <= 0 and self.is_failover_possible(self.cluster.members):
+                    logger.info("Demoting self because crash recovery is taking too long")
+                    self.state_handler.cancellable.cancel(True)
+                    self.demote('immediate')
+                    return 'terminated crash recovery because of startup timeout'
+
             return 'updated leader lock during ' + self._async_executor.scheduled_action
-        elif not self.state_handler.bootstrapping:
+        elif not self.state_handler.bootstrapping and not self.is_paused():
-            # Don't have lock, make sure we are not starting up a master in the background
+            # Don't have lock, make sure we are not promoting or starting up a master in the background
+            if self._async_executor.scheduled_action == 'promote':
+                with self._async_response:
+                    cancel = self._async_response.cancel()
+                if cancel:
+                    self.state_handler.cancellable.cancel()
+                    return 'lost leader before promote'
+
             if self.state_handler.role == 'master':
                 logger.info("Demoting master during " + self._async_executor.scheduled_action)
                 if self._async_executor.scheduled_action == 'restart':
@@ -1112,7 +1198,6 @@ class Ha(object):
                             self.state_handler.terminate_starting_postmaster(postmaster=task.result)
                 self.demote('immediate-nolock')
                 return 'lost leader lock during ' + self._async_executor.scheduled_action
-
         if self.cluster.is_unlocked():
             logger.info('not healthy enough for leader race')
 
@@ -1129,41 +1214,42 @@ class Ha(object):
         if not self.state_handler.is_running():
             self.watchdog.disable()
             if self.has_lock():
-                self.state_handler.set_role('demoted')
+                if self.state_handler.role in ('master', 'standby_leader'):
+                    self.state_handler.set_role('demoted')
                 self._delete_leader()
                 return 'removed leader key after trying and failing to start postgres'
             return 'failed to start postgres'
-        self._crash_recovery_executed = False
         return None
 
     def cancel_initialization(self):
         logger.info('removing initialize key after failed attempt to bootstrap the cluster')
         self.dcs.cancel_initialization()
-        self.state_handler.stop('immediate')
+        self.state_handler.stop('immediate', stop_timeout=self.patroni.config['retry_timeout'])
         self.state_handler.move_data_directory()
-        raise PatroniException('Failed to bootstrap cluster')
+        raise PatroniFatalException('Failed to bootstrap cluster')
 
     def post_bootstrap(self):
+        with self._async_response:
+            result = self._async_response.result
         # bootstrap has failed if postgres is not running
-        if not self.state_handler.is_running() or self._post_bootstrap_task.result is False:
+        if not self.state_handler.is_running() or result is False:
             self.cancel_initialization()
 
-        if self._post_bootstrap_task.result is None:
+        if result is None:
             if not self.state_handler.is_leader():
                 return 'waiting for end of recovery after bootstrap'
 
             self.state_handler.set_role('master')
-            self._async_executor.schedule('post_bootstrap')
+            ret = self._async_executor.try_run_async('post_bootstrap', self.state_handler.bootstrap.post_bootstrap,
-            self._async_executor.run_async(self.state_handler.bootstrap.post_bootstrap,
+                                                     args=(self.patroni.config['bootstrap'], self._async_response))
-                                           args=(self.patroni.config['bootstrap'], self._post_bootstrap_task))
+            return ret or 'running post_bootstrap'
-            return 'running post_bootstrap'
 
         self.state_handler.bootstrapping = False
-        self.dcs.set_config_value(json.dumps(self.patroni.config.dynamic_configuration, separators=(',', ':')))
         if not self.watchdog.activate():
             logger.error('Cancelling bootstrap because watchdog activation failed')
             self.cancel_initialization()
-        self.state_handler.slots_handler.sync_replication_slots(self.cluster)
+        self.dcs.initialize(create_new=(self.cluster.initialize is None), sysid=self.state_handler.sysid)
+        self.dcs.set_config_value(json.dumps(self.patroni.config.dynamic_configuration, separators=(',', ':')))
         self.dcs.take_leader()
         self.set_is_leader(True)
         self.state_handler.call_nowait(ACTION_ON_START)
@@ -1219,8 +1305,12 @@ class Ha(object):
     def _run_cycle(self):
         dcs_failed = False
         try:
-            self.state_handler.reset_cluster_info_state()
+            try:
-            self.load_cluster_from_dcs()
+                self.load_cluster_from_dcs()
+                self.state_handler.reset_cluster_info_state(self.cluster, self.patroni.nofailover)
+            except Exception:
+                self.state_handler.reset_cluster_info_state(None, self.patroni.nofailover)
+                raise
 
             if self.is_paused():
                 self.watchdog.disable()
@@ -1252,17 +1342,29 @@ class Ha(object):
             if self.state_handler.bootstrapping:
                 return self.post_bootstrap()
 
-            if self.recovering and not self._rewind.is_needed:
+            if self.recovering:
                 self.recovering = False
-                # Check if we tried to recover and failed
+
-                msg = self.post_recover()
+                if not self._rewind.is_needed:
-                if msg is not None:
+                    # Check if we tried to recover from postgres crash and failed
-                    return msg
+                    msg = self.post_recover()
+                    if msg is not None:
+                        return msg
+
+                # Reset some states after postgres successfully started up
+                self._crash_recovery_executed = False
+                if self._rewind.executed and not self._rewind.failed:
+                    self._rewind.reset_state()
+
+                # The Raft cluster without a quorum takes a bit of time to stabilize.
+                # Therefore we want to postpone the leader race if we just started up.
+                if self.cluster.is_unlocked() and self.dcs.__class__.__name__ == 'Raft':
+                    return 'started as a secondary'
 
             # is data directory empty?
             if self.state_handler.data_directory_empty():
                 self.state_handler.set_role('uninitialized')
-                self.state_handler.stop('immediate')
+                self.state_handler.stop('immediate', stop_timeout=self.patroni.config['retry_timeout'])
                 # In case datadir went away while we were master.
                 self.watchdog.disable()
 
@@ -1271,30 +1373,42 @@ class Ha(object):
                     self.release_leader_key_voluntarily()
                     return 'released leader key voluntarily as data dir empty and currently leader'
 
+                if self.is_paused():
+                    return 'running with empty data directory'
                 return self.bootstrap()  # new node
-            # "bootstrap", but data directory is not empty
-            elif not self.sysid_valid(self.cluster.initialize) and self.cluster.is_unlocked() and not self.is_paused():
-                if not self.state_handler.cb_called and self.state_handler.is_running() \
-                        and not self.state_handler.is_leader():
-                    self._join_aborted = True
-                    logger.error('No initialize key in DCS and PostgreSQL is running as replica, aborting start')
-                    logger.error('Please first start Patroni on the node running as master')
-                    sys.exit(1)
-                self.dcs.initialize(create_new=(self.cluster.initialize is None), sysid=self.state_handler.sysid)
             else:
                 # check if we are allowed to join
                 data_sysid = self.state_handler.sysid
                 if not self.sysid_valid(data_sysid):
                     # data directory is not empty, but no valid sysid, cluster must be broken, suggest reinit
-                    return "data dir for the cluster is not empty, but system ID is invalid; consider doing reinitalize"
+                    return ("data dir for the cluster is not empty, "
+                            "but system ID is invalid; consider doing reinitialize")
 
-                if self.sysid_valid(self.cluster.initialize) and self.cluster.initialize != self.state_handler.sysid:
+                if self.sysid_valid(self.cluster.initialize):
-                    logger.fatal("system ID mismatch, node %s belongs to a different cluster: %s != %s",
+                    if self.cluster.initialize != data_sysid:
-                                 self.state_handler.name, self.cluster.initialize, self.state_handler.sysid)
+                        if self.is_paused():
-                    sys.exit(1)
+                            logger.warning('system ID has changed while in paused mode. Patroni will exit when resuming'
+                                           ' unless system ID is reset: %s != %s', self.cluster.initialize, data_sysid)
+                            if self.has_lock():
+                                self.release_leader_key_voluntarily()
+                                return 'released leader key voluntarily due to the system ID mismatch'
+                        else:
+                            logger.fatal('system ID mismatch, node %s belongs to a different cluster: %s != %s',
+                                         self.state_handler.name, self.cluster.initialize, data_sysid)
+                            sys.exit(1)
+                elif self.cluster.is_unlocked() and not self.is_paused():
+                    # "bootstrap", but data directory is not empty
+                    if not self.state_handler.cb_called and self.state_handler.is_running() \
+                            and not self.state_handler.is_leader():
+                        self._join_aborted = True
+                        logger.error('No initialize key in DCS and PostgreSQL is running as replica, aborting start')
+                        logger.error('Please first start Patroni on the node running as master')
+                        sys.exit(1)
+                    self.dcs.initialize(create_new=(self.cluster.initialize is None), sysid=data_sysid)
 
             if not self.state_handler.is_healthy():
                 if self.is_paused():
+                    self.state_handler.set_state('stopped')
                     if self.has_lock():
                         self._delete_leader()
                         return 'removed leader lock because postgres is not running'
@@ -1307,25 +1421,35 @@ class Ha(object):
                             (self._rewind.is_needed and self._rewind.can_rewind_or_reinitialize_allowed):
                         return 'postgres is not running'
 
+                if self.state_handler.state in ('running', 'starting'):
+                    self.state_handler.set_state('crashed')
                 # try to start dead postgres
                 return self.recover()
 
             try:
                 if self.cluster.is_unlocked():
-                    return self.process_unhealthy_cluster()
+                    ret = self.process_unhealthy_cluster()
                 else:
                     msg = self.process_healthy_cluster()
-                    return self.evaluate_scheduled_restart() or msg
+                    ret = self.evaluate_scheduled_restart() or msg
             finally:
                 # we might not have a valid PostgreSQL connection here if another thread
                 # stops PostgreSQL, therefore, we only reload replication slots if no
                 # asynchronous processes are running (should be always the case for the master)
                 if not self._async_executor.busy and not self.state_handler.is_starting():
-                    self.state_handler.slots_handler.sync_replication_slots(self.cluster)
+                    create_slots = self.state_handler.slots_handler.sync_replication_slots(self.cluster,
+                                                                                           self.patroni.nofailover)
                     if not self.state_handler.cb_called:
                         if not self.state_handler.is_leader():
                             self._rewind.trigger_check_diverged_lsn()
                         self.state_handler.call_nowait(ACTION_ON_START)
+                    if create_slots and self.cluster.leader:
+                        err = self._async_executor.try_run_async('copy_logical_slots',
+                                                                 self.state_handler.slots_handler.copy_logical_slots,
+                                                                 args=(self.cluster.leader, create_slots))
+                        if not err:
+                            ret = 'Copying logical slots {0} from the primary'.format(create_slots)
+            return ret
         except DCSError:
             dcs_failed = True
             logger.error('Error communicating with DCS')
@@ -1341,8 +1465,14 @@ class Ha(object):
 
     def run_cycle(self):
         with self._async_executor:
-            info = self._run_cycle()
+            try:
-            return (self.is_paused() and 'PAUSE: ' or '') + info
+                info = self._run_cycle()
+                return (self.is_paused() and 'PAUSE: ' or '') + info
+            except PatroniFatalException:
+                raise
+            except Exception:
+                logger.exception('Unexpected exception')
+                return 'Unexpected exception raised, please report it as a BUG'
 
     def shutdown(self):
         if self.is_paused():
@@ -1350,14 +1480,16 @@ class Ha(object):
             self.watchdog.disable()
         elif not self._join_aborted:
             # FIXME: If stop doesn't reach safepoint quickly enough keepalive is triggered. If shutdown checkpoint
-            # takes longer than ttl, then leader key is lost and replication might not have sent out all xlog.
+            # takes longer than ttl, then leader key is lost and replication might not have sent out all WAL.
             # This might not be the desired behavior of users, as a graceful shutdown of the host can mean lost data.
             # We probably need to something smarter here.
             disable_wd = self.watchdog.disable if self.watchdog.is_running else None
-            self.while_not_sync_standby(lambda: self.state_handler.stop(checkpoint=False, on_safepoint=disable_wd))
+            self.while_not_sync_standby(lambda: self.state_handler.stop(checkpoint=False, on_safepoint=disable_wd,
+                                                                        stop_timeout=self.master_stop_timeout()))
             if not self.state_handler.is_running():
-                if self.has_lock():
+                if self.is_leader():
-                    self.dcs.delete_leader()
+                    checkpoint_location = self.state_handler.latest_checkpoint_location()
+                    self.dcs.delete_leader(checkpoint_location)
                 self.touch_member()
             else:
                 # XXX: what about when Patroni is started as the wrong user that has access to the watchdog device
@@ -1368,7 +1500,7 @@ class Ha(object):
 
     def watch(self, timeout):
         # watch on leader key changes if the postgres is running and leader is known and current node is not lock owner
-        if self._async_executor.busy or self.cluster.is_unlocked() or self.has_lock(False):
+        if self._async_executor.busy or not self.cluster or self.cluster.is_unlocked() or self.has_lock(False):
             leader_index = None
         else:
             leader_index = self.cluster.leader.index

patroni/log.py

@@ -11,6 +11,20 @@ from threading import Lock, Thread
 _LOGGER = logging.getLogger(__name__)
 
 
+def debug_exception(logger_obj, msg, *args, **kwargs):
+    kwargs.pop("exc_info", False)
+    if logger_obj.isEnabledFor(logging.DEBUG):
+        logger_obj.debug(msg, *args, exc_info=True, **kwargs)
+    else:
+        msg = "{0}, DETAIL: '{1}'".format(msg, sys.exc_info()[1])
+        logger_obj.error(msg, *args, exc_info=False, **kwargs)
+
+
+def error_exception(logger_obj, msg, *args, **kwargs):
+    exc_info = kwargs.pop("exc_info", True)
+    logger_obj.error(msg, *args, exc_info=exc_info, **kwargs)
+
+
 class QueueHandler(logging.Handler):
 
     def __init__(self):
@@ -48,9 +62,20 @@ class QueueHandler(logging.Handler):
         return self._records_lost
 
 
+class ProxyHandler(logging.Handler):
+
+    def __init__(self, patroni_logger):
+        logging.Handler.__init__(self)
+        self.patroni_logger = patroni_logger
+
+    def emit(self, record):
+        self.patroni_logger.log_handler.handle(record)
+
+
 class PatroniLogger(Thread):
 
     DEFAULT_LEVEL = 'INFO'
+    DEFAULT_TRACEBACK_LEVEL = 'ERROR'
     DEFAULT_FORMAT = '%(asctime)s %(levelname)s: %(message)s'
 
     NORMAL_LOG_QUEUE_SIZE = 2  # When everything goes normal Patroni writes only 2 messages per HA loop
@@ -59,16 +84,18 @@ class PatroniLogger(Thread):
 
     def __init__(self):
         super(PatroniLogger, self).__init__()
-        self.daemon = True
         self._queue_handler = QueueHandler()
         self._root_logger = logging.getLogger()
-        self._root_logger.addHandler(self._queue_handler)
         self._config = None
-        self._log_handler = None
+        self.log_handler = None
+        self.log_handler_lock = Lock()
         self._old_handlers = []
-        self._log_handler_lock = Lock()
         self.reload_config({'level': 'DEBUG'})
-        self.start()
+        # We will switch to the QueueHandler only when thread was started.
+        # This is necessary to protect from the cases when Patroni constructor
+        # failed and PatroniLogger thread remain running and prevent shutdown.
+        self._proxy_handler = ProxyHandler(self)
+        self._root_logger.addHandler(self._proxy_handler)
 
     def update_loggers(self):
         loggers = deepcopy(self._config.get('loggers') or {})
@@ -87,18 +114,22 @@ class PatroniLogger(Thread):
                 self._queue_handler.queue.maxsize = config.get('max_queue_size', self.DEFAULT_MAX_QUEUE_SIZE)
 
             self._root_logger.setLevel(config.get('level', PatroniLogger.DEFAULT_LEVEL))
+            if config.get('traceback_level', PatroniLogger.DEFAULT_TRACEBACK_LEVEL).lower() == 'debug':
+                logging.Logger.exception = debug_exception
+            else:
+                logging.Logger.exception = error_exception
 
             new_handler = None
             if 'dir' in config:
-                if not isinstance(self._log_handler, RotatingFileHandler):
+                if not isinstance(self.log_handler, RotatingFileHandler):
                     new_handler = RotatingFileHandler(os.path.join(config['dir'], __name__))
-                handler = new_handler or self._log_handler
+                handler = new_handler or self.log_handler
                 handler.maxBytes = int(config.get('file_size', 25000000))
                 handler.backupCount = int(config.get('file_num', 4))
             else:
-                if self._log_handler is None or isinstance(self._log_handler, RotatingFileHandler):
+                if self.log_handler is None or isinstance(self.log_handler, RotatingFileHandler):
                     new_handler = logging.StreamHandler()
-                handler = new_handler or self._log_handler
+                handler = new_handler or self.log_handler
 
             oldlogformat = (self._config or {}).get('format', PatroniLogger.DEFAULT_FORMAT)
             logformat = config.get('format', PatroniLogger.DEFAULT_FORMAT)
@@ -110,17 +141,17 @@ class PatroniLogger(Thread):
                 handler.setFormatter(logging.Formatter(logformat, dateformat))
 
             if new_handler:
-                with self._log_handler_lock:
+                with self.log_handler_lock:
-                    if self._log_handler:
+                    if self.log_handler:
-                        self._old_handlers.append(self._log_handler)
+                        self._old_handlers.append(self.log_handler)
-                    self._log_handler = new_handler
+                    self.log_handler = new_handler
 
             self._config = config.copy()
             self.update_loggers()
 
     def _close_old_handlers(self):
         while True:
-            with self._log_handler_lock:
+            with self.log_handler_lock:
                 if not self._old_handlers:
                     break
                 handler = self._old_handlers.pop()
@@ -130,6 +161,13 @@ class PatroniLogger(Thread):
                 _LOGGER.exception('Failed to close the old log handler %s', handler)
 
     def run(self):
+        # switch to QueueHandler only when the thread was started
+        with self.log_handler_lock:
+            self._root_logger.addHandler(self._queue_handler)
+            self._root_logger.removeHandler(self._proxy_handler)
+
+        prev_record = None
+
         while True:
             self._close_old_handlers()
 
@@ -137,7 +175,18 @@ class PatroniLogger(Thread):
             if record is None:
                 break
 
-            self._log_handler.handle(record)
+            if self._root_logger.level == logging.INFO:
+                if record.msg.startswith('Lock owner: '):
+                    prev_record, record = record, None
+                else:
+                    if prev_record and prev_record.thread == record.thread:
+                        if not (record.msg.startswith('no action. ') or record.msg.startswith('PAUSE: no action')):
+                            self.log_handler.handle(prev_record)
+                        prev_record = None
+
+            if record:
+                self.log_handler.handle(record)
+
             self._queue_handler.queue.task_done()
 
     def shutdown(self):

patroni/postgresql/__init__.py

@@ -1,26 +1,31 @@
 import logging
 import os
 import psycopg2
+import re
 import shlex
 import shutil
+import six
 import subprocess
 import time
 
 from contextlib import contextmanager
 from copy import deepcopy
-from patroni.postgresql.callback_executor import CallbackExecutor
+from datetime import datetime
-from patroni.postgresql.bootstrap import Bootstrap
+from dateutil import tz
-from patroni.postgresql.cancellable import CancellableSubprocess
+from psutil import TimeoutExpired
-from patroni.postgresql.config import ConfigHandler
-from patroni.postgresql.connection import Connection, get_connection_cursor
-from patroni.postgresql.misc import parse_history, postgres_major_version_to_int
-from patroni.postgresql.postmaster import PostmasterProcess
-from patroni.postgresql.slots import SlotsHandler
-from patroni.exceptions import PostgresConnectionException
-from patroni.utils import Retry, RetryFailedError, polling_loop
-from patroni.dcs import slot_name_from_member_name, RemoteMember
 from threading import current_thread, Lock
 
+from .bootstrap import Bootstrap
+from .callback_executor import CallbackExecutor
+from .cancellable import CancellableSubprocess
+from .config import ConfigHandler, mtime
+from .connection import Connection, get_connection_cursor
+from .misc import parse_history, parse_lsn, postgres_major_version_to_int
+from .postmaster import PostmasterProcess
+from .slots import SlotsHandler
+from ..exceptions import PostgresConnectionException
+from ..utils import Retry, RetryFailedError, polling_loop, data_directory_is_empty, parse_int
+
 
 logger = logging.getLogger(__name__)
 
@@ -38,15 +43,6 @@ STATE_UNKNOWN = 'unknown'
 
 STOP_POLLING_INTERVAL = 1
 
-cluster_info_query = ("SELECT CASE WHEN pg_catalog.pg_is_in_recovery() THEN 0 "
-                      "ELSE ('x' || pg_catalog.substr(pg_catalog.pg_{0}file_name("
-                      "pg_catalog.pg_current_{0}_{1}()), 1, 8))::bit(32)::int END, "
-                      "CASE WHEN pg_catalog.pg_is_in_recovery() THEN GREATEST("
-                      " pg_catalog.pg_{0}_{1}_diff(COALESCE("
-                      "pg_catalog.pg_last_{0}_receive_{1}(), '0/0'), '0/0')::bigint,"
-                      " pg_catalog.pg_{0}_{1}_diff(pg_catalog.pg_last_{0}_replay_{1}(), '0/0')::bigint)"
-                      "ELSE pg_catalog.pg_{0}_{1}_diff(pg_catalog.pg_current_{0}_{1}(), '0/0')::bigint END")
-
 
 @contextmanager
 def null_context():
@@ -55,12 +51,23 @@ def null_context():
 
 class Postgresql(object):
 
+    POSTMASTER_START_TIME = "pg_catalog.pg_postmaster_start_time()"
+    TL_LSN = ("CASE WHEN pg_catalog.pg_is_in_recovery() THEN 0 "
+              "ELSE ('x' || pg_catalog.substr(pg_catalog.pg_{0}file_name("
+              "pg_catalog.pg_current_{0}_{1}()), 1, 8))::bit(32)::int END, "  # master timeline
+              "CASE WHEN pg_catalog.pg_is_in_recovery() THEN 0 "
+              "ELSE pg_catalog.pg_{0}_{1}_diff(pg_catalog.pg_current_{0}_{1}(), '0/0')::bigint END, "  # write_lsn
+              "pg_catalog.pg_{0}_{1}_diff(pg_catalog.pg_last_{0}_replay_{1}(), '0/0')::bigint, "
+              "pg_catalog.pg_{0}_{1}_diff(COALESCE(pg_catalog.pg_last_{0}_receive_{1}(), '0/0'), '0/0')::bigint, "
+              "pg_catalog.pg_is_in_recovery() AND pg_catalog.pg_is_{0}_replay_paused()")
+
     def __init__(self, config):
         self.name = config['name']
         self.scope = config['scope']
         self._data_dir = config['data_dir']
         self._database = config.get('database', 'postgres')
         self._version_file = os.path.join(self._data_dir, 'PG_VERSION')
+        self._pg_control = os.path.join(self._data_dir, 'global', 'pg_control')
         self._major_version = self.get_major_version()
 
         self._state_lock = Lock()
@@ -69,6 +76,7 @@ class Postgresql(object):
         self._pending_restart = False
         self._connection = Connection()
         self.config = ConfigHandler(self, config)
+        self.config.check_directories()
 
         self._bin_dir = config.get('bin_dir') or ''
         self.bootstrap = Bootstrap(self)
@@ -77,7 +85,6 @@ class Postgresql(object):
 
         self.slots_handler = SlotsHandler(self)
 
-        self._pgpass = config.get('pgpass') or os.path.join(os.path.expanduser('~'), 'pgpass')
         self._callback_executor = CallbackExecutor()
         self.__cb_called = False
         self.__cb_pending = None
@@ -97,15 +104,19 @@ class Postgresql(object):
         self._state_entry_timestamp = None
 
         self._cluster_info_state = {}
+        self._has_permanent_logical_slots = True
+        self._enforce_hot_standby_feedback = False
         self._cached_replica_timeline = None
 
         # Last known running process
         self._postmaster_proc = None
 
-        if self.is_running():
+        if self.is_running():  # we are "joining" already running postgres
             self.set_state('running')
             self.set_role('master' if self.is_leader() else 'replica')
-            self.config.write_postgresql_conf()  # we are "joining" already running postgres
+            # postpone writing postgresql.conf for 12+ because recovery parameters are not yet known
+            if self.major_version < 120000 or self.is_leader():
+                self.config.write_postgresql_conf()
             hba_saved = self.config.replace_pg_hba()
             ident_saved = self.config.replace_pg_ident()
             if hba_saved or ident_saved:
@@ -133,6 +144,10 @@ class Postgresql(object):
     def callback(self):
         return self.config.get('callbacks') or {}
 
+    @property
+    def wal_dir(self):
+        return os.path.join(self._data_dir, 'pg_' + self.wal_name)
+
     @property
     def wal_name(self):
         return 'wal' if self._major_version >= 100000 else 'xlog'
@@ -141,6 +156,24 @@ class Postgresql(object):
     def lsn_name(self):
         return 'lsn' if self._major_version >= 100000 else 'location'
 
+    @property
+    def cluster_info_query(self):
+        if self._major_version >= 90600:
+            extra = "(SELECT pg_catalog.json_agg(s.*) FROM (SELECT slot_name, slot_type as type, datoid::bigint, " +\
+                    "plugin, catalog_xmin, pg_catalog.pg_wal_lsn_diff(confirmed_flush_lsn, '0/0')::bigint" + \
+                    " AS confirmed_flush_lsn FROM pg_catalog.pg_get_replication_slots()) AS s)"\
+                            if self._has_permanent_logical_slots and self._major_version >= 110000 else "NULL"
+            extra = (", CASE WHEN latest_end_lsn IS NULL THEN NULL ELSE received_tli END,"
+                     " slot_name, conninfo, {0} FROM pg_catalog.pg_stat_get_wal_receiver()").format(extra)
+            if self.role == 'standby_leader':
+                extra = "timeline_id" + extra + ", pg_catalog.pg_control_checkpoint()"
+            else:
+                extra = "0" + extra
+        else:
+            extra = "0, NULL, NULL, NULL, NULL"
+
+        return ("SELECT " + self.TL_LSN + ", {2}").format(self.wal_name, self.lsn_name, extra)
+
     def _version_file_exists(self):
         return not self.data_directory_empty() and os.path.isfile(self._version_file)
 
@@ -188,8 +221,8 @@ class Postgresql(object):
                         3: STATE_UNKNOWN}
         return return_codes.get(ret, STATE_UNKNOWN)
 
-    def reload_config(self, config):
+    def reload_config(self, config, sighup=False):
-        self.config.reload_config(config)
+        self.config.reload_config(config, sighup)
         self._is_leader_retry.deadline = self.retry.deadline = config['retry_timeout']/2.0
 
     @property
@@ -207,7 +240,7 @@ class Postgresql(object):
         return self._sysid
 
     def get_postgres_role_from_data_directory(self):
-        if self.data_directory_empty():
+        if self.data_directory_empty() or not self.controldata():
             return 'uninitialized'
         elif self.config.recovery_conf_exists():
             return 'replica'
@@ -255,21 +288,13 @@ class Postgresql(object):
         except RetryFailedError as e:
             raise PostgresConnectionException(str(e))
 
+    def pg_control_exists(self):
+        return os.path.isfile(self._pg_control)
+
     def data_directory_empty(self):
-        return not os.path.exists(self._data_dir) or os.listdir(self._data_dir) == []
+        if self.pg_control_exists():
-
+            return False
-    def write_pgpass(self, record):
+        return data_directory_is_empty(self._data_dir)
-        if 'user' not in record or 'password' not in record:
-            return os.environ.copy()
-
-        with open(self._pgpass, 'w') as f:
-            if os.name != 'nt':
-                os.fchmod(f.fileno(), 0o600)
-            f.write('{host}:{port}:*:{user}:{password}\n'.format(**record))
-
-        env = os.environ.copy()
-        env['PGPASSFILE'] = self._pgpass
-        return env
 
     def replica_method_options(self, method):
         return deepcopy(self.config.get(method, {}))
@@ -285,15 +310,36 @@ class Postgresql(object):
             replica_methods = self.create_replica_methods
         return any(self.replica_method_can_work_without_replication_connection(m) for m in replica_methods)
 
-    def reset_cluster_info_state(self):
+    @property
+    def enforce_hot_standby_feedback(self):
+        return self._enforce_hot_standby_feedback
+
+    def set_enforce_hot_standby_feedback(self, value):
+        # If we enable or disable the hot_standby_feedback we need to update postgresql.conf and reload
+        if self._enforce_hot_standby_feedback != value:
+            self._enforce_hot_standby_feedback = value
+            if self.is_running():
+                self.config.write_postgresql_conf()
+                self.reload()
+
+    def reset_cluster_info_state(self, cluster, nofailover=None):
         self._cluster_info_state = {}
+        if cluster and cluster.config and cluster.config.modify_index:
+            self._has_permanent_logical_slots =\
+                cluster.has_permanent_logical_slots(self.name, nofailover, self.major_version)
+            self.set_enforce_hot_standby_feedback(
+                self._has_permanent_logical_slots or
+                cluster.should_enforce_hot_standby_feedback(self.name, nofailover, self.major_version))
 
     def _cluster_info_state_get(self, name):
         if not self._cluster_info_state:
-            stmt = cluster_info_query.format(self.wal_name, self.lsn_name)
             try:
-                result = self._is_leader_retry(self._query, stmt).fetchone()
+                result = self._is_leader_retry(self._query, self.cluster_info_query).fetchone()
-                self._cluster_info_state = dict(zip(['timeline', 'wal_position'], result))
+                cluster_info_state = dict(zip(['timeline', 'wal_position', 'replayed_location',
+                                               'received_location', 'replay_paused', 'pg_control_timeline',
+                                               'received_tli', 'slot_name', 'conninfo', 'slots'], result))
+                cluster_info_state['slots'] = self.slots_handler.process_permanent_slots(cluster_info_state['slots'])
+                self._cluster_info_state = cluster_info_state
             except RetryFailedError as e:  # SELECT failed two times
                 self._cluster_info_state = {'error': str(e)}
                 if not self.is_starting() and self.pg_isready() == STATE_REJECT:
@@ -304,9 +350,80 @@ class Postgresql(object):
 
         return self._cluster_info_state.get(name)
 
+    def replayed_location(self):
+        return self._cluster_info_state_get('replayed_location')
+
+    def received_location(self):
+        return self._cluster_info_state_get('received_location')
+
+    def slots(self):
+        return self._cluster_info_state_get('slots')
+
+    def primary_slot_name(self):
+        return self._cluster_info_state_get('slot_name')
+
+    def primary_conninfo(self):
+        return self._cluster_info_state_get('conninfo')
+
+    def received_timeline(self):
+        return self._cluster_info_state_get('received_tli')
+
     def is_leader(self):
         return bool(self._cluster_info_state_get('timeline'))
 
+    def replay_paused(self):
+        return self._cluster_info_state_get('replay_paused')
+
+    def resume_wal_replay(self):
+        self._query('SELECT pg_catalog.pg_{0}_replay_resume()'.format(self.wal_name))
+
+    def handle_parameter_change(self):
+        if self.major_version >= 140000 and self.replay_paused():
+            logger.info('Resuming paused WAL replay for PostgreSQL 14+')
+            self.resume_wal_replay()
+
+    def pg_control_timeline(self):
+        try:
+            return int(self.controldata().get("Latest checkpoint's TimeLineID"))
+        except (TypeError, ValueError):
+            logger.exception('Failed to parse timeline from pg_controldata output')
+
+    def parse_wal_record(self, timeline, lsn):
+        out, err = self.waldump(timeline, lsn, 1)
+        if out and not err:
+            match = re.match(r'^rmgr:\s+(.+?)\s+len \(rec/tot\):\s+\d+/\s+\d+, tx:\s+\d+, '
+                             r'lsn: ([0-9A-Fa-f]+/[0-9A-Fa-f]+), prev ([0-9A-Fa-f]+/[0-9A-Fa-f]+), '
+                             r'.*?desc: (.+)', out.decode('utf-8'))
+            if match:
+                return match.groups()
+        return None, None, None, None
+
+    def latest_checkpoint_location(self):
+        """Returns checkpoint location for the cleanly shut down primary.
+           But, if we know that the checkpoint was written to the new WAL
+           due to the archive_mode=on, we will return the LSN of prev wal record (SWITCH)."""
+
+        data = self.controldata()
+        timeline = data.get("Latest checkpoint's TimeLineID")
+        lsn = checkpoint_lsn = data.get('Latest checkpoint location')
+        if data.get('Database cluster state') == 'shut down' and lsn and timeline:
+            try:
+                checkpoint_lsn = parse_lsn(checkpoint_lsn)
+                rm_name, lsn, prev, desc = self.parse_wal_record(timeline, lsn)
+                desc = desc.strip().lower()
+                if rm_name == 'XLOG' and parse_lsn(lsn) == checkpoint_lsn and prev and\
+                        desc.startswith('checkpoint') and desc.endswith('shutdown'):
+                    _, lsn, _, desc = self.parse_wal_record(timeline, prev)
+                    prev = parse_lsn(prev)
+                    # If the cluster is shutdown with archive_mode=on, WAL is switched before writing the checkpoint.
+                    # In this case we want to take the LSN of previous record (switch) as the last known WAL location.
+                    if parse_lsn(lsn) == prev and desc.strip() in ('xlog switch', 'SWITCH'):
+                        return str(prev)
+            except Exception as e:
+                logger.error('Exception when parsing WAL pg_%sdump output: %r', self.wal_name, e)
+            if isinstance(checkpoint_lsn, six.integer_types):
+                return str(checkpoint_lsn)
+
     def is_running(self):
         """Returns PostmasterProcess if one is running on the data directory or None. If most recently seen process
         is running updates the cached process based on pid file."""
@@ -315,7 +432,7 @@ class Postgresql(object):
                 return self._postmaster_proc
             self._postmaster_proc = None
 
-        # we noticed that postgres was restarted, force syncing of replication
+        # we noticed that postgres was restarted, force syncing of replication slots and check of logical slots
         self.slots_handler.schedule()
 
         self._postmaster_proc = PostmasterProcess.from_pidfile(self._data_dir)
@@ -411,14 +528,21 @@ class Postgresql(object):
         self.set_state('starting')
         self._pending_restart = False
 
-        configuration = self.config.effective_configuration
+        try:
+            if not self._major_version:
+                self.configure_server_parameters()
+            configuration = self.config.effective_configuration
+        except Exception:
+            return None
+
+        self.config.check_directories()
         self.config.write_postgresql_conf(configuration)
         self.config.resolve_connection_addresses()
         self.config.replace_pg_hba()
         self.config.replace_pg_ident()
 
         options = ['--{0}={1}'.format(p, configuration[p]) for p in self.config.CMDLINE_OPTIONS
-                   if p in configuration and p != 'wal_keep_segments']
+                   if p in configuration and p not in ('wal_keep_segments', 'wal_keep_size')]
 
         if self.cancellable.is_cancelled:
             return False
@@ -455,11 +579,13 @@ class Postgresql(object):
         else:
             return None
 
-    def checkpoint(self, connect_kwargs=None):
+    def checkpoint(self, connect_kwargs=None, timeout=None):
         check_not_is_in_recovery = connect_kwargs is not None
         connect_kwargs = connect_kwargs or self.config.local_connect_kwargs
         for p in ['connect_timeout', 'options']:
             connect_kwargs.pop(p, None)
+        if timeout:
+            connect_kwargs['connect_timeout'] = timeout
         try:
             with get_connection_cursor(**connect_kwargs) as cur:
                 cur.execute("SET statement_timeout = 0")
@@ -472,7 +598,7 @@ class Postgresql(object):
             logger.exception('Exception during CHECKPOINT')
             return 'not accessible or not healty'
 
-    def stop(self, mode='fast', block_callbacks=False, checkpoint=None, on_safepoint=None):
+    def stop(self, mode='fast', block_callbacks=False, checkpoint=None, on_safepoint=None, stop_timeout=None):
         """Stop PostgreSQL
 
         Supports a callback when a safepoint is reached. A safepoint is when no user backend can return a successful
@@ -484,7 +610,7 @@ class Postgresql(object):
         if checkpoint is None:
             checkpoint = False if mode == 'immediate' else True
 
-        success, pg_signaled = self._do_stop(mode, block_callbacks, checkpoint, on_safepoint)
+        success, pg_signaled = self._do_stop(mode, block_callbacks, checkpoint, on_safepoint, stop_timeout)
         if success:
             # block_callbacks is used during restart to avoid
             # running start/stop callbacks in addition to restart ones
@@ -497,7 +623,7 @@ class Postgresql(object):
             self.set_state('stop failed')
         return success
 
-    def _do_stop(self, mode, block_callbacks, checkpoint, on_safepoint):
+    def _do_stop(self, mode, block_callbacks, checkpoint, on_safepoint, stop_timeout):
         postmaster = self.is_running()
         if not postmaster:
             if on_safepoint:
@@ -505,13 +631,13 @@ class Postgresql(object):
             return True, False
 
         if checkpoint and not self.is_starting():
-            self.checkpoint()
+            self.checkpoint(timeout=stop_timeout)
 
         if not block_callbacks:
             self.set_state('stopping')
 
         # Send signal to postmaster to stop
-        success = postmaster.signal_stop(mode)
+        success = postmaster.signal_stop(mode, self.pgcommand('pg_ctl'))
         if success is not None:
             if success and on_safepoint:
                 on_safepoint()
@@ -524,15 +650,32 @@ class Postgresql(object):
             postmaster.wait_for_user_backends_to_close()
             on_safepoint()
 
-        postmaster.wait()
+        try:
+            postmaster.wait(timeout=stop_timeout)
+        except TimeoutExpired:
+            logger.warning("Timeout during postmaster stop, aborting Postgres.")
+            if not self.terminate_postmaster(postmaster, mode, stop_timeout):
+                postmaster.wait()
 
         return True, True
 
-    @staticmethod
+    def terminate_postmaster(self, postmaster, mode, stop_timeout):
-    def terminate_starting_postmaster(postmaster):
+        if mode in ['fast', 'smart']:
+            try:
+                success = postmaster.signal_stop('immediate', self.pgcommand('pg_ctl'))
+                if success:
+                    return True
+                postmaster.wait(timeout=stop_timeout)
+                return True
+            except TimeoutExpired:
+                pass
+        logger.warning("Sending SIGKILL to Postmaster and its children")
+        return postmaster.signal_kill()
+
+    def terminate_starting_postmaster(self, postmaster):
         """Terminates a postmaster that has not yet opened ports or possibly even written a pid file. Blocks
         until the process goes away."""
-        postmaster.signal_stop('immediate')
+        postmaster.signal_stop('immediate', self.pgcommand('pg_ctl'))
         postmaster.wait()
 
     def _wait_for_connection_close(self, postmaster):
@@ -544,9 +687,9 @@ class Postgresql(object):
         except psycopg2.Error:
             pass
 
-    def reload(self):
+    def reload(self, block_callbacks=False):
         ret = self.pg_ctl('reload')
-        if ret:
+        if ret and not block_callbacks:
             self.call_nowait(ACTION_ON_RELOAD)
         return ret
 
@@ -566,10 +709,12 @@ class Postgresql(object):
         if ready == STATE_REJECT:
             return False
         elif ready == STATE_NO_RESPONSE:
-            self.set_state('start failed')
+            ret = not self.is_running()
-            self.slots_handler.schedule(False)  # TODO: can remove this?
+            if ret:
-            self.config.save_configuration_files(True)  # TODO: maybe remove this?
+                self.set_state('start failed')
-            return True
+                self.slots_handler.schedule(False)  # TODO: can remove this?
+                self.config.save_configuration_files(True)  # TODO: maybe remove this?
+            return ret
         else:
             if ready != STATE_RUNNING:
                 # Bad configuration or unexpected OS error. No idea of PostgreSQL status.
@@ -624,45 +769,60 @@ class Postgresql(object):
             return False
         return True
 
+    def get_guc_value(self, name):
+        cmd = [self.pgcommand('postgres'), '-D', self._data_dir, '-C', name]
+        try:
+            data = subprocess.check_output(cmd)
+            if data:
+                return data.decode('utf-8').strip()
+        except Exception as e:
+            logger.error('Failed to execute %s: %r', cmd, e)
+
     def controldata(self):
         """ return the contents of pg_controldata, or non-True value if pg_controldata call failed """
-        result = {}
         # Don't try to call pg_controldata during backup restore
         if self._version_file_exists() and self.state != 'creating replica':
             try:
-                env = {'LANG': 'C', 'LC_ALL': 'C', 'PATH': os.getenv('PATH')}
+                env = os.environ.copy()
-                if os.getenv('SYSTEMROOT') is not None:
+                env.update(LANG='C', LC_ALL='C')
-                    env['SYSTEMROOT'] = os.getenv('SYSTEMROOT')
                 data = subprocess.check_output([self.pgcommand('pg_controldata'), self._data_dir], env=env)
                 if data:
-                    data = data.decode('utf-8').splitlines()
+                    data = filter(lambda e: ':' in e, data.decode('utf-8').splitlines())
-                    # pg_controldata output depends on major verion. Some of parameters are prefixed by 'Current '
+                    # pg_controldata output depends on major version. Some of parameters are prefixed by 'Current '
-                    result = {l.split(':')[0].replace('Current ', '', 1): l.split(':', 1)[1].strip() for l in data
+                    return {k.replace('Current ', '', 1): v.strip() for k, v in map(lambda e: e.split(':', 1), data)}
-                              if l and ':' in l}
             except subprocess.CalledProcessError:
                 logger.exception("Error when calling pg_controldata")
-        return result
+        return {}
+
+    def waldump(self, timeline, lsn, limit):
+        cmd = self.pgcommand('pg_{0}dump'.format(self.wal_name))
+        env = os.environ.copy()
+        env.update(LANG='C', LC_ALL='C', PGDATA=self._data_dir)
+        try:
+            waldump = subprocess.Popen([cmd, '-t', str(timeline), '-s', str(lsn), '-n', str(limit)],
+                                       stdout=subprocess.PIPE, stderr=subprocess.PIPE, env=env)
+            out, err = waldump.communicate()
+            waldump.wait()
+            return out, err
+        except Exception as e:
+            logger.error('Failed to execute `%s -t %s -s %s -n %s`: %r', cmd, timeline, lsn, limit, e)
+            return None, None
 
     @contextmanager
-    def get_replication_connection_cursor(self, host='localhost', port=5432, database=None, **kwargs):
+    def get_replication_connection_cursor(self, host='localhost', port=5432, **kwargs):
-        replication = self.config.replication
+        conn_kwargs = self.config.replication.copy()
-        with get_connection_cursor(host=host, port=int(port), database=database or self._database, replication=1,
+        conn_kwargs.update(host=host, port=int(port) if port else None, user=conn_kwargs.pop('username'),
-                                   user=replication['username'], password=replication.get('password'),
+                           connect_timeout=3, replication=1, options='-c statement_timeout=2000')
-                                   connect_timeout=3, options='-c statement_timeout=2000') as cur:
+        with get_connection_cursor(**conn_kwargs) as cur:
             yield cur
 
-    def get_local_timeline_lsn_from_replication_connection(self):
+    def get_replica_timeline(self):
-        timeline = lsn = None
         try:
             with self.get_replication_connection_cursor(**self.config.local_replication_address) as cur:
                 cur.execute('IDENTIFY_SYSTEM')
-                timeline, lsn = cur.fetchone()[1:3]
+                return cur.fetchone()[1]
         except Exception:
             logger.exception('Can not fetch local timeline and lsn from replication connection')
-        return timeline, lsn
-
-    def get_replica_timeline(self):
-        return self.get_local_timeline_lsn_from_replication_connection()[0]
 
     def replica_cached_timeline(self, master_timeline):
         if not self._cached_replica_timeline or not master_timeline or self._cached_replica_timeline != master_timeline:
@@ -673,57 +833,40 @@ class Postgresql(object):
         return self._cluster_info_state_get('timeline')
 
     def get_history(self, timeline):
-        history_path = 'pg_{0}/{1:08X}.history'.format(self.wal_name, timeline)
+        history_path = os.path.join(self.wal_dir, '{0:08X}.history'.format(timeline))
-        try:
+        history_mtime = mtime(history_path)
-            cursor = self._connection.cursor()
+        if history_mtime:
-            cursor.execute('SELECT isdir, modification FROM pg_catalog.pg_stat_file(%s)', (history_path,))
+            try:
-            isdir, modification = cursor.fetchone()
+                with open(history_path, 'r') as f:
-            if not isdir:
+                    history = f.read()
-                cursor.execute('SELECT pg_catalog.pg_read_file(%s)', (history_path,))
+                history = list(parse_history(history))
-                history = list(parse_history(cursor.fetchone()[0]))
                 if history[-1][0] == timeline - 1:
-                    history[-1].append(modification.isoformat())
+                    history_mtime = datetime.fromtimestamp(history_mtime).replace(tzinfo=tz.tzlocal())
+                    history[-1].append(history_mtime.isoformat())
                 return history
-        except Exception:
+            except Exception:
-            logger.exception('Failed to read and parse %s', (history_path,))
+                logger.exception('Failed to read and parse %s', (history_path,))
-
-    def follow(self, member, role='replica', timeout=None):
-        is_remote_master = isinstance(member, RemoteMember)
-        no_replication_slot = is_remote_master and member.no_replication_slot
-        restore_command = is_remote_master and member.restore_command
-        min_apply_delay = is_remote_master and member.recovery_min_apply_delay
-        archive_cleanup = is_remote_master and member.archive_cleanup_command
-
-        primary_conninfo = self.config.primary_conninfo(member)
-        change_role = self.cb_called and (self.role in ('master', 'demoted') or
-                                          not {'standby_leader', 'replica'} - {self.role, role})
-
-        recovery_params = self.config.get('recovery_conf', {}).copy()
-        recovery_params.update({'standby_mode': 'on', 'recovery_target_timeline': 'latest'})
-        if primary_conninfo:
-            recovery_params['primary_conninfo'] = primary_conninfo
-        if self.slots_handler.use_slots and not no_replication_slot:
-            required_name = is_remote_master and member.data.get('primary_slot_name')
-            name = required_name or slot_name_from_member_name(self.name)
-            recovery_params['primary_slot_name'] = name
-        if restore_command:
-            recovery_params['restore_command'] = restore_command
-        if min_apply_delay:
-            recovery_params['recovery_min_apply_delay'] = min_apply_delay
-        if archive_cleanup:
-            recovery_params['archive_cleanup_command'] = archive_cleanup
 
+    def follow(self, member, role='replica', timeout=None, do_reload=False):
+        recovery_params = self.config.build_recovery_params(member)
         self.config.write_recovery_conf(recovery_params)
 
         # When we demoting the master or standby_leader to replica or promoting replica to a standby_leader
         # and we know for sure that postgres was already running before, we will only execute on_role_change
         # callback and prevent execution of on_restart/on_start callback.
         # If the role remains the same (replica or standby_leader), we will execute on_start or on_restart
+        change_role = self.cb_called and (self.role in ('master', 'demoted') or
+                                          not {'standby_leader', 'replica'} - {self.role, role})
         if change_role:
             self.__cb_pending = ACTION_NOOP
 
         if self.is_running():
-            self.restart(block_callbacks=change_role, role=role)
+            if do_reload:
+                self.config.write_postgresql_conf()
+                if self.reload(block_callbacks=change_role) and change_role:
+                    self.set_role(role)
+            else:
+                self.restart(block_callbacks=change_role, role=role)
         else:
             self.start(timeout=timeout, block_callbacks=change_role, role=role)
 
@@ -738,49 +881,136 @@ class Postgresql(object):
             if data.get('Database cluster state') == 'in production':
                 return True
 
-    def promote(self, wait_seconds, on_success=None, access_is_restricted=False):
+    def _pre_promote(self):
+        """
+        Runs a fencing script after the leader lock is acquired but before the replica is promoted.
+        If the script exits with a non-zero code, promotion does not happen and the leader key is removed from DCS.
+        """
+
+        cmd = self.config.get('pre_promote')
+        if not cmd:
+            return True
+
+        ret = self.cancellable.call(shlex.split(cmd))
+        if ret is not None:
+            logger.info('pre_promote script `%s` exited with %s', cmd, ret)
+        return ret == 0
+
+    def promote(self, wait_seconds, task, on_success=None):
         if self.role == 'master':
             return True
+
+        ret = self._pre_promote()
+        with task:
+            if task.is_cancelled:
+                return False
+            task.complete(ret)
+
+        if ret is False:
+            return False
+
+        if self.cancellable.is_cancelled:
+            logger.info("PostgreSQL promote cancelled.")
+            return False
+
+        self.slots_handler.on_promote()
+
         ret = self.pg_ctl('promote', '-W')
         if ret:
             self.set_role('master')
             if on_success is not None:
                 on_success()
-            if not access_is_restricted:
+            self.call_nowait(ACTION_ON_ROLE_CHANGE)
-                self.call_nowait(ACTION_ON_ROLE_CHANGE)
             ret = self._wait_promote(wait_seconds)
         return ret
 
+    @staticmethod
+    def _wal_position(is_leader, wal_position, received_location, replayed_location):
+        return wal_position if is_leader else max(received_location or 0, replayed_location or 0)
+
     def timeline_wal_position(self):
         # This method could be called from different threads (simultaneously with some other `_query` calls).
         # If it is called not from main thread we will create a new cursor to execute statement.
         if current_thread().ident == self.__thread_ident:
-            return self._cluster_info_state_get('timeline'), self._cluster_info_state_get('wal_position')
+            timeline = self._cluster_info_state_get('timeline')
+            wal_position = self._cluster_info_state_get('wal_position')
+            replayed_location = self.replayed_location()
+            received_location = self.received_location()
+            pg_control_timeline = self._cluster_info_state_get('pg_control_timeline')
+        else:
+            with self.connection().cursor() as cursor:
+                cursor.execute(self.cluster_info_query)
+                (timeline, wal_position, replayed_location,
+                 received_location, _, pg_control_timeline) = cursor.fetchone()[:6]
 
-        with self.connection().cursor() as cursor:
+        wal_position = self._wal_position(timeline, wal_position, received_location, replayed_location)
-            cursor.execute(cluster_info_query.format(self.wal_name, self.lsn_name))
+        return (timeline, wal_position, pg_control_timeline)
-            return cursor.fetchone()[:2]
 
     def postmaster_start_time(self):
         try:
-            cursor = self.query("SELECT pg_catalog.to_char(pg_catalog.pg_postmaster_start_time(),"
+            query = "SELECT " + self.POSTMASTER_START_TIME
-                                " 'YYYY-MM-DD HH24:MI:SS.MS TZ')")
+            if current_thread().ident == self.__thread_ident:
-            return cursor.fetchone()[0]
+                return self.query(query).fetchone()[0].isoformat(sep=' ')
+            with self.connection().cursor() as cursor:
+                cursor.execute(query)
+                return cursor.fetchone()[0].isoformat(sep=' ')
         except psycopg2.Error:
             return None
 
     def last_operation(self):
-        return str(self._cluster_info_state_get('wal_position'))
+        return self._wal_position(self.is_leader(), self._cluster_info_state_get('wal_position'),
+                                  self.received_location(), self.replayed_location())
 
     def configure_server_parameters(self):
         self._major_version = self.get_major_version()
         self.config.setup_server_parameters()
         return True
 
+    def pg_wal_realpath(self):
+        """Returns a dict containing the symlink (key) and target (value) for the wal directory"""
+        links = {}
+        for pg_wal_dir in ('pg_xlog', 'pg_wal'):
+            pg_wal_path = os.path.join(self._data_dir, pg_wal_dir)
+            if os.path.exists(pg_wal_path) and os.path.islink(pg_wal_path):
+                pg_wal_realpath = os.path.realpath(pg_wal_path)
+                links[pg_wal_path] = pg_wal_realpath
+        return links
+
+    def pg_tblspc_realpaths(self):
+        """Returns a dict containing the symlink (key) and target (values) for the tablespaces"""
+        links = {}
+        pg_tblsp_dir = os.path.join(self._data_dir, 'pg_tblspc')
+        if os.path.exists(pg_tblsp_dir):
+            for tsdn in os.listdir(pg_tblsp_dir):
+                pg_tsp_path = os.path.join(pg_tblsp_dir, tsdn)
+                if parse_int(tsdn) and os.path.islink(pg_tsp_path):
+                    pg_tsp_rpath = os.path.realpath(pg_tsp_path)
+                    links[pg_tsp_path] = pg_tsp_rpath
+        return links
+
     def move_data_directory(self):
         if os.path.isdir(self._data_dir) and not self.is_running():
             try:
-                new_name = '{0}_{1}'.format(self._data_dir, time.strftime('%Y-%m-%d-%H-%M-%S'))
+                postfix = time.strftime('%Y-%m-%d-%H-%M-%S')
+
+                # let's see if the wal directory is a symlink, in this case we
+                # should move the target
+                for (source, pg_wal_realpath) in self.pg_wal_realpath().items():
+                    logger.info('renaming WAL directory and updating symlink: %s', pg_wal_realpath)
+                    new_name = '{0}_{1}'.format(pg_wal_realpath, postfix)
+                    os.rename(pg_wal_realpath, new_name)
+                    os.unlink(source)
+                    os.symlink(new_name, source)
+
+                # Move user defined tablespace directory
+                for (source, pg_tsp_rpath) in self.pg_tblspc_realpaths().items():
+                    logger.info('renaming user defined tablespace directory and updating symlink: %s', pg_tsp_rpath)
+                    new_name = '{0}_{1}'.format(pg_tsp_rpath, postfix)
+                    os.rename(pg_tsp_rpath, new_name)
+                    os.unlink(source)
+                    os.symlink(new_name, source)
+
+                new_name = '{0}_{1}'.format(self._data_dir, postfix)
                 logger.info('renaming data directory to %s', new_name)
                 os.rename(self._data_dir, new_name)
             except OSError:
@@ -798,105 +1028,81 @@ class Postgresql(object):
                 os.remove(self._data_dir)
             elif os.path.isdir(self._data_dir):
 
-                # let's see if pg_xlog|pg_wal is a symlink, in this case we
+                # let's see if wal directory is a symlink, in this case we
                 # should clean the target
-                for pg_wal_dir in ('pg_xlog', 'pg_wal'):
+                for pg_wal_realpath in self.pg_wal_realpath().values():
-                    pg_wal_path = os.path.join(self._data_dir, pg_wal_dir)
+                    logger.info('Removing WAL directory: %s', pg_wal_realpath)
-                    if os.path.exists(pg_wal_path) and os.path.islink(pg_wal_path):
+                    shutil.rmtree(pg_wal_realpath)
-                        pg_wal_realpath = os.path.realpath(pg_wal_path)
+
-                        logger.info('Removing WAL directory: %s', pg_wal_realpath)
+                # Remove user defined tablespace directories
-                        shutil.rmtree(pg_wal_realpath)
+                for pg_tsp_rpath in self.pg_tblspc_realpaths().values():
+                    logger.info('Removing user defined tablespace directory: %s', pg_tsp_rpath)
+                    shutil.rmtree(pg_tsp_rpath, ignore_errors=True)
 
                 shutil.rmtree(self._data_dir)
         except (IOError, OSError):
             logger.exception('Could not remove data directory %s', self._data_dir)
             self.move_data_directory()
 
-    def pick_synchronous_standby(self, cluster):
+    def _get_synchronous_commit_param(self):
+        return self.query("SHOW synchronous_commit").fetchone()[0]
+
+    def pick_synchronous_standby(self, cluster, sync_node_count=1, sync_node_maxlag=-1):
         """Finds the best candidate to be the synchronous standby.
 
         Current synchronous standby is always preferred, unless it has disconnected or does not want to be a
         synchronous standby any longer.
+        Parameter sync_node_maxlag(maximum_lag_on_syncnode) would help swapping unhealthy sync replica in case
+        if it stops responding (or hung). Please set the value high enough so it won't unncessarily swap sync
+        standbys during high loads. Any less or equal of 0 value keep the behavior backward compatible and
+        will not swap. Please note that it will not also swap sync standbys in case where all replicas are hung.
 
-        :returns tuple of candidate name or None, and bool showing if the member is the active synchronous standby.
+        :returns tuple of candidates list and synchronous standby list.
         """
-        current = cluster.sync.sync_standby
+        if self._major_version < 90600:
-        current = current.lower() if current else current
+            sync_node_count = 1
         members = {m.name.lower(): m for m in cluster.members}
         candidates = []
-        # Pick candidates based on who has flushed WAL farthest.
+        sync_nodes = []
-        # TODO: for synchronous_commit = remote_write we actually want to order on write_location
+        replica_list = []
-        for app_name, state, sync_state in self.query(
+        # Pick candidates based on who has higher replay/remote_write/flush lsn.
-                "SELECT pg_catalog.lower(application_name), state, sync_state"
+        sync_commit_par = self._get_synchronous_commit_param()
+        sort_col = {'remote_apply': 'replay', 'remote_write': 'write'}.get(sync_commit_par, 'flush')
+        # pg_stat_replication.sync_state has 4 possible states - async, potential, quorum, sync.
+        # Sort clause "ORDER BY sync_state DESC" is to get the result in required order and to keep
+        # the result consistent in case if a synchronous standby member is slowed down OR async node
+        # receiving changes faster than the sync member (very rare but possible). Such cases would
+        # trigger sync standby member swapping frequently and the sort on sync_state desc should
+        # help in keeping the query result consistent.
+        for app_name, sync_state, replica_lsn in self.query(
+                "SELECT pg_catalog.lower(application_name), sync_state, pg_{2}_{1}_diff({0}_{1}, '0/0')::bigint"
                 " FROM pg_catalog.pg_stat_replication"
-                " ORDER BY flush_{0} DESC".format(self.lsn_name)):
+                " WHERE state = 'streaming'"
+                " ORDER BY sync_state DESC, {0}_{1} DESC".format(sort_col, self.lsn_name, self.wal_name)):
             member = members.get(app_name)
-            if state != 'streaming' or not member or member.tags.get('nosync', False):
+            if member and not member.tags.get('nosync', False):
-                continue
+                replica_list.append((member.name, sync_state, replica_lsn))
-            if sync_state == 'sync':
+
-                return app_name, True
+        max_lsn = max(replica_list, key=lambda x: x[2])[2] if len(replica_list) > 1 else int(str(self.last_operation()))
-            if sync_state == 'potential' and app_name == current:
+
-                # Prefer current even if not the best one any more to avoid indecisivness and spurious swaps.
+        for app_name, sync_state, replica_lsn in replica_list:
-                return current, False
+            if sync_node_maxlag <= 0 or max_lsn - replica_lsn <= sync_node_maxlag:
-            if sync_state in ('async', 'potential'):
                 candidates.append(app_name)
+                if sync_state == 'sync':
+                    sync_nodes.append(app_name)
+            if len(candidates) >= sync_node_count:
+                break
 
-        if candidates:
+        return candidates, sync_nodes
-            return candidates[0], False
-        return None, False
-
-    def read_postmaster_opts(self):
-        """returns the list of option names/values from postgres.opts, Empty dict if read failed or no file"""
-        result = {}
-        try:
-            with open(os.path.join(self._data_dir, 'postmaster.opts')) as f:
-                data = f.read()
-                for opt in data.split('" "'):
-                    if '=' in opt and opt.startswith('--'):
-                        name, val = opt.split('=', 1)
-                        result[name.strip('-')] = val.rstrip('"\n')
-        except IOError:
-            logger.exception('Error when reading postmaster.opts')
-        return result
-
-    def single_user_mode(self, command=None, options=None):
-        """run a given command in a single-user mode. If the command is empty - then just start and stop"""
-        cmd = [self.pgcommand('postgres'), '--single', '-D', self._data_dir]
-        for opt, val in sorted((options or {}).items()):
-            cmd.extend(['-c', '{0}={1}'.format(opt, val)])
-        # need a database name to connect
-        cmd.append(self._database)
-        return self.cancellable.call(cmd, communicate_input=command)
-
-    def cleanup_archive_status(self):
-        status_dir = os.path.join(self._data_dir, 'pg_' + self.wal_name, 'archive_status')
-        try:
-            for f in os.listdir(status_dir):
-                path = os.path.join(status_dir, f)
-                try:
-                    if os.path.islink(path):
-                        os.unlink(path)
-                    elif os.path.isfile(path):
-                        os.remove(path)
-                except OSError:
-                    logger.exception('Unable to remove %s', path)
-        except OSError:
-            logger.exception('Unable to list %s', status_dir)
-
-    def fix_cluster_state(self):
-        self.cleanup_archive_status()
-
-        # Start in a single user mode and stop to produce a clean shutdown
-        opts = self.read_postmaster_opts()
-        opts.update({'archive_mode': 'on', 'archive_command': 'false'})
-        self.config.remove_recovery_conf()
-        return self.single_user_mode(options=opts) == 0 or None
 
     def schedule_sanity_checks_after_pause(self):
         """
             After coming out of pause we have to:
-            1. sync replication slots, because it might happen that slots were removed
+            1. configure server parameters if necessary
-            2. get new 'Database system identifier' to make sure that it wasn't changed
+            2. sync replication slots, because it might happen that slots were removed
+            3. get new 'Database system identifier' to make sure that it wasn't changed
         """
+        if not self._major_version:
+            self.configure_server_parameters()
         self.slots_handler.schedule()
         self._sysid = None

patroni/postgresql/bootstrap.py

@@ -5,9 +5,8 @@ import tempfile
 import time
 
 from patroni.dcs import RemoteMember
-from patroni.utils import deep_compare, uri
+from patroni.utils import deep_compare
 from six import string_types
-from six.moves.urllib.parse import quote_plus
 
 logger = logging.getLogger(__name__)
 
@@ -54,7 +53,7 @@ class Bootstrap(object):
                     error_handler('Error when parsing {0} option {1}: value should be string value'
                                   ' or a single key-value pair'.format(tool, opt))
         else:
-            error_handler('{0} options must be list ot dict'.format(tool))
+            error_handler('{0} options must be list or dict'.format(tool))
         return user_options
 
     def _initdb(self, config):
@@ -91,15 +90,16 @@ class Bootstrap(object):
         self._postgresql.configure_server_parameters()
 
         # make sure there is no trigger file or postgres will be automatically promoted
-        trigger_file = 'promote_trigger_file' if self._postgresql.major_version >= 120000 else 'trigger_file'
+        trigger_file = self._postgresql.config.triggerfile_good_name
-        trigger_file = self._postgresql.config.get('recovery_conf', {}).get(trigger_file) or 'promote'
+        trigger_file = (self._postgresql.config.get('recovery_conf') or {}).get(trigger_file) or 'promote'
         trigger_file = os.path.abspath(os.path.join(self._postgresql.data_dir, trigger_file))
         if os.path.exists(trigger_file):
             os.unlink(trigger_file)
 
     def _custom_bootstrap(self, config):
         self._postgresql.set_state('running custom bootstrap script')
-        params = ['--scope=' + self._postgresql.scope, '--datadir=' + self._postgresql.data_dir]
+        params = [] if config.get('no_params') else ['--scope=' + self._postgresql.scope,
+                                                     '--datadir=' + self._postgresql.data_dir]
         try:
             logger.info('Running custom bootstrap script: %s', config['command'])
             if self._postgresql.cancellable.call(shlex.split(config['command']) + params) != 0:
@@ -123,28 +123,15 @@ class Bootstrap(object):
         cmd = config.get('post_bootstrap') or config.get('post_init')
         if cmd:
             r = self._postgresql.config.local_connect_kwargs
-
+            connstring = self._postgresql.config.format_dsn(r, True)
-            if 'host' in r:
+            if 'host' not in r:
-                # '/tmp' => '%2Ftmp' for unix socket path
-                host = quote_plus(r['host']) if r['host'].startswith('/') else r['host']
-            else:
-                host = ''
-
                 # https://www.postgresql.org/docs/current/static/libpq-pgpass.html
                 # A host name of localhost matches both TCP (host name localhost) and Unix domain socket
                 # (pghost empty or the default socket directory) connections coming from the local machine.
                 r['host'] = 'localhost'  # set it to localhost to write into pgpass
 
-            if 'user' in r:
+            env = self._postgresql.config.write_pgpass(r)
-                user = r['user']
+            env['PGOPTIONS'] = '-c synchronous_commit=local'
-            else:
-                user = ''
-                if 'password' in r:
-                    import getpass
-                    r.setdefault('user', os.environ.get('PGUSER', getpass.getuser()))
-
-            connstring = uri('postgres', (host, r['port']), r['database'], user)
-            env = self._postgresql.write_pgpass(r) if 'password' in r else None
 
             try:
                 ret = self._postgresql.cancellable.call(shlex.split(cmd) + [connstring], env=env)
@@ -176,9 +163,9 @@ class Bootstrap(object):
 
         if clone_member and clone_member.conn_url:
             r = clone_member.conn_kwargs(self._postgresql.config.replication)
-            connstring = uri('postgres', (r['host'], r['port']), r['database'], r['user'])
             # add the credentials to connect to the replica origin to pgpass.
-            env = self._postgresql.write_pgpass(r)
+            env = self._postgresql.config.write_pgpass(r)
+            connstring = self._postgresql.config.format_dsn(r, True)
         else:
             connstring = ''
             env = os.environ.copy()
@@ -325,7 +312,15 @@ BEGIN
         CREATE ROLE "{0}" WITH {1};
     END IF;
 END;$$""".format(name, ' '.join(options))
-        self._postgresql.query(sql, *params)
+        self._postgresql.query('SET log_statement TO none')
+        self._postgresql.query('SET log_min_duration_statement TO -1')
+        self._postgresql.query("SET log_min_error_statement TO 'log'")
+        try:
+            self._postgresql.query(sql, *params)
+        finally:
+            self._postgresql.query('RESET log_min_error_statement')
+            self._postgresql.query('RESET log_min_duration_statement')
+            self._postgresql.query('RESET log_statement')
 
     def post_bootstrap(self, config, task):
         try:
@@ -361,7 +356,8 @@ END;$$""".format(f, rewind['username'])
                     self._running_custom_bootstrap = False
                     # If we don't have custom configuration for pg_hba.conf we need to restore original file
                     if not postgresql.config.get('pg_hba'):
-                        os.unlink(postgresql.config.pg_hba_conf)
+                        if os.path.exists(postgresql.config.pg_hba_conf):
+                            os.unlink(postgresql.config.pg_hba_conf)
                         postgresql.config.restore_configuration_files()
                     postgresql.config.write_postgresql_conf()
                     postgresql.config.replace_pg_ident()
@@ -369,7 +365,7 @@ END;$$""".format(f, rewind['username'])
                     # at this point there should be no recovery.conf
                     postgresql.config.remove_recovery_conf()
 
-                    if postgresql.config.hba_file and postgresql.config.hba_file != postgresql.config.pg_hba_conf:
+                    if postgresql.config.hba_file:
                         postgresql.restart()
                     else:
                         postgresql.config.replace_pg_hba()

patroni/postgresql/callback_executor.py

@@ -1,40 +1,36 @@
 import logging
-import subprocess
+
-from threading import Event, Lock, Thread
+from patroni.postgresql.cancellable import CancellableExecutor
+from threading import Condition, Thread
 
 logger = logging.getLogger(__name__)
 
 
-class CallbackExecutor(Thread):
+class CallbackExecutor(CancellableExecutor, Thread):
 
     def __init__(self):
-        super(CallbackExecutor, self).__init__()
+        CancellableExecutor.__init__(self)
+        Thread.__init__(self)
         self.daemon = True
-        self._lock = Lock()
         self._cmd = None
-        self._process = None
+        self._condition = Condition()
-        self._callback_event = Event()
         self.start()
 
     def call(self, cmd):
-        with self._lock:
+        self._kill_process()
-            if self._process and self._process.poll() is None:
+        with self._condition:
-                try:
+            self._cmd = cmd
-                    self._process.kill()
+            self._condition.notify()
-                    logger.warning('Killed the old callback process because it was still running: %s', self._cmd)
-                except OSError:
-                    logger.exception('Failed to kill the old callback')
-        self._cmd = cmd
-        self._callback_event.set()
 
     def run(self):
         while True:
-            self._callback_event.wait()
+            with self._condition:
-            self._callback_event.clear()
+                if self._cmd is None:
+                    self._condition.wait()
+                cmd, self._cmd = self._cmd, None
+
             with self._lock:
-                try:
+                if not self._start_process(cmd, close_fds=True):
-                    self._process = subprocess.Popen(self._cmd, close_fds=True)
-                except Exception:
-                    logger.exception('Failed to execute %s',  self._cmd)
                     continue
             self._process.wait()
+            self._kill_children()

patroni/postgresql/cancellable.py

@@ -1,36 +1,92 @@
 import logging
-import os
+import psutil
 import subprocess
 
 from patroni.exceptions import PostgresException
 from patroni.utils import polling_loop
-from six import string_types
 from threading import Lock
 
 logger = logging.getLogger(__name__)
 
 
-class CancellableSubprocess(object):
+class CancellableExecutor(object):
+
+    """
+    There must be only one such process so that AsyncExecutor can easily cancel it.
+    """
 
     def __init__(self):
-        self._is_cancelled = False
         self._process = None
+        self._process_cmd = None
+        self._process_children = []
         self._lock = Lock()
 
+    def _start_process(self, cmd, *args, **kwargs):
+        """This method must be executed only when the `_lock` is acquired"""
+
+        try:
+            self._process_children = []
+            self._process_cmd = cmd
+            self._process = psutil.Popen(cmd, *args, **kwargs)
+        except Exception:
+            return logger.exception('Failed to execute %s',  cmd)
+        return True
+
+    def _kill_process(self):
+        with self._lock:
+            if self._process is not None and self._process.is_running() and not self._process_children:
+                try:
+                    self._process.suspend()  # Suspend the process before getting list of children
+                except psutil.Error as e:
+                    logger.info('Failed to suspend the process: %s', e.msg)
+
+                try:
+                    self._process_children = self._process.children(recursive=True)
+                except psutil.Error:
+                    pass
+
+                try:
+                    self._process.kill()
+                    logger.warning('Killed %s because it was still running', self._process_cmd)
+                except psutil.NoSuchProcess:
+                    pass
+                except psutil.AccessDenied as e:
+                    logger.warning('Failed to kill the process: %s', e.msg)
+
+    def _kill_children(self):
+        waitlist = []
+        with self._lock:
+            for child in self._process_children:
+                try:
+                    child.kill()
+                except psutil.NoSuchProcess:
+                    continue
+                except psutil.AccessDenied as e:
+                    logger.info('Failed to kill child process: %s', e.msg)
+                waitlist.append(child)
+        psutil.wait_procs(waitlist)
+
+
+class CancellableSubprocess(CancellableExecutor):
+
+    def __init__(self):
+        super(CancellableSubprocess, self).__init__()
+        self._is_cancelled = False
+
     def call(self, *args, **kwargs):
         for s in ('stdin', 'stdout', 'stderr'):
             kwargs.pop(s, None)
 
-        communicate_input = 'communicate_input' in kwargs
+        communicate = kwargs.pop('communicate', None)
-        if communicate_input:
+        if isinstance(communicate, dict):
-            input_data = kwargs.pop('communicate_input', None)
+            input_data = communicate.get('input')
-            if not isinstance(input_data, string_types):
+            if input_data:
-                input_data = ''
+                if input_data[-1] != '\n':
-            if input_data and input_data[-1] != '\n':
+                    input_data += '\n'
-                input_data += '\n'
+                input_data = input_data.encode('utf-8')
             kwargs['stdin'] = subprocess.PIPE
-            kwargs['stdout'] = open(os.devnull, 'w')
+            kwargs['stdout'] = subprocess.PIPE
-            kwargs['stderr'] = subprocess.STDOUT
+            kwargs['stderr'] = subprocess.PIPE
 
         try:
             with self._lock:
@@ -38,17 +94,16 @@ class CancellableSubprocess(object):
                     raise PostgresException('cancelled')
 
                 self._is_cancelled = False
-                self._process = subprocess.Popen(*args, **kwargs)
+                started = self._start_process(*args, **kwargs)
 
-            if communicate_input:
+            if started:
-                if input_data:
+                if isinstance(communicate, dict):
-                    self._process.communicate(input_data)
+                    communicate['stdout'], communicate['stderr'] = self._process.communicate(input_data)
-                self._process.stdin.close()
+                return self._process.wait()
-
-            return self._process.wait()
         finally:
             with self._lock:
                 self._process = None
+            self._kill_children()
 
     def reset_is_cancelled(self):
         with self._lock:
@@ -59,18 +114,20 @@ class CancellableSubprocess(object):
         with self._lock:
             return self._is_cancelled
 
-    def cancel(self):
+    def cancel(self, kill=False):
         with self._lock:
             self._is_cancelled = True
-            if self._process is None or self._process.returncode is not None:
+            if self._process is None or not self._process.is_running():
                 return
+
+            logger.info('Terminating %s', self._process_cmd)
             self._process.terminate()
 
         for _ in polling_loop(10):
             with self._lock:
-                if self._process is None or self._process.returncode is not None:
+                if self._process is None or not self._process.is_running():
                     return
+            if kill:
+                break
 
-        with self._lock:
+        self._kill_process()
-            if self._process is not None and self._process.returncode is None:
-                self._process.kill()

patroni/postgresql/connection.py

@@ -40,7 +40,8 @@ class Connection(object):
 
 @contextmanager
 def get_connection_cursor(**kwargs):
-    with psycopg2.connect(**kwargs) as conn:
+    conn = psycopg2.connect(**kwargs)
-        conn.autocommit = True
+    conn.autocommit = True
-        with conn.cursor() as cur:
+    with conn.cursor() as cur:
-            yield cur
+        yield cur
+    conn.close()

patroni/postgresql/misc.py

@@ -37,7 +37,7 @@ def postgres_version_to_int(pg_version):
         raise PostgresException('Invalid PostgreSQL version format: X.Y or X.Y.Z is accepted: {0}'.format(pg_version))
 
     if len(components) == 2:
-        # new style verion numbers, i.e. 10.1 becomes 100001
+        # new style version numbers, i.e. 10.1 becomes 100001
         components.insert(1, 0)
 
     return int(''.join('{0:02d}'.format(c) for c in components))
@@ -68,3 +68,8 @@ def parse_history(data):
                 yield values
             except (IndexError, ValueError):
                 logger.exception('Exception when parsing timeline history line "%s"', values)
+
+
+def format_lsn(lsn, full=False):
+    template = '{0:X}/{1:08X}' if full else '{0:X}/{1:X}'
+    return template.format(lsn >> 32, lsn & 0xFFFFFFFF)

patroni/postgresql/postmaster.py

@@ -5,13 +5,24 @@ import psutil
 import re
 import signal
 import subprocess
+import sys
+
+from patroni import PATRONI_ENV_PREFIX, KUBERNETES_ENV_PREFIX
+
+# avoid spawning the resource tracker process
+if sys.version_info >= (3, 8):  # pragma: no cover
+    import multiprocessing.resource_tracker
+    multiprocessing.resource_tracker.getfd = lambda: 0
+elif sys.version_info >= (3, 4):  # pragma: no cover
+    import multiprocessing.semaphore_tracker
+    multiprocessing.semaphore_tracker.getfd = lambda: 0
 
 logger = logging.getLogger(__name__)
 
 STOP_SIGNALS = {
-    'smart': signal.SIGTERM,
+    'smart': 'TERM',
-    'fast': signal.SIGINT,
+    'fast': 'INT',
-    'immediate': signal.SIGQUIT if os.name != 'nt' else signal.SIGABRT,
+    'immediate': 'QUIT',
 }
 
 
@@ -94,7 +105,43 @@ class PostmasterProcess(psutil.Process):
         except psutil.NoSuchProcess:
             return None
 
-    def signal_stop(self, mode):
+    def signal_kill(self):
+        """to suspend and kill postmaster and all children
+
+        :returns True if postmaster and children are killed, False if error
+        """
+        try:
+            self.suspend()
+        except psutil.NoSuchProcess:
+            return True
+        except psutil.Error as e:
+            logger.warning('Failed to suspend postmaster: %s', e)
+
+        try:
+            children = self.children(recursive=True)
+        except psutil.NoSuchProcess:
+            return True
+        except psutil.Error as e:
+            logger.warning('Failed to get a list of postmaster children: %s', e)
+            children = []
+
+        try:
+            self.kill()
+        except psutil.NoSuchProcess:
+            return True
+        except psutil.Error as e:
+            logger.warning('Could not kill postmaster: %s', e)
+            return False
+
+        for child in children:
+            try:
+                child.kill()
+            except psutil.Error:
+                pass
+        psutil.wait_procs(children + [self])
+        return True
+
+    def signal_stop(self, mode, pg_ctl='pg_ctl'):
         """Signal postmaster process to stop
 
         :returns None if signaled, True if process is already gone, False if error
@@ -102,8 +149,10 @@ class PostmasterProcess(psutil.Process):
         if self.is_single_user:
             logger.warning("Cannot stop server; single-user server is running (PID: {0})".format(self.pid))
             return False
+        if os.name != 'posix':
+            return self.pg_ctl_kill(mode, pg_ctl)
         try:
-            self.send_signal(STOP_SIGNALS[mode])
+            self.send_signal(getattr(signal, 'SIG' + STOP_SIGNALS[mode]))
         except psutil.NoSuchProcess:
             return True
         except psutil.AccessDenied as e:
@@ -112,6 +161,16 @@ class PostmasterProcess(psutil.Process):
 
         return None
 
+    def pg_ctl_kill(self, mode, pg_ctl):
+        try:
+            status = subprocess.call([pg_ctl, "kill", STOP_SIGNALS[mode], str(self.pid)])
+        except OSError:
+            return False
+        if status == 0:
+            return None
+        else:
+            return not self.is_running()
+
     def wait_for_user_backends_to_close(self):
         # These regexps are cross checked against versions PostgreSQL 9.1 .. 11
         aux_proc_re = re.compile("(?:postgres:)( .*:)? (?:(?:archiver|startup|autovacuum launcher|autovacuum worker|"
@@ -128,10 +187,10 @@ class PostmasterProcess(psutil.Process):
         user_backends_cmdlines = []
         for child in children:
             try:
-                cmdline = child.cmdline()[0]
+                cmdline = child.cmdline()
-                if not aux_proc_re.match(cmdline):
+                if cmdline and not aux_proc_re.match(cmdline[0]):
                     user_backends.append(child)
-                    user_backends_cmdlines.append(cmdline)
+                    user_backends_cmdlines.append(cmdline[0])
             except psutil.NoSuchProcess:
                 pass
         if user_backends:
@@ -153,7 +212,8 @@ class PostmasterProcess(psutil.Process):
         # In order to make everything portable we can't use fork&exec approach here, so  we will call
         # ourselves and pass list of arguments which must be used to start postgres.
         # On Windows, in order to run a side-by-side assembly the specified env must include a valid SYSTEMROOT.
-        env = {p: os.environ[p] for p in ('PATH', 'LD_LIBRARY_PATH', 'LC_ALL', 'LANG', 'SYSTEMROOT') if p in os.environ}
+        env = {p: os.environ[p] for p in os.environ if not p.startswith(
+            PATRONI_ENV_PREFIX) and not p.startswith(KUBERNETES_ENV_PREFIX)}
         try:
             proc = PostmasterProcess._from_pidfile(data_dir)
             if proc and not proc._is_postmaster_process():
@@ -170,8 +230,9 @@ class PostmasterProcess(psutil.Process):
             pass
         cmdline = [pgcommand, '-D', data_dir, '--config-file={}'.format(conf)] + options
         logger.debug("Starting postgres: %s", " ".join(cmdline))
-        parent_conn, child_conn = multiprocessing.Pipe(False)
+        ctx = multiprocessing.get_context('spawn') if sys.version_info >= (3, 4) else multiprocessing
-        proc = multiprocessing.Process(target=pg_ctl_start, args=(child_conn, cmdline, env))
+        parent_conn, child_conn = ctx.Pipe(False)
+        proc = ctx.Process(target=pg_ctl_start, args=(child_conn, cmdline, env))
         proc.start()
         pid = parent_conn.recv()
         proc.join()

patroni/postgresql/rewind.py

@@ -1,10 +1,15 @@
 import logging
 import os
+import shlex
+import six
 import subprocess
 
-from patroni.dcs import Leader
+from threading import Lock, Thread
-from patroni.postgresql.connection import get_connection_cursor
+
-from patroni.postgresql.misc import parse_history, parse_lsn
+from .connection import get_connection_cursor
+from .misc import format_lsn, parse_history, parse_lsn
+from ..async_executor import CriticalTask
+from ..dcs import Leader
 
 logger = logging.getLogger(__name__)
 
@@ -16,6 +21,7 @@ class Rewind(object):
 
     def __init__(self, postgresql):
         self._postgresql = postgresql
+        self._checkpoint_task_lock = Lock()
         self.reset_state()
 
     @staticmethod
@@ -48,11 +54,10 @@ class Rewind(object):
         if self.can_rewind_or_reinitialize_allowed and self._state != REWIND_STATUS.NEED:
             self._state = REWIND_STATUS.CHECK
 
-    def check_leader_is_not_in_recovery(self, **kwargs):
+    @staticmethod
-        if not kwargs.get('database'):
+    def check_leader_is_not_in_recovery(conn_kwargs):
-            kwargs['database'] = self._postgresql.database
         try:
-            with get_connection_cursor(connect_timeout=3, options='-c statement_timeout=2000', **kwargs) as cur:
+            with get_connection_cursor(connect_timeout=3, options='-c statement_timeout=2000', **conn_kwargs) as cur:
                 cur.execute('SELECT pg_catalog.pg_is_in_recovery()')
                 if not cur.fetchone()[0]:
                     return True
@@ -60,32 +65,99 @@ class Rewind(object):
         except Exception:
             return logger.exception('Exception when working with leader')
 
+    def _get_checkpoint_end(self, timeline, lsn):
+        """The checkpoint record size in WAL depends on postgres major version and platform (memory alignment).
+        Hence, the only reliable way to figure out where it ends, read the record from file with the help of pg_waldump
+        and parse the output. We are trying to read two records, and expect that it will fail to read the second one:
+        `pg_waldump: fatal: error in WAL record at 0/182E220: invalid record length at 0/182E298: wanted 24, got 0`
+        The error message contains information about LSN of the next record, which is exactly where checkpoint ends."""
+
+        lsn8 = format_lsn(lsn, True)
+        lsn = format_lsn(lsn)
+        out, err = self._postgresql.waldump(timeline, lsn, 2)
+        if out is not None and err is not None:
+            out = out.decode('utf-8').rstrip().split('\n')
+            err = err.decode('utf-8').rstrip().split('\n')
+            pattern = 'error in WAL record at {0}: invalid record length at '.format(lsn)
+
+            if len(out) == 1 and len(err) == 1 and ', lsn: {0}, prev '.format(lsn8) in out[0] and pattern in err[0]:
+                i = err[0].find(pattern) + len(pattern)
+                j = err[0].find(": wanted ", i)
+                if j > -1:
+                    try:
+                        return parse_lsn(err[0][i:j])
+                    except Exception as e:
+                        logger.error('Failed to parse lsn %s: %r', err[0][i:j], e)
+            logger.error('Failed to parse pg_%sdump output', self._postgresql.wal_name)
+            logger.error(' stdout=%s', '\n'.join(out))
+            logger.error(' stderr=%s', '\n'.join(err))
+
+        return 0
+
     def _get_local_timeline_lsn_from_controldata(self):
-        timeline = lsn = None
+        in_recovery = timeline = lsn = None
         data = self._postgresql.controldata()
         try:
             if data.get('Database cluster state') == 'shut down in recovery':
+                in_recovery = True
                 lsn = data.get('Minimum recovery ending location')
                 timeline = int(data.get("Min recovery ending loc's timeline"))
                 if lsn == '0/0' or timeline == 0:  # it was a master when it crashed
                     data['Database cluster state'] = 'shut down'
             if data.get('Database cluster state') == 'shut down':
+                in_recovery = False
                 lsn = data.get('Latest checkpoint location')
                 timeline = int(data.get("Latest checkpoint's TimeLineID"))
         except (TypeError, ValueError):
             logger.exception('Failed to get local timeline and lsn from pg_controldata output')
-        return timeline, lsn
+
+        if lsn is not None:
+            try:
+                lsn = parse_lsn(lsn)
+            except (IndexError, ValueError) as e:
+                logger.error('Exception when parsing lsn %s: %r', lsn, e)
+                lsn = None
+
+        return in_recovery, timeline, lsn
 
     def _get_local_timeline_lsn(self):
-        if self._postgresql.is_running():  # if postgres is running - get timeline and lsn from replication connection
+        if self._postgresql.is_running():  # if postgres is running - get timeline from replication connection
-            timeline, lsn = self._postgresql.get_local_timeline_lsn_from_replication_connection()
+            in_recovery = True
+            timeline = self._postgresql.received_timeline() or self._postgresql.get_replica_timeline()
+            lsn = self._postgresql.replayed_location()
         else:  # otherwise analyze pg_controldata output
-            timeline, lsn = self._get_local_timeline_lsn_from_controldata()
+            in_recovery, timeline, lsn = self._get_local_timeline_lsn_from_controldata()
-        logger.info('Local timeline=%s lsn=%s', timeline, lsn)
+
-        return timeline, lsn
+        log_lsn = format_lsn(lsn) if isinstance(lsn, six.integer_types) else lsn
+        logger.info('Local timeline=%s lsn=%s', timeline, log_lsn)
+        return in_recovery, timeline, lsn
+
+    @staticmethod
+    def _log_master_history(history, i):
+        start = max(0, i - 3)
+        end = None if i + 4 >= len(history) else i + 2
+        history_show = []
+
+        def format_history_line(line):
+            return '{0}\t{1}\t{2}'.format(line[0], format_lsn(line[1]), line[2])
+
+        for line in history[start:end]:
+            history_show.append(format_history_line(line))
+
+        if line != history[-1]:
+            history_show.append('...')
+            history_show.append(format_history_line(history[-1]))
+
+        logger.info('master: history=%s', '\n'.join(history_show))
+
+    def _conn_kwargs(self, member, auth):
+        ret = member.conn_kwargs(auth)
+        if not ret.get('database'):
+            ret['database'] = self._postgresql.database
+        return ret
 
     def _check_timeline_and_lsn(self, leader):
-        local_timeline, local_lsn = self._get_local_timeline_lsn()
+        in_recovery, local_timeline, local_lsn = self._get_local_timeline_lsn()
         if local_timeline is None or local_lsn is None:
             return
 
@@ -93,7 +165,7 @@ class Rewind(object):
             if leader.member.data.get('role') != 'master':
                 return
         # standby cluster
-        elif not self.check_leader_is_not_in_recovery(**leader.conn_kwargs(self._postgresql.config.replication)):
+        elif not self.check_leader_is_not_in_recovery(self._conn_kwargs(leader, self._postgresql.config.replication)):
             return
 
         history = need_rewind = None
@@ -104,25 +176,34 @@ class Rewind(object):
                 logger.info('master_timeline=%s', master_timeline)
                 if local_timeline > master_timeline:  # Not always supported by pg_rewind
                     need_rewind = True
+                elif local_timeline == master_timeline:
+                    need_rewind = False
                 elif master_timeline > 1:
                     cur.execute('TIMELINE_HISTORY %s', (master_timeline,))
-                    history = bytes(cur.fetchone()[1]).decode('utf-8')
+                    history = cur.fetchone()[1]
-                    logger.info('master: history=%s', history)
+                    if not isinstance(history, six.string_types):
-                else:  # local_timeline == master_timeline == 1
+                        history = bytes(history).decode('utf-8')
-                    need_rewind = False
+                    logger.debug('master: history=%s', history)
         except Exception:
             return logger.exception('Exception when working with master via replication connection')
 
         if history is not None:
-            for parent_timeline, switchpoint, _ in parse_history(history):
+            history = list(parse_history(history))
+            for i, (parent_timeline, switchpoint, _) in enumerate(history):
                 if parent_timeline == local_timeline:
-                    try:
+                    # We don't need to rewind when:
-                        need_rewind = parse_lsn(local_lsn) >= switchpoint
+                    # 1. for replica: replayed location is not ahead of switchpoint
-                    except (IndexError, ValueError):
+                    # 2. for the former primary: end of checkpoint record is the same as switchpoint
-                        logger.exception('Exception when parsing lsn')
+                    if in_recovery:
+                        need_rewind = local_lsn > switchpoint
+                    elif local_lsn >= switchpoint:
+                        need_rewind = True
+                    else:
+                        need_rewind = switchpoint != self._get_checkpoint_end(local_timeline, local_lsn)
                     break
                 elif parent_timeline > local_timeline:
                     break
+            self._log_master_history(history, i)
 
         self._state = need_rewind and REWIND_STATUS.NEED or REWIND_STATUS.NOT_NEED
 
@@ -131,44 +212,121 @@ class Rewind(object):
             self._check_timeline_and_lsn(leader)
         return leader and leader.conn_url and self._state == REWIND_STATUS.NEED
 
-    def check_for_checkpoint_after_promote(self):
+    def __checkpoint(self, task, wakeup):
+        try:
+            result = self._postgresql.checkpoint()
+        except Exception as e:
+            result = 'Exception: ' + str(e)
+        with task:
+            task.complete(not bool(result))
+            if task.result:
+                wakeup()
+
+    def ensure_checkpoint_after_promote(self, wakeup):
+        """After promote issue a CHECKPOINT from a new thread and asynchronously check the result.
+        In case if CHECKPOINT failed, just check that timeline in pg_control was updated."""
+
         if self._state == REWIND_STATUS.INITIAL and self._postgresql.is_leader():
-            try:
+            with self._checkpoint_task_lock:
-                timeline = int(self._postgresql.controldata().get("Latest checkpoint's TimeLineID"))
+                if self._checkpoint_task:
-                if self._postgresql.get_master_timeline() == timeline:
+                    with self._checkpoint_task:
-                    self._state = REWIND_STATUS.CHECKPOINT
+                        if self._checkpoint_task.result:
-            except (TypeError, ValueError):
+                            self._state = REWIND_STATUS.CHECKPOINT
-                logger.exception('Failed to parse timeline from pg_controldata output')
+                        if self._checkpoint_task.result is not False:
+                            return
+                else:
+                    self._checkpoint_task = CriticalTask()
+                    return Thread(target=self.__checkpoint, args=(self._checkpoint_task, wakeup)).start()
+
+            if self._postgresql.get_master_timeline() == self._postgresql.pg_control_timeline():
+                self._state = REWIND_STATUS.CHECKPOINT
 
     def checkpoint_after_promote(self):
         return self._state == REWIND_STATUS.CHECKPOINT
 
+    def _fetch_missing_wal(self, restore_command, wal_filename):
+        cmd = ''
+        length = len(restore_command)
+        i = 0
+        while i < length:
+            if restore_command[i] == '%' and i + 1 < length:
+                i += 1
+                if restore_command[i] == 'p':
+                    cmd += os.path.join(self._postgresql.wal_dir, wal_filename)
+                elif restore_command[i] == 'f':
+                    cmd += wal_filename
+                elif restore_command[i] == 'r':
+                    cmd += '000000010000000000000001'
+                elif restore_command[i] == '%':
+                    cmd += '%'
+                else:
+                    cmd += '%'
+                    i -= 1
+            else:
+                cmd += restore_command[i]
+            i += 1
+
+        logger.info('Trying to fetch the missing wal: %s', cmd)
+        return self._postgresql.cancellable.call(shlex.split(cmd)) == 0
+
+    def _find_missing_wal(self, data):
+        # could not open file "$PGDATA/pg_wal/0000000A00006AA100000068": No such file or directory
+        pattern = 'could not open file "'
+        for line in data.decode('utf-8').split('\n'):
+            b = line.find(pattern)
+            if b > -1:
+                b += len(pattern)
+                e = line.find('": ', b)
+                if e > -1 and '/' in line[b:e]:
+                    waldir, wal_filename = line[b:e].rsplit('/', 1)
+                    if waldir.endswith('/pg_' + self._postgresql.wal_name) and len(wal_filename) == 24:
+                        return wal_filename
+
     def pg_rewind(self, r):
         # prepare pg_rewind connection
-        env = self._postgresql.write_pgpass(r)
+        env = self._postgresql.config.write_pgpass(r)
-        env['PGOPTIONS'] = '-c statement_timeout=0'
+        env.update(LANG='C', LC_ALL='C', PGOPTIONS='-c statement_timeout=0')
-        dsn_attrs = [
+        dsn = self._postgresql.config.format_dsn(r, True)
-            ('user', r.get('user')),
-            ('host', r.get('host')),
-            ('port', r.get('port')),
-            ('dbname', r.get('database') or self._postgresql.database),
-            ('sslmode', 'prefer'),
-            ('sslcompression', '1'),
-        ]
-        dsn = " ".join("{0}={1}".format(k, v) for k, v in dsn_attrs if v is not None)
         logger.info('running pg_rewind from %s', dsn)
-        try:
+
-            return self._postgresql.cancellable.call([self._postgresql.pgcommand('pg_rewind'), '-D',
+        restore_command = self._postgresql.config.get('recovery_conf', {}).get('restore_command') \
-                                                      self._postgresql.data_dir, '--source-server', dsn], env=env) == 0
+            if self._postgresql.major_version < 120000 else self._postgresql.get_guc_value('restore_command')
-        except OSError:
+
-            return False
+        cmd = [self._postgresql.pgcommand('pg_rewind')]
+        if self._postgresql.major_version >= 130000 and restore_command:
+            cmd.append('--restore-target-wal')
+        cmd.extend(['-D', self._postgresql.data_dir, '--source-server', dsn])
+
+        while True:
+            results = {}
+            ret = self._postgresql.cancellable.call(cmd, env=env, communicate=results)
+
+            logger.info('pg_rewind exit code=%s', ret)
+            if ret is None:
+                return False
+
+            logger.info(' stdout=%s', results['stdout'].decode('utf-8'))
+            logger.info(' stderr=%s', results['stderr'].decode('utf-8'))
+            if ret == 0:
+                return True
+
+            if not restore_command or self._postgresql.major_version >= 130000:
+                return False
+
+            missing_wal = self._find_missing_wal(results['stderr']) or self._find_missing_wal(results['stdout'])
+            if not missing_wal:
+                return False
+
+            if not self._fetch_missing_wal(restore_command, missing_wal):
+                logger.info('Failed to fetch WAL segment %s required for pg_rewind', missing_wal)
+                return False
 
     def execute(self, leader):
         if self._postgresql.is_running() and not self._postgresql.stop(checkpoint=False):
             return logger.warning('Can not run pg_rewind because postgres is still running')
 
         # prepare pg_rewind connection
-        r = leader.conn_kwargs(self._postgresql.config.rewind_credentials)
+        r = self._conn_kwargs(leader, self._postgresql.config.rewind_credentials)
 
         # 1. make sure that we are really trying to rewind from the master
         # 2. make sure that pg_control contains the new timeline by:
@@ -176,17 +334,17 @@ class Rewind(object):
         #   waiting until Patroni on the master will expose checkpoint_after_promote=True
         checkpoint_status = leader.checkpoint_after_promote if isinstance(leader, Leader) else None
         if checkpoint_status is None:  # master still runs the old Patroni
-            leader_status = self._postgresql.checkpoint(leader.conn_kwargs(self._postgresql.config.superuser))
+            leader_status = self._postgresql.checkpoint(self._conn_kwargs(leader, self._postgresql.config.superuser))
             if leader_status:
                 return logger.warning('Can not use %s for rewind: %s', leader.name, leader_status)
         elif not checkpoint_status:
             return logger.info('Waiting for checkpoint on %s before rewind', leader.name)
-        elif not self.check_leader_is_not_in_recovery(**r):
+        elif not self.check_leader_is_not_in_recovery(r):
             return
 
         if self.pg_rewind(r):
             self._state = REWIND_STATUS.SUCCESS
-        elif not self.check_leader_is_not_in_recovery(**r):
+        elif not self.check_leader_is_not_in_recovery(r):
             logger.warning('Failed to rewind because master %s become unreachable', leader.name)
         else:
             logger.error('Failed to rewind from healty master: %s', leader.name)
@@ -203,6 +361,8 @@ class Rewind(object):
 
     def reset_state(self):
         self._state = REWIND_STATUS.INITIAL
+        with self._checkpoint_task_lock:
+            self._checkpoint_task = None
 
     @property
     def is_needed(self):
@@ -215,3 +375,56 @@ class Rewind(object):
     @property
     def failed(self):
         return self._state == REWIND_STATUS.FAILED
+
+    def read_postmaster_opts(self):
+        """returns the list of option names/values from postgres.opts, Empty dict if read failed or no file"""
+        result = {}
+        try:
+            with open(os.path.join(self._postgresql.data_dir, 'postmaster.opts')) as f:
+                data = f.read()
+                for opt in data.split('" "'):
+                    if '=' in opt and opt.startswith('--'):
+                        name, val = opt.split('=', 1)
+                        result[name.strip('-')] = val.rstrip('"\n')
+        except IOError:
+            logger.exception('Error when reading postmaster.opts')
+        return result
+
+    def single_user_mode(self, communicate=None, options=None):
+        """run a given command in a single-user mode. If the command is empty - then just start and stop"""
+        cmd = [self._postgresql.pgcommand('postgres'), '--single', '-D', self._postgresql.data_dir]
+        for opt, val in sorted((options or {}).items()):
+            cmd.extend(['-c', '{0}={1}'.format(opt, val)])
+        # need a database name to connect
+        cmd.append('template1')
+        return self._postgresql.cancellable.call(cmd, communicate=communicate)
+
+    def cleanup_archive_status(self):
+        status_dir = os.path.join(self._postgresql.wal_dir, 'archive_status')
+        try:
+            for f in os.listdir(status_dir):
+                path = os.path.join(status_dir, f)
+                try:
+                    if os.path.islink(path):
+                        os.unlink(path)
+                    elif os.path.isfile(path):
+                        os.remove(path)
+                except OSError:
+                    logger.exception('Unable to remove %s', path)
+        except OSError:
+            logger.exception('Unable to list %s', status_dir)
+
+    def ensure_clean_shutdown(self):
+        self.cleanup_archive_status()
+
+        # Start in a single user mode and stop to produce a clean shutdown
+        opts = self.read_postmaster_opts()
+        opts.update({'archive_mode': 'on', 'archive_command': 'false'})
+        self._postgresql.config.remove_recovery_conf()
+        output = {}
+        ret = self.single_user_mode(communicate=output, options=opts)
+        if ret != 0:
+            logger.error('Crash recovery finished with code=%s', ret)
+            logger.info(' stdout=%s', output['stdout'].decode('utf-8'))
+            logger.info(' stderr=%s', output['stderr'].decode('utf-8'))
+        return ret == 0 or None

patroni/postgresql/slots.py

@@ -1,42 +1,109 @@
+import errno
 import logging
+import os
+import shutil
 
-from patroni.postgresql.connection import get_connection_cursor
 from collections import defaultdict
+from contextlib import contextmanager
+from psycopg2.errors import UndefinedFile
+
+from .connection import get_connection_cursor
+from .misc import format_lsn
 
 logger = logging.getLogger(__name__)
 
 
-def compare_slots(s1, s2):
+def compare_slots(s1, s2, dbid='database'):
     return s1['type'] == s2['type'] and (s1['type'] == 'physical' or
-                                         s1['database'] == s2['database'] and s1['plugin'] == s2['plugin'])
+                                         s1.get(dbid) == s2.get(dbid) and s1['plugin'] == s2['plugin'])
+
+
+def fsync_dir(path):
+    if os.name != 'nt':
+        fd = os.open(path, os.O_DIRECTORY)
+        try:
+            os.fsync(fd)
+        except OSError as e:
+            # Some filesystems don't like fsyncing directories and raise EINVAL. Ignoring it is usually safe.
+            if e.errno != errno.EINVAL:
+                raise
+        finally:
+            os.close(fd)
 
 
 class SlotsHandler(object):
 
     def __init__(self, postgresql):
         self._postgresql = postgresql
-        self._use_slots = postgresql.config.get('use_slots', True)
         self._replication_slots = {}  # already existing replication slots
+        self._unready_logical_slots = set()
         self.schedule()
 
-    @property
-    def use_slots(self):
-        return self._use_slots and self._postgresql.major_version >= 90400
-
     def _query(self, sql, *params):
         return self._postgresql.query(sql, *params, retry=False)
 
+    @staticmethod
+    def _copy_items(src, dst, keys=None):
+        dst.update({key: src[key] for key in keys or ('datoid', 'catalog_xmin', 'confirmed_flush_lsn')})
+
+    def process_permanent_slots(self, slots):
+        """This methods solves three problems at once (I know, it is weird).
+
+        The cluster_info_query from `Postgresql` is executed every HA loop and returns
+        information about all replication slots that exists on the current host.
+        Based on this information we perform the following actions:
+        1. For the primary we want to expose to DCS permanent logical slots, therefore the method
+           builds (and returns) a dict, that maps permanent logical slot names and confirmed_flush_lsns.
+        2. This method also detects if one of the previously known permanent slots got missing and schedules resync.
+        3. Updates the local cache with the fresh catalog_xmin and confirmed_flush_lsn for every known slot.
+           This info is used when performing the check of logical slot readiness on standbys.
+        """
+        ret = {}
+
+        slots = {slot['slot_name']: slot for slot in slots or []}
+        if slots:
+            for name, value in slots.items():
+                if name in self._replication_slots:
+                    if compare_slots(value, self._replication_slots[name], 'datoid'):
+                        if value['type'] == 'logical':
+                            ret[name] = value['confirmed_flush_lsn']
+                            self._copy_items(value, self._replication_slots[name])
+                    else:
+                        self._schedule_load_slots = True
+
+        # It could happen that the slots was deleted in the background, we want to detect this case
+        if any(name not in slots for name in self._replication_slots.keys()):
+            self._schedule_load_slots = True
+
+        return ret
+
     def load_replication_slots(self):
-        if self.use_slots and self._schedule_load_slots:
+        if self._postgresql.major_version >= 90400 and self._schedule_load_slots:
             replication_slots = {}
-            cursor = self._query('SELECT slot_name, slot_type, plugin, database FROM pg_catalog.pg_replication_slots')
+            extra = ", catalog_xmin, pg_catalog.pg_wal_lsn_diff(confirmed_flush_lsn, '0/0')::bigint"\
+                if self._postgresql.major_version >= 100000 else ""
+            cursor = self._query('SELECT slot_name, slot_type, plugin, database, datoid'
+                                 '{0} FROM pg_catalog.pg_replication_slots'.format(extra))
             for r in cursor:
                 value = {'type': r[1]}
                 if r[1] == 'logical':
-                    value.update({'plugin': r[2], 'database': r[3]})
+                    value.update(plugin=r[2], database=r[3], datoid=r[4])
+                    if self._postgresql.major_version >= 100000:
+                        value.update(catalog_xmin=r[5], confirmed_flush_lsn=r[6])
                 replication_slots[r[0]] = value
             self._replication_slots = replication_slots
             self._schedule_load_slots = False
+            if self._force_readiness_check:
+                self._unready_logical_slots = set(n for n, v in replication_slots.items() if v['type'] == 'logical')
+                self._force_readiness_check = False
+
+    def ignore_replication_slot(self, cluster, name):
+        slot = self._replication_slots[name]
+        for matcher in cluster.config.ignore_slots_matchers:
+            if ((matcher.get("name") is None or matcher["name"] == name)
+               and all(not matcher.get(a) or matcher[a] == slot.get(a) for a in ('database', 'plugin', 'type'))):
+                return True
+        return False
 
     def drop_replication_slot(self, name):
         cursor = self._query(('SELECT pg_catalog.pg_drop_replication_slot(%s) WHERE EXISTS (SELECT 1 ' +
@@ -44,65 +111,192 @@ class SlotsHandler(object):
         # In normal situation rowcount should be 1, otherwise either slot doesn't exists or it is still active
         return cursor.rowcount == 1
 
-    def sync_replication_slots(self, cluster):
+    def _drop_incorrect_slots(self, cluster, slots):
-        if self.use_slots:
+        # drop old replication slots which are not presented in desired slots
+        for name in set(self._replication_slots) - set(slots):
+            if not self.ignore_replication_slot(cluster, name) and not self.drop_replication_slot(name):
+                logger.error("Failed to drop replication slot '%s'", name)
+                self._schedule_load_slots = True
+
+        for name, value in slots.items():
+            if name in self._replication_slots and not compare_slots(value, self._replication_slots[name]):
+                logger.info("Trying to drop replication slot '%s' because value is changing from %s to %s",
+                            name, self._replication_slots[name], value)
+                if self.drop_replication_slot(name):
+                    self._replication_slots.pop(name)
+                else:
+                    logger.error("Failed to drop replication slot '%s'", name)
+                    self._schedule_load_slots = True
+
+    def _ensure_physical_slots(self, slots):
+        immediately_reserve = ', true' if self._postgresql.major_version >= 90600 else ''
+        for name, value in slots.items():
+            if name not in self._replication_slots and value['type'] == 'physical':
+                try:
+                    self._query(("SELECT pg_catalog.pg_create_physical_replication_slot(%s{0})" +
+                                 " WHERE NOT EXISTS (SELECT 1 FROM pg_catalog.pg_replication_slots" +
+                                 " WHERE slot_type = 'physical' AND slot_name = %s)").format(
+                                     immediately_reserve), name, name)
+                except Exception:
+                    logger.exception("Failed to create physical replication slot '%s'", name)
+                self._schedule_load_slots = True
+
+    @contextmanager
+    def _get_local_connection_cursor(self, database):
+        conn_kwargs = self._postgresql.config.local_connect_kwargs
+        conn_kwargs['database'] = database
+        with get_connection_cursor(**conn_kwargs) as cur:
+            yield cur
+
+    def _ensure_logical_slots_primary(self, slots):
+        # Group logical slots to be created by database name
+        logical_slots = defaultdict(dict)
+        for name, value in slots.items():
+            if value['type'] == 'logical':
+                # If the logical already exists, copy some information about it into the original structure
+                if self._replication_slots.get(name, {}).get('datoid'):
+                    self._copy_items(self._replication_slots[name], value)
+                else:
+                    logical_slots[value['database']][name] = value
+
+        # Create new logical slots
+        for database, values in logical_slots.items():
+            with self._get_local_connection_cursor(database) as cur:
+                for name, value in values.items():
+                    try:
+                        cur.execute("SELECT pg_catalog.pg_create_logical_replication_slot(%s, %s)" +
+                                    " WHERE NOT EXISTS (SELECT 1 FROM pg_catalog.pg_replication_slots" +
+                                    " WHERE slot_type = 'logical' AND slot_name = %s)",
+                                    (name, value['plugin'], name))
+                    except Exception as e:
+                        logger.error("Failed to create logical replication slot '%s' plugin='%s': %r",
+                                     name, value['plugin'], e)
+                        slots.pop(name)
+                    self._schedule_load_slots = True
+
+    def _ensure_logical_slots_replica(self, cluster, slots):
+        advance_slots = defaultdict(dict)  # Group logical slots to be advanced by database name
+        create_slots = []  # And collect logical slots to be created on the replica
+        for name, value in slots.items():
+            if value['type'] == 'logical':
+                # If the logical already exists, copy some information about it into the original structure
+                if self._replication_slots.get(name, {}).get('datoid'):
+                    self._copy_items(self._replication_slots[name], value)
+                    if name in cluster.slots:
+                        try:  # Skip slots that doesn't need to be advanced
+                            if value['confirmed_flush_lsn'] < int(cluster.slots[name]):
+                                advance_slots[value['database']][name] = value
+                        except Exception as e:
+                            logger.error('Failed to parse "%s": %r', cluster.slots[name], e)
+                elif name in cluster.slots:  # We want to copy only slots with feedback in a DCS
+                    create_slots.append(name)
+
+        # Advance logical slots
+        for database, values in advance_slots.items():
+            with self._get_local_connection_cursor(database) as cur:
+                for name, value in values.items():
+                    try:
+                        cur.execute("SELECT pg_catalog.pg_replication_slot_advance(%s, %s)",
+                                    (name, format_lsn(int(cluster.slots[name]))))
+                    except Exception as e:
+                        logger.error("Failed to advance logical replication slot '%s': %r", name, e)
+                        if isinstance(e, UndefinedFile):
+                            create_slots.append(name)
+                        self._schedule_load_slots = True
+        return create_slots
+
+    def sync_replication_slots(self, cluster, nofailover, replicatefrom=None):
+        ret = None
+        if self._postgresql.major_version >= 90400 and cluster.config:
             try:
                 self.load_replication_slots()
 
-                slots = cluster.get_replication_slots(self._postgresql.name, self._postgresql.role)
+                slots = cluster.get_replication_slots(self._postgresql.name, self._postgresql.role,
+                                                      nofailover, self._postgresql.major_version, True)
 
-                # drop old replication slots which are not presented in desired slots
+                self._drop_incorrect_slots(cluster, slots)
-                for name in set(self._replication_slots) - set(slots):
-                    if not self.drop_replication_slot(name):
-                        logger.error("Failed to drop replication slot '%s'", name)
-                        self._schedule_load_slots = True
 
-                immediately_reserve = ', true' if self._postgresql.major_version >= 90600 else ''
+                self._ensure_physical_slots(slots)
 
-                logical_slots = defaultdict(dict)
+                if self._postgresql.is_leader():
-                for name, value in slots.items():
+                    self._unready_logical_slots.clear()
-                    if name in self._replication_slots and not compare_slots(value, self._replication_slots[name]):
+                    self._ensure_logical_slots_primary(slots)
-                        logger.info("Trying to drop replication slot '%s' because value is changing from %s to %s",
+                elif cluster.slots and slots:
-                                    name, self._replication_slots[name], value)
+                    self.check_logical_slots_readiness(cluster, nofailover, replicatefrom)
-                        if not self.drop_replication_slot(name):
+
-                            logger.error("Failed to drop replication slot '%s'", name)
+                    ret = self._ensure_logical_slots_replica(cluster, slots)
-                            self._schedule_load_slots = True
-                            continue
-                        self._replication_slots.pop(name)
-                    if name not in self._replication_slots:
-                        if value['type'] == 'physical':
-                            try:
-                                self._query(("SELECT pg_catalog.pg_create_physical_replication_slot(%s{0})" +
-                                             " WHERE NOT EXISTS (SELECT 1 FROM pg_catalog.pg_replication_slots" +
-                                             " WHERE slot_type = 'physical' AND slot_name = %s)").format(
-                                                 immediately_reserve), name, name)
-                            except Exception:
-                                logger.exception("Failed to create physical replication slot '%s'", name)
-                                self._schedule_load_slots = True
-                        elif value['type'] == 'logical' and name not in self._replication_slots:
-                            logical_slots[value['database']][name] = value
 
-                # create new logical slots
-                for database, values in logical_slots.items():
-                    conn_kwargs = self._postgresql.config.local_connect_kwargs
-                    conn_kwargs['database'] = database
-                    with get_connection_cursor(**conn_kwargs) as cur:
-                        for name, value in values.items():
-                            try:
-                                cur.execute("SELECT pg_catalog.pg_create_logical_replication_slot(%s, %s)" +
-                                            " WHERE NOT EXISTS (SELECT 1 FROM pg_catalog.pg_replication_slots" +
-                                            " WHERE slot_type = 'logical' AND slot_name = %s)",
-                                            (name, value['plugin'], name))
-                            except Exception:
-                                logger.exception("Failed to create logical replication slot '%s' plugin='%s'",
-                                                 name, value['plugin'])
-                                self._schedule_load_slots = True
                 self._replication_slots = slots
             except Exception:
                 logger.exception('Exception when changing replication slots')
                 self._schedule_load_slots = True
+        return ret
+
+    @contextmanager
+    def _get_leader_connection_cursor(self, leader):
+        conn_kwargs = leader.conn_kwargs(self._postgresql.config.rewind_credentials)
+        conn_kwargs['database'] = self._postgresql.database
+        with get_connection_cursor(connect_timeout=3, options="-c statement_timeout=2000", **conn_kwargs) as cur:
+            yield cur
+
+    def check_logical_slots_readiness(self, cluster, nofailover, replicatefrom):
+        if self._unready_logical_slots:
+            slot_name = cluster.get_my_slot_name_on_primary(self._postgresql.name, replicatefrom)
+            try:
+                with self._get_leader_connection_cursor(cluster.leader) as cur:
+                    cur.execute("SELECT catalog_xmin FROM pg_catalog.pg_get_replication_slots()"
+                                " WHERE NOT pg_catalog.pg_is_in_recovery() AND slot_name = %s", (slot_name,))
+                    if cur.rowcount < 1:
+                        return logger.warning('Physical slot %s does not exist on the primary', slot_name)
+                    catalog_xmin = cur.fetchone()[0]
+            except Exception as e:
+                return logger.error("Failed to check %s physical slot on the primary: %r", slot_name, e)
+        for name in list(self._unready_logical_slots):
+            value = self._replication_slots.get(name)
+            if not value or catalog_xmin <= value['catalog_xmin']:
+                self._unready_logical_slots.remove(name)
+                if value:
+                    logger.info('Logical slot %s is safe to be used after a failover', name)
+
+    def copy_logical_slots(self, leader, slots):
+        with self._get_leader_connection_cursor(leader) as cur:
+            try:
+                cur.execute("SELECT slot_name, catalog_xmin, "
+                            "pg_catalog.pg_wal_lsn_diff(confirmed_flush_lsn, '0/0')::bigint, "
+                            "pg_catalog.pg_read_binary_file('pg_replslot/' || slot_name || '/state')"
+                            " FROM pg_catalog.pg_get_replication_slots() WHERE NOT pg_catalog.pg_is_in_recovery()"
+                            " AND slot_name = ANY(%s)", (slots,))
+                slots = {r[0]: {'catalog_xmin': r[1], 'confirmed_flush_lsn': r[2], 'data': r[3]} for r in cur}
+            except Exception as e:
+                logger.error("Failed to copy logical slots from the %s via postgresql connection: %r", leader.name, e)
+
+        if isinstance(slots, dict) and self._postgresql.stop():
+            pg_replslot_dir = os.path.join(self._postgresql.data_dir, 'pg_replslot')
+            for name, value in slots.items():
+                slot_dir = os.path.join(pg_replslot_dir, name)
+                slot_tmp_dir = slot_dir + '.tmp'
+                if os.path.exists(slot_tmp_dir):
+                    shutil.rmtree(slot_tmp_dir)
+                os.makedirs(slot_tmp_dir)
+                fsync_dir(slot_tmp_dir)
+                with open(os.path.join(slot_tmp_dir, 'state'), 'wb') as f:
+                    f.write(value['data'])
+                    f.flush()
+                    os.fsync(f.fileno())
+                if os.path.exists(slot_dir):
+                    shutil.rmtree(slot_dir)
+                os.rename(slot_tmp_dir, slot_dir)
+                fsync_dir(slot_dir)
+                self._unready_logical_slots.add(name)
+            fsync_dir(pg_replslot_dir)
+            self._postgresql.start()
 
     def schedule(self, value=None):
         if value is None:
-            value = self.use_slots
+            value = self._postgresql.major_version >= 90400
-        self._schedule_load_slots = value
+        self._schedule_load_slots = self._force_readiness_check = value
+
+    def on_promote(self):
+        if self._unready_logical_slots:
+            logger.warning('Logical replication slots that might be unsafe to use after promote: %s',
+                           self._unready_logical_slots)

patroni/postgresql/validator.py

@@ -0,0 +1,537 @@
+import abc
+import logging
+import six
+
+from collections import namedtuple
+from urllib3.response import HTTPHeaderDict
+
+from ..utils import parse_bool, parse_int, parse_real
+
+logger = logging.getLogger(__name__)
+
+
+class CaseInsensitiveDict(HTTPHeaderDict):
+
+    def add(self, key, val):
+        self[key] = val
+
+    def __getitem__(self, key):
+        return self._container[key.lower()][1]
+
+    def __repr__(self):
+        return str(dict(self.items()))
+
+    def copy(self):
+        return CaseInsensitiveDict(self._container.values())
+
+
+class Bool(namedtuple('Bool', 'version_from,version_till')):
+
+    @staticmethod
+    def transform(name, value):
+        if parse_bool(value) is not None:
+            return value
+        logger.warning('Removing bool parameter=%s from the config due to the invalid value=%s', name, value)
+
+
+@six.add_metaclass(abc.ABCMeta)
+class Number(namedtuple('Number', 'version_from,version_till,min_val,max_val,unit')):
+
+    @staticmethod
+    @abc.abstractmethod
+    def parse(value, unit):
+        """parse value"""
+
+    def transform(self, name, value):
+        num_value = self.parse(value, self.unit)
+        if num_value is not None:
+            if num_value < self.min_val:
+                logger.warning('Value=%s of parameter=%s is too low, increasing to %s%s',
+                               value, name, self.min_val, self.unit or '')
+                return self.min_val
+            if num_value > self.max_val:
+                logger.warning('Value=%s of parameter=%s is too big, decreasing to %s%s',
+                               value, name, self.max_val, self.unit or '')
+                return self.max_val
+            return value
+        logger.warning('Removing %s parameter=%s from the config due to the invalid value=%s',
+                       self.__class__.__name__.lower(), name, value)
+
+
+class Integer(Number):
+
+    @staticmethod
+    def parse(value, unit):
+        return parse_int(value, unit)
+
+
+class Real(Number):
+
+    @staticmethod
+    def parse(value, unit):
+        return parse_real(value, unit)
+
+
+class Enum(namedtuple('Enum', 'version_from,version_till,possible_values')):
+
+    def transform(self, name, value):
+        if str(value).lower() in self.possible_values:
+            return value
+        logger.warning('Removing enum parameter=%s from the config due to the invalid value=%s', name, value)
+
+
+class EnumBool(Enum):
+
+    def transform(self, name, value):
+        if parse_bool(value) is not None:
+            return value
+        return super(EnumBool, self).transform(name, value)
+
+
+class String(namedtuple('String', 'version_from,version_till')):
+
+    @staticmethod
+    def transform(name, value):
+        return value
+
+
+# Format:
+#  key - parameter name
+#  value - tuple or multiple tuples if something was changing in GUC across postgres versions
+parameters = CaseInsensitiveDict({
+    'allow_system_table_mods': Bool(90300, None),
+    'application_name': String(90300, None),
+    'archive_command': String(90300, None),
+    'archive_mode': (
+        Bool(90300, 90500),
+        EnumBool(90500, None, ('always',))
+    ),
+    'archive_timeout': Integer(90300, None, 0, 1073741823, 's'),
+    'array_nulls': Bool(90300, None),
+    'authentication_timeout': Integer(90300, None, 1, 600, 's'),
+    'autovacuum': Bool(90300, None),
+    'autovacuum_analyze_scale_factor': Real(90300, None, 0, 100, None),
+    'autovacuum_analyze_threshold': Integer(90300, None, 0, 2147483647, None),
+    'autovacuum_freeze_max_age': Integer(90300, None, 100000, 2000000000, None),
+    'autovacuum_max_workers': (
+        Integer(90300, 90600, 1, 8388607, None),
+        Integer(90600, None, 1, 262143, None)
+    ),
+    'autovacuum_multixact_freeze_max_age': Integer(90300, None, 10000, 2000000000, None),
+    'autovacuum_naptime': Integer(90300, None, 1, 2147483, 's'),
+    'autovacuum_vacuum_cost_delay': (
+        Integer(90300, 120000, -1, 100, 'ms'),
+        Real(120000, None, -1, 100, 'ms')
+    ),
+    'autovacuum_vacuum_cost_limit': Integer(90300, None, -1, 10000, None),
+    'autovacuum_vacuum_insert_scale_factor': Real(130000, None, 0, 100, None),
+    'autovacuum_vacuum_insert_threshold': Integer(130000, None, -1, 2147483647, None),
+    'autovacuum_vacuum_scale_factor': Real(90300, None, 0, 100, None),
+    'autovacuum_vacuum_threshold': Integer(90300, None, 0, 2147483647, None),
+    'autovacuum_work_mem': Integer(90400, None, -1, 2147483647, 'kB'),
+    'backend_flush_after': Integer(90600, None, 0, 256, '8kB'),
+    'backslash_quote': EnumBool(90300, None, ('safe_encoding',)),
+    'backtrace_functions': String(130000, None),
+    'bgwriter_delay': Integer(90300, None, 10, 10000, 'ms'),
+    'bgwriter_flush_after': Integer(90600, None, 0, 256, '8kB'),
+    'bgwriter_lru_maxpages': (
+        Integer(90300, 100000, 0, 1000, None),
+        Integer(100000, None, 0, 1073741823, None)
+    ),
+    'bgwriter_lru_multiplier': Real(90300, None, 0, 10, None),
+    'bonjour': Bool(90300, None),
+    'bonjour_name': String(90300, None),
+    'bytea_output': Enum(90300, None, ('escape', 'hex')),
+    'check_function_bodies': Bool(90300, None),
+    'checkpoint_completion_target': Real(90300, None, 0, 1, None),
+    'checkpoint_flush_after': Integer(90600, None, 0, 256, '8kB'),
+    'checkpoint_segments': Integer(90300, 90500, 1, 2147483647, None),
+    'checkpoint_timeout': (
+        Integer(90300, 90600, 30, 3600, 's'),
+        Integer(90600, None, 30, 86400, 's')
+    ),
+    'checkpoint_warning': Integer(90300, None, 0, 2147483647, 's'),
+    'client_connection_check_interval': Integer(140000, None, '0', '2147483647', 'ms'),
+    'client_encoding': String(90300, None),
+    'client_min_messages': Enum(90300, None, ('debug5', 'debug4', 'debug3', 'debug2',
+                                              'debug1', 'log', 'notice', 'warning', 'error')),
+    'cluster_name': String(90500, None),
+    'commit_delay': Integer(90300, None, 0, 100000, None),
+    'commit_siblings': Integer(90300, None, 0, 1000, None),
+    'compute_query_id': EnumBool(140000, None, ('auto',)),
+    'config_file': String(90300, None),
+    'constraint_exclusion': EnumBool(90300, None, ('partition',)),
+    'cpu_index_tuple_cost': Real(90300, None, 0, 1.79769e+308, None),
+    'cpu_operator_cost': Real(90300, None, 0, 1.79769e+308, None),
+    'cpu_tuple_cost': Real(90300, None, 0, 1.79769e+308, None),
+    'cursor_tuple_fraction': Real(90300, None, 0, 1, None),
+    'data_directory': String(90300, None),
+    'data_sync_retry': Bool(90400, None),
+    'DateStyle': String(90300, None),
+    'db_user_namespace': Bool(90300, None),
+    'deadlock_timeout': Integer(90300, None, 1, 2147483647, 'ms'),
+    'debug_invalidate_system_caches_always': Integer(140000, None, '0', '0', None),
+    'debug_pretty_print': Bool(90300, None),
+    'debug_print_parse': Bool(90300, None),
+    'debug_print_plan': Bool(90300, None),
+    'debug_print_rewritten': Bool(90300, None),
+    'default_statistics_target': Integer(90300, None, 1, 10000, None),
+    'default_table_access_method': String(120000, None),
+    'default_tablespace': String(90300, None),
+    'default_text_search_config': String(90300, None),
+    'default_toast_compression': Enum(140000, None, ('pglz', 'lz4')),
+    'default_transaction_deferrable': Bool(90300, None),
+    'default_transaction_isolation': Enum(90300, None, ('serializable', 'repeatable read',
+                                                        'read committed', 'read uncommitted')),
+    'default_transaction_read_only': Bool(90300, None),
+    'default_with_oids': Bool(90300, 120000),
+    'dynamic_library_path': String(90300, None),
+    'dynamic_shared_memory_type': (
+        Enum(90400, 120000, ('posix', 'sysv', 'mmap', 'none')),
+        Enum(120000, None, ('posix', 'sysv', 'mmap'))
+    ),
+    'effective_cache_size': Integer(90300, None, 1, 2147483647, '8kB'),
+    'effective_io_concurrency': Integer(90300, None, 0, 1000, None),
+    'enable_async_append': Bool(140000, None),
+    'enable_bitmapscan': Bool(90300, None),
+    'enable_gathermerge': Bool(100000, None),
+    'enable_hashagg': Bool(90300, None),
+    'enable_hashjoin': Bool(90300, None),
+    'enable_incremental_sort': Bool(130000, None),
+    'enable_indexonlyscan': Bool(90300, None),
+    'enable_indexscan': Bool(90300, None),
+    'enable_material': Bool(90300, None),
+    'enable_mergejoin': Bool(90300, None),
+    'enable_nestloop': Bool(90300, None),
+    'enable_parallel_append': Bool(110000, None),
+    'enable_parallel_hash': Bool(110000, None),
+    'enable_partition_pruning': Bool(110000, None),
+    'enable_partitionwise_aggregate': Bool(110000, None),
+    'enable_partitionwise_join': Bool(110000, None),
+    'enable_resultcache': Bool(140000, None),
+    'enable_seqscan': Bool(90300, None),
+    'enable_sort': Bool(90300, None),
+    'enable_tidscan': Bool(90300, None),
+    'escape_string_warning': Bool(90300, None),
+    'event_source': String(90300, None),
+    'exit_on_error': Bool(90300, None),
+    'extension_destdir': String(140000, None),
+    'external_pid_file': String(90300, None),
+    'extra_float_digits': Integer(90300, None, -15, 3, None),
+    'force_parallel_mode': EnumBool(90600, None, ('regress',)),
+    'from_collapse_limit': Integer(90300, None, 1, 2147483647, None),
+    'fsync': Bool(90300, None),
+    'full_page_writes': Bool(90300, None),
+    'geqo': Bool(90300, None),
+    'geqo_effort': Integer(90300, None, 1, 10, None),
+    'geqo_generations': Integer(90300, None, 0, 2147483647, None),
+    'geqo_pool_size': Integer(90300, None, 0, 2147483647, None),
+    'geqo_seed': Real(90300, None, 0, 1, None),
+    'geqo_selection_bias': Real(90300, None, 1.5, 2, None),
+    'geqo_threshold': Integer(90300, None, 2, 2147483647, None),
+    'gin_fuzzy_search_limit': Integer(90300, None, 0, 2147483647, None),
+    'gin_pending_list_limit': Integer(90500, None, 64, 2147483647, 'kB'),
+    'hash_mem_multiplier': Real(130000, None, 1, 1000, None),
+    'hba_file': String(90300, None),
+    'hot_standby': Bool(90300, None),
+    'hot_standby_feedback': Bool(90300, None),
+    'huge_pages': EnumBool(90400, None, ('try',)),
+    'huge_page_size': Integer(140000, None, '0', '2147483647', 'kB'),
+    'ident_file': String(90300, None),
+    'idle_in_transaction_session_timeout': Integer(90600, None, 0, 2147483647, 'ms'),
+    'idle_session_timeout': Integer(140000, None, '0', '2147483647', 'ms'),
+    'ignore_checksum_failure': Bool(90300, None),
+    'ignore_invalid_pages': Bool(130000, None),
+    'ignore_system_indexes': Bool(90300, None),
+    'IntervalStyle': Enum(90300, None, ('postgres', 'postgres_verbose', 'sql_standard', 'iso_8601')),
+    'jit': Bool(110000, None),
+    'jit_above_cost': Real(110000, None, -1, 1.79769e+308, None),
+    'jit_debugging_support': Bool(110000, None),
+    'jit_dump_bitcode': Bool(110000, None),
+    'jit_expressions': Bool(110000, None),
+    'jit_inline_above_cost': Real(110000, None, -1, 1.79769e+308, None),
+    'jit_optimize_above_cost': Real(110000, None, -1, 1.79769e+308, None),
+    'jit_profiling_support': Bool(110000, None),
+    'jit_provider': String(110000, None),
+    'jit_tuple_deforming': Bool(110000, None),
+    'join_collapse_limit': Integer(90300, None, 1, 2147483647, None),
+    'krb_caseins_users': Bool(90300, None),
+    'krb_server_keyfile': String(90300, None),
+    'krb_srvname': String(90300, 90400),
+    'lc_messages': String(90300, None),
+    'lc_monetary': String(90300, None),
+    'lc_numeric': String(90300, None),
+    'lc_time': String(90300, None),
+    'listen_addresses': String(90300, None),
+    'local_preload_libraries': String(90300, None),
+    'lock_timeout': Integer(90300, None, 0, 2147483647, 'ms'),
+    'lo_compat_privileges': Bool(90300, None),
+    'log_autovacuum_min_duration': Integer(90300, None, -1, 2147483647, 'ms'),
+    'log_checkpoints': Bool(90300, None),
+    'log_connections': Bool(90300, None),
+    'log_destination': String(90300, None),
+    'log_directory': String(90300, None),
+    'log_disconnections': Bool(90300, None),
+    'log_duration': Bool(90300, None),
+    'log_error_verbosity': Enum(90300, None, ('terse', 'default', 'verbose')),
+    'log_executor_stats': Bool(90300, None),
+    'log_file_mode': Integer(90300, None, 0, 511, None),
+    'log_filename': String(90300, None),
+    'logging_collector': Bool(90300, None),
+    'log_hostname': Bool(90300, None),
+    'logical_decoding_work_mem': Integer(130000, None, 64, 2147483647, 'kB'),
+    'log_line_prefix': String(90300, None),
+    'log_lock_waits': Bool(90300, None),
+    'log_min_duration_sample': Integer(130000, None, -1, 2147483647, 'ms'),
+    'log_min_duration_statement': Integer(90300, None, -1, 2147483647, 'ms'),
+    'log_min_error_statement': Enum(90300, None, ('debug5', 'debug4', 'debug3', 'debug2', 'debug1', 'info',
+                                                  'notice', 'warning', 'error', 'log', 'fatal', 'panic')),
+    'log_min_messages': Enum(90300, None, ('debug5', 'debug4', 'debug3', 'debug2', 'debug1', 'info',
+                                           'notice', 'warning', 'error', 'log', 'fatal', 'panic')),
+    'log_parameter_max_length': Integer(130000, None, -1, 1073741823, 'B'),
+    'log_parameter_max_length_on_error': Integer(130000, None, -1, 1073741823, 'B'),
+    'log_parser_stats': Bool(90300, None),
+    'log_planner_stats': Bool(90300, None),
+    'log_recovery_conflict_waits': Bool(140000, None),
+    'log_replication_commands': Bool(90500, None),
+    'log_rotation_age': Integer(90300, None, 0, 35791394, 'min'),
+    'log_rotation_size': Integer(90300, None, 0, 2097151, 'kB'),
+    'log_statement': Enum(90300, None, ('none', 'ddl', 'mod', 'all')),
+    'log_statement_sample_rate': Real(130000, None, 0, 1, None),
+    'log_statement_stats': Bool(90300, None),
+    'log_temp_files': Integer(90300, None, -1, 2147483647, 'kB'),
+    'log_timezone': String(90300, None),
+    'log_transaction_sample_rate': Real(120000, None, 0, 1, None),
+    'log_truncate_on_rotation': Bool(90300, None),
+    'maintenance_io_concurrency': Integer(130000, None, 0, 1000, None),
+    'maintenance_work_mem': Integer(90300, None, 1024, 2147483647, 'kB'),
+    'max_connections': (
+        Integer(90300, 90600, 1, 8388607, None),
+        Integer(90600, None, 1, 262143, None)
+    ),
+    'max_files_per_process': (
+        Integer(90300, 130000, 25, 2147483647, None),
+        Integer(130000, None, 64, 2147483647, None)
+    ),
+    'max_locks_per_transaction': Integer(90300, None, 10, 2147483647, None),
+    'max_logical_replication_workers': Integer(100000, None, 0, 262143, None),
+    'max_parallel_maintenance_workers': Integer(110000, None, 0, 1024, None),
+    'max_parallel_workers': Integer(100000, None, 0, 1024, None),
+    'max_parallel_workers_per_gather': Integer(90600, None, 0, 1024, None),
+    'max_pred_locks_per_page': Integer(100000, None, 0, 2147483647, None),
+    'max_pred_locks_per_relation': Integer(100000, None, -2147483648, 2147483647, None),
+    'max_pred_locks_per_transaction': Integer(90300, None, 10, 2147483647, None),
+    'max_prepared_transactions': (
+        Integer(90300, 90600, 0, 8388607, None),
+        Integer(90600, None, 0, 262143, None)
+    ),
+    'max_replication_slots': (
+        Integer(90400, 90600, 0, 8388607, None),
+        Integer(90600, None, 0, 262143, None)
+    ),
+    'max_slot_wal_keep_size': Integer(130000, None, -1, 2147483647, 'MB'),
+    'max_stack_depth': Integer(90300, None, 100, 2147483647, 'kB'),
+    'max_standby_archive_delay': Integer(90300, None, -1, 2147483647, 'ms'),
+    'max_standby_streaming_delay': Integer(90300, None, -1, 2147483647, 'ms'),
+    'max_sync_workers_per_subscription': Integer(100000, None, 0, 262143, None),
+    'max_wal_senders': (
+        Integer(90300, 90600, 0, 8388607, None),
+        Integer(90600, None, 0, 262143, None)
+    ),
+    'max_wal_size': (
+        Integer(90500, 100000, 2, 2147483647, '16MB'),
+        Integer(100000, None, 2, 2147483647, 'MB')
+    ),
+    'max_worker_processes': (
+        Integer(90400, 90600, 1, 8388607, None),
+        Integer(90600, None, 0, 262143, None)
+    ),
+    'min_dynamic_shared_memory': Integer(140000, None, '0', '2147483647', 'MB'),
+    'min_parallel_index_scan_size': Integer(100000, None, 0, 715827882, '8kB'),
+    'min_parallel_relation_size': Integer(90600, 100000, 0, 715827882, '8kB'),
+    'min_parallel_table_scan_size': Integer(100000, None, 0, 715827882, '8kB'),
+    'min_wal_size': (
+        Integer(90500, 100000, 2, 2147483647, '16MB'),
+        Integer(100000, None, 2, 2147483647, 'MB')
+    ),
+    'old_snapshot_threshold': Integer(90600, None, -1, 86400, 'min'),
+    'operator_precedence_warning': Bool(90500, 140000),
+    'parallel_leader_participation': Bool(110000, None),
+    'parallel_setup_cost': Real(90600, None, 0, 1.79769e+308, None),
+    'parallel_tuple_cost': Real(90600, None, 0, 1.79769e+308, None),
+    'password_encryption': (
+        Bool(90300, 100000),
+        Enum(100000, None, ('md5', 'scram-sha-256'))
+    ),
+    'plan_cache_mode': Enum(120000, None, ('auto', 'force_generic_plan', 'force_custom_plan')),
+    'port': Integer(90300, None, 1, 65535, None),
+    'post_auth_delay': Integer(90300, None, 0, 2147, 's'),
+    'pre_auth_delay': Integer(90300, None, 0, 60, 's'),
+    'quote_all_identifiers': Bool(90300, None),
+    'random_page_cost': Real(90300, None, 0, 1.79769e+308, None),
+    'recovery_init_sync_method': Enum(140000, None, ('fsync', 'syncfs')),
+    'remove_temp_files_after_crash': Bool(140000, None),
+    'replacement_sort_tuples': Integer(90600, 110000, 0, 2147483647, None),
+    'restart_after_crash': Bool(90300, None),
+    'row_security': Bool(90500, None),
+    'search_path': String(90300, None),
+    'seq_page_cost': Real(90300, None, 0, 1.79769e+308, None),
+    'session_preload_libraries': String(90400, None),
+    'session_replication_role': Enum(90300, None, ('origin', 'replica', 'local')),
+    'shared_buffers': Integer(90300, None, 16, 1073741823, '8kB'),
+    'shared_memory_type': Enum(120000, None, ('sysv', 'mmap')),
+    'shared_preload_libraries': String(90300, None),
+    'sql_inheritance': Bool(90300, 100000),
+    'ssl': Bool(90300, None),
+    'ssl_ca_file': String(90300, None),
+    'ssl_cert_file': String(90300, None),
+    'ssl_ciphers': String(90300, None),
+    'ssl_crl_dir': String(140000, None),
+    'ssl_crl_file': String(90300, None),
+    'ssl_dh_params_file': String(100000, None),
+    'ssl_ecdh_curve': String(90400, None),
+    'ssl_key_file': String(90300, None),
+    'ssl_max_protocol_version': Enum(120000, None, ('', 'tlsv1', 'tlsv1.1', 'tlsv1.2', 'tlsv1.3')),
+    'ssl_min_protocol_version': Enum(120000, None, ('tlsv1', 'tlsv1.1', 'tlsv1.2', 'tlsv1.3')),
+    'ssl_passphrase_command': String(110000, None),
+    'ssl_passphrase_command_supports_reload': Bool(110000, None),
+    'ssl_prefer_server_ciphers': Bool(90400, None),
+    'ssl_renegotiation_limit': Integer(90300, 90500, 0, 2147483647, 'kB'),
+    'standard_conforming_strings': Bool(90300, None),
+    'statement_timeout': Integer(90300, None, 0, 2147483647, 'ms'),
+    'stats_temp_directory': String(90300, None),
+    'superuser_reserved_connections': (
+        Integer(90300, 90600, 0, 8388607, None),
+        Integer(90600, None, 0, 262143, None)
+    ),
+    'synchronize_seqscans': Bool(90300, None),
+    'synchronous_commit': (
+        EnumBool(90300, 90600, ('local', 'remote_write')),
+        EnumBool(90600, None, ('local', 'remote_write', 'remote_apply'))
+    ),
+    'synchronous_standby_names': String(90300, None),
+    'syslog_facility': Enum(90300, None, ('local0', 'local1', 'local2', 'local3',
+                                          'local4', 'local5', 'local6', 'local7')),
+    'syslog_ident': String(90300, None),
+    'syslog_sequence_numbers': Bool(90600, None),
+    'syslog_split_messages': Bool(90600, None),
+    'tcp_keepalives_count': Integer(90300, None, 0, 2147483647, None),
+    'tcp_keepalives_idle': Integer(90300, None, 0, 2147483647, 's'),
+    'tcp_keepalives_interval': Integer(90300, None, 0, 2147483647, 's'),
+    'tcp_user_timeout': Integer(120000, None, 0, 2147483647, 'ms'),
+    'temp_buffers': Integer(90300, None, 100, 1073741823, '8kB'),
+    'temp_file_limit': Integer(90300, None, -1, 2147483647, 'kB'),
+    'temp_tablespaces': String(90300, None),
+    'TimeZone': String(90300, None),
+    'timezone_abbreviations': String(90300, None),
+    'trace_notify': Bool(90300, None),
+    'trace_recovery_messages': Enum(90300, None, ('debug5', 'debug4', 'debug3', 'debug2',
+                                                  'debug1', 'log', 'notice', 'warning', 'error')),
+    'trace_sort': Bool(90300, None),
+    'track_activities': Bool(90300, None),
+    'track_activity_query_size': (
+        Integer(90300, 110000, 100, 102400, None),
+        Integer(110000, 130000, 100, 102400, 'B'),
+        Integer(130000, None, 100, 1048576, 'B')
+    ),
+    'track_commit_timestamp': Bool(90500, None),
+    'track_counts': Bool(90300, None),
+    'track_functions': Enum(90300, None, ('none', 'pl', 'all')),
+    'track_io_timing': Bool(90300, None),
+    'track_wal_io_timing': Bool(140000, None),
+    'transaction_deferrable': Bool(90300, None),
+    'transaction_isolation': Enum(90300, None, ('serializable', 'repeatable read',
+                                                'read committed', 'read uncommitted')),
+    'transaction_read_only': Bool(90300, None),
+    'transform_null_equals': Bool(90300, None),
+    'unix_socket_directories': String(90300, None),
+    'unix_socket_group': String(90300, None),
+    'unix_socket_permissions': Integer(90300, None, 0, 511, None),
+    'update_process_title': Bool(90300, None),
+    'vacuum_cleanup_index_scale_factor': Real(110000, 140000, 0, 1e+10, None),
+    'vacuum_cost_delay': (
+        Integer(90300, 120000, 0, 100, 'ms'),
+        Real(120000, None, 0, 100, 'ms')
+    ),
+    'vacuum_cost_limit': Integer(90300, None, 1, 10000, None),
+    'vacuum_cost_page_dirty': Integer(90300, None, 0, 10000, None),
+    'vacuum_cost_page_hit': Integer(90300, None, 0, 10000, None),
+    'vacuum_cost_page_miss': Integer(90300, None, 0, 10000, None),
+    'vacuum_defer_cleanup_age': Integer(90300, None, 0, 1000000, None),
+    'vacuum_failsafe_age': Integer(140000, None, '0', '2100000000', None),
+    'vacuum_freeze_min_age': Integer(90300, None, 0, 1000000000, None),
+    'vacuum_freeze_table_age': Integer(90300, None, 0, 2000000000, None),
+    'vacuum_multixact_failsafe_age': Integer(140000, None, '0', '2100000000', None),
+    'vacuum_multixact_freeze_min_age': Integer(90300, None, 0, 1000000000, None),
+    'vacuum_multixact_freeze_table_age': Integer(90300, None, 0, 2000000000, None),
+    'wal_buffers': Integer(90300, None, -1, 262143, '8kB'),
+    'wal_compression': Bool(90500, None),
+    'wal_consistency_checking': String(100000, None),
+    'wal_init_zero': Bool(120000, None),
+    'wal_keep_segments': Integer(90300, 130000, 0, 2147483647, None),
+    'wal_keep_size': Integer(130000, None, 0, 2147483647, 'MB'),
+    'wal_level': (
+        Enum(90300, 90400, ('minimal', 'archive', 'hot_standby')),
+        Enum(90400, 90600, ('minimal', 'archive', 'hot_standby', 'logical')),
+        Enum(90600, None, ('minimal', 'replica', 'logical'))
+    ),
+    'wal_log_hints': Bool(90400, None),
+    'wal_receiver_create_temp_slot': Bool(130000, None),
+    'wal_receiver_status_interval': Integer(90300, None, 0, 2147483, 's'),
+    'wal_receiver_timeout': Integer(90300, None, 0, 2147483647, 'ms'),
+    'wal_recycle': Bool(120000, None),
+    'wal_retrieve_retry_interval': Integer(90500, None, 1, 2147483647, 'ms'),
+    'wal_sender_timeout': Integer(90300, None, 0, 2147483647, 'ms'),
+    'wal_skip_threshold': Integer(130000, None, 0, 2147483647, 'kB'),
+    'wal_sync_method': Enum(90300, None, ('fsync', 'fdatasync', 'open_sync', 'open_datasync')),
+    'wal_writer_delay': Integer(90300, None, 1, 10000, 'ms'),
+    'wal_writer_flush_after': Integer(90600, None, 0, 2147483647, '8kB'),
+    'work_mem': Integer(90300, None, 64, 2147483647, 'kB'),
+    'xmlbinary': Enum(90300, None, ('base64', 'hex')),
+    'xmloption': Enum(90300, None, ('content', 'document')),
+    'zero_damaged_pages': Bool(90300, None)
+})
+
+
+recovery_parameters = CaseInsensitiveDict({
+    'archive_cleanup_command': String(90300, None),
+    'pause_at_recovery_target': Bool(90300, 90500),
+    'primary_conninfo': String(90300, None),
+    'primary_slot_name': String(90400, None),
+    'promote_trigger_file': String(120000, None),
+    'recovery_end_command': String(90300, None),
+    'recovery_min_apply_delay': Integer(90400, None, 0, 2147483647, 'ms'),
+    'recovery_target': Enum(90400, None, ('immediate', '')),
+    'recovery_target_action': Enum(90500, None, ('pause', 'promote', 'shutdown')),
+    'recovery_target_inclusive': Bool(90300, None),
+    'recovery_target_lsn': String(100000, None),
+    'recovery_target_name': String(90400, None),
+    'recovery_target_time': String(90300, None),
+    'recovery_target_timeline': String(90300, None),
+    'recovery_target_xid': String(90300, None),
+    'restore_command': String(90300, None),
+    'standby_mode': Bool(90300, 120000),
+    'trigger_file': String(90300, 120000)
+})
+
+
+def _transform_parameter_value(validators, version, name, value):
+    validators = validators.get(name)
+    if validators:
+        for validator in (validators if isinstance(validators[0], tuple) else [validators]):
+            if version >= validator.version_from and\
+                    (validator.version_till is None or version < validator.version_till):
+                return validator.transform(name, value)
+    logger.warning('Removing unexpected parameter=%s value=%s from the config', name, value)
+
+
+def transform_postgresql_parameter_value(version, name, value):
+    if '.' in name:
+        return value
+    if name in recovery_parameters:
+        return None
+    return _transform_parameter_value(parameters, version, name, value)
+
+
+def transform_recovery_parameter_value(version, name, value):
+    return _transform_parameter_value(recovery_parameters, version, name, value)

patroni/raft_controller.py

@@ -0,0 +1,29 @@
+import logging
+
+from .daemon import AbstractPatroniDaemon, abstract_main
+from .dcs.raft import KVStoreTTL
+
+logger = logging.getLogger(__name__)
+
+
+class RaftController(AbstractPatroniDaemon):
+
+    def __init__(self, config):
+        super(RaftController, self).__init__(config)
+
+        config = self.config.get('raft')
+        assert 'self_addr' in config
+        self._raft = KVStoreTTL(None, None, None, **config)
+
+    def _run_cycle(self):
+        try:
+            self._raft.doTick(self._raft.conf.autoTickPeriod)
+        except Exception:
+            logger.exception('doTick')
+
+    def _shutdown(self):
+        self._raft.destroy()
+
+
+def main():
+    abstract_main(RaftController)

patroni/request.py

@@ -0,0 +1,58 @@
+import json
+import urllib3
+import six
+
+from six.moves.urllib_parse import urlparse, urlunparse
+
+from .utils import USER_AGENT
+
+
+class PatroniRequest(object):
+
+    def __init__(self, config, insecure=False):
+        cert_reqs = 'CERT_NONE' if insecure or config.get('ctl', {}).get('insecure', False) else 'CERT_REQUIRED'
+        self._pool = urllib3.PoolManager(num_pools=10, maxsize=10, cert_reqs=cert_reqs)
+        self.reload_config(config)
+
+    @staticmethod
+    def _get_cfg_value(config, name):
+        return config.get('ctl', {}).get(name) or config.get('restapi', {}).get(name)
+
+    def _apply_pool_param(self, param, value):
+        if value:
+            self._pool.connection_pool_kw[param] = value
+        else:
+            self._pool.connection_pool_kw.pop(param, None)
+
+    def _apply_ssl_file_param(self, config, name):
+        value = self._get_cfg_value(config, name + 'file')
+        self._apply_pool_param(name + '_file', value)
+        return value
+
+    def reload_config(self, config):
+        self._pool.headers = urllib3.make_headers(basic_auth=self._get_cfg_value(config, 'auth'), user_agent=USER_AGENT)
+
+        if self._apply_ssl_file_param(config, 'cert'):
+            self._apply_ssl_file_param(config, 'key')
+        else:
+            self._pool.connection_pool_kw.pop('key_file', None)
+
+        cacert = config.get('ctl', {}).get('cacert') or config.get('restapi', {}).get('cafile')
+        self._apply_pool_param('ca_certs', cacert)
+
+    def request(self, method, url, body=None, **kwargs):
+        if body is not None and not isinstance(body, six.string_types):
+            body = json.dumps(body)
+        return self._pool.request(method.upper(), url, body=body, **kwargs)
+
+    def __call__(self, member, method='GET', endpoint=None, data=None, **kwargs):
+        url = member.api_url
+        if endpoint:
+            scheme, netloc, _, _, _, _ = urlparse(url)
+            url = urlunparse((scheme, netloc, endpoint, '', '', ''))
+        return self.request(method, url, data, **kwargs)
+
+
+def get(url, verify=True, **kwargs):
+    http = PatroniRequest({}, not verify)
+    return http.request('GET', url, **kwargs)

patroni/scripts/aws.py

@@ -1,12 +1,12 @@
 #!/usr/bin/env python
 
+import json
 import logging
-import requests
-from requests.exceptions import RequestException
 import sys
 import boto.ec2
 
 from patroni.utils import Retry, RetryFailedError
+from patroni.request import get as requests_get
 
 logger = logging.getLogger(__name__)
 
@@ -19,14 +19,14 @@ class AWSConnection(object):
         self._retry = Retry(deadline=300, max_delay=30, max_tries=-1, retry_exceptions=(boto.exception.StandardError,))
         try:
             # get the instance id
-            r = requests.get('http://169.254.169.254/latest/dynamic/instance-identity/document', timeout=2.1)
+            r = requests_get('http://169.254.169.254/latest/dynamic/instance-identity/document', timeout=2.1)
-        except RequestException:
+        except Exception:
             logger.error('cannot query AWS meta-data')
             return
 
-        if r.ok:
+        if r.status < 400:
             try:
-                content = r.json()
+                content = json.loads(r.data.decode('utf-8'))
                 self.instance_id = content['instanceId']
                 self.region = content['region']
             except Exception:

patroni/scripts/wale_restore.py

@@ -250,7 +250,7 @@ class WALERestore(object):
                 diff_in_bytes = 0
             break
 
-        # if the size of the accumulated WAL segments is more than a certan percentage of the backup size
+        # if the size of the accumulated WAL segments is more than a certain percentage of the backup size
         # or exceeds the pre-determined size - pg_basebackup is chosen instead.
         is_size_thresh_ok = diff_in_bytes < int(threshold_megabytes) * 1048576
         threshold_pct_bytes = backup_size * threshold_percent / 100.0
@@ -308,7 +308,7 @@ class WALERestore(object):
             try:
                 os.mkdir(path)
             except OSError:
-                logger.exception("coud not create missing %s directory path", dirname)
+                logger.exception("could not create missing %s directory path", dirname)
                 return False
         return True
 

patroni/utils.py

@@ -1,15 +1,25 @@
+import errno
+import json.decoder as json_decoder
 import logging
+import os
+import platform
 import random
 import re
+import socket
+import sys
+import tempfile
 import time
 
 from dateutil import tz
-from patroni.exceptions import PatroniException
+
+from .exceptions import PatroniException
+from .version import __version__
 
 tzutc = tz.tzutc()
 
 logger = logging.getLogger(__name__)
 
+USER_AGENT = 'Patroni/{0} Python/{1} {2}'.format(__version__, platform.python_version(), platform.system())
 OCT_RE = re.compile(r'^[-+]?0[0-7]*')
 DEC_RE = re.compile(r'^[-+]?(0|[1-9][0-9]*)')
 HEX_RE = re.compile(r'^[-+]?0x[0-9a-fA-F]+')
@@ -296,6 +306,17 @@ class Retry(object):
                      max_jitter=self.max_jitter / 100.0, max_delay=self.max_delay, sleep_func=self.sleep_func,
                      deadline=self.deadline, retry_exceptions=self.retry_exceptions)
 
+    @property
+    def sleeptime(self):
+        return self._cur_delay + (random.randint(0, self.max_jitter) / 100.0)
+
+    def update_delay(self):
+        self._cur_delay = min(self._cur_delay * self.backoff, self.max_delay)
+
+    @property
+    def stoptime(self):
+        return self._cur_stoptime
+
     def __call__(self, func, *args, **kwargs):
         """Call a function with arguments until it completes without throwing a `retry_exceptions`
 
@@ -317,14 +338,14 @@ class Retry(object):
                     logger.warning('Retry got exception: %s', e)
                     raise RetryFailedError("Too many retry attempts")
                 self._attempts += 1
-                sleeptime = self._cur_delay + (random.randint(0, self.max_jitter) / 100.0)
+                sleeptime = hasattr(e, 'sleeptime') and e.sleeptime or self.sleeptime
 
                 if self._cur_stoptime is not None and time.time() + sleeptime >= self._cur_stoptime:
                     logger.warning('Retry got exception: %s', e)
                     raise RetryFailedError("Exceeded retry deadline")
                 logger.debug('Retry got exception: %s', e)
                 self.sleep_func(sleeptime)
-                self._cur_delay = min(self._cur_delay * self.backoff, self.max_delay)
+                self.update_delay()
 
 
 def polling_loop(timeout, interval=1):
@@ -340,6 +361,8 @@ def polling_loop(timeout, interval=1):
 
 def split_host_port(value, default_port):
     t = value.rsplit(':', 1)
+    if ':' in t[0]:
+        t[0] = t[0].strip('[]')
     t.append(default_port)
     return t[0], int(t[1])
 
@@ -352,3 +375,160 @@ def uri(proto, netloc, path='', user=None):
     path = '/{0}'.format(path) if path and not path.startswith('/') else path
     user = '{0}@'.format(user) if user else ''
     return '{0}://{1}{2}{3}{4}'.format(proto, user, host, port, path)
+
+
+def iter_response_objects(response):
+    prev = ''
+    decoder = json_decoder.JSONDecoder()
+    for chunk in response.read_chunked(decode_content=False):
+        if isinstance(chunk, bytes):
+            chunk = chunk.decode('utf-8')
+        chunk = prev + chunk
+
+        length = len(chunk)
+        idx = json_decoder.WHITESPACE.match(chunk, 0).end()
+        while idx < length:
+            try:
+                message, idx = decoder.raw_decode(chunk, idx)
+            except ValueError:  # malformed or incomplete JSON, unlikely to happen
+                break
+            else:
+                yield message
+                idx = json_decoder.WHITESPACE.match(chunk, idx).end()
+        prev = chunk[idx:]
+
+
+def is_standby_cluster(config):
+    # Check whether or not provided configuration describes a standby cluster
+    return isinstance(config, dict) and (config.get('host') or config.get('port') or config.get('restore_command'))
+
+
+def cluster_as_json(cluster):
+    leader_name = cluster.leader.name if cluster.leader else None
+    cluster_lsn = cluster.last_lsn or 0
+
+    ret = {'members': []}
+    for m in cluster.members:
+        if m.name == leader_name:
+            config = cluster.config.data if cluster.config and cluster.config.modify_index else {}
+            role = 'standby_leader' if is_standby_cluster(config.get('standby_cluster')) else 'leader'
+        elif m.name in cluster.sync.members:
+            role = 'sync_standby'
+        else:
+            role = 'replica'
+
+        member = {'name': m.name, 'role': role, 'state': m.data.get('state', ''), 'api_url': m.api_url}
+        conn_kwargs = m.conn_kwargs()
+        if conn_kwargs.get('host'):
+            member['host'] = conn_kwargs['host']
+            if conn_kwargs.get('port'):
+                member['port'] = int(conn_kwargs['port'])
+        optional_attributes = ('timeline', 'pending_restart', 'scheduled_restart', 'tags')
+        member.update({n: m.data[n] for n in optional_attributes if n in m.data})
+
+        if m.name != leader_name:
+            lsn = m.data.get('xlog_location')
+            if lsn is None:
+                member['lag'] = 'unknown'
+            elif cluster_lsn >= lsn:
+                member['lag'] = cluster_lsn - lsn
+            else:
+                member['lag'] = 0
+
+        ret['members'].append(member)
+
+    # sort members by name for consistency
+    ret['members'].sort(key=lambda m: m['name'])
+    if cluster.is_paused():
+        ret['pause'] = True
+    if cluster.failover and cluster.failover.scheduled_at:
+        ret['scheduled_switchover'] = {'at': cluster.failover.scheduled_at.isoformat()}
+        if cluster.failover.leader:
+            ret['scheduled_switchover']['from'] = cluster.failover.leader
+        if cluster.failover.candidate:
+            ret['scheduled_switchover']['to'] = cluster.failover.candidate
+    return ret
+
+
+def is_subpath(d1, d2):
+    real_d1 = os.path.realpath(d1) + os.path.sep
+    real_d2 = os.path.realpath(os.path.join(real_d1, d2))
+    return os.path.commonprefix([real_d1, real_d2 + os.path.sep]) == real_d1
+
+
+def validate_directory(d, msg="{} {}"):
+    if not os.path.exists(d):
+        try:
+            os.makedirs(d)
+        except OSError as e:
+            logger.error(e)
+            if e.errno != errno.EEXIST:
+                raise PatroniException(msg.format(d, "couldn't create the directory"))
+    elif os.path.isdir(d):
+        try:
+            fd, tmpfile = tempfile.mkstemp(dir=d)
+            os.close(fd)
+            os.remove(tmpfile)
+        except OSError:
+            raise PatroniException(msg.format(d, "the directory is not writable"))
+    else:
+        raise PatroniException(msg.format(d, "is not a directory"))
+
+
+def data_directory_is_empty(data_dir):
+    if not os.path.exists(data_dir):
+        return True
+    return all(os.name != 'nt' and (n.startswith('.') or n == 'lost+found') for n in os.listdir(data_dir))
+
+
+def keepalive_intvl(timeout, idle, cnt=3):
+    return max(1, int(float(timeout - idle) / cnt))
+
+
+def keepalive_socket_options(timeout, idle, cnt=3):
+    yield (socket.SOL_SOCKET, socket.SO_KEEPALIVE, 1)
+
+    if sys.platform.startswith('linux'):
+        yield (socket.SOL_TCP, 18, int(timeout * 1000))  # TCP_USER_TIMEOUT
+        TCP_KEEPIDLE = getattr(socket, 'TCP_KEEPIDLE', None)
+        TCP_KEEPINTVL = getattr(socket, 'TCP_KEEPINTVL', None)
+        TCP_KEEPCNT = getattr(socket, 'TCP_KEEPCNT', None)
+    elif sys.platform.startswith('darwin'):
+        TCP_KEEPIDLE = 0x10  # (named "TCP_KEEPALIVE" in C)
+        TCP_KEEPINTVL = 0x101
+        TCP_KEEPCNT = 0x102
+    else:
+        return
+
+    intvl = keepalive_intvl(timeout, idle, cnt)
+    yield (socket.IPPROTO_TCP, TCP_KEEPIDLE, idle)
+    yield (socket.IPPROTO_TCP, TCP_KEEPINTVL, intvl)
+    yield (socket.IPPROTO_TCP, TCP_KEEPCNT, cnt)
+
+
+def enable_keepalive(sock, timeout, idle, cnt=3):
+    SIO_KEEPALIVE_VALS = getattr(socket, 'SIO_KEEPALIVE_VALS', None)
+    if SIO_KEEPALIVE_VALS is not None:  # Windows
+        intvl = keepalive_intvl(timeout, idle, cnt)
+        return sock.ioctl(SIO_KEEPALIVE_VALS, (1, idle * 1000, intvl * 1000))
+
+    for opt in keepalive_socket_options(timeout, idle, cnt):
+        sock.setsockopt(*opt)
+
+
+def find_executable(executable, path=None):
+    _, ext = os.path.splitext(executable)
+
+    if (sys.platform == 'win32') and (ext != '.exe'):
+        executable = executable + '.exe'
+
+    if os.path.isfile(executable):
+        return executable
+
+    if path is None:
+        path = os.environ.get('PATH', os.defpath)
+
+    for p in path.split(os.pathsep):
+        f = os.path.join(p, executable)
+        if os.path.isfile(f):
+            return f

patroni/validator.py

@@ -0,0 +1,393 @@
+#!/usr/bin/env python3
+import os
+import socket
+import re
+import subprocess
+
+from six import string_types
+
+from .utils import find_executable, split_host_port, data_directory_is_empty
+from .dcs import dcs_modules
+from .exceptions import ConfigParseError
+
+
+def data_directory_empty(data_dir):
+    if os.path.isfile(os.path.join(data_dir, "global", "pg_control")):
+        return False
+    return data_directory_is_empty(data_dir)
+
+
+def validate_connect_address(address):
+    try:
+        host, _ = split_host_port(address, 1)
+    except (AttributeError, TypeError, ValueError):
+        raise ConfigParseError("contains a wrong value")
+    if host in ["127.0.0.1", "0.0.0.0", "*", "::1", "localhost"]:
+        raise ConfigParseError('must not contain "127.0.0.1", "0.0.0.0", "*", "::1", "localhost"')
+    return True
+
+
+def validate_host_port(host_port, listen=False, multiple_hosts=False):
+    try:
+        hosts, port = split_host_port(host_port, None)
+    except (ValueError, TypeError):
+        raise ConfigParseError("contains a wrong value")
+    else:
+        if multiple_hosts:
+            hosts = hosts.split(",")
+        else:
+            hosts = [hosts]
+        for host in hosts:
+            proto = socket.getaddrinfo(host, "", 0, socket.SOCK_STREAM, 0, socket.AI_PASSIVE)
+            s = socket.socket(proto[0][0], socket.SOCK_STREAM)
+            try:
+                if s.connect_ex((host, port)) == 0:
+                    if listen:
+                        raise ConfigParseError("Port {} is already in use.".format(port))
+                elif not listen:
+                    raise ConfigParseError("{} is not reachable".format(host_port))
+            except socket.gaierror as e:
+                raise ConfigParseError(e)
+            finally:
+                s.close()
+    return True
+
+
+def validate_host_port_list(value):
+    assert all([validate_host_port(v) for v in value]), "didn't pass the validation"
+    return True
+
+
+def comma_separated_host_port(string):
+    return validate_host_port_list([s.strip() for s in string.split(",")])
+
+
+def validate_host_port_listen(host_port):
+    return validate_host_port(host_port, listen=True)
+
+
+def validate_host_port_listen_multiple_hosts(host_port):
+    return validate_host_port(host_port, listen=True, multiple_hosts=True)
+
+
+def is_ipv4_address(ip):
+    try:
+        socket.inet_aton(ip)
+    except Exception:
+        raise ConfigParseError("Is not a valid ipv4 address")
+    return True
+
+
+def is_ipv6_address(ip):
+    try:
+        socket.inet_pton(socket.AF_INET6, ip)
+    except Exception:
+        raise ConfigParseError("Is not a valid ipv6 address")
+    return True
+
+
+def get_major_version(bin_dir=None):
+    if not bin_dir:
+        binary = 'postgres'
+    else:
+        binary = os.path.join(bin_dir, 'postgres')
+    version = subprocess.check_output([binary, '--version']).decode()
+    version = re.match(r'^[^\s]+ [^\s]+ (\d+)(\.(\d+))?', version)
+    return '.'.join([version.group(1), version.group(3)]) if int(version.group(1)) < 10 else version.group(1)
+
+
+def validate_data_dir(data_dir):
+    if not data_dir:
+        raise ConfigParseError("is an empty string")
+    elif os.path.exists(data_dir) and not os.path.isdir(data_dir):
+        raise ConfigParseError("is not a directory")
+    elif not data_directory_empty(data_dir):
+        if not os.path.exists(os.path.join(data_dir, "PG_VERSION")):
+            raise ConfigParseError("doesn't look like a valid data directory")
+        else:
+            with open(os.path.join(data_dir, "PG_VERSION"), "r") as version:
+                pgversion = version.read().strip()
+            waldir = ("pg_wal" if float(pgversion) >= 10 else "pg_xlog")
+            if not os.path.isdir(os.path.join(data_dir, waldir)):
+                raise ConfigParseError("data dir for the cluster is not empty, but doesn't contain"
+                                       " \"{}\" directory".format(waldir))
+            bin_dir = schema.data.get("postgresql", {}).get("bin_dir", None)
+            major_version = get_major_version(bin_dir)
+            if pgversion != major_version:
+                raise ConfigParseError("data_dir directory postgresql version ({}) doesn't match with "
+                                       "'postgres --version' output ({})".format(pgversion, major_version))
+    return True
+
+
+class Result(object):
+    def __init__(self, status, error="didn't pass validation", level=0, path="", data=""):
+        self.status = status
+        self.path = path
+        self.data = data
+        self.level = level
+        self._error = error
+        if not self.status:
+            self.error = error
+        else:
+            self.error = None
+
+    def __repr__(self):
+        return self.path + (" " + str(self.data) + " " + self._error if self.error else "")
+
+
+class Case(object):
+    def __init__(self, schema):
+        self._schema = schema
+
+
+class Or(object):
+    def __init__(self, *args):
+        self.args = args
+
+
+class Optional(object):
+    def __init__(self, name):
+        self.name = name
+
+
+class Directory(object):
+    def __init__(self, contains=None, contains_executable=None):
+        self.contains = contains
+        self.contains_executable = contains_executable
+
+    def validate(self, name):
+        if not name:
+            yield Result(False, "is an empty string")
+        elif not os.path.exists(name):
+            yield Result(False, "Directory '{}' does not exist.".format(name))
+        elif not os.path.isdir(name):
+            yield Result(False, "'{}' is not a directory.".format(name))
+        else:
+            if self.contains:
+                for path in self.contains:
+                    if not os.path.exists(os.path.join(name, path)):
+                        yield Result(False, "'{}' does not contain '{}'".format(name, path))
+            if self.contains_executable:
+                for program in self.contains_executable:
+                    if not find_executable(program, name):
+                        yield Result(False, "'{}' does not contain '{}'".format(name, program))
+
+
+class Schema(object):
+    def __init__(self, validator):
+        self.validator = validator
+
+    def __call__(self, data):
+        for i in self.validate(data):
+            if not i.status:
+                print(i)
+
+    def validate(self, data):
+        self.data = data
+        if isinstance(self.validator, string_types):
+            yield Result(isinstance(self.data, string_types), "is not a string", level=1, data=self.data)
+        elif issubclass(type(self.validator), type):
+            validator = self.validator
+            if self.validator == str:
+                validator = string_types
+            yield Result(isinstance(self.data, validator),
+                         "is not {}".format(_get_type_name(self.validator)), level=1, data=self.data)
+        elif callable(self.validator):
+            if hasattr(self.validator, "expected_type"):
+                if not isinstance(data, self.validator.expected_type):
+                    yield Result(False, "is not {}"
+                                 .format(_get_type_name(self.validator.expected_type)), level=1, data=self.data)
+                    return
+            try:
+                self.validator(data)
+                yield Result(True, data=self.data)
+            except Exception as e:
+                yield Result(False, "didn't pass validation: {}".format(e), data=self.data)
+        elif isinstance(self.validator, dict):
+            if not len(self.validator):
+                yield Result(isinstance(self.data, dict), "is not a dictionary", level=1, data=self.data)
+        elif isinstance(self.validator, list):
+            if not isinstance(self.data, list):
+                yield Result(isinstance(self.data, list), "is not a list", level=1, data=self.data)
+                return
+        for i in self.iter():
+            yield i
+
+    def iter(self):
+        if isinstance(self.validator, dict):
+            if not isinstance(self.data, dict):
+                yield Result(False, "is not a dictionary.", level=1)
+            else:
+                for i in self.iter_dict():
+                    yield i
+        elif isinstance(self.validator, list):
+            if len(self.data) == 0:
+                yield Result(False, "is an empty list", data=self.data)
+            if len(self.validator) > 0:
+                for key, value in enumerate(self.data):
+                    for v in Schema(self.validator[0]).validate(value):
+                        yield Result(v.status, v.error,
+                                     path=(str(key) + ("." + v.path if v.path else "")), level=v.level, data=value)
+        elif isinstance(self.validator, Directory):
+            for v in self.validator.validate(self.data):
+                yield v
+        elif isinstance(self.validator, Or):
+            for i in self.iter_or():
+                yield i
+
+    def iter_dict(self):
+        for key in self.validator.keys():
+            for d in self._data_key(key):
+                if d not in self.data and not isinstance(key, Optional):
+                    yield Result(False, "is not defined.", path=d)
+                elif d not in self.data and isinstance(key, Optional):
+                    continue
+                else:
+                    validator = self.validator[key]
+                    if isinstance(key, Or) and isinstance(self.validator[key], Case):
+                        validator = self.validator[key]._schema[d]
+                    for v in Schema(validator).validate(self.data[d]):
+                        yield Result(v.status, v.error,
+                                     path=(d + ("." + v.path if v.path else "")), level=v.level, data=v.data)
+
+    def iter_or(self):
+        results = []
+        for a in self.validator.args:
+            r = []
+            for v in Schema(a).validate(self.data):
+                r.append(v)
+            if any([x.status for x in r]) and not all([x.status for x in r]):
+                results += filter(lambda x: not x.status, r)
+            else:
+                results += r
+        if not any([x.status for x in results]):
+            max_level = 3
+            for v in sorted(results, key=lambda x: x.level):
+                if v.level > max_level:
+                    break
+                max_level = v.level
+                yield Result(v.status, v.error, path=v.path, level=v.level, data=v.data)
+
+    def _data_key(self, key):
+        if isinstance(self.data, dict) and isinstance(key, str):
+            yield key
+        elif isinstance(key, Optional):
+            yield key.name
+        elif isinstance(key, Or):
+            if any([i in self.data for i in key.args]):
+                for i in key.args:
+                    if i in self.data:
+                        yield i
+            else:
+                for i in key.args:
+                    yield i
+
+
+def _get_type_name(python_type):
+    return {str: 'a string', int: 'and integer', float: 'a number', bool: 'a boolean',
+            list: 'an array', dict: 'a dictionary', string_types: "a string"}.get(
+                    python_type, getattr(python_type, __name__, "unknown type"))
+
+
+def assert_(condition, message="Wrong value"):
+    assert condition, message
+
+
+userattributes = {"username": "", Optional("password"): ""}
+available_dcs = [m.split(".")[-1] for m in dcs_modules()]
+validate_host_port_list.expected_type = list
+comma_separated_host_port.expected_type = string_types
+validate_connect_address.expected_type = string_types
+validate_host_port_listen.expected_type = string_types
+validate_host_port_listen_multiple_hosts.expected_type = string_types
+validate_data_dir.expected_type = string_types
+validate_etcd = {
+    Or("host", "hosts", "srv", "url", "proxy"): Case({
+        "host": validate_host_port,
+        "hosts": Or(comma_separated_host_port, [validate_host_port]),
+        "srv": str,
+        "url": str,
+        "proxy": str})
+}
+
+schema = Schema({
+  "name": str,
+  "scope": str,
+  "restapi": {
+    "listen": validate_host_port_listen,
+    "connect_address": validate_connect_address
+  },
+  Optional("bootstrap"): {
+    "dcs": {
+        Optional("ttl"): int,
+        Optional("loop_wait"): int,
+        Optional("retry_timeout"): int,
+        Optional("maximum_lag_on_failover"): int
+        },
+    "pg_hba": [str],
+    "initdb": [Or(str, dict)]
+  },
+  Or(*available_dcs): Case({
+      "consul": {
+          Or("host", "url"): Case({
+              "host": validate_host_port,
+              "url": str})
+          },
+      "etcd": validate_etcd,
+      "etcd3": validate_etcd,
+      "exhibitor": {
+          "hosts": [str],
+          "port": lambda i: assert_(int(i) <= 65535),
+          Optional("pool_interval"): int
+          },
+      "raft": {
+          "self_addr": validate_connect_address,
+          Optional("bind_addr"): validate_host_port_listen,
+          "partner_addrs": validate_host_port_list,
+          Optional("data_dir"): str,
+          Optional("password"): str
+          },
+      "zookeeper": {
+          "hosts": Or(comma_separated_host_port, [validate_host_port]),
+          },
+      "kubernetes": {
+          "labels": {},
+          Optional("namespace"): str,
+          Optional("scope_label"): str,
+          Optional("role_label"): str,
+          Optional("use_endpoints"): bool,
+          Optional("pod_ip"): Or(is_ipv4_address, is_ipv6_address),
+          Optional("ports"): [{"name": str, "port": int}],
+          },
+      }),
+  "postgresql": {
+    "listen": validate_host_port_listen_multiple_hosts,
+    "connect_address": validate_connect_address,
+    "authentication": {
+      "replication": userattributes,
+      "superuser": userattributes,
+      "rewind":  userattributes
+    },
+    "data_dir": validate_data_dir,
+    Optional("bin_dir"): Directory(contains_executable=["pg_ctl", "initdb", "pg_controldata", "pg_basebackup",
+                                                        "postgres", "pg_isready"]),
+    Optional("parameters"): {
+      Optional("unix_socket_directories"): lambda s: assert_(all([isinstance(s, string_types), len(s)]))
+    },
+    Optional("pg_hba"): [str],
+    Optional("pg_ident"): [str],
+    Optional("pg_ctl_timeout"): int,
+    Optional("use_pg_rewind"): bool
+  },
+  Optional("watchdog"): {
+    Optional("mode"): lambda m: assert_(m in ["off", "automatic", "required"]),
+    Optional("device"): str
+  },
+  Optional("tags"): {
+    Optional("nofailover"): bool,
+    Optional("clonefrom"): bool,
+    Optional("noloadbalance"): bool,
+    Optional("replicatefrom"): str,
+    Optional("nosync"): bool
+  }
+})

patroni/version.py

@@ -1 +1 @@
-__version__ = '1.6.0'
+__version__ = '2.1.0'

patroni/watchdog/base.py

@@ -57,7 +57,7 @@ class WatchdogConfig(object):
         return not self == other
 
     def get_impl(self):
-        if self.driver == 'testing':
+        if self.driver == 'testing':  # pragma: no cover
             from patroni.watchdog.linux import TestingWatchdogDevice
             return TestingWatchdogDevice.from_config(self.driver_config)
         elif platform.system() == 'Linux' and self.driver == 'default':

patroni/watchdog/linux.py

@@ -16,11 +16,11 @@ IOC_DIRBITS = 2
 
 # Non-generic platform special cases
 machine = platform.machine()
-if machine in ['mips', 'sparc', 'powerpc', 'ppc64']:
+if machine in ['mips', 'sparc', 'powerpc', 'ppc64']:  # pragma: no cover
     IOC_SIZEBITS = 13
     IOC_DIRBITS = 3
     IOC_NONE, IOC_WRITE, IOC_READ = 1, 2, 4
-elif machine == 'parisc':
+elif machine == 'parisc':  # pragma: no cover
     IOC_WRITE, IOC_READ = 2, 1
 
 IOC_NRSHIFT = 0
@@ -218,7 +218,7 @@ class LinuxWatchdogDevice(WatchdogBase):
         return timeout.value
 
 
-class TestingWatchdogDevice(LinuxWatchdogDevice):
+class TestingWatchdogDevice(LinuxWatchdogDevice):  # pragma: no cover
     """Converts timeout ioctls to regular writes that can be intercepted from a named pipe."""
     timeout = 60
 

patroni_raft_controller.py

@@ -0,0 +1,6 @@
+#!/usr/bin/env python
+from patroni.raft_controller import main
+
+
+if __name__ == '__main__':
+    main()

postgres0.yml

@@ -17,7 +17,25 @@ restapi:
 #   cacert: /etc/ssl/certs/ssl-cacert-snakeoil.pem
 
 etcd:
+  #Provide host to do the initial discovery of the cluster topology:
   host: 127.0.0.1:2379
+  #Or use "hosts" to provide multiple endpoints
+  #Could be a comma separated string:
+  #hosts: host1:port1,host2:port2
+  #or an actual yaml list:
+  #hosts:
+  #- host1:port1
+  #- host2:port2
+  #Once discovery is complete Patroni will use the list of advertised clientURLs
+  #It is possible to change this behavior through by setting:
+  #use_proxies: true
+
+#raft:
+#  data_dir: .
+#  self_addr: 127.0.0.1:2222
+#  partner_addrs:
+#  - 127.0.0.1:2223
+#  - 127.0.0.1:2224
 
 bootstrap:
   # this section will be written into Etcd:/<namespace>/<scope>/config after initializing new cluster
@@ -39,10 +57,15 @@ bootstrap:
       parameters:
 #        wal_level: hot_standby
 #        hot_standby: "on"
+#        max_connections: 100
+#        max_worker_processes: 8
 #        wal_keep_segments: 8
 #        max_wal_senders: 10
 #        max_replication_slots: 10
+#        max_prepared_transactions: 0
+#        max_locks_per_transaction: 64
 #        wal_log_hints: "on"
+#        track_commit_timestamp: "off"
 #        archive_mode: "on"
 #        archive_timeout: 1800s
 #        archive_command: mkdir -p ../wal_archive && test ! -f ../wal_archive/%f && cp %p ../wal_archive/%f
@@ -97,6 +120,8 @@ postgresql:
     # same as KRB5CCNAME used by the GSS
 #   krb_server_keyfile: /var/spool/keytabs/postgres
     unix_socket_directories: '.'
+  # Additional fencing script executed after acquiring the leader lock but before promoting the replica
+  #pre_promote: /path/to/pre_promote.sh
 
 #watchdog:
 #  mode: automatic # Allowed values: off, automatic, required

postgres1.yml

@@ -17,7 +17,25 @@ restapi:
 #   cacert: /etc/ssl/certs/ssl-cacert-snakeoil.pem
 
 etcd:
+  #Provide host to do the initial discovery of the cluster topology:
   host: 127.0.0.1:2379
+  #Or use "hosts" to provide multiple endpoints
+  #Could be a comma separated string:
+  #hosts: host1:port1,host2:port2
+  #or an actual yaml list:
+  #hosts:
+  #- host1:port1
+  #- host2:port2
+  #Once discovery is complete Patroni will use the list of advertised clientURLs
+  #It is possible to change this behavior through by setting:
+  #use_proxies: true
+
+#raft:
+#  data_dir: .
+#  self_addr: 127.0.0.1:2223
+#  partner_addrs:
+#  - 127.0.0.1:2222
+#  - 127.0.0.1:2224
 
 bootstrap:
   # this section will be written into Etcd:/<namespace>/<scope>/config after initializing new cluster
@@ -33,10 +51,15 @@ bootstrap:
       parameters:
 #        wal_level: hot_standby
 #        hot_standby: "on"
+#        max_connections: 100
+#        max_worker_processes: 8
 #        wal_keep_segments: 8
 #        max_wal_senders: 10
 #        max_replication_slots: 10
+#        max_prepared_transactions: 0
+#        max_locks_per_transaction: 64
 #        wal_log_hints: "on"
+#        track_commit_timestamp: "off"
 #        archive_mode: "on"
 #        archive_timeout: 1800s
 #        archive_command: mkdir -p ../wal_archive && test ! -f ../wal_archive/%f && cp %p ../wal_archive/%f
@@ -94,6 +117,10 @@ postgresql:
   basebackup:
       - verbose
       - max-rate: 100M
+#      - waldir: /pg-wal-mount/external-waldir # only needed in case pg_wal is symlinked outside of data_dir
+  # Additional fencing script executed after acquiring the leader lock but before promoting the replica
+  #pre_promote: /path/to/pre_promote.sh
+
 tags:
     nofailover: false
     noloadbalance: false

postgres2.yml

@@ -17,7 +17,25 @@ restapi:
 #   cacert: /etc/ssl/certs/ssl-cacert-snakeoil.pem
 
 etcd:
+  #Provide host to do the initial discovery of the cluster topology:
   host: 127.0.0.1:2379
+  #Or use "hosts" to provide multiple endpoints
+  #Could be a comma separated string:
+  #hosts: host1:port1,host2:port2
+  #or an actual yaml list:
+  #hosts:
+  #- host1:port1
+  #- host2:port2
+  #Once discovery is complete Patroni will use the list of advertised clientURLs
+  #It is possible to change this behavior through by setting:
+  #use_proxies: true
+
+#raft:
+#  data_dir: .
+#  self_addr: 127.0.0.1:2224
+#  partner_addrs:
+#  - 127.0.0.1:2222
+#  - 127.0.0.1:2223
 
 bootstrap:
   # this section will be written into Etcd:/<namespace>/<scope>/config after initializing new cluster
@@ -33,10 +51,15 @@ bootstrap:
       parameters:
 #        wal_level: hot_standby
 #        hot_standby: "on"
+#        max_connections: 100
+#        max_worker_processes: 8
 #        wal_keep_segments: 8
 #        max_wal_senders: 10
 #        max_replication_slots: 10
+#        max_prepared_transactions: 0
+#        max_locks_per_transaction: 64
 #        wal_log_hints: "on"
+#        track_commit_timestamp: "off"
 #        archive_mode: "on"
 #        archive_timeout: 1800s
 #        archive_command: mkdir -p ../wal_archive && test ! -f ../wal_archive/%f && cp %p ../wal_archive/%f

requirements.dev.txt

@@ -0,0 +1,8 @@
+psycopg2-binary
+behave
+coverage
+flake8
+mock
+pytest-cov
+pytest
+setuptools

requirements.txt

@@ -1,15 +1,15 @@
 urllib3>=1.19.1,!=1.21
+ipaddress; python_version=="2.7"
 boto
 PyYAML
-requests
 six >= 1.7
 kazoo>=1.3.1
 python-etcd>=0.4.3,<0.5
-python-consul>=0.7.0
+python-consul>=0.7.1
 click>=4.1
 prettytable>=0.7
-tzlocal
 python-dateutil
+pysyncobj>=0.3.8
+cryptography>=1.4
 psutil>=2.0.0
-cdiff
+ydiff>=1.2.0
-kubernetes>=2.0.0,<=7.0.0,!=4.0.*,!=5.0.*

setup.py

@@ -8,22 +8,12 @@ import inspect
 import os
 import sys
 
-from patroni import check_psycopg2, fatal
+from setuptools import Command, find_packages, setup
-from patroni.version import __version__ as VERSION
-from setuptools.command.test import test as TestCommand
-from setuptools import find_packages, setup
-
-if sys.version_info < (2, 7, 0):
-    fatal('patroni needs to be run with Python 2.7+')
-check_psycopg2()
-del sys.modules['patroni']
-del sys.modules['patroni.version']
 
 __location__ = os.path.join(os.getcwd(), os.path.dirname(inspect.getfile(inspect.currentframe())))
 
 NAME = 'patroni'
 MAIN_PACKAGE = NAME
-SCRIPTS = 'scripts'
 DESCRIPTION = 'PostgreSQL High-Available orchestrator and CLI'
 LICENSE = 'The MIT License'
 URL = 'https://github.com/zalando/patroni'
@@ -32,9 +22,11 @@ AUTHOR_EMAIL = '[email protected], [email protected], alexk
 KEYWORDS = 'etcd governor patroni postgresql postgres ha haproxy confd' +\
     ' zookeeper exhibitor consul streaming replication kubernetes k8s'
 
+EXTRAS_REQUIRE = {'aws': ['boto'], 'etcd': ['python-etcd'], 'etcd3': ['python-etcd'],
+                  'consul': ['python-consul'], 'exhibitor': ['kazoo'], 'zookeeper': ['kazoo'],
+                  'kubernetes': [], 'raft': ['pysyncobj', 'cryptography']}
 COVERAGE_XML = True
 COVERAGE_HTML = False
-JUNIT_XML = True
 
 # Add here all kinds of additional classifiers as defined under
 # https://pypi.python.org/pypi?%3Aaction=list_classifiers
@@ -55,87 +47,144 @@ CLASSIFIERS = [
     'Programming Language :: Python :: 3.5',
     'Programming Language :: Python :: 3.6',
     'Programming Language :: Python :: 3.7',
+    'Programming Language :: Python :: 3.8',
+    'Programming Language :: Python :: 3.9',
     'Programming Language :: Python :: Implementation :: CPython',
 ]
 
 CONSOLE_SCRIPTS = ['patroni = patroni:main',
                    'patronictl = patroni.ctl:ctl',
+                   'patroni_raft_controller = patroni.raft_controller:main',
                    "patroni_wale_restore = patroni.scripts.wale_restore:main",
                    "patroni_aws = patroni.scripts.aws:main"]
 
 
-class PyTest(TestCommand):
+class Flake8(Command):
 
-    user_options = [('cov=', None, 'Run coverage'), ('cov-xml=', None, 'Generate junit xml report'), ('cov-html=',
+    user_options = []
-                    None, 'Generate junit html report'), ('junitxml=', None, 'Generate xml of test results')]
 
     def initialize_options(self):
-        TestCommand.initialize_options(self)
+        from flake8.main import application
+
+        self.flake8 = application.Application()
+        self.flake8.initialize([])
+
+    def finalize_options(self):
+        pass
+
+    def package_files(self):
+        seen_package_directories = ()
+        directories = self.distribution.package_dir or {}
+        empty_directory_exists = "" in directories
+        packages = self.distribution.packages or []
+        for package in packages:
+            if package in directories:
+                package_directory = directories[package]
+            elif empty_directory_exists:
+                package_directory = os.path.join(directories[""], package)
+            else:
+                package_directory = package
+
+            if not package_directory.startswith(seen_package_directories):
+                seen_package_directories += (package_directory + ".",)
+                yield package_directory
+
+    def targets(self):
+        return [package for package in self.package_files()] + ['tests', 'setup.py']
+
+    def run(self):
+        self.flake8.run_checks(self.targets())
+        self.flake8.formatter.start()
+        self.flake8.report_errors()
+        self.flake8.report_statistics()
+        self.flake8.report_benchmarks()
+        self.flake8.formatter.stop()
+        try:
+            self.flake8.exit()
+        except SystemExit as e:
+            # Cause system exit only if exit code is not zero (terminates
+            # other possibly remaining/pending setuptools commands).
+            if e.code:
+                raise
+
+
+class PyTest(Command):
+
+    user_options = [('cov=', None, 'Run coverage'), ('cov-xml=', None, 'Generate junit xml report'),
+                    ('cov-html=', None, 'Generate junit html report')]
+
+    def initialize_options(self):
+        self.cov = []
         self.cov_xml = False
         self.cov_html = False
-        self.junitxml = None
 
     def finalize_options(self):
-        TestCommand.finalize_options(self)
         if self.cov_xml or self.cov_html:
             self.cov = ['--cov', MAIN_PACKAGE, '--cov-report', 'term-missing']
             if self.cov_xml:
                 self.cov.extend(['--cov-report', 'xml'])
             if self.cov_html:
                 self.cov.extend(['--cov-report', 'html'])
-        if self.junitxml is not None:
-            self.junitxml = ['--junitxml', self.junitxml]
 
     def run_tests(self):
         try:
             import pytest
         except Exception:
             raise RuntimeError('py.test is not installed, run: pip install pytest')
-        params = {'args': self.test_args}
-        if self.cov:
-            params['args'] += self.cov
-        if self.junitxml:
-            params['args'] += self.junitxml
-        params['args'] += ['--doctest-modules', MAIN_PACKAGE, '-vv']
 
         import logging
         silence = logging.WARNING
         logging.basicConfig(format='%(asctime)s %(levelname)s: %(message)s', level=os.getenv('LOGLEVEL', silence))
-        params['args'] += ['-s' if logging.getLogger().getEffectiveLevel() < silence else '--capture=fd']
+
-        if not os.getenv('SYSTEMROOT'):
+        args = ['--verbose', 'tests', '--doctest-modules', MAIN_PACKAGE] +\
-            os.environ['SYSTEMROOT'] = '/'
+            ['-s' if logging.getLogger().getEffectiveLevel() < silence else '--capture=fd']
-        errno = pytest.main(**params)
+        if self.cov:
+            args += self.cov
+
+        errno = pytest.main(args=args)
         sys.exit(errno)
 
+    def run(self):
+        from pkg_resources import evaluate_marker
+
+        requirements = set(self.distribution.install_requires + ['mock>=2.0.0', 'pytest-cov', 'pytest'])
+        for k, v in self.distribution.extras_require.items():
+            if not k.startswith(':') or evaluate_marker(k[1:]):
+                requirements.update(v)
+
+        self.distribution.fetch_build_eggs(list(requirements))
+        self.run_tests()
+
 
 def read(fname):
     with open(os.path.join(__location__, fname)) as fd:
         return fd.read()
 
 
-def setup_package():
+def setup_package(version):
     # Assemble additional setup commands
-    cmdclass = {'test': PyTest}
+    cmdclass = {'test': PyTest, 'flake8': Flake8}
 
     install_requires = []
-    extras_require = {'aws': ['boto'], 'etcd': ['python-etcd'], 'consul': ['python-consul'],
-                      'exhibitor': ['kazoo'], 'zookeeper': ['kazoo'], 'kubernetes': ['kubernetes']}
 
     for r in read('requirements.txt').split('\n'):
         r = r.strip()
         if r == '':
             continue
         extra = False
-        for e, v in extras_require.items():
+        for e, deps in EXTRAS_REQUIRE.items():
-            if r.startswith(v[0]):
+            for i, v in enumerate(deps):
-                extras_require[e] = [r]
+                if r.startswith(v):
-                extra = True
+                    deps[i] = r
+                    EXTRAS_REQUIRE[e] = deps
+                    extra = True
+                    break
+            if extra:
+                break
         if not extra:
             install_requires.append(r)
 
-    command_options = {'test': {'test_suite': ('setup.py', 'tests')}}
+    command_options = {'test': {}}
-    if JUNIT_XML:
-        command_options['test']['junitxml'] = 'setup.py', 'junit.xml'
     if COVERAGE_XML:
         command_options['test']['cov_xml'] = 'setup.py', True
     if COVERAGE_HTML:
@@ -143,7 +192,7 @@ def setup_package():
 
     setup(
         name=NAME,
-        version=VERSION,
+        version=version,
         url=URL,
         author=AUTHOR,
         author_email=AUTHOR_EMAIL,
@@ -152,17 +201,28 @@ def setup_package():
         keywords=KEYWORDS,
         long_description=read('README.rst'),
         classifiers=CLASSIFIERS,
-        test_suite='tests',
         packages=find_packages(exclude=['tests', 'tests.*']),
         package_data={MAIN_PACKAGE: ["*.json"]},
+        python_requires='>=2.7',
         install_requires=install_requires,
-        extras_require=extras_require,
+        extras_require=EXTRAS_REQUIRE,
+        setup_requires='flake8',
         cmdclass=cmdclass,
-        tests_require=['flake8', 'mock>=2.0.0', 'pytest-cov', 'pytest'],
         command_options=command_options,
         entry_points={'console_scripts': CONSOLE_SCRIPTS},
     )
 
 
 if __name__ == '__main__':
-    setup_package()
+    old_modules = sys.modules.copy()
+    try:
+        from patroni import check_psycopg2, fatal, __version__
+    finally:
+        sys.modules.clear()
+        sys.modules.update(old_modules)
+
+    if sys.version_info < (2, 7, 0):
+        fatal('Patroni needs to be run with Python 2.7+')
+    check_psycopg2()
+
+    setup_package(__version__)

tests/__init__.py

@@ -1,19 +1,17 @@
 import datetime
-import json
 import os
 import shutil
 import unittest
 
 from mock import Mock, patch
-from tempfile import gettempdir
 
 import psycopg2
-import requests
+import urllib3
 
 from patroni.dcs import Leader, Member
 from patroni.postgresql import Postgresql
 from patroni.postgresql.config import ConfigHandler
-from patroni.utils import RetryFailedError
+from patroni.utils import RetryFailedError, tzutc
 
 
 class SleepException(Exception):
@@ -25,19 +23,12 @@ class MockResponse(object):
     def __init__(self, status_code=200):
         self.status_code = status_code
         self.content = '{}'
-        self.ok = True
+        self.reason = 'Not Found'
-
-    def json(self):
-        return json.loads(self.content)
 
     @property
     def data(self):
         return self.content.encode('utf-8')
 
-    @property
-    def text(self):
-        return self.content
-
     @property
     def status(self):
         return self.status_code
@@ -46,13 +37,17 @@ class MockResponse(object):
     def getheader(*args):
         return ''
 
+    @staticmethod
+    def getheaders():
+        return {'content-type': 'json'}
+
 
 def requests_get(url, **kwargs):
     members = '[{"id":14855829450254237642,"peerURLs":["http://localhost:2380","http://localhost:7001"],' +\
               '"name":"default","clientURLs":["http://localhost:2379","http://localhost:4001"]}]'
     response = MockResponse()
     if url.startswith('http://local'):
-        raise requests.exceptions.RequestException()
+        raise urllib3.exceptions.HTTPError()
     elif ':8011/patroni' in url:
         response.content = '{"role": "replica", "xlog": {"received_location": 0}, "tags": {}}'
     elif url.endswith('/members'):
@@ -63,11 +58,9 @@ def requests_get(url, **kwargs):
         data = kwargs.get('data', '')
         if ' false}' in data:
             response.status_code = 503
-            response.ok = False
             response.content = 'restarting after failure already in progress'
     else:
         response.status_code = 404
-        response.ok = False
     return response
 
 
@@ -78,6 +71,7 @@ class MockPostmaster(object):
         self.wait_for_user_backends_to_close = Mock()
         self.signal_stop = Mock(return_value=None)
         self.wait = Mock()
+        self.signal_kill = Mock(return_value=False)
 
 
 class MockCursor(object):
@@ -87,6 +81,7 @@ class MockCursor(object):
         self.closed = False
         self.rowcount = 0
         self.results = []
+        self.description = [Mock()]
 
     def execute(self, sql, *params):
         if sql.startswith('blabla'):
@@ -95,30 +90,33 @@ class MockCursor(object):
             raise psycopg2.OperationalError()
         elif sql.startswith('RetryFailedError'):
             raise RetryFailedError('retry')
+        elif sql.startswith('SELECT catalog_xmin'):
+            self.results = [(100, 501)]
+        elif sql.startswith('SELECT slot_name, catalog_xmin'):
+            self.results = [('ls', 100, 500, b'123456')]
         elif sql.startswith('SELECT slot_name'):
-            self.results = [('blabla', 'physical'), ('foobar', 'physical'), ('ls', 'logical', 'a', 'b')]
+            self.results = [('blabla', 'physical'), ('foobar', 'physical'), ('ls', 'logical', 'a', 'b', 5, 100, 500)]
         elif sql.startswith('SELECT CASE WHEN pg_catalog.pg_is_in_recovery()'):
-            self.results = [(1, 2)]
+            self.results = [(1, 2, 1, 0, False, 1, 1, None, None, [{"slot_name": "ls", "confirmed_flush_lsn": 12345}])]
         elif sql.startswith('SELECT pg_catalog.pg_is_in_recovery()'):
             self.results = [(False, 2)]
-        elif sql.startswith('WITH replication_info AS ('):
+        elif sql.startswith('SELECT pg_catalog.pg_postmaster_start_time'):
             replication_info = '[{"application_name":"walreceiver","client_addr":"1.2.3.4",' +\
                                '"state":"streaming","sync_state":"async","sync_priority":0}]'
-            self.results = [('', 0, '', '', '', '', False, replication_info)]
+            now = datetime.datetime.now(tzutc)
+            self.results = [(now, 0, '', 0, '', False, now, replication_info)]
         elif sql.startswith('SELECT name, setting'):
             self.results = [('wal_segment_size', '2048', '8kB', 'integer', 'internal'),
+                            ('wal_block_size', '8192', None, 'integer', 'internal'),
+                            ('shared_buffers', '16384', '8kB', 'integer', 'postmaster'),
+                            ('wal_buffers', '-1', '8kB', 'integer', 'postmaster'),
                             ('search_path', 'public', None, 'string', 'user'),
                             ('port', '5433', None, 'integer', 'postmaster'),
                             ('listen_addresses', '*', None, 'string', 'postmaster'),
                             ('autovacuum', 'on', None, 'bool', 'sighup'),
                             ('unix_socket_directories', '/tmp', None, 'string', 'postmaster')]
         elif sql.startswith('IDENTIFY_SYSTEM'):
-            self.results = [('1', 2, '0/402EEC0', '')]
+            self.results = [('1', 3, '0/402EEC0', '')]
-        elif sql.startswith('SELECT isdir, modification'):
-            self.results = [(False, datetime.datetime.now())]
-        elif sql.startswith('SELECT pg_catalog.pg_read_file'):
-            self.results = [('1\t0/40159C0\tno recovery target specified\n\n'
-                             '2\t1/40159C0\tno recovery target specified\n',)]
         elif sql.startswith('TIMELINE_HISTORY '):
             self.results = [('', b'x\t0/40159C0\tno recovery target specified\n\n'
                                  b'1\t0/40159C0\tno recovery target specified\n\n'
@@ -173,17 +171,22 @@ class PostgresInit(unittest.TestCase):
                    'search_path': 'public', 'hot_standby': 'on', 'max_wal_senders': 5,
                    'wal_keep_segments': 8, 'wal_log_hints': 'on', 'max_locks_per_transaction': 64,
                    'max_worker_processes': 8, 'max_connections': 100, 'max_prepared_transactions': 0,
-                   'track_commit_timestamp': 'off', 'unix_socket_directories': '/tmp', 'trigger_file': 'bla'}
+                   'track_commit_timestamp': 'off', 'unix_socket_directories': '/tmp',
+                   'trigger_file': 'bla', 'stats_temp_directory': '/tmp', 'zero_damaged_pages': '',
+                   'force_parallel_mode': '1', 'constraint_exclusion': '',
+                   'max_stack_depth': 'Z', 'vacuum_cost_limit': -1, 'vacuum_cost_delay': 200}
 
     @patch('psycopg2.connect', psycopg2_connect)
+    @patch('patroni.postgresql.CallbackExecutor', Mock())
     @patch.object(ConfigHandler, 'write_postgresql_conf', Mock())
     @patch.object(ConfigHandler, 'replace_pg_hba', Mock())
     @patch.object(ConfigHandler, 'replace_pg_ident', Mock())
+    @patch.object(Postgresql, 'get_postgres_role_from_data_directory', Mock(return_value='master'))
     def setUp(self):
-        data_dir = 'data/test0'
+        data_dir = os.path.join('data', 'test0')
         self.p = Postgresql({'name': 'postgresql0', 'scope': 'batman', 'data_dir': data_dir,
                              'config_dir': data_dir, 'retry_timeout': 10,
-                             'krbsrvname': 'postgres', 'pgpass': os.path.join(gettempdir(), 'pgpass0'),
+                             'krbsrvname': 'postgres', 'pgpass': os.path.join(data_dir, 'pgpass0'),
                              'listen': '127.0.0.2, 127.0.0.3:5432', 'connect_address': '127.0.0.2:5432',
                              'authentication': {'superuser': {'username': 'foo', 'password': 'test'},
                                                 'replication': {'username': '', 'password': 'rep-pass'}},

tests/test_api.py

@@ -12,6 +12,7 @@ from patroni.utils import tzutc
 from six import BytesIO as IO
 from six.moves import BaseHTTPServer
 from . import psycopg2_connect, MockCursor
+from .test_ha import get_cluster_initialized_without_leader
 
 
 future_restart_time = datetime.datetime.now(tzutc) + datetime.timedelta(days=5)
@@ -29,6 +30,8 @@ class MockPostgresql(object):
     pending_restart = True
     wal_name = 'wal'
     lsn_name = 'lsn'
+    POSTMASTER_START_TIME = 'pg_catalog.pg_postmaster_start_time()'
+    TL_LSN = 'CASE WHEN pg_catalog.pg_is_in_recovery()'
 
     @staticmethod
     def connection():
@@ -36,7 +39,7 @@ class MockPostgresql(object):
 
     @staticmethod
     def postmaster_start_time():
-        return str(postmaster_start_time)
+        return postmaster_start_time
 
     @staticmethod
     def replica_cached_timeline(_):
@@ -115,7 +118,7 @@ class MockPatroni(object):
     postgresql = ha.state_handler
     dcs = Mock()
     logger = MockLogger()
-    tags = {}
+    tags = {"key1": True, "key2": False, "key3": 1, "key4": 1.4, "key5": "RandomTag"}
     version = '0.00'
     noloadbalance = PropertyMock(return_value=False)
     scheduled_restart = {'schedule': future_restart_time,
@@ -145,7 +148,8 @@ class MockRestApiServer(RestApiServer):
         self.serve_forever = Mock()
         MockRestApiServer._BaseServer__is_shut_down = Mock()
         MockRestApiServer._BaseServer__shutdown_request = True
-        config = config or {'listen': '127.0.0.1:8008', 'auth': 'test:test', 'certfile': 'dumb'}
+        config = config or {'listen': '127.0.0.1:8008', 'auth': 'test:test', 'certfile': 'dumb', 'verify_client': 'a',
+                            'http_extra_headers': {'foo': 'bar'}, 'https_extra_headers': {'foo': 'sbar'}}
         super(MockRestApiServer, self).__init__(MockPatroni(), config)
         Handler(MockRequest(request), ('0.0.0.0', 8080), self)
 
@@ -158,23 +162,30 @@ class TestRestApiHandler(unittest.TestCase):
     _authorization = '\nAuthorization: Basic dGVzdDp0ZXN0'
 
     def test_do_GET(self):
+        MockPatroni.dcs.cluster.last_lsn = 20
         MockRestApiServer(RestApiHandler, 'GET /replica')
+        MockRestApiServer(RestApiHandler, 'GET /replica?lag=1M')
+        MockRestApiServer(RestApiHandler, 'GET /replica?lag=10MB')
+        MockRestApiServer(RestApiHandler, 'GET /replica?lag=10485760')
+        MockRestApiServer(RestApiHandler, 'GET /read-only')
         with patch.object(RestApiHandler, 'get_postgresql_status', Mock(return_value={})):
             MockRestApiServer(RestApiHandler, 'GET /replica')
         with patch.object(RestApiHandler, 'get_postgresql_status', Mock(return_value={'role': 'master'})):
             MockRestApiServer(RestApiHandler, 'GET /replica')
         with patch.object(RestApiHandler, 'get_postgresql_status', Mock(return_value={'state': 'running'})):
             MockRestApiServer(RestApiHandler, 'GET /health')
-        MockRestApiServer(RestApiHandler, 'GET /master')
+        MockRestApiServer(RestApiHandler, 'GET /leader')
-        MockPatroni.dcs.cluster.sync.sync_standby = MockPostgresql.name
+        MockPatroni.dcs.cluster.sync.members = [MockPostgresql.name]
         MockPatroni.dcs.cluster.is_synchronous_mode = Mock(return_value=True)
         with patch.object(RestApiHandler, 'get_postgresql_status', Mock(return_value={'role': 'replica'})):
             MockRestApiServer(RestApiHandler, 'GET /synchronous')
         with patch.object(RestApiHandler, 'get_postgresql_status', Mock(return_value={'role': 'replica'})):
-            MockPatroni.dcs.cluster.sync.sync_standby = ''
+            MockPatroni.dcs.cluster.sync.members = []
             MockRestApiServer(RestApiHandler, 'GET /asynchronous')
-        MockPatroni.ha.is_leader = Mock(return_value=True)
+        with patch.object(MockHa, 'is_leader', Mock(return_value=True)):
-        MockRestApiServer(RestApiHandler, 'GET /replica')
+            MockRestApiServer(RestApiHandler, 'GET /replica')
+            with patch.object(MockHa, 'is_standby_cluster', Mock(return_value=True)):
+                MockRestApiServer(RestApiHandler, 'GET /standby_leader')
         MockPatroni.dcs.cluster = None
         with patch.object(RestApiHandler, 'get_postgresql_status', Mock(return_value={'role': 'master'})):
             MockRestApiServer(RestApiHandler, 'GET /master')
@@ -186,9 +197,102 @@ class TestRestApiHandler(unittest.TestCase):
         with patch.object(MockHa, 'is_standby_cluster', Mock(return_value=True)):
             MockRestApiServer(RestApiHandler, 'GET /standby_leader')
 
+        # test tags
+        #
+        MockRestApiServer(RestApiHandler, 'GET /master?lag=1M&'
+                                          'tag_key1=true&tag_key2=false&'
+                                          'tag_key3=1&tag_key4=1.4&tag_key5=RandomTag')
+        MockRestApiServer(RestApiHandler, 'GET /master?lag=1M&'
+                                          'tag_key1=true&tag_key2=False&'
+                                          'tag_key3=1&tag_key4=1.4&tag_key5=RandomTag')
+        MockRestApiServer(RestApiHandler, 'GET /master?lag=1M&'
+                                          'tag_key1=true&tag_key2=false&'
+                                          'tag_key3=1.0&tag_key4=1.4&tag_key5=RandomTag')
+        MockRestApiServer(RestApiHandler, 'GET /master?lag=1M&'
+                                          'tag_key1=true&tag_key2=false&'
+                                          'tag_key3=1&tag_key4=1.4&tag_key5=RandomTag&tag_key6=RandomTag2')
+        #
+        with patch.object(RestApiHandler, 'get_postgresql_status', Mock(return_value={'role': 'master'})):
+            MockRestApiServer(RestApiHandler, 'GET /master?lag=1M&'
+                                              'tag_key1=true&tag_key2=false&'
+                                              'tag_key3=1&tag_key4=1.4&tag_key5=RandomTag')
+            MockRestApiServer(RestApiHandler, 'GET /master?lag=1M&'
+                                              'tag_key1=true&tag_key2=False&'
+                                              'tag_key3=1&tag_key4=1.4&tag_key5=RandomTag')
+            MockRestApiServer(RestApiHandler, 'GET /master?lag=1M&'
+                                              'tag_key1=true&tag_key2=false&'
+                                              'tag_key3=1.0&tag_key4=1.4&tag_key5=RandomTag')
+            MockRestApiServer(RestApiHandler, 'GET /master?lag=1M&'
+                                              'tag_key1=true&tag_key2=false&'
+                                              'tag_key3=1&tag_key4=1.4&tag_key5=RandomTag&tag_key6=RandomTag2')
+        #
+        with patch.object(RestApiHandler, 'get_postgresql_status', Mock(return_value={'role': 'standby_leader'})):
+            MockRestApiServer(RestApiHandler, 'GET /standby_leader?lag=1M&'
+                                              'tag_key1=true&tag_key2=false&'
+                                              'tag_key3=1&tag_key4=1.4&tag_key5=RandomTag')
+            MockRestApiServer(RestApiHandler, 'GET /standby_leader?lag=1M&'
+                                              'tag_key1=true&tag_key2=False&'
+                                              'tag_key3=1&tag_key4=1.4&tag_key5=RandomTag')
+            MockRestApiServer(RestApiHandler, 'GET /standby_leader?lag=1M&'
+                                              'tag_key1=true&tag_key2=false&'
+                                              'tag_key3=1.0&tag_key4=1.4&tag_key5=RandomTag')
+            MockRestApiServer(RestApiHandler, 'GET /standby_leader?lag=1M&'
+                                              'tag_key1=true&tag_key2=false&'
+                                              'tag_key3=1&tag_key4=1.4&tag_key5=RandomTag&tag_key6=RandomTag2')
+        #
+        MockRestApiServer(RestApiHandler, 'GET /replica?lag=1M&'
+                                          'tag_key1=true&tag_key2=false&'
+                                          'tag_key3=1&tag_key4=1.4&tag_key5=RandomTag')
+        MockRestApiServer(RestApiHandler, 'GET /replica?lag=1M&'
+                                          'tag_key1=true&tag_key2=False&'
+                                          'tag_key3=1&tag_key4=1.4&tag_key5=RandomTag')
+        MockRestApiServer(RestApiHandler, 'GET /replica?lag=1M&'
+                                          'tag_key1=true&tag_key2=false&'
+                                          'tag_key3=1.0&tag_key4=1.4&tag_key5=RandomTag')
+        MockRestApiServer(RestApiHandler, 'GET /replica?lag=1M&'
+                                          'tag_key1=true&tag_key2=false&'
+                                          'tag_key3=1&tag_key4=1.4&tag_key5=RandomTag&tag_key6=RandomTag2')
+        #
+        with patch.object(RestApiHandler, 'get_postgresql_status', Mock(return_value={'role': 'master'})):
+            MockRestApiServer(RestApiHandler, 'GET /replica?lag=1M&'
+                                              'tag_key1=true&tag_key2=false&'
+                                              'tag_key3=1&tag_key4=1.4&tag_key5=RandomTag')
+            MockRestApiServer(RestApiHandler, 'GET /replica?lag=1M&'
+                                              'tag_key1=true&tag_key2=False&'
+                                              'tag_key3=1&tag_key4=1.4&tag_key5=RandomTag')
+            MockRestApiServer(RestApiHandler, 'GET /replica?lag=1M&'
+                                              'tag_key1=true&tag_key2=false&'
+                                              'tag_key3=1.0&tag_key4=1.4&tag_key5=RandomTag')
+            MockRestApiServer(RestApiHandler, 'GET /replica?lag=1M&'
+                                              'tag_key1=true&tag_key2=false&'
+                                              'tag_key3=1&tag_key4=1.4&tag_key5=RandomTag&tag_key6=RandomTag2')
+        #
+        MockRestApiServer(RestApiHandler, 'GET /read-write?lag=1M&'
+                                          'tag_key1=true&tag_key2=false&'
+                                          'tag_key3=1&tag_key4=1.4&tag_key5=RandomTag')
+        MockRestApiServer(RestApiHandler, 'GET /read-write?lag=1M&'
+                                          'tag_key1=true&tag_key2=False&'
+                                          'tag_key3=1&tag_key4=1.4&tag_key5=RandomTag')
+        MockRestApiServer(RestApiHandler, 'GET /read-write?lag=1M&'
+                                          'tag_key1=true&tag_key2=false&'
+                                          'tag_key3=1.0&tag_key4=1.4&tag_key5=RandomTag')
+        MockRestApiServer(RestApiHandler, 'GET /read-write?lag=1M&'
+                                          'tag_key1=true&tag_key2=false&'
+                                          'tag_key3=1&tag_key4=1.4&tag_key5=RandomTag&tag_key6=RandomTag2')
+
     def test_do_OPTIONS(self):
         self.assertIsNotNone(MockRestApiServer(RestApiHandler, 'OPTIONS / HTTP/1.0'))
 
+    def test_do_GET_liveness(self):
+        self.assertIsNotNone(MockRestApiServer(RestApiHandler, 'GET /liveness HTTP/1.0'))
+
+    def test_do_GET_readiness(self):
+        self.assertIsNotNone(MockRestApiServer(RestApiHandler, 'GET /readiness HTTP/1.0'))
+        with patch.object(MockHa, 'is_leader', Mock(return_value=True)):
+            self.assertIsNotNone(MockRestApiServer(RestApiHandler, 'GET /readiness HTTP/1.0'))
+        with patch.object(MockPostgresql, 'state', PropertyMock(return_value='stopped')):
+            self.assertIsNotNone(MockRestApiServer(RestApiHandler, 'GET /readiness HTTP/1.0'))
+
     @patch.object(MockPostgresql, 'state', PropertyMock(return_value='stopped'))
     def test_do_GET_patroni(self):
         self.assertIsNotNone(MockRestApiServer(RestApiHandler, 'GET /patroni'))
@@ -197,6 +301,17 @@ class TestRestApiHandler(unittest.TestCase):
         self.assertIsNotNone(MockRestApiServer(RestApiHandler, 'POST /restart HTTP/1.0'))
         MockRestApiServer(RestApiHandler, 'POST /restart HTTP/1.0\nAuthorization:')
 
+    @patch.object(MockPatroni, 'dcs')
+    def test_do_GET_cluster(self, mock_dcs):
+        mock_dcs.get_cluster.return_value = get_cluster_initialized_without_leader()
+        mock_dcs.get_cluster.return_value.members[1].data['xlog_location'] = 11
+        self.assertIsNotNone(MockRestApiServer(RestApiHandler, 'GET /cluster'))
+
+    @patch.object(MockPatroni, 'dcs')
+    def test_do_GET_history(self, mock_dcs):
+        mock_dcs.cluster = get_cluster_initialized_without_leader()
+        self.assertIsNotNone(MockRestApiServer(RestApiHandler, 'GET /history'))
+
     @patch.object(MockPatroni, 'dcs')
     def test_do_GET_config(self, mock_dcs):
         mock_dcs.cluster.config.data = {}
@@ -204,6 +319,10 @@ class TestRestApiHandler(unittest.TestCase):
         mock_dcs.cluster.config = None
         self.assertIsNotNone(MockRestApiServer(RestApiHandler, 'GET /config'))
 
+    @patch.object(MockPatroni, 'dcs')
+    def test_do_GET_metrics(self, mock_dcs):
+        self.assertIsNotNone(MockRestApiServer(RestApiHandler, 'GET /metrics'))
+
     @patch.object(MockPatroni, 'dcs')
     def test_do_PATCH_config(self, mock_dcs):
         config = {'postgresql': {'use_slots': False, 'use_pg_rewind': True, 'parameters': {'wal_level': 'logical'}}}
@@ -219,6 +338,8 @@ class TestRestApiHandler(unittest.TestCase):
         MockRestApiServer(RestApiHandler, request)
         mock_dcs.set_config_value.return_value = False
         MockRestApiServer(RestApiHandler, request)
+        mock_dcs.get_cluster.return_value.config = None
+        MockRestApiServer(RestApiHandler, request)
 
     @patch.object(MockPatroni, 'dcs')
     def test_do_PUT_config(self, mock_dcs):
@@ -233,11 +354,8 @@ class TestRestApiHandler(unittest.TestCase):
         mock_dcs.get_cluster.return_value.config = ClusterConfig.from_node(1, config)
         MockRestApiServer(RestApiHandler, request)
 
-    @patch.object(MockPatroni, 'sighup_handler', Mock(side_effect=Exception))
+    @patch.object(MockPatroni, 'sighup_handler', Mock())
     def test_do_POST_reload(self):
-        with patch.object(MockPatroni, 'config') as mock_config:
-            mock_config.reload_local_configuration.return_value = False
-            MockRestApiServer(RestApiHandler, 'POST /reload HTTP/1.0' + self._authorization)
         self.assertIsNotNone(MockRestApiServer(RestApiHandler, 'POST /reload HTTP/1.0' + self._authorization))
 
     @patch.object(MockPatroni, 'dcs')
@@ -297,6 +415,15 @@ class TestRestApiHandler(unittest.TestCase):
                 request = 'DELETE /restart HTTP/1.0' + self._authorization
                 self.assertIsNotNone(MockRestApiServer(RestApiHandler, request))
 
+    @patch.object(MockPatroni, 'dcs')
+    def test_do_DELETE_switchover(self, mock_dcs):
+        request = 'DELETE /switchover HTTP/1.0' + self._authorization
+        self.assertIsNotNone(MockRestApiServer(RestApiHandler, request))
+        mock_dcs.manual_failover.return_value = False
+        self.assertIsNotNone(MockRestApiServer(RestApiHandler, request))
+        mock_dcs.get_cluster.return_value.failover = None
+        self.assertIsNotNone(MockRestApiServer(RestApiHandler, request))
+
     @patch.object(MockPatroni, 'dcs')
     def test_do_POST_reinitialize(self, mock_dcs):
         cluster = mock_dcs.get_cluster.return_value
@@ -403,22 +530,76 @@ class TestRestApiHandler(unittest.TestCase):
         MockRestApiServer(RestApiHandler, post + '37\n\n{"candidate":"2","scheduled_at": "1"}')
 
 
-@patch('ssl.SSLContext.load_cert_chain', Mock())
-@patch('ssl.SSLContext.wrap_socket', Mock(return_value=0))
-@patch.object(BaseHTTPServer.HTTPServer, '__init__', Mock())
 class TestRestApiServer(unittest.TestCase):
 
+    @patch('ssl.SSLContext.load_cert_chain', Mock())
+    @patch('ssl.SSLContext.set_ciphers', Mock())
+    @patch('ssl.SSLContext.wrap_socket', Mock(return_value=0))
+    @patch.object(BaseHTTPServer.HTTPServer, '__init__', Mock())
+    def setUp(self):
+        self.srv = MockRestApiServer(Mock(), '', {'listen': '*:8008', 'certfile': 'a', 'verify_client': 'required',
+                                                  'ciphers': '!SSLv1:!SSLv2:!SSLv3:!TLSv1:!TLSv1.1',
+                                                  'allowlist': ['127.0.0.1', '::1/128', '::1/zxc'],
+                                                  'allowlist_include_members': True})
+
+    @patch.object(BaseHTTPServer.HTTPServer, '__init__', Mock())
     def test_reload_config(self):
         bad_config = {'listen': 'foo'}
         self.assertRaises(ValueError, MockRestApiServer, None, '', bad_config)
-        srv = MockRestApiServer(lambda a1, a2, a3: None, '')
+        self.assertRaises(ValueError, self.srv.reload_config, bad_config)
-        self.assertRaises(ValueError, srv.reload_config, bad_config)
+        self.assertRaises(ValueError, self.srv.reload_config, {})
-        self.assertRaises(ValueError, srv.reload_config, {})
         with patch.object(socket.socket, 'setsockopt', Mock(side_effect=socket.error)):
-            srv.reload_config({'listen': ':8008'})
+            self.srv.reload_config({'listen': ':8008'})
+
+    @patch.object(MockPatroni, 'dcs')
+    def test_check_access(self, mock_dcs):
+        mock_dcs.cluster = get_cluster_initialized_without_leader()
+        mock_dcs.cluster.members[1].data['api_url'] = 'http://127.0.0.1z:8011/patroni'
+        mock_dcs.cluster.members.append(Member(0, 'bad-api-url', 30, {'api_url': 123}))
+        mock_rh = Mock()
+        mock_rh.client_address = ('127.0.0.2',)
+        self.assertIsNot(self.srv.check_access(mock_rh), True)
+        mock_rh.client_address = ('127.0.0.1',)
+        mock_rh.request.getpeercert.return_value = None
+        self.assertIsNot(self.srv.check_access(mock_rh), True)
 
     def test_handle_error(self):
         try:
             raise Exception()
         except Exception:
             self.assertIsNone(MockRestApiServer.handle_error(None, ('127.0.0.1', 55555)))
+
+    @patch.object(BaseHTTPServer.HTTPServer, '__init__', Mock(side_effect=socket.error))
+    def test_socket_error(self):
+        self.assertRaises(socket.error, MockRestApiServer, Mock(), '', {'listen': '*:8008'})
+
+    @patch.object(MockRestApiServer, 'finish_request', Mock())
+    def test_process_request_thread(self):
+        mock_socket = Mock()
+        self.srv.process_request_thread((mock_socket, 1), '2')
+        mock_socket.context.wrap_socket.side_effect = socket.error
+        self.srv.process_request_thread((mock_socket, 1), '2')
+
+    @patch.object(socket.socket, 'accept')
+    def test_get_request(self, mock_accept):
+        newsock = Mock()
+        mock_accept.return_value = (newsock, '2')
+        self.srv.socket = Mock()
+        self.assertEqual(self.srv.get_request(), ((self.srv.socket, newsock), '2'))
+
+    @patch.object(MockRestApiServer, 'process_request', Mock(side_effect=RuntimeError))
+    def test_process_request_error(self):
+        mock_address = ('127.0.0.1', 55555)
+        mock_socket = Mock()
+        mock_ssl_socket = (Mock(), Mock())
+        for mock_request in (mock_socket, mock_ssl_socket):
+            with patch.object(
+                MockRestApiServer,
+                'get_request',
+                Mock(return_value=(mock_request, mock_address))
+            ):
+                self.srv._handle_request_noblock()
+
+    @patch('ssl._ssl._test_decode_cert', Mock())
+    def test_reload_local_certificate(self):
+        self.assertTrue(self.srv.reload_local_certificate())

tests/test_aws.py

@@ -1,11 +1,11 @@
 import boto.ec2
 import sys
 import unittest
+import urllib3
 
 from mock import Mock, patch
 from collections import namedtuple
 from patroni.scripts.aws import AWSConnection, main as _main
-from requests.exceptions import RequestException
 
 
 class MockEc2Connection(object):
@@ -22,28 +22,11 @@ class MockEc2Connection(object):
         return True
 
 
-class MockResponse(object):
-    ok = True
-
-    def __init__(self, content):
-        self.content = content
-
-    def json(self):
-        return self.content
-
-
-def requests_get(url, **kwargs):
-    if url.split('/')[-1] == 'document':
-        result = {"instanceId": "012345", "region": "eu-west-1"}
-    else:
-        result = 'foo'
-    return MockResponse(result)
-
-
 @patch('boto.ec2.connect_to_region', Mock(return_value=MockEc2Connection()))
 class TestAWSConnection(unittest.TestCase):
 
-    @patch('requests.get', requests_get)
+    @patch('patroni.scripts.aws.requests_get', Mock(return_value=urllib3.HTTPResponse(
+        status=200, body=b'{"instanceId": "012345", "region": "eu-west-1"}')))
     def setUp(self):
         self.conn = AWSConnection('test')
 
@@ -53,17 +36,18 @@ class TestAWSConnection(unittest.TestCase):
             self.conn._retry.max_tries = 1
             self.assertFalse(self.conn.on_role_change('master'))
 
-    @patch('requests.get', Mock(side_effect=RequestException('foo')))
+    @patch('patroni.scripts.aws.requests_get', Mock(side_effect=Exception('foo')))
     def test_non_aws(self):
         conn = AWSConnection('test')
         self.assertFalse(conn.on_role_change("master"))
 
-    @patch('requests.get', Mock(return_value=MockResponse('foo')))
+    @patch('patroni.scripts.aws.requests_get', Mock(return_value=urllib3.HTTPResponse(status=200, body=b'foo')))
     def test_aws_bizare_response(self):
         conn = AWSConnection('test')
         self.assertFalse(conn.aws_available())
 
-    @patch('requests.get', requests_get)
+    @patch('patroni.scripts.aws.requests_get', Mock(return_value=urllib3.HTTPResponse(
+        status=200, body=b'{"instanceId": "012345", "region": "eu-west-1"}')))
     @patch('sys.exit', Mock())
     def test_main(self):
         self.assertIsNone(_main())

tests/test_bootstrap.py

@@ -16,6 +16,7 @@ from . import psycopg2_connect, BaseTestPostgresql
 @patch('os.rename', Mock())
 class TestBootstrap(BaseTestPostgresql):
 
+    @patch('patroni.postgresql.CallbackExecutor', Mock())
     def setUp(self):
         super(TestBootstrap, self).setUp()
         self.b = self.p.bootstrap
@@ -101,6 +102,9 @@ class TestBootstrap(BaseTestPostgresql):
     @patch.object(CancellableSubprocess, 'call', Mock())
     @patch.object(Postgresql, 'is_running', Mock(return_value=True))
     @patch.object(Postgresql, 'data_directory_empty', Mock(return_value=False))
+    @patch.object(Postgresql, 'controldata', Mock(return_value={'max_connections setting': 100,
+                                                                'max_prepared_xacts setting': 0,
+                                                                'max_locks_per_xact setting': 64}))
     def test_bootstrap(self):
         with patch('subprocess.call', Mock(return_value=1)):
             self.assertFalse(self.b.bootstrap({}))
@@ -108,7 +112,8 @@ class TestBootstrap(BaseTestPostgresql):
         config = {'users': {'replicator': {'password': 'rep-pass', 'options': ['replication']}}}
 
         with patch.object(Postgresql, 'is_running', Mock(return_value=False)),\
-                patch('multiprocessing.Process', Mock(side_effect=Exception)):
+                patch('multiprocessing.Process', Mock(side_effect=Exception)),\
+                patch('multiprocessing.get_context', Mock(side_effect=Exception), create=True):
             self.assertRaises(Exception, self.b.bootstrap, config)
         with open(os.path.join(self.p.data_dir, 'pg_hba.conf')) as f:
             lines = f.readlines()
@@ -126,6 +131,7 @@ class TestBootstrap(BaseTestPostgresql):
 
     @patch.object(CancellableSubprocess, 'call')
     @patch.object(Postgresql, 'get_major_version', Mock(return_value=90600))
+    @patch.object(Postgresql, 'controldata',  Mock(return_value={'Database cluster state': 'in production'}))
     def test_custom_bootstrap(self, mock_cancellable_subprocess_call):
         self.p.config._config.pop('pg_hba')
         config = {'method': 'foo', 'foo': {'command': 'bar'}}
@@ -135,6 +141,7 @@ class TestBootstrap(BaseTestPostgresql):
 
         mock_cancellable_subprocess_call.return_value = 0
         with patch('multiprocessing.Process', Mock(side_effect=Exception("42"))),\
+                patch('multiprocessing.get_context', Mock(side_effect=Exception("42")), create=True),\
                 patch('os.path.isfile', Mock(return_value=True)),\
                 patch('os.unlink', Mock()),\
                 patch.object(ConfigHandler, 'save_configuration_files', Mock()),\
@@ -202,13 +209,13 @@ class TestBootstrap(BaseTestPostgresql):
         mock_cancellable_subprocess_call.assert_called()
         args, kwargs = mock_cancellable_subprocess_call.call_args
         self.assertTrue('PGPASSFILE' in kwargs['env'])
-        self.assertEqual(args[0], ['/bin/false', 'postgres://127.0.0.2:5432/postgres'])
+        self.assertEqual(args[0], ['/bin/false', 'dbname=postgres host=127.0.0.2 port=5432'])
 
         mock_cancellable_subprocess_call.reset_mock()
         self.p.config._local_address.pop('host')
         self.assertTrue(self.b.call_post_bootstrap({'post_init': '/bin/false'}))
         mock_cancellable_subprocess_call.assert_called()
-        self.assertEqual(mock_cancellable_subprocess_call.call_args[0][0], ['/bin/false', 'postgres://:5432/postgres'])
+        self.assertEqual(mock_cancellable_subprocess_call.call_args[0][0], ['/bin/false', 'dbname=postgres port=5432'])
 
         mock_cancellable_subprocess_call.side_effect = OSError
         self.assertFalse(self.b.call_post_bootstrap({'post_init': '/bin/false'}))

tests/test_callback_executor.py

@@ -1,3 +1,4 @@
+import psutil
 import unittest
 
 from mock import Mock, patch
@@ -6,22 +7,30 @@ from patroni.postgresql.callback_executor import CallbackExecutor
 
 class TestCallbackExecutor(unittest.TestCase):
 
-    @patch('subprocess.Popen')
+    @patch('psutil.Popen')
     def test_callback_executor(self, mock_popen):
-        mock_popen.return_value.wait.side_effect = Exception
+        mock_popen.return_value.children.return_value = []
-        mock_popen.return_value.poll.return_value = None
+        mock_popen.return_value.is_running.return_value = True
 
         ce = CallbackExecutor()
+        ce._kill_children = Mock(side_effect=Exception)
+        ce._invoke_excepthook = Mock()
         self.assertIsNone(ce.call([]))
         ce.join()
 
         self.assertIsNone(ce.call([]))
 
-        mock_popen.return_value.kill.side_effect = OSError
+        mock_popen.return_value.kill.side_effect = psutil.AccessDenied()
+        self.assertIsNone(ce.call([]))
+
+        ce._process_children = []
+        mock_popen.return_value.children.side_effect = psutil.Error()
+        mock_popen.return_value.kill.side_effect = psutil.NoSuchProcess(123)
         self.assertIsNone(ce.call([]))
 
         mock_popen.side_effect = Exception
         ce = CallbackExecutor()
-        ce._callback_event.wait = Mock(side_effect=[None, Exception])
+        ce._condition.wait = Mock(side_effect=[None, Exception])
+        ce._invoke_excepthook = Mock()
         self.assertIsNone(ce.call([]))
         ce.join()

tests/test_cancellable.py

@@ -1,6 +1,7 @@
+import psutil
 import unittest
 
-from mock import Mock, PropertyMock, patch
+from mock import Mock, patch
 from patroni.exceptions import PostgresException
 from patroni.postgresql.cancellable import CancellableSubprocess
 
@@ -12,12 +13,22 @@ class TestCancellableSubprocess(unittest.TestCase):
 
     def test_call(self):
         self.c.cancel()
-        self.assertRaises(PostgresException, self.c.call, communicate_input=None)
+        self.assertRaises(PostgresException, self.c.call)
+
+    def test__kill_children(self):
+        self.c._process_children = [Mock()]
+        self.c._kill_children()
+        self.c._process_children[0].kill.side_effect = psutil.AccessDenied()
+        self.c._kill_children()
+        self.c._process_children[0].kill.side_effect = psutil.NoSuchProcess(123)
+        self.c._kill_children()
 
     @patch('patroni.postgresql.cancellable.polling_loop', Mock(return_value=[0, 0]))
     def test_cancel(self):
         self.c._process = Mock()
-        self.c._process.returncode = None
+        self.c._process.is_running.return_value = True
+        self.c._process.children.side_effect = psutil.Error()
+        self.c._process.suspend.side_effect = psutil.Error()
         self.c.cancel()
-        type(self.c._process).returncode = PropertyMock(side_effect=[None, -15])
+        self.c._process.is_running.side_effect = [True, False]
         self.c.cancel()

tests/test_config.py

@@ -1,9 +1,10 @@
 import os
 import sys
 import unittest
+import io
 
 from mock import MagicMock, Mock, patch
-from patroni.config import Config
+from patroni.config import Config, ConfigParseError
 from six.moves import builtins
 
 
@@ -15,10 +16,7 @@ class TestConfig(unittest.TestCase):
     def setUp(self):
         sys.argv = ['patroni.py']
         os.environ[Config.PATRONI_CONFIG_VARIABLE] = 'restapi: {}\npostgresql: {data_dir: foo}'
-        self.config = Config()
+        self.config = Config(None)
-
-    def test_no_config(self):
-        self.assertRaises(SystemExit, Config)
 
     def test_set_dynamic_configuration(self):
         with patch.object(Config, '_build_effective_configuration', Mock(side_effect=Exception)):
@@ -32,12 +30,14 @@ class TestConfig(unittest.TestCase):
             'PATRONI_SCOPE': 'batman2',
             'PATRONI_LOGLEVEL': 'ERROR',
             'PATRONI_LOG_LOGGERS': 'patroni.postmaster: WARNING, urllib3: DEBUG',
+            'PATRONI_LOG_FILE_NUM': '5',
             'PATRONI_RESTAPI_USERNAME': 'username',
             'PATRONI_RESTAPI_PASSWORD': 'password',
             'PATRONI_RESTAPI_LISTEN': '0.0.0.0:8008',
             'PATRONI_RESTAPI_CONNECT_ADDRESS': '127.0.0.1:8008',
             'PATRONI_RESTAPI_CERTFILE': '/certfile',
             'PATRONI_RESTAPI_KEYFILE': '/keyfile',
+            'PATRONI_RESTAPI_ALLOWLIST_INCLUDE_MEMBERS': 'on',
             'PATRONI_POSTGRESQL_LISTEN': '0.0.0.0:5432',
             'PATRONI_POSTGRESQL_CONNECT_ADDRESS': '127.0.0.1:5432',
             'PATRONI_POSTGRESQL_DATA_DIR': 'data/postgres0',
@@ -58,6 +58,7 @@ class TestConfig(unittest.TestCase):
             'PATRONI_ZOOKEEPER_HOSTS': "'host1:2181','host2:2181'",
             'PATRONI_EXHIBITOR_HOSTS': 'host1,host2',
             'PATRONI_EXHIBITOR_PORT': '8181',
+            'PATRONI_RAFT_PARTNER_ADDRS': "'host1:1234','host2:1234'",
             'PATRONI_foo_HOSTS': '[host1,host2',  # Exception in parse_list
             'PATRONI_SUPERUSER_USERNAME': 'postgres',
             'PATRONI_SUPERUSER_PASSWORD': 'zalando',
@@ -66,12 +67,10 @@ class TestConfig(unittest.TestCase):
             'PATRONI_admin_PASSWORD': 'admin',
             'PATRONI_admin_OPTIONS': 'createrole,createdb'
         })
-        sys.argv = ['patroni.py', 'postgres0.yml']
+        config = Config('postgres0.yml')
-        config = Config()
         with patch.object(Config, '_load_config_file', Mock(return_value={'restapi': {}})):
             with patch.object(Config, '_build_effective_configuration', Mock(side_effect=Exception)):
-                self.assertRaises(Exception, config.reload_local_configuration, True)
+                config.reload_local_configuration()
-            self.assertTrue(config.reload_local_configuration(True))
             self.assertTrue(config.reload_local_configuration())
             self.assertIsNone(config.reload_local_configuration())
 
@@ -99,3 +98,46 @@ class TestConfig(unittest.TestCase):
         self.config.set_dynamic_configuration(dynamic_configuration)
         for name, value in dynamic_configuration['standby_cluster'].items():
             self.assertEqual(self.config['standby_cluster'][name], value)
+
+    @patch('os.path.exists', Mock(return_value=True))
+    @patch('os.path.isfile', Mock(side_effect=lambda fname: fname != 'postgres0'))
+    @patch('os.path.isdir', Mock(return_value=True))
+    @patch('os.listdir', Mock(return_value=['01-specific.yml', '00-base.yml']))
+    def test_configuration_directory(self):
+        def open_mock(fname, *args, **kwargs):
+            if fname.endswith('00-base.yml'):
+                return io.StringIO(
+                    u'''
+                    test: True
+                    test2:
+                      child-1: somestring
+                      child-2: 5
+                      child-3: False
+                    test3: True
+                    test4:
+                     - abc: 3
+                     - abc: 4
+                    ''')
+            elif fname.endswith('01-specific.yml'):
+                return io.StringIO(
+                    u'''
+                    test: False
+                    test2:
+                      child-2: 10
+                      child-3: !!null
+                    test4:
+                     - ab: 5
+                    new-attr: True
+                    ''')
+
+        with patch.object(builtins, 'open', MagicMock(side_effect=open_mock)):
+            config = Config('postgres0')
+            self.assertEqual(config._local_configuration,
+                             {'test': False, 'test2': {'child-1': 'somestring', 'child-2': 10},
+                              'test3': True, 'test4': [{'ab': 5}], 'new-attr': True})
+
+    @patch('os.path.exists', Mock(return_value=True))
+    @patch('os.path.isfile', Mock(return_value=False))
+    @patch('os.path.isdir', Mock(return_value=False))
+    def test_invalid_path(self):
+        self.assertRaises(ConfigParseError, Config, 'postgres0')