Merge branch 'master' of github.com:zalando/patroni into feature/terminaltables

PostgreSQL 13 finally introduced the possibility to change the `primary_conninfo` without a restart. Just doing reload is enough, but in case if the role is changing from the `replica` to the `standby_leader` we want to call only `on_role_change` callback and skip `on_reload`, because they duplicate each other.

.gitignore

@@ -51,3 +51,9 @@ scm-source.json
 docs/build/
 docs/source/_static/
 docs/source/_templates/
+
+# Pycharm IDE
+.idea/
+
+#VSCode IDE
+.vscode/

.travis.yml

@@ -1,9 +1,13 @@
 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
+  - ETCDVERSION=3.0.17 ZKVERSION=3.4.14 CONSULVERSION=0.7.4
   - PYVERSIONS="2.7 3.5 3.6"
   - EXCLUDE_BEHAVE="3.5"
   - BOTO_CONFIG=/doesnotexist
@@ -115,7 +119,7 @@ install:
             fi
             source ~/virtualenv/python${pv}/bin/activate
             # explicitly install all needed python modules to cache them
-            for p in '-r requirements.txt' 'behave codacy-coverage coverage coveralls flake8 mock pytest-cov pytest setuptools'; do
+            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
@@ -131,13 +135,13 @@ script:
 
         if [[ $TEST_SUITE != "behave" ]]; then
             echo Running unit tests using python${pv}
-            $TEST_SUITE test
+            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 $TEST_SUITE; then
+            if ! PATH=.:/usr/lib/postgresql/9.6/bin:$PATH unbuffer $TEST_SUITE; then
                 # output all log files when tests are failing
-                grep . features/output/*/*postgres?.*
+                grep . features/output/*_failed/*postgres?.*
                 exit 1
             fi
         fi

Dockerfile

@@ -1,45 +1,158 @@
 ## This Dockerfile is meant to aid in the building and debugging patroni whilst developing on your local machine
 ## It has all the necessary components to play/debug with a single node appliance, running etcd
-FROM postgres:10
-MAINTAINER Alexander Kukushkin <[email protected]>
+ARG PG_MAJOR=10
+ARG COMPRESS=false
+ARG PGHOME=/home/postgres
+ARG PGDATA=$PGHOME/data
+ARG LC_ALL=C.UTF-8
+ARG LANG=C.UTF-8
 
-RUN export DEBIAN_FRONTEND=noninteractive \
+FROM postgres:$PG_MAJOR as builder
+
+ARG PGHOME
+ARG PGDATA
+ARG LC_ALL
+ARG LANG
+
+ENV ETCDVERSION=2.3.8 CONFDVERSION=0.16.0
+
+RUN set -ex \
+    && export DEBIAN_FRONTEND=noninteractive \
     && echo 'APT::Install-Recommends "0";\nAPT::Install-Suggests "0";' > /etc/apt/apt.conf.d/01norecommend \
     && apt-get update -y \
-    && apt-get upgrade -y \
-    # postgres:10 is based on debian, which has patroni package. We will install all required dependencies
-    && apt-get install -s patroni | sed -n -e '/^Inst patroni /d' -e 's/^Inst \([^ ]\+\) .*$/\1/p' \
-            | xargs apt-get install -y curl jq haproxy locales python3-etcd python3-kazoo \
-
-    ## Make sure we have a en_US.UTF-8 locale available
+    # postgres:10 is based on debian, which has the patroni package. We will install all required dependencies
+    && apt-cache depends patroni | sed -n -e 's/.*Depends: \(python3-.\+\)$/\1/p' \
+            | 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 \
+    && echo 'en_US.UTF-8 UTF-8' > /usr/share/i18n/SUPPORTED \
+\
+    # Make sure we have a en_US.UTF-8 locale available
     && localedef -i en_US -c -f UTF-8 -A /usr/share/locale/locale.alias en_US.UTF-8 \
-
-    && mkdir -p /home/postgres \
-    && chown postgres:postgres /home/postgres \
-
-    # Clean up
-    && apt-get purge -y libpython2.7-stdlib libpython2.7-minimal \
+\
+    # haproxy dummy config
+    && echo 'global\n        stats socket /run/haproxy/admin.sock mode 660 level admin' > /etc/haproxy/haproxy.cfg \
+\
+    # vim config
+    && echo 'syntax on\nfiletype plugin indent on\nset mouse-=a\nautocmd FileType yaml setlocal ts=2 sts=2 sw=2 expandtab' > /etc/vim/vimrc.local \
+\
+    # Prepare postgres/patroni/haproxy environment
+    && mkdir -p $PGHOME/.config/patroni /patroni /run/haproxy \
+    && ln -s ../../postgres0.yml $PGHOME/.config/patroni/patronictl.yaml \
+    && ln -s /patronictl.py /usr/local/bin/patronictl \
+    && sed -i "s|/var/lib/postgresql.*|$PGHOME:/bin/bash|" /etc/passwd \
+    && chown -R postgres:postgres /var/log \
+\
+    # Download etcd
+    && curl -sL https://github.com/coreos/etcd/releases/download/v${ETCDVERSION}/etcd-v${ETCDVERSION}-linux-amd64.tar.gz \
+            | tar xz -C /usr/local/bin --strip=1 --wildcards --no-anchored etcd etcdctl \
+\
+    # Download confd
+    && curl -sL https://github.com/kelseyhightower/confd/releases/download/v${CONFDVERSION}/confd-${CONFDVERSION}-linux-amd64 \
+            > /usr/local/bin/confd && chmod +x /usr/local/bin/confd \
+\
+    # Clean up all useless packages and some files
+    && apt-get purge -y --allow-remove-essential python3-pip gzip bzip2 util-linux e2fsprogs \
+                libmagic1 bsdmainutils login ncurses-bin libmagic-mgc e2fslibs bsdutils \
+                exim4-config gnupg-agent dirmngr libpython2.7-stdlib libpython2.7-minimal \
     && apt-get autoremove -y \
     && apt-get clean -y \
-    && rm -rf /var/lib/apt/lists/* /root/.cache
+    && rm -rf /var/lib/apt/lists/* \
+        /root/.cache \
+        /var/cache/debconf/* \
+        /etc/rc?.d \
+        /etc/systemd \
+        /docker-entrypoint* \
+        /sbin/pam* \
+        /sbin/swap* \
+        /sbin/unix* \
+        /usr/local/bin/gosu \
+        /usr/sbin/[acgipr]* \
+        /usr/sbin/*user* \
+        /usr/share/doc* \
+        /usr/share/man \
+        /usr/share/info \
+        /usr/share/i18n/locales/translit_hangul \
+        /usr/share/locale/?? \
+        /usr/share/locale/??_?? \
+        /usr/share/postgresql/*/man \
+        /usr/share/postgresql-common/pg_wrapper \
+        /usr/share/vim/vim80/doc \
+        /usr/share/vim/vim80/lang \
+        /usr/share/vim/vim80/tutor \
+#        /var/lib/dpkg/info/* \
+    && find /usr/bin -xtype l -delete \
+    && find /var/log -type f -exec truncate --size 0 {} \; \
+    && find /usr/lib/python3/dist-packages -name '*test*' | xargs rm -fr \
+    && find /lib/x86_64-linux-gnu/security -type f ! -name pam_env.so ! -name pam_permit.so ! -name pam_unix.so -delete
 
-ENV ETCDVERSION 3.2.23
-RUN curl -L https://github.com/coreos/etcd/releases/download/v${ETCDVERSION}/etcd-v${ETCDVERSION}-linux-amd64.tar.gz \
-    | tar xz -C /usr/local/bin --strip=1 --wildcards --no-anchored etcd etcdctl
+# perform compression if it is necessary
+ARG COMPRESS
+RUN if [ "$COMPRESS" = "true" ]; then \
+        set -ex \
+        # Allow certain sudo commands from postgres
+        && echo 'postgres ALL=(ALL) NOPASSWD: /bin/tar xpJf /a.tar.xz -C /, /bin/rm /a.tar.xz, /bin/ln -snf dash /bin/sh' >> /etc/sudoers \
+        && ln -snf busybox /bin/sh \
+        && files="/bin/sh /usr/bin/sudo /usr/lib/sudo/sudoers.so /lib/x86_64-linux-gnu/security/pam_*.so" \
+        && libs="$(ldd $files | awk '{print $3;}' | grep '^/' | sort -u) /lib/x86_64-linux-gnu/ld-linux-x86-64.so.* /lib/x86_64-linux-gnu/libnsl.so.* /lib/x86_64-linux-gnu/libnss_compat.so.*" \
+        && (echo /var/run $files $libs | tr ' ' '\n' && realpath $files $libs) | sort -u | sed 's/^\///' > /exclude \
+        && find /etc/alternatives -xtype l -delete \
+        && save_dirs="usr lib var bin sbin etc/ssl etc/init.d etc/alternatives etc/apt" \
+        && XZ_OPT=-e9v tar -X /exclude -cpJf a.tar.xz $save_dirs \
+        # we call "cat /exclude" to avoid including files from the $save_dirs that are also among
+        # the exceptions listed in the /exclude, as "uniq -u" eliminates all non-unique lines.
+        # By calling "cat /exclude" a second time we guarantee that there will be at least two lines
+        # for each exception and therefore they will be excluded from the output passed to 'rm'.
+        && /bin/busybox sh -c "(find $save_dirs -not -type d && cat /exclude /exclude && echo exclude) | sort | uniq -u | xargs /bin/busybox rm" \
+        && /bin/busybox --install -s \
+        && /bin/busybox sh -c "find $save_dirs -type d -depth -exec rmdir -p {} \; 2> /dev/null"; \
+    fi
 
-ENV CONFDVERSION 0.16.0
-RUN curl -L https://github.com/kelseyhightower/confd/releases/download/v${CONFDVERSION}/confd-${CONFDVERSION}-linux-amd64 > /usr/local/bin/confd \
-    && chmod +x /usr/local/bin/confd
+FROM scratch
+COPY --from=builder / /
 
-ADD patronictl.py patroni.py docker/entrypoint.sh /
-ADD patroni /patroni/
-ADD extras/confd /etc/confd
+LABEL maintainer="Alexander Kukushkin <[email protected]>"
 
-RUN sed -i 's/env python/&3/' patroni*.py && ln -s /patronictl.py /usr/local/bin/patronictl && mkdir /data/ /run/haproxy \
-    && touch /pgpass /patroni.yml && chown postgres:postgres -R /patroni/ /data/ /pgpass /patroni.yml /etc/haproxy /var/run/ /var/lib/ /var/log/
+ARG PG_MAJOR
+ARG COMPRESS
+ARG PGHOME
+ARG PGDATA
+ARG LC_ALL
+ARG LANG
 
-EXPOSE 2379 5432 8008
+ARG PGBIN=/usr/lib/postgresql/$PG_MAJOR/bin
+
+ENV LC_ALL=$LC_ALL LANG=$LANG EDITOR=/usr/bin/editor
+ENV PGDATA=$PGDATA PATH=$PATH:$PGBIN
+
+COPY patroni /patroni/
+COPY extras/confd/conf.d/haproxy.toml /etc/confd/conf.d/
+COPY extras/confd/templates/haproxy.tmpl /etc/confd/templates/
+COPY patroni*.py docker/entrypoint.sh /
+COPY postgres?.yml $PGHOME/
+
+WORKDIR $PGHOME
+
+RUN sed -i 's/env python/&3/' /patroni*.py \
+    # "fix" patroni configs
+    && sed -i 's/^\(  connect_address:\|  - host\)/#&/' postgres?.yml \
+    && sed -i 's/^  listen: 127.0.0.1/  listen: 0.0.0.0/' postgres?.yml \
+    && sed -i "s|^\(  data_dir: \).*|\1$PGDATA|" postgres?.yml \
+    && sed -i "s|^#\(  bin_dir: \).*|\1$PGBIN|" postgres?.yml \
+    && sed -i 's/^  - encoding: UTF8/  - locale: en_US.UTF-8\n&/' postgres?.yml \
+    && sed -i 's/^\(scope\|name\|etcd\|  host\|  authentication\|  pg_hba\|  parameters\):/#&/' postgres?.yml \
+    && 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
 
-ENV LC_ALL=en_US.UTF-8 LANG=en_US.UTF-8
-ENTRYPOINT ["/bin/bash", "/entrypoint.sh"]
 USER postgres
+
+ENTRYPOINT ["/bin/sh", "/entrypoint.sh"]

README.rst

@@ -59,6 +59,33 @@ To install requirements on a Mac, run the following:
 
     brew install postgresql etcd haproxy libyaml python
 
+**Psycopg2**
+
+Starting from `psycopg2-2.8 <http://initd.org/psycopg/articles/2019/04/04/psycopg-28-released/>`__ the binary version of psycopg2 will no longer be installed by default. Installing it from the source code requires C compiler and postgres+python dev packages.
+Since in the python world it is not possible to specify dependency as ``psycopg2 OR psycopg2-binary`` you will have to decide how to install it.
+
+There are a few options available:
+
+1. Use the package manager from your distro
+
+::
+
+    sudo apt-get install python-psycopg2   # install python2 psycopg2 module on Debian/Ubuntu
+    sudo apt-get install python3-psycopg2  # install python3 psycopg2 module on Debian/Ubuntu
+    sudo yum install python-psycopg2       # install python2 psycopg2 on RedHat/Fedora/CentOS
+
+2. Install psycopg2 from the binary package
+
+::
+
+    pip install psycopg2-binary
+
+3. Install psycopg2 from source
+
+::
+
+    pip install psycopg2>=2.5.4
+
 **General installation for pip**
 
 Patroni can be installed with pip:
@@ -97,7 +124,7 @@ 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
 

docker-compose.yml

@@ -1,58 +1,69 @@
 # docker compose file for running a 3-node PostgreSQL cluster
-# with etcd as the SIS
+# with 3-node etcd cluster as the DCS and one haproxy node
+version: "2"
 
-patroni_etcd:
-    container_name: patroni_etcd
-    image: patroni
-    command: --etcd
+networks:
+    demo:
 
-dbnode1:
-    image: patroni
-    hostname: dbnode1
-    links:
-        - patroni_etcd:patroni_etcd
-    volumes:
-        - ./patroni:/patroni
-    env_file: docker/patroni-secrets.env
-    environment:
-        PATRONI_ETCD_URL: http://patroni_etcd:2379
-        PATRONI_NAME: dbnode1
-        PATRONI_SCOPE: testcluster
+services:
+    etcd1:
+        image: patroni
+        networks: [ demo ]
+        env_file: docker/etcd.env
+        container_name: demo-etcd1
+        hostname: etcd1
+        command: etcd -name etcd1 -initial-advertise-peer-urls http://etcd1:2380
 
-dbnode2:
-    image: patroni
-    hostname: dbnode2
-    links:
-        - patroni_etcd:patroni_etcd
-    volumes:
-        - ./patroni:/patroni
-    env_file: docker/patroni-secrets.env
-    environment:
-        PATRONI_ETCD_URL: http://patroni_etcd:2379
-        PATRONI_NAME: dbnode2
-        PATRONI_SCOPE: testcluster
+    etcd2:
+        image: patroni
+        networks: [ demo ]
+        env_file: docker/etcd.env
+        container_name: demo-etcd2
+        hostname: etcd2
+        command: etcd -name etcd2 -initial-advertise-peer-urls http://etcd2:2380
 
-dbnode3:
-    image: patroni
-    hostname: dbnode3
-    links:
-        - patroni_etcd:patroni_etcd
-    volumes:
-        - ./patroni:/patroni
-    env_file: docker/patroni-secrets.env
-    environment:
-        PATRONI_ETCD_URL: http://patroni_etcd:2379
-        PATRONI_NAME: dbnode3
-        PATRONI_SCOPE: testcluster
+    etcd3:
+        image: patroni
+        networks: [ demo ]
+        env_file: docker/etcd.env
+        container_name: demo-etcd3
+        hostname: etcd3
+        command: etcd -name etcd3 -initial-advertise-peer-urls http://etcd3:2380
 
-haproxy:
-    image: patroni
-    links:
-        - patroni_etcd:patroni_etcd
-    ports:
-        - "5000:5000"
-        - "5001:5001"
-    environment:
-        PATRONI_ETCD_URL: http://patroni_etcd:2379
-        PATRONI_SCOPE: testcluster
-    command: --confd
+    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 ]
+        env_file: docker/patroni.env
+        hostname: haproxy
+        container_name: demo-haproxy
+        ports:
+            - "5000:5000"
+            - "5001:5001"
+        command: haproxy

docker/README.md

@@ -1,47 +1,117 @@
 # Patroni Dockerfile
-You can run Patroni in a docker container using this Dockerfile, or by using one of the Docker image at
-
-    https://registry.opensource.zalan.do/v1/repositories/acid/patroni/tags
+You can run Patroni in a docker container using this Dockerfile
 
 This Dockerfile is meant in aiding development of Patroni and quick testing of features. It is not a production-worthy
 Dockerfile
 
+    docker build -t patroni .
+
 # Examples
 
 ## Standalone Patroni
 
-    docker run -d registry.opensource.zalan.do/acid/patroni:1.0-SNAPSHOT
+    docker run -d patroni
 
-## Multiple Patroni's communicating with a standalone etcd inside Docker
-
-Basically what you would do would be: 
-
-* Run 1 container which provides etcd
-
-        docker run -d <IMAGE> --etcd-only
-
-* Run n containers running Patroni, passing the `--etcd` option to the `docker run` command
-
-        docker run -d <IMAGE> --etcd=<IP FROM etcd CONTAINER:PORT>
-
-To automate this you can run the following script:
-
-    dev_patroni_cluster.sh [OPTIONS]
-
-    Options:
-
-        --image     IMAGE    The Docker image to use for the cluster
-        --members   INT      The number of members for the cluster
-        --name      NAME     The name of the new cluster
+## Three-node Patroni cluster with three-node etcd cluster and one haproxy container using docker-compose
 
 Example session:
 
-    $ ./dev_patroni_cluster.sh --image registry.opensource.zalan.do/acid/patroni:1.0-SNAPSHOT --members=2 --name=bravo
-    The etcd container is 6be871a11cb373406ca5ea1c6b39e1.0-SNAPSHOTfdde9fb1d6177212d6ad0c0d1bd9b563, ip=172.17.1.24
-    Started Patroni container 67e611f2eca7c40f9e6e0e24a4a8f2cba7e3e56d22a420e15ab9240a37a9d7a4, ip=172.17.1.25
-    Started Patroni container 47dd12ae635ab83b039f5889e250048b606ed5e48e3650b69e365e7e1d4acbcf, ip=172.17.1.26
+    $ docker-compose up -d
+    Creating demo-haproxy ...
+    Creating demo-patroni2 ...
+    Creating demo-patroni1 ...
+    Creating demo-patroni3 ...
+    Creating demo-etcd2 ...
+    Creating demo-etcd1 ...
+    Creating demo-etcd3 ...
+    Creating demo-haproxy
+    Creating demo-patroni2
+    Creating demo-patroni1
+    Creating demo-patroni3
+    Creating demo-etcd1
+    Creating demo-etcd2
+    Creating demo-etcd2 ... done
+
     $ docker ps
-    CONTAINER ID        IMAGE                                         COMMAND                CREATED             STATUS              PORTS                          NAMES
-    47dd12ae635a        registry.opensource.zalan.do/acid/patroni:1.0-SNAPSHOT   "/bin/bash /entrypoi   10 seconds ago      Up 8 seconds        4001/tcp, 5432/tcp, 2380/tcp   bravo_OR64g8bx
-    67e611f2eca7        registry.opensource.zalan.do/acid/patroni:1.0-SNAPSHOT   "/bin/bash /entrypoi   11 seconds ago      Up 10 seconds       2380/tcp, 4001/tcp, 5432/tcp   bravo_si9no8iz
-    6be871a11cb3        registry.opensource.zalan.do/acid/patroni:1.0-SNAPSHOT   "/bin/bash /entrypoi   12 seconds ago      Up 10 seconds       4001/tcp, 5432/tcp, 2380/tcp   bravo_etcd
+    CONTAINER ID        IMAGE               COMMAND                  CREATED             STATUS              PORTS                              NAMES
+    5b7a90b4cfbf        patroni             "/bin/sh /entrypoint…"   29 seconds ago      Up 27 seconds                                          demo-etcd2
+    e30eea5222f2        patroni             "/bin/sh /entrypoint…"   29 seconds ago      Up 27 seconds                                          demo-etcd1
+    83bcf3cb208f        patroni             "/bin/sh /entrypoint…"   29 seconds ago      Up 27 seconds                                          demo-etcd3
+    922532c56e7d        patroni             "/bin/sh /entrypoint…"   29 seconds ago      Up 28 seconds                                          demo-patroni3
+    14f875e445f3        patroni             "/bin/sh /entrypoint…"   29 seconds ago      Up 28 seconds                                          demo-patroni2
+    110d1073b383        patroni             "/bin/sh /entrypoint…"   29 seconds ago      Up 28 seconds                                          demo-patroni1
+    5af5e6e36028        patroni             "/bin/sh /entrypoint…"   29 seconds ago      Up 28 seconds       0.0.0.0:5000-5001->5000-5001/tcp   demo-haproxy
+
+    $ docker logs demo-patroni1
+    2019-02-20 08:19:32,714 INFO: Failed to import patroni.dcs.consul
+    2019-02-20 08:19:32,737 INFO: Selected new etcd server http://etcd3:2379
+    2019-02-20 08:19:35,140 INFO: Lock owner: None; I am patroni1
+    2019-02-20 08:19:35,174 INFO: trying to bootstrap a new cluster
+    ...
+    2019-02-20 08:19:39,310 INFO: postmaster pid=37
+    2019-02-20 08:19:39.314 UTC [37] LOG:  listening on IPv4 address "0.0.0.0", port 5432
+    2019-02-20 08:19:39.321 UTC [37] LOG:  listening on Unix socket "/var/run/postgresql/.s.PGSQL.5432"
+    2019-02-20 08:19:39.353 UTC [39] LOG:  database system was shut down at 2019-02-20 08:19:36 UTC
+    2019-02-20 08:19:39.354 UTC [40] FATAL:  the database system is starting up
+    localhost:5432 - rejecting connections
+    2019-02-20 08:19:39.369 UTC [37] LOG:  database system is ready to accept connections
+    localhost:5432 - accepting connections
+    2019-02-20 08:19:39,383 INFO: establishing a new patroni connection to the postgres cluster
+    2019-02-20 08:19:39,408 INFO: running post_bootstrap
+    2019-02-20 08:19:39,432 WARNING: Could not activate Linux watchdog device: "Can't open watchdog device: [Errno 2] No such file or directory: '/dev/watchdog'"
+    2019-02-20 08:19:39,515 INFO: initialized a new cluster
+    2019-02-20 08:19:49,424 INFO: Lock owner: patroni1; I am patroni1
+    2019-02-20 08:19:49,447 INFO: Lock owner: patroni1; I am patroni1
+    2019-02-20 08:19:49,480 INFO: no action.  i am the leader with the lock
+    2019-02-20 08:19:59,422 INFO: Lock owner: patroni1; I am patroni1
+
+    $ docker exec -ti demo-patroni1 bash
+    postgres@patroni1:~$ patronictl list
+    +---------+----------+------------+--------+---------+----+-----------+
+    | Cluster |  Member  |    Host    |  Role  |  State  | TL | Lag in MB |
+    +---------+----------+------------+--------+---------+----+-----------+
+    |   demo  | patroni1 | 172.22.0.3 | Leader | running |  1 |         0 |
+    |   demo  | patroni2 | 172.22.0.7 |        | running |  1 |         0 |
+    |   demo  | patroni3 | 172.22.0.4 |        | running |  1 |         0 |
+    +---------+----------+------------+--------+---------+----+-----------+
+
+    postgres@patroni1:~$ etcdctl ls --recursive --sort -p /service/demo
+    /service/demo/config
+    /service/demo/initialize
+    /service/demo/leader
+    /service/demo/members/
+    /service/demo/members/patroni1
+    /service/demo/members/patroni2
+    /service/demo/members/patroni3
+    /service/demo/optime/
+    /service/demo/optime/leader
+
+    postgres@patroni1:~$ etcdctl member list
+    1bab629f01fa9065: name=etcd3 peerURLs=http://etcd3:2380 clientURLs=http://etcd3:2379 isLeader=false
+    8ecb6af518d241cc: name=etcd2 peerURLs=http://etcd2:2380 clientURLs=http://etcd2:2379 isLeader=true
+    b2e169fcb8a34028: name=etcd1 peerURLs=http://etcd1:2380 clientURLs=http://etcd1:2379 isLeader=false
+    postgres@patroni1:~$ exit
+
+    $ psql -h localhost -p 5000 -U postgres -W
+    Password: postgres
+    psql (11.2 (Ubuntu 11.2-1.pgdg18.04+1), server 10.7 (Debian 10.7-1.pgdg90+1))
+    Type "help" for help.
+
+    localhost/postgres=# select pg_is_in_recovery();
+     pg_is_in_recovery
+    ───────────────────
+     f
+    (1 row)
+
+    localhost/postgres=# \q
+
+    $ psql -h localhost -p 5001 -U postgres -W
+    Password: postgres
+    psql (11.2 (Ubuntu 11.2-1.pgdg18.04+1), server 10.7 (Debian 10.7-1.pgdg90+1))
+    Type "help" for help.
+
+    localhost/postgres=# select pg_is_in_recovery();
+     pg_is_in_recovery
+    ───────────────────
+     t
+    (1 row)

docker/dev_patroni_cluster.sh

@@ -1,104 +0,0 @@
-#!/bin/bash
-
-DOCKER_IMAGE="registry.opensource.zalan.do/acid/patroni:1.0-SNAPSHOT"
-MEMBERS=3
-
-
-function usage()
-{
-    cat <<__EOF__
-Usage: $0 
-
-Options:
-
-    --image     IMAGE    The Docker image to use for the cluster
-    --members   INT      The number of members for the cluster
-    --name      NAME     The name of the new cluster
-
-Examples:
-
-    $0 --image ${DOCKER_IMAGE}
-    $0
-    $0 --image ${DOCKER_IMAGE} --members=2
-__EOF__
-}
-
-
-optspec=":-:"
-while getopts "$optspec" optchar; do
-    case "${optchar}" in
-        -)
-            case "${OPTARG}" in
-                help)
-                    usage
-                    exit 0
-                    ;;
-                name)
-                    PATRONI_SCOPE="${!OPTIND}"; OPTIND=$(( $OPTIND + 1 ))
-                    ;;
-                name=*)
-                    PATRONI_SCOPE="${OPTARG#*=}"
-                    ;;
-                image)
-                    DOCKER_IMAGE="${!OPTIND}"; OPTIND=$(( $OPTIND + 1 ))
-                    ;;
-                image=*)
-                    DOCKER_IMAGE="${OPTARG#*=}"
-                    ;;
-                members)
-                    MEMBERS="${!OPTIND}"; OPTIND=$(( $OPTIND + 1 ))
-                    ;;
-                members=*)
-                    MEMBERS="${OPTARG#*=}"
-                    ;;
-                *)
-                    if [ "$OPTERR" = 1 ] && [ "${optspec:0:1}" != ":" ]; then
-                        echo "Unknown option --${OPTARG}" >&2
-                    fi
-                    ;;
-            esac;;
-        *)
-            if [ "$OPTERR" != 1 ] || [ "${optspec:0:1}" = ":" ]; then
-                echo "Non-option argument: '-${OPTARG}'" >&2
-                usage
-                exit 1
-            fi
-            ;;
-    esac
-done
-
-if [ -z ${PATRONI_SCOPE} ]; then
-    PATRONI_SCOPE=$(cat /dev/urandom | LC_ALL=C tr -dc 'a-z0-9' | head -c 8)
-fi
-
-function docker_run()
-{
-    local name=$1
-    shift
-    container=$(docker run -d --name=$name $*)
-    container_ip=$(docker inspect --format '{{ .NetworkSettings.IPAddress }}' ${container})
-    echo "Started container ${name}, ip=${container_ip}"
-}
-
-
-ETCD_CONTAINER="${PATRONI_SCOPE}_etcd"
-docker_run ${ETCD_CONTAINER} ${DOCKER_IMAGE} --etcd
-
-DOCKER_ARGS="--link=${ETCD_CONTAINER}:${ETCD_CONTAINER} -e PATRONI_SCOPE=${PATRONI_SCOPE} -e PATRONI_ETCD_URL=http://${ETCD_CONTAINER}:2379"
-PATRONI_ENV=$(sed 's/#.*//g' docker/patroni-secrets.env | sed -n 's/^PATRONI_.*$/-e &/p' | tr '\n' ' ')
-PATRONI_VOLUME="-v $(dirname $(dirname $(realpath $0)))/patroni:/patroni"
-
-for i in $(seq 1 "${MEMBERS}"); do
-    container_name=postgres${i}
-    docker_run "${PATRONI_SCOPE}_${container_name}" \
-        $PATRONI_VOLUME \
-        $DOCKER_ARGS \
-        $PATRONI_ENV \
-        -e PATRONI_NAME=${container_name} \
-        ${DOCKER_IMAGE}
-done
-
-docker_run "${PATRONI_SCOPE}_haproxy" \
-    -p=5000 -p=5001 \
-    $DOCKER_ARGS \
-    ${DOCKER_IMAGE} --confd

docker/entrypoint.sh

@@ -1,115 +1,60 @@
-#!/bin/bash
+#!/bin/sh
 
-function usage()
-{
-    cat <<__EOF__
-Usage: $0
+if [ -f /a.tar.xz ]; then
+    echo "decompressing image..."
+    sudo tar xpJf /a.tar.xz -C / > /dev/null 2>&1
+    sudo rm /a.tar.xz
+    sudo ln -snf dash /bin/sh
+fi
 
-Options:
+readonly PATRONI_SCOPE=${PATRONI_SCOPE:-batman}
+PATRONI_NAMESPACE=${PATRONI_NAMESPACE:-/service}
+readonly PATRONI_NAMESPACE=${PATRONI_NAMESPACE%/}
+readonly DOCKER_IP=$(hostname --ip-address)
 
-    --etcd      Do not run Patroni, run a standalone etcd
-    --confd     Do not run Patroni, run a standalone confd
-    --zookeeper Do not run Patroni, run a standalone zookeeper
-
-Examples:
-
-    $0 --etcd
-    $0 --confd
-    $0 --zookeeper
-    $0
-__EOF__
-}
-
-DOCKER_IP=$(hostname --ip-address)
-PATRONI_SCOPE=${PATRONI_SCOPE:-batman}
-ETCD_ARGS="--data-dir /tmp/etcd.data -advertise-client-urls=http://${DOCKER_IP}:2379 -listen-client-urls=http://0.0.0.0:2379"
-
-optspec=":vh-:"
-while getopts "$optspec" optchar; do
-    case "${optchar}" in
-        -)
-            case "${OPTARG}" in
-                confd)
-                    haproxy -f /etc/haproxy/haproxy.cfg -p /var/run/haproxy.pid -D
-                    CONFD="confd -prefix=${PATRONI_NAMESPACE:-/service}/$PATRONI_SCOPE -interval=10 -backend"
-                    if [ ! -z ${PATRONI_ZOOKEEPER_HOSTS} ]; then
-                        while ! /usr/share/zookeeper/bin/zkCli.sh -server ${PATRONI_ZOOKEEPER_HOSTS} ls /; do
-                            sleep 1
-                        done
-                        exec $CONFD zookeeper -node ${PATRONI_ZOOKEEPER_HOSTS}
-                    else
-                        while ! curl -s ${PATRONI_ETCD_URL}/v2/members | jq -r '.members[0].clientURLs[0]' | grep -q http; do
-                            sleep 1
-                        done
-                        exec $CONFD etcd -node $PATRONI_ETCD_URL
-                    fi
-                    ;;
-                etcd)
-                    exec etcd $ETCD_ARGS
-                    ;;
-                zookeeper)
-                    exec /usr/share/zookeeper/bin/zkServer.sh start-foreground
-                    ;;
-                cheat)
-                    CHEAT=1
-                    ;;
-                help)
-                    usage
-                    exit 0
-                    ;;
-                *)
-                    if [ "$OPTERR" = 1 ] && [ "${optspec:0:1}" != ":" ]; then
-                        echo "Unknown option --${OPTARG}" >&2
-                    fi
-                    ;;
-            esac;;
-        *)
-            if [ "$OPTERR" != 1 ] || [ "${optspec:0:1}" = ":" ]; then
-                echo "Non-option argument: '-${OPTARG}'" >&2
-                usage
-                exit 1
-            fi
-            ;;
-    esac
-done
+case "$1" in
+    haproxy)
+        haproxy -f /etc/haproxy/haproxy.cfg -p /var/run/haproxy.pid -D
+        CONFD="confd -prefix=$PATRONI_NAMESPACE/$PATRONI_SCOPE -interval=10 -backend"
+        if [ ! -z "$PATRONI_ZOOKEEPER_HOSTS" ]; then
+            while ! /usr/share/zookeeper/bin/zkCli.sh -server $PATRONI_ZOOKEEPER_HOSTS ls /; do
+                sleep 1
+            done
+            exec dumb-init $CONFD zookeeper -node $PATRONI_ZOOKEEPER_HOSTS
+        else
+            while ! etcdctl cluster-health 2> /dev/null; do
+                sleep 1
+            done
+            exec dumb-init $CONFD etcd -node $(echo $ETCDCTL_ENDPOINTS | sed 's/,/ -node /g')
+        fi
+        ;;
+    etcd)
+        exec "$@" -advertise-client-urls http://$DOCKER_IP:2379
+        ;;
+    zookeeper)
+        exec /usr/share/zookeeper/bin/zkServer.sh start-foreground
+        ;;
+esac
 
 ## We start an etcd
-if [[ -z ${PATRONI_ETCD_URL} && -z ${PATRONI_ZOOKEEPER_HOSTS} ]]; then
-    etcd $ETCD_ARGS > /var/log/etcd.log 2> /var/log/etcd.err &
+if [ -z "$PATRONI_ETCD_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
 
 export PATRONI_SCOPE
-export PATRONI_NAME="${PATRONI_NAME:-${HOSTNAME}}"
-export PATRONI_RESTAPI_CONNECT_ADDRESS="${DOCKER_IP}:8008"
+export PATRONI_NAMESPACE
+export PATRONI_NAME="${PATRONI_NAME:-$(hostname)}"
+export PATRONI_RESTAPI_CONNECT_ADDRESS="$DOCKER_IP:8008"
 export PATRONI_RESTAPI_LISTEN="0.0.0.0:8008"
-export PATRONI_admin_PASSWORD="${PATRONI_admin_PASSWORD:=admin}"
+export PATRONI_admin_PASSWORD="${PATRONI_admin_PASSWORD:-admin}"
 export PATRONI_admin_OPTIONS="${PATRONI_admin_OPTIONS:-createdb, createrole}"
-export PATRONI_POSTGRESQL_CONNECT_ADDRESS="${DOCKER_IP}:5432"
+export PATRONI_POSTGRESQL_CONNECT_ADDRESS="$DOCKER_IP:5432"
 export PATRONI_POSTGRESQL_LISTEN="0.0.0.0:5432"
-export PATRONI_POSTGRESQL_DATA_DIR="data/${PATRONI_SCOPE}"
+export PATRONI_POSTGRESQL_DATA_DIR="${PATRONI_POSTGRESQL_DATA_DIR:-$PGDATA}"
 export PATRONI_REPLICATION_USERNAME="${PATRONI_REPLICATION_USERNAME:-replicator}"
-export PATRONI_REPLICATION_PASSWORD="${PATRONI_REPLICATION_PASSWORD:-abcd}"
+export PATRONI_REPLICATION_PASSWORD="${PATRONI_REPLICATION_PASSWORD:-replicate}"
 export PATRONI_SUPERUSER_USERNAME="${PATRONI_SUPERUSER_USERNAME:-postgres}"
 export PATRONI_SUPERUSER_PASSWORD="${PATRONI_SUPERUSER_PASSWORD:-postgres}"
-export PATRONI_POSTGRESQL_PGPASS="$HOME/.pgpass"
 
-cat > /patroni.yml <<__EOF__
-bootstrap:
-  dcs:
-    postgresql:
-      use_pg_rewind: true
-
-  pg_hba:
-  - host all all 0.0.0.0/0 md5
-  - host replication replicator ${DOCKER_IP}/16    md5
-__EOF__
-
-mkdir -p "$HOME/.config/patroni"
-[ -h "$HOME/.config/patroni/patronictl.yaml" ] || ln -s /patroni.yml "$HOME/.config/patroni/patronictl.yaml"
-
-[ -z $CHEAT ] && exec python3 /patroni.py /patroni.yml
-
-while true; do
-    sleep 60
-done
+exec python3 /patroni.py postgres0.yml

docker/etcd.env

@@ -0,0 +1,5 @@
+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

@@ -1,3 +1,6 @@
+PATRONI_SCOPE=demo
+PATRONI_ETCD_HOSTS='etcd1:2379','etcd2:2379','etcd3:2379'
+
 PATRONI_RESTAPI_USERNAME=admin
 PATRONI_RESTAPI_PASSWORD=admin
 PATRONI_SUPERUSER_USERNAME=postgres
@@ -6,3 +9,6 @@ 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/ENVIRONMENT.rst

@@ -11,8 +11,18 @@ 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
--  **PATRONI\_LOGLEVEL**: sets the general logging level (see `the docs for Python logging <https://docs.python.org/3.6/library/logging.html#levels>`_)
--  **PATRONI\_REQUESTS_LOGLEVEL**: sets the logging level for all HTTP requests e.g. Kubernetes API calls (see `the docs for Python logging <https://docs.python.org/3.6/library/logging.html#levels>`_)
+
+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.
+-  **PATRONI\_LOG\_DIR**: Directory to write application logs to. The directory must exist and be writable by the user executing Patroni. If you set this env variable, the application will retain 4 25MB logs by default. You can tune those retention values with `PATRONI_LOG_FILE_NUM` and `PATRONI_LOG_FILE_SIZE` (see below).
+-  **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}"``
 
 Bootstrap configuration
 -----------------------
@@ -25,8 +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\_URL**: url for the Consul, in format: http(s)://host:port
+-  **PATRONI\_CONSUL\_HOST**: the host:port for the Consul local agent.
+-  **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
@@ -35,19 +45,31 @@ Consul
 -  **PATRONI\_CONSUL\_CERT**: (optional) File with the client certificate
 -  **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\_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\_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. 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
 
 Etcd
 ----
--  **PATRONI\_ETCD\_HOST**: the host:port for the etcd endpoint.
--  **PATRONI\_ETCD\_HOSTS**: list of etcd endpoints in format host1:port1,host2:port2,etc...
--  **PATRONI\_ETCD\_URL**: url for the etcd, in format: http(s)://(username:password@)host:port
+
 -  **PATRONI\_ETCD\_PROXY**: proxy url for the etcd. If you are connecting to the etcd using proxy, use this parameter instead of **PATRONI\_ETCD\_URL**
+-  **PATRONI\_ETCD\_URL**: url for the etcd, in format: http(s)://(username:password@)host:port
+-  **PATRONI\_ETCD\_HOSTS**: list of etcd endpoints in format 'host1:port1','host2:port2',etc...
+-  **PATRONI\_ETCD\_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 but stick to a fixed list of **hosts**.
+-  **PATRONI\_ETCD\_PROTOCOL**: http or https, if not specified http is used. If the **url** or **proxy** is specified - will take protocol from them.
+-  **PATRONI\_ETCD\_HOST**: the host:port for the etcd endpoint.
 -  **PATRONI\_ETCD\_SRV**: Domain to search the SRV record(s) for cluster autodiscovery.
+-  **PATRONI\_ETCD\_USERNAME**: username for etcd authentication.
+-  **PATRONI\_ETCD\_PASSWORD**: password for etcd authentication.
 -  **PATRONI\_ETCD\_CACERT**: The ca certificate. If present it will enable validation.
--  **PATRONI\_ETCD\_CERT**: File with the client certificate
+-  **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.
 
+ZooKeeper
+---------
+-  **PATRONI\_ZOOKEEPER\_HOSTS**: comma separated list of ZooKeeper cluster members: "'host1:port1','host2:port2','etc...'". It is important to quote every single entity!
+
 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.
@@ -63,7 +85,7 @@ Kubernetes
 -  **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.
 
 PostgreSQL
 ----------
@@ -75,18 +97,41 @@ 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\_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\_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\_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\_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\_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.
 
 REST API
--------- 
+--------
 -  **PATRONI\_RESTAPI\_CONNECT\_ADDRESS**: IP address and port to access the REST API.
 -  **PATRONI\_RESTAPI\_LISTEN**: IP address and port that Patroni will listen to, to provide health-check information for HAProxy.
 -  **PATRONI\_RESTAPI\_USERNAME**: Basic-auth username to protect unsafe REST API endpoints.
 -  **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\_CAFILE**: Specifies the file with the CA_BUNDLE with certificates of trusted CAs to use while verifying client certs.
+-  **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. 
 
-ZooKeeper
----------
--  **PATRONI\_ZOOKEEPER\_HOSTS**: comma separated list of ZooKeeper cluster members: "'host1:port1','host2:port2','etc...'". It is important to quote every single entity!
+CTL
+---
+-  **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

@@ -33,6 +33,35 @@ To install requirements on a Mac, run the following:
 
     brew install postgresql etcd haproxy libyaml python
 
+.. _psycopg2_install_options:
+
+**Psycopg2**
+
+Starting from `psycopg2-2.8 <http://initd.org/psycopg/articles/2019/04/04/psycopg-28-released/>`__ the binary version of psycopg2 will no longer be installed by default. Installing it from the source code requires C compiler and postgres+python dev packages.
+Since in the python world it is not possible to specify dependency as ``psycopg2 OR psycopg2-binary`` you will have to decide how to install it.
+
+There are a few options available:
+
+1. Use the package manager from your distro
+
+::
+
+    sudo apt-get install python-psycopg2   # install python2 psycopg2 module on Debian/Ubuntu
+    sudo apt-get install python3-psycopg2  # install python3 psycopg2 module on Debian/Ubuntu
+    sudo yum install python-psycopg2       # install python2 psycopg2 on RedHat/Fedora/CentOS
+
+2. Install psycopg2 from the binary package
+
+::
+
+    pip install psycopg2-binary
+
+3. Install psycopg2 from source
+
+::
+
+    pip install psycopg2>=2.5.4
+
 **General installation for pip**
 
 Patroni can be installed with pip:
@@ -66,6 +95,8 @@ Note that external tools to call in the replica creation or custom bootstrap scr
 independently of Patroni.
 
 
+.. _running_configuring:
+
 Running and Configuring
 -----------------------
 
@@ -76,7 +107,7 @@ 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
 

docs/SETTINGS.rst

@@ -4,53 +4,85 @@
 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.
+-  **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. Patroni will try to create slots before opening connections to the cluster.
+    -  **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.
+
 Global/Universal
 ----------------
 -  **name**: the name of the host. Must be unique for the cluster.
 -  **namespace**: path within the configuration store where Patroni will keep information about the cluster. Default value: "/service"
 -  **scope**: cluster name
 
+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.
+-  **dir**: Directory to write application logs to. The directory must exist and be writable by the user executing Patroni. If you set this value, the application will retain 4 25MB logs by default. You can tune those retention values with `file_num` and `file_size` (see below).
+-  **file\_num**: The number of application logs to retain.
+-  **file\_size**: Size of patroni.log file (in bytes) that triggers a log rolling.
+-  **loggers**: This section allows redefining logging level per python module
+    -  **patroni.postmaster: WARNING**
+    -  **urllib3: DEBUG**
+
+.. _bootstrap_settings:
+
 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.
-    -  **loop\_wait**: the number of seconds the loop will sleep. Default value: 10
-    -  **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
-    -  **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
-    -  **maximum\_lag\_on\_failover**: the maximum bytes a follower may lag to be able to participate in leader election.
-    -  **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.
-    -  **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
-        -  **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.
-        -  **recovery\_conf**: additional configuration settings written to recovery.conf when configuring follower. 
-        -  **parameters**: list of configuration settings for Postgres. Many of these are required for replication to work.
-    -  **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
--  **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.
+-  **bootstrap**:
+    -  **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>`.
+    -  **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:
 
@@ -58,17 +90,20 @@ 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
--  **url**: url for the Consul endpoint
--  **port**: (optional) Consul port
--  **scheme**: (optional) **http** or **https**, defaults to **http**
--  **token**: (optional) ACL token
--  **verify**: (optional) whether to verify the SSL certificate for HTTPS requests
+-  **host**: the host:port for the Consul local agent.
+-  **url**: url for the Consul local agent, in format: http(s)://host:port.
+-  **port**: (optional) Consul port.
+-  **scheme**: (optional) **http** or **https**, defaults to **http**.
+-  **token**: (optional) ACL token.
+-  **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.
--  **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
+-  **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. 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**.
+-  **service\_check\_interval**: (optional) how often to perform health check against registered url.
 
 Etcd
 ----
@@ -76,20 +111,25 @@ 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.
--  **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**
+-  **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.
+-  **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
+-  **username**: (optional) username for etcd authentication.
 -  **password**: (optional) password for etcd authentication.
 -  **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**.
 
+ZooKeeper
+----------
+-  **hosts**: list of ZooKeeper cluster members in format: ['host1:port1', 'host2:port2', 'etc...'].
+
 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:
@@ -102,73 +142,104 @@ Kubernetes
 -  **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.
 
 .. _postgresql_settings:
 
 PostgreSQL
 ----------
--  **authentication**:
-    -  **superuser**:
-        -  **username**: name for the superuser, set during initialization (initdb) and later used by Patroni to connect to the postgres.
-        -  **password**: password for the superuser, set during initialization (initdb).
-    -  **replication**:
-        -  **username**: replication username; the user will be created during initialization. Replicas will use this user to access master via streaming replication
-        -  **password**: replication password; the user will be created during initialization.
--  **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.)
-        -  **on\_reload**: run this script when configuration reload is triggered.
-        -  **on\_restart**: run this script when the cluster restarts.
-        -  **on\_role\_change**: run this script when the cluster is being promoted or demoted.
-        -  **on\_start**: run this script when the cluster starts.
-        -  **on\_stop**: run this script when the cluster 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 existing 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 definded, Patroni will use 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 default value should be used and omit ``host`` from 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\_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 postgres data directory and recreate replica. Otherwise it will try to follow the new leader. 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".
+-  **postgresql**:
+    -  **authentication**:
+        -  **superuser**:
+            -  **username**: name for the superuser, set during initialization (initdb) and later used by Patroni to connect to the postgres.
+            -  **password**: password for the superuser, set during initialization (initdb).
+            -  **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``.
+            -  **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.
+            -  **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.
+            -  **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.
+            -  **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.
+        -  **replication**:
+            -  **username**: replication username; the user will be created during initialization. Replicas will use this user to access master via streaming replication
+            -  **password**: replication password; the user will be created during initialization.
+            -  **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``.
+            -  **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.
+            -  **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.
+            -  **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.
+            -  **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.
+        -  **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.
+            -  **password**: password for the user for ``pg_rewind``; the user will be created during initialization.
+            -  **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``.
+            -  **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.
+            -  **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.
+            -  **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.
+            -  **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.
+    -  **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.)
+            -  **on\_reload**: run this script when configuration reload is triggered.
+            -  **on\_restart**: run this script when the postgres restarts (without changing role).
+            -  **on\_role\_change**: run this script when the postgres is being promoted or demoted.
+            -  **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.
+    -  **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".
 
 REST API
--------- 
--  **connect\_address**: IP address (or hostname) and port, to access the Patroni's REST API. 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.
--  **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**:
+--------
+-  **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).
+
+        -  **authentication**: (optional)
             -  **username**: Basic-auth username to protect unsafe REST API endpoints.
             -  **password**: Basic-auth password to protect unsafe REST API endpoints.
-
-        -  **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.
-        -  **keyfile**: Specifies the file with the secret key in the PEM format.
+        -  **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.
+        -  **cafile**: (optional): Specifies the file with the CA_BUNDLE with certificates of trusted CAs to use while verifying client certs.
+        -  **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.
 
 .. _patronictl_settings:
 
 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.
-    -  **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.
-
-ZooKeeper
-----------
--  **hosts**: list of ZooKeeper cluster members in format: ['host1:port1', 'host2:port2', 'etc...'].
+- **ctl**: (optional)
+    -  **insecure**: Allow connections to REST API without verifying 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 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.
 
 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
+----
+- **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.

docs/dynamic_configuration.rst

@@ -14,7 +14,7 @@ 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.
@@ -76,153 +76,11 @@ 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
 
 Upon changing these options, Patroni will read the relevant section of the configuration stored in DCS and change its
 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

@@ -0,0 +1,36 @@
+.. _existing_data:
+
+Convert a Standalone to a Patroni Cluster
+=========================================
+
+This section describes the process for converting a standalone PostgreSQL instance into a Patroni cluster.
+
+To deploy a Patroni cluster without using a pre-existing PostgreSQL instance, see :ref:`Running and Configuring <running_configuring>` instead.
+
+Procedure
+---------
+
+A Patroni cluster can be started with a data directory from a single-node PostgreSQL database. This is achieved by following closely these steps:
+
+1. Manually start PostgreSQL daemon
+2. Create Patroni superuser and replication users as defined in the :ref:`authentication <postgresql_settings>` section of the Patroni configuration. If this user is created in SQL, the following queries achieve this:
+
+.. code-block:: sql
+
+   CREATE USER $PATRONI_SUPERUSER_USERNAME WITH SUPERUSER ENCRYPTED PASSWORD '$PATRONI_SUPERUSER_PASSWORD';
+   CREATE USER $PATRONI_REPLICATION_USERNAME WITH REPLICATION ENCRYPTED PASSWORD '$PATRONI_REPLICATION_PASSWORD';
+
+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.
+
+
+FAQ
+---
+
+- During Patroni startup, Patroni complains that it cannot bind to the PostgreSQL port.
+
+  You need to verify ``listen_addresses`` and ``port`` in ``postgresql.conf`` and ``postgresql.listen`` in ``patroni.yml``. Don't forget that ``pg_hba.conf`` should allow such access.
+
+- After asking Patroni to restart the node, PostgreSQL displays the error message ``could not open configuration file "/etc/postgresql/10/main/pg_hba.conf": No such file or directory``
+
+  It can mean various things depending on how you manage PostgreSQL configuration. If you specified `postgresql.config_dir`, Patroni generates the ``pg_hba.conf`` based on the settings in the :ref:`bootstrap <bootstrap_settings>` section only when it bootstraps a new cluster. In this scenario the ``PGDATA`` was not empty, therefore no bootstrap happened. This file must exist beforehand.

docs/index.rst

@@ -19,6 +19,7 @@ We call Patroni a "template" because it is far from being a one-size-fits-all or
 
    README
    dynamic_configuration
+   rest_api
    ENVIRONMENT
    SETTINGS
    replica_bootstrap

docs/releases.rst

@@ -3,19 +3,702 @@
 Release notes
 =============
 
-Version 1.5.0                                                                                                                           
--------------                                                                                                                           
-                                                                                                                                        
+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 threre 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 contaning 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
+-------------
+
+This version adds compatibility with PostgreSQL 12, makes is possible to run pg_rewind without superuser on PostgreSQL 11 and newer, and enables IPv6 support.
+
+
+**New features**
+
+- Psycopg2 was removed from requirements and must be installed independently (Alexander Kukushkin)
+
+  Starting from 2.8.0 ``psycopg2`` was split into two different packages, ``psycopg2``, and ``psycopg2-binary``, which could be installed at the same time into the same place on the filesystem. In order to decrease dependency hell problem, we let a user choose how to install it. There are a few options available, please consult the :ref:`documentation <psycopg2_install_options>`.
+
+- Compatibility with PostgreSQL 12 (Alexander Kukushkin)
+
+  Starting from PostgreSQL 12 there is no ``recovery.conf`` anymore and all former recovery parameters are converted into `GUC <https://www.enterprisedb.com/blog/what-is-a-guc-variable>`_. In order to protect from ``ALTER SYSTEM SET primary_conninfo`` or similar, Patroni will parse ``postgresql.auto.conf`` and remove all standby and recovery parameters from there. Patroni config remains backward compatible. For example despite ``restore_command`` being a GUC, one can still specify it in the ``postgresql.recovery_conf.restore_command`` section and Patroni will write it into ``postgresql.conf`` for PostgreSQL 12.
+
+- Make it possible to use ``pg_rewind`` without superuser on PostgreSQL 11 and newer (Alexander Kukushkin)
+
+  If you want to use this feature please define ``username`` and ``password`` in the ``postgresql.authentication.rewind`` section of Patroni configuration file. For an already existing cluster you will have to create the user manually and ``GRANT EXECUTE`` permission on a few functions. You can find more details in the PostgreSQL `documentation <https://www.postgresql.org/docs/11/app-pgrewind.html#id-1.9.5.8.8>`__.
+
+- Do a smart comparison of actual and desired ``primary_conninfo`` values on replicas (Alexander Kukushkin)
+
+  It might help to avoid replica restart when you are converting an already existing primary-standby cluster to one managed by Patroni
+
+- IPv6 support (Alexander Kukushkin)
+
+  There were two major issues. Patroni REST API service was listening only on ``0.0.0.0`` and IPv6 IP addresses used in the ``api_url`` and ``conn_url`` were not properly quoted.
+
+- Kerberos support (Ajith Vilas, Alexander Kukushkin)
+
+  It makes possible using Kerberos authentication between Postgres nodes instead of defining passwords in Patroni configuration file
+
+- Manage ``pg_ident.conf`` (Alexander Kukushkin)
+
+  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**
+
+- Added ``/health`` endpoint (Wilfried Roset)
+
+  It will return an HTTP status code only if PostgreSQL is running
+
+- Added ``/read-only`` and ``/read-write`` endpoints (Julien Riou)
+
+  The ``/read-only`` endpoint enables reads balanced across replicas and the primary. The ``/read-write`` endpoint is an alias for ``/primary``, ``/leader`` and ``/master``.
+
+- Use ``SSLContext`` to wrap the REST API socket (Julien Riou)
+
+  Usage of ``ssl.wrap_socket()`` is deprecated and was still allowing soon-to-be-deprecated protocols like TLS 1.1.
+
+
+**Logging improvements**
+
+- Two-step logging (Alexander Kukushkin)
+
+  All log messages are first written into the in-memory queue and later they are asynchronously flushed into the stderr or file from a separate thread. The maximum queue size is limited (configurable). If the limit is reached, Patroni will start losing logs, which is still better than blocking the HA loop.
+
+- Enable debug logging for GET/OPTIONS API calls together with latency (Jan Tomsa)
+
+  It will help with debugging of health-checks performed by HAProxy, Consul or other tooling that decides which node is the primary/replica.
+
+- Log exceptions caught in Retry (Daniel Kucera)
+
+  Log the final exception when either the number of attempts or the timeout were reached. It will hopefully help to debug some issues when communication to DCS fails.
+
+
+**Improvements in patronictl**
+
+- Enhance dialogues for scheduled switchover and restart (Rafia Sabih)
+
+  Previously dialogues did not take into account scheduled actions and therefore were misleading.
+
+- Check if config file exists (Wilfried Roset)
+
+  Be verbose about configuration file when the given filename does not exists, instead of ignoring silently (which can lead to misunderstanding).
+
+- Add fallback value for ``EDITOR`` (Wilfried Roset)
+
+  When the ``EDITOR`` environment variable was not defined, ``patronictl edit-config`` was failing with `PatroniCtlException`. The new strategy is to try ``editor`` and than ``vi``, which should be available on most systems.
+
+
+**Improvements in Consul support**
+
+- Allow to specify Consul consistency mode (Jan Tomsa)
+
+  You can read more about consistency mode `here <https://www.consul.io/api/features/consistency.html>`__.
+
+- Reload Consul config on SIGHUP (Cameron Daniel, Alexander Kukushkin)
+
+  It is especially useful when somebody is changing the value of ``token``.
+
+
+**Bugfixes**
+
+- Fix corner case in switchover/failover (Sharoon Thomas)
+
+  The variable ``scheduled_at`` may be undefined if REST API is not accessible and we are using DCS as a fallback.
+
+- Open trust to localhost in ``pg_hba.conf`` during custom bootstrap (Alexander Kukushkin)
+
+  Previously it was open only to unix_socket, which was causing a lot of errors: ``FATAL:  no pg_hba.conf entry for replication connection from host "127.0.0.1", user "replicator"``
+
+- Consider synchronous node as healthy even when the former leader is ahead (Alexander Kukushkin)
+
+  If the primary loses access to the DCS, it restarts Postgres in read-only, but it might happen that other nodes can still access the old primary via the REST API. Such a situation was causing the synchronous standby not to promote because the old primary was reporting WAL position ahead of the synchronous standby.
+
+- Standby cluster bugfixes (Alexander Kukushkin)
+
+  Make it possible to bootstrap a replica in a standby cluster when the standby_leader is not accessible and a few other minor fixes.
+
+
+Version 1.5.6
+-------------
+
+**New features**
+
+- Support work with etcd cluster via set of proxies (Alexander Kukushkin)
+
+  It might happen that etcd cluster is not accessible directly but via set of proxies. In this case Patroni will not perform etcd topology discovery but just round-robin via proxy hosts. Behavior is controlled by `etcd.use_proxies`.
+
+- Changed callbacks behavior when role on the node is changed (Alexander)
+
+  If the role was changed from `master` or `standby_leader` to `replica` or from `replica` to `standby_leader`, `on_restart` callback will not be called anymore in favor of `on_role_change` callback.
+
+- Change the way how we start postgres (Alexander)
+
+  Use `multiprocessing.Process` instead of executing itself and `multiprocessing.Pipe` to transmit the postmaster pid to the Patroni process. Before that we were using pipes, what was leaving postmaster process with stdin closed.
+
+**Bug fixes**
+
+- Fix role returned by REST API for the standby leader (Alexander)
+
+  It was incorrectly returning `replica` instead of `standby_leader`
+
+- Wait for callback end if it could not be killed (Julien Tachoires)
+
+  Patroni doesn't have enough privileges to terminate the callback script running under `sudo` what was cancelling the new callback. If the running script could not be killed, Patroni will wait until it finishes and then run the next callback.
+
+- Reduce lock time taken by dcs.get_cluster method (Alexander)
+
+  Due to the lock being held DCS slowness was affecting the REST API health checks causing false positives.
+
+- Improve cleaning of PGDATA when `pg_wal`/`pg_xlog` is a symlink (Julien)
+
+  In this case Patroni will explicitly remove files from the target directory.
+
+- Remove unnecessary usage of os.path.relpath (Ants Aasma)
+
+  It depends on being able to resolve the working directory, what will fail if Patroni is started in a directory that is later unlinked from the filesystem.
+
+- Do not enforce ssl version when communicating with Etcd (Alexander)
+
+  For some unknown reason python3-etcd on debian and ubuntu are not based on the latest version of the package and therefore it enforces TLSv1 which is not supported by Etcd v3. We solved this problem on Patroni side.
+
+Version 1.5.5
+-------------
+
+This version introduces the possibility of automatic reinit of the former master, improves patronictl list output and fixes a number of bugs.
+
+**New features**
+
+- Add support of `PATRONI_ETCD_PROTOCOL`, `PATRONI_ETCD_USERNAME` and `PATRONI_ETCD_PASSWORD` environment variables (Étienne M)
+
+  Before it was possible to configure them only in the config file or as a part of `PATRONI_ETCD_URL`, which is not always convenient.
+
+- Make it possible to automatically reinit the former master (Alexander Kukushkin)
+
+  If the pg_rewind is disabled or can't be used, the former master could fail to start as a new replica due to diverged timelines. In this case, the only way to fix it is wiping the data directory and reinitializing. This behavior could be changed by setting `postgresql.remove_data_directory_on_diverged_timelines`. When it is set, Patroni will wipe the data directory and reinitialize the former master automatically.
+
+- Show information about timelines in patronictl list (Alexander)
+
+  It helps to detect stale replicas. In addition to that, `Host` will include ':{port}' if the port value isn't default or there is more than one member running on the same host.
+
+- Create a headless service associated with the $SCOPE-config endpoint (Alexander)
+
+  The "config" endpoint keeps information about the cluster-wide Patroni and Postgres configuration, history file, and last but the most important, it holds the `initialize` key. When the Kubernetes master node is restarted or upgraded, it removes endpoints without services. The headless service will prevent it from being removed.
+
+**Bug fixes**
+
+- Adjust the read timeout for the leader watch blocking query (Alexander)
+
+  According to the Consul documentation, the actual response timeout is increased by a small random amount of additional wait time added to the supplied maximum wait time to spread out the wake up time of any concurrent requests. It adds up to `wait / 16` additional time to the maximum duration. In our case we are adding `wait / 15` or 1 second depending on what is bigger.
+
+- Always use replication=1 when connecting via replication protocol to the postgres (Alexander)
+
+  Starting from Postgres 10 the line in the pg_hba.conf with database=replication doesn't accept connections with the parameter replication=database.
+
+- Don't write primary_conninfo into recovery.conf for wal-only standby cluster (Alexander)
+
+  Despite not having neither `host` nor `port` defined in the `standby_cluster` config, Patroni was putting the `primary_conninfo` into the `recovery.conf`, which is useless and generating a lot of errors.
+
+
+Version 1.5.4
+-------------
+
+This version implements flexible logging and fixes a number of bugs.
+
+**New features**
+
+- Improvements in logging infrastructure (Alexander Kukushkin, Lucas Capistrant, Alexander Anikin)
+
+  Logging configuration could be configured not only from environment variables but also from Patroni config file. It makes it possible to change logging configuration in runtime by updating config and doing reload or sending SIGHUP to the Patroni process. By default Patroni writes logs to stderr, but now it becomes possible to write logs directly into the file and rotate when it reaches a certain size. In addition to that added support of custom dateformat and the possibility to fine-tune log level for each python module.
+
+- Make it possible to take into account the current timeline during leader elections (Alexander Kukushkin)
+
+  It could happen that the node is considering itself as a healthiest one although it is currently not on the latest known timeline. In some cases we want to avoid promoting of such node, which could be achieved by setting `check_timeline` parameter to `true` (default behavior remains unchanged).
+
+- Relaxed requirements on superuser credentials
+
+  Libpq allows opening connections without explicitly specifying neither username nor password. Depending on situation it relies either on pgpass file or trust authentication method in pg_hba.conf. Since pg_rewind is also using libpq, it will work the same way.
+
+- Implemented possibility to configure Consul Service registration and check interval via environment variables (Alexander Kukushkin)
+
+  Registration of service in Consul was added in the 1.5.0, but so far it was only possible to turn it on via patroni.yaml.
+
+**Stability Improvements**
+
+- Set archive_mode to off during the custom bootstrap (Alexander Kukushkin)
+
+  We want to avoid archiving wals and history files until the cluster is fully functional.  It really helps if the custom bootstrap involves pg_upgrade.
+
+- Apply five seconds backoff when loading global config on start (Alexander Kukushkin)
+
+  It helps to avoid hammering DCS when Patroni just starting up.
+
+- Reduce amount of error messages generated on shutdown (Alexander Kukushkin)
+
+  They were harmless but rather annoying and sometimes scary.
+
+- Explicitly secure rw perms for recovery.conf at creation time (Lucas)
+
+  We don't want anybody except patroni/postgres user reading this file, because it contains replication user and password.
+
+- Redirect HTTPServer exceptions to logger (Julien Riou)
+
+  By default, such exceptions were logged on standard output messing with regular logs.
+
+**Bug fixes**
+
+- Removed stderr pipe to stdout on pg_ctl process (Cody Coons)
+
+  Inheriting stderr from the main Patroni process allows all Postgres logs to be seen along with all patroni logs. This is very useful in a container environment as Patroni and Postgres logs may be consumed using standard tools (docker logs, kubectl, etc). In addition to that, this change fixes a bug with Patroni not being able to catch postmaster pid when postgres writing some warnings into stderr.
+
+- Set Consul service check deregister timeout in Go time format (Pavel Kirillov)
+
+  Without explicitly mentioned time unit registration was failing.
+
+- Relax checks of standby_cluster cluster configuration (Dmitry Dolgov, Alexander Kukushkin)
+
+  It was accepting only strings as valid values and therefore it was not possible to specify the port as integer and create_replica_methods as a list.
+
+Version 1.5.3
+-------------
+
+Compatibility and bugfix release.
+
+- Improve stability when running with python3 against zookeeper (Alexander Kukushkin)
+
+  Change of `loop_wait` was causing Patroni to disconnect from zookeeper and never reconnect back.
+
+- 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'
+
+- Make sure we refresh Consul session at least once per HA loop and improve handling of consul sessions exceptions (Alexander)
+
+  Restart of local consul agent invalidates all sessions related to the node. Not calling session refresh on time and not doing proper handling of session errors was causing demote of the primary.
+
+Version 1.5.2
+-------------
+
+Compatibility and bugfix release.
+
+- Compatibility with kazoo-2.6.0 (Alexander Kukushkin)
+
+  In order to make sure that requests are performed with an appropriate timeout, Patroni redefines create_connection method from python-kazoo module. The last release of kazoo slightly changed the way how create_connection method is called.
+
+- Fix Patroni crash when Consul cluster loses the leader (Alexander)
+
+  The crash was happening due to incorrect implementation of touch_member method, it should return boolean and not raise any exceptions.
+
+Version 1.5.1
+-------------
+
+This version implements support of permanent replication slots, adds support of pgBackRest and fixes number of bugs.
+
+**New features**
+
+- Permanent replication slots (Alexander Kukushkin)
+
+  Permanent replication slots are preserved on failover/switchover, that is, Patroni on the new primary will create configured replication slots right after doing promote. Slots could be configured with the help of `patronictl edit-config`. The initial configuration could be also done in the :ref:`bootstrap.dcs <settings>`.
+
+- Add pgbackrest support (Yogesh Sharma)
+
+  pgBackrest can restore in existing $PGDATA folder, this allows speedy restore as files which have not changed since last backup are skipped, to support this feature new parameter `keep_data` has been introduced. See :ref:`replica creation method <custom_replica_creation>` section for additional examples.
+
+**Bug fixes**
+
+- A few bugfixes in the "standby cluster" workflow (Alexander)
+
+  Please see https://github.com/zalando/patroni/pull/823 for more details.
+
+- Fix REST API health check when cluster management is paused and DCS is not accessible (Alexander)
+
+  Regression was introduced in https://github.com/zalando/patroni/commit/90cf930036a9d5249265af15d2b787ec7517cf57
+
+Version 1.5.0
+-------------
+
 This version enables Patroni HA cluster to operate in a standby mode, introduces experimental support for running on Windows, and provides a new configuration parameter to register PostgreSQL service in Consul.
 
-**New features**                                                                                                                        
+**New features**
 
-- Standby cluster (Dmitry Dolgov)                                                                                                       
+- Standby cluster (Dmitry Dolgov)
 
   One or more Patroni nodes can form a standby cluster that runs alongside the primary one (i.e. in another datacenter) and consists of standby nodes that replicate from the master in the primary cluster. All PostgreSQL nodes in the standby cluster are replicas; one of those replicas elects itself to replicate directly from the remote master, while the others replicate from it in a cascading manner. More detailed description of this feature and some configuration examples can be found at :ref:`here <standby_cluster>`.
 
 - Register Services in Consul (Pavel Kirillov, Alexander Kukushkin)
-                                                                                                                                        
+
   If `register_service` parameter in the consul :ref:`configuration <consul_settings>` is enabled, the node will register a service with the name `scope` and the tag `master`, `replica` or `standby-leader`.
 
 - Experimental Windows support (Pavel Golub)
@@ -299,8 +982,8 @@ In addition to using Endpoints, Patroni supports ConfigMaps. You can find more i
 
 - Minimize the amount of SELECT's issued by Patroni on every loop of HA cylce (Alexander Kukushkin)
 
-  On every iteration of HA loop Patroni needs to know recovery status and absolute wal position. From now on Patroni will run only single SELECT to get this information instead of two on the replica and three on the master. 
-  
+  On every iteration of HA loop Patroni needs to know recovery status and absolute wal position. From now on Patroni will run only single SELECT to get this information instead of two on the replica and three on the master.
+
 - Remove leader key on shutdown only when we have the lock (Ants)
 
   Unconditional removal was generating unnecessary and missleading exceptions.
@@ -310,7 +993,7 @@ In addition to using Endpoints, Patroni supports ConfigMaps. You can find more i
 - Add version command to patronictl (Ants)
 
   It will show the version of installed Patroni and versions of running Patroni instances (if the cluster name is specified).
-  
+
 - Make optional specifying cluster_name argument for some of patronictl commands (Alexander, Ants)
 
   It will work if patronictl is using usual Patroni configuration file with the ``scope`` defined.
@@ -318,11 +1001,11 @@ In addition to using Endpoints, Patroni supports ConfigMaps. You can find more i
 - Show information about scheduled switchover and maintenance mode (Alexander)
 
   Before that it was possible to get this information only from Patroni logs or directly from DCS.
-  
+
 - Improve ``patronictl reinit`` (Alexander)
 
   Sometimes ``patronictl reinit`` refused to proceed when Patroni was busy with other actions, namely trying to start postgres. `patronictl` didn't provide any commands to cancel such long running actions and the only (dangerous) workarond was removing a data directory manually. The new implementation of `reinit` forcefully cancells other long-running actions before proceeding with reinit.
-  
+
 - Implement ``--wait`` flag in ``patronictl pause`` and ``patronictl resume`` (Alexander)
 
   It will make ``patronictl`` wait until the requested action is acknowledged by all nodes in the cluster.
@@ -369,7 +1052,7 @@ Version 1.3.6
   After a crash that doesn't clean up postmaster.pid there could be a new process with the same pid, resulting in a false positive for is_running(), which will lead to all kinds of bad behavior.
 
 - Shutdown postgresql before bootstrap when we lost data directory (ainlolcat)
-  
+
   When data directory on the master is forcefully removed, postgres process can still stay alive for some time and prevent the replica created in place of that former master from starting or replicating.
   The fix makes Patroni cache the postmaster pid and its start time and let it terminate the old postmaster in case it is still running after the corresponding data directory has been removed.
 
@@ -439,7 +1122,7 @@ 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>`__.
-  
+
 
 - Advanced configuration for Consul (Alexander Kukushkin)
 

docs/replica_bootstrap.rst

@@ -65,6 +65,19 @@ for outdated backup files. Some people prefer other backup solutions, such as ``
 others, or simply roll their own scripts. In order to accommodate all those use-cases Patroni supports running custom
 scripts to clone a new replica. Those are configured in the ``postgresql`` configuration block:
 
+.. code:: YAML
+
+    postgresql:
+        create_replica_methods:
+            - <method name>
+        <method name>:
+            command: <command name>
+            keep_data: True
+            no_params: True
+            no_master: 1
+
+example: wal_e
+
 .. code:: YAML
 
     postgresql:
@@ -79,6 +92,21 @@ scripts to clone a new replica. Those are configured in the ``postgresql`` confi
         basebackup:
             max-rate: '100M'
 
+example: pgbackrest
+
+.. code:: YAML
+
+    postgresql:
+        create_replica_methods:
+            - pgbackrest
+            - basebackup
+        pgbackrest:
+            command: /usr/bin/pgbackrest --stanza=<scope> --delta restore
+            keep_data: True
+            no_params: True
+        basebackup:
+            max-rate: '100M'
+
 
 The ``create_replica_methods`` defines available replica creation methods and the order of executing them. Patroni will
 stop on the first one that returns 0. Each method should define a separate section in the configuration file, listing the command
@@ -99,6 +127,10 @@ 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 instruct Patroni to  not clean PGDATA folder before calling restore.
+
+A special ``no_params`` parameter, if defined, restricts passing parameters to custom command.
+
 A ``basebackup`` method is a special case: it will be used if
 ``create_replica_methods`` is empty, although it is possible
 to list it explicitly among the ``create_replica_methods`` methods. This method initializes a new replica with the
@@ -145,9 +177,15 @@ standby nodes replicating from some remote master. This type of clusters has:
 
 Standby leader holds and updates a leader lock in DCS. If the leader lock
 expires, cascade replicas will perform an election to choose another leader
-from the standbys. For the sake of flexibility, you can specify different
-methods of creating a replica and recovery WAL records when a cluster is in the
-"standby mode", and after it was detached to function as a normal cluster.
+from the standbys.
+
+For the sake of flexibility, you can specify methods of creating a replica and
+recovery WAL records when a cluster is in the "standby mode" by providing
+`create_replica_methods` key in `standby_cluster` section. It is distinct from
+creating replicas, when cluster is detached and functions as a normal cluster,
+which is controlled by `create_replica_methods` in `postgresql` section. Both
+"standby" and "normal" `create_replica_methods` reference  keys in `postgresql`
+section.
 
 To configure such cluster you need to specify the section ``standby_cluster``
 in a patroni configuration:
@@ -160,6 +198,10 @@ in a patroni configuration:
                 host: 1.2.3.4
                 port: 5432
                 primary_slot_name: patroni
+                create_replica_methods:
+                - basebackup
 
 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 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

@@ -13,6 +13,8 @@ In asynchronous mode the cluster is allowed to lose some committed transactions
 
 The amount of transactions that can be lost is controlled via ``maximum_lag_on_failover`` parameter. Because the primary transaction log position is not sampled in real time, in reality the amount of lost data on failover is worst case bounded by  ``maximum_lag_on_failover`` bytes of transaction log plus the amount that is written in the last ``ttl`` seconds (``loop_wait``/2 seconds in the average case). However typical steady state replication delay is well under a second.
 
+By default, when running leader elections, Patroni does not take into account the current timeline of replicas, what in some cases could be undesirable behavior. You can prevent the node not having the same timeline as a former master become the new leader by changing the value of ``check_timeline`` parameter to ``true``.
+
 PostgreSQL synchronous replication
 ----------------------------------
 
@@ -44,7 +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``),
@@ -55,6 +57,8 @@ 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 mode implementation
 -------------------------------

docs/rest_api.rst

@@ -0,0 +1,339 @@
+.. _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 leader:
+
+  - ``GET /``
+  - ``GET /master``
+  - ``GET /leader``
+  - ``GET /primary``
+  - ``GET /read-write``
+
+- ``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 /read-only``: like the above endpoint, but also includes the primary.
+
+- ``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 /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 /health``: returns HTTP status code **200** only when PostgreSQL is up and running.
+
+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_keep_segments": 8,
+	      "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_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 "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_keep_segments": 8,
+	      "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_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
+	}
+
+
+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``.

extras/README.md

@@ -1,11 +1,11 @@
 ### confd
 
-`confd` directory contains haproxy template files for the [confd](https://github.com/kelseyhightower/confd) -- lightweight configuration management tool
+`confd` directory contains haproxy and pgbouncer template files for the [confd](https://github.com/kelseyhightower/confd) -- lightweight configuration management tool
 You need to copy content of `confd` directory into /etcd/confd and run confd service:
 ```bash
 $ confd -prefix=/service/$PATRONI_SCOPE -backend etcd -node $PATRONI_ETCD_URL -interval=10
 ```
-It will periodically update haproxy.cfg with the actual list of Patroni nodes from `etcd` and "reload" haproxy when it is necessary.
+It will periodically update haproxy.cfg and pgbouncer.ini with the actual list of Patroni nodes from `etcd` and "reload" haproxy and pgbouncer.ini when it is necessary.
 
 
 ### startup-scripts

extras/confd/conf.d/pgbouncer.toml

@@ -0,0 +1,12 @@
+[template]
+prefix = "/service/batman"
+owner = "postgres"
+mode = "0644"
+src = "pgbouncer.tmpl"
+dest = "/etc/pgbouncer/pgbouncer.ini"
+
+reload_cmd = "systemctl reload pgbouncer"
+
+keys = [
+    "/members/","/leader"
+]

extras/confd/templates/pgbouncer.tmpl

@@ -0,0 +1,17 @@
+[databases]
+{{with get "/leader"}}{{$leader := .Value}}{{$leadkey := printf "/members/%s" $leader}}{{with get $leadkey}}{{$data := json .Value}}{{$hostport := base (replace (index (split $data.conn_url "/") 2) "@" "/" -1)}}{{ $host := base (index (split $hostport ":") 0)}}{{ $port := base (index (split $hostport ":") 1)}}* = host={{ $host }} port={{ $port }} pool_size=10{{end}}{{end}}
+
+[pgbouncer]
+logfile = /var/log/postgresql/pgbouncer.log
+pidfile = /var/run/postgresql/pgbouncer.pid
+listen_addr = *
+listen_port = 6432
+unix_socket_dir = /var/run/postgresql
+auth_type = trust
+auth_file = /etc/pgbouncer/userlist.txt
+auth_hba_file = /etc/pgbouncer/pg_hba.txt
+admin_users = pgbouncer
+stats_users = pgbouncer
+pool_mode = session
+max_client_conn = 100
+default_pool_size = 20

extras/startup-scripts/patroni.service

@@ -18,8 +18,14 @@ WorkingDirectory=~
 
 # Where to send early-startup messages from the server
 # This is normally controlled by the global default set by systemd
-# StandardOutput=syslog
+#StandardOutput=syslog
 
+# Pre-commands to start watchdog device
+# Uncomment if watchdog is part of your patroni setup
+#ExecStartPre=-/usr/bin/sudo /sbin/modprobe softdog
+#ExecStartPre=-/usr/bin/sudo /bin/chown postgres /dev/watchdog
+
+# Start the patroni process
 ExecStart=/bin/patroni /etc/patroni.yml
 
 # Send HUP to reload from patroni.yml

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
@@ -35,10 +36,12 @@ 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.py

@@ -0,0 +1,17 @@
+#!/usr/bin/env python
+import os
+import psycopg2
+import sys
+
+
+if __name__ == '__main__':
+    if not (len(sys.argv) >= 3 and sys.argv[3] == "master"):
+        sys.exit(1)
+
+    os.environ['PGPASSWORD'] = 'zalando'
+    connection = psycopg2.connect(host='127.0.0.1', port=sys.argv[1], user='postgres')
+    cursor = connection.cursor()
+    cursor.execute("SELECT slot_name FROM pg_replication_slots WHERE slot_type = 'logical'")
+
+    with open("data/postgres0/label", "w") as label:
+        label.write(next(iter(cursor.fetchone()), ""))

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

@@ -1,17 +1,13 @@
 import abc
-import consul
 import datetime
-import etcd
-import kazoo.client
-import kazoo.exceptions
 import os
-import psutil
 import psycopg2
 import json
 import shutil
 import signal
 import six
 import subprocess
+import sys
 import tempfile
 import threading
 import time
@@ -84,7 +80,7 @@ class AbstractController(object):
 
 
 class PatroniController(AbstractController):
-    __PORT = 5440
+    __PORT = 5360
     PATRONI_CONFIG = '{}.yml'
     """ starts and stops individual patronis"""
 
@@ -110,19 +106,31 @@ class PatroniController(AbstractController):
         with open(os.path.join(self._data_dir, 'label'), 'w') as f:
             f.write(content)
 
-    def read_label(self):
+    def read_label(self, label):
         try:
-            with open(os.path.join(self._data_dir, 'label'), 'r') as f:
+            with open(os.path.join(self._data_dir, label), 'r') as f:
                 return f.read().strip()
         except IOError:
             return None
 
-    def add_tag_to_config(self, tag, value):
+    @staticmethod
+    def recursive_update(dst, src):
+        for k, v in src.items():
+            if k in dst and isinstance(dst[k], dict):
+                PatroniController.recursive_update(dst[k], v)
+            else:
+                dst[k] = v
+
+    def update_config(self, custom_config):
         with open(self._config) as r:
             config = yaml.safe_load(r)
-            config['tags']['tag'] = value
+            self.recursive_update(config, custom_config)
             with open(self._config, 'w') as w:
                 yaml.safe_dump(config, w, default_flow_style=False)
+        self._scope = config.get('scope', 'batman')
+
+    def add_tag_to_config(self, tag, value):
+        self.update_config({'tags': {tag: value}})
 
     def _start(self):
         if self.watchdog:
@@ -130,7 +138,8 @@ class PatroniController(AbstractController):
         if isinstance(self._context.dcs_ctl, KubernetesController):
             self._context.dcs_ctl.create_pod(self._name[8:], self._scope)
             os.environ['PATRONI_KUBERNETES_POD_IP'] = '10.0.0.' + self._name[-1]
-        return subprocess.Popen(['coverage', 'run', '--source=patroni', '-p', 'patroni.py', self._config],
+        return subprocess.Popen([sys.executable, '-m', 'coverage', 'run',
+                                '--source=patroni', '-p', 'patroni.py', self._config],
                                 stdout=self._log, stderr=subprocess.STDOUT, cwd=self._work_directory)
 
     def stop(self, kill=False, timeout=15, postgres=False):
@@ -162,7 +171,8 @@ class PatroniController(AbstractController):
 
         config['name'] = name
         config['postgresql']['data_dir'] = self._data_dir
-        config['postgresql']['use_unix_socket'] = True
+        config['postgresql']['use_unix_socket'] = 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',
@@ -174,13 +184,10 @@ class PatroniController(AbstractController):
                 config['bootstrap']['initdb'].extend([{'auth': 'md5'}, {'auth-host': 'md5'}])
 
         if custom_config is not None:
-            def recursive_update(dst, src):
-                for k, v in src.items():
-                    if k in dst and isinstance(dst[k], dict):
-                        recursive_update(dst[k], v)
-                    else:
-                        dst[k] = v
-            recursive_update(config, custom_config)
+            self.recursive_update(config, custom_config)
+
+        if config['postgresql'].get('callbacks', {}).get('on_role_change'):
+            config['postgresql']['callbacks']['on_role_change'] += ' ' + str(self.__PORT)
 
         with open(patroni_config_path, 'w') as f:
             yaml.safe_dump(config, f, default_flow_style=False)
@@ -238,59 +245,24 @@ 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='basebackup'):
-        subprocess.call([PatroniPoolController.BACKUP_SCRIPT, '--walmethod=none',
-                         '--datadir=' + os.path.join(self._output_dir, dest),
-                         '--dbname=' + self.backup_source])
+    def backup(self, dest=os.path.join('data', 'basebackup')):
+        subprocess.call(PatroniPoolController.BACKUP_SCRIPT + ['--walmethod=none',
+                        '--datadir=' + os.path.join(self._work_directory, dest),
+                        '--dbname=' + self.backup_source])
 
 
 class ProcessHang(object):
@@ -363,9 +335,12 @@ class ConsulController(AbstractDcsController):
     def __init__(self, context):
         super(ConsulController, self).__init__(context)
         os.environ['PATRONI_CONSUL_HOST'] = 'localhost:8500'
-        self._client = consul.Consul()
+        os.environ['PATRONI_CONSUL_REGISTER_SERVICE'] = 'on'
         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:
@@ -405,6 +380,8 @@ class EtcdController(AbstractDcsController):
     def __init__(self, context):
         super(EtcdController, self).__init__(context)
         os.environ['PATRONI_ETCD_HOST'] = 'localhost:2379'
+
+        import etcd
         self._client = etcd.Client(port=2379)
 
     def _start(self):
@@ -412,12 +389,14 @@ class EtcdController(AbstractDcsController):
                                 stdout=self._log, stderr=subprocess.STDOUT)
 
     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):
@@ -461,13 +440,13 @@ class KubernetesController(AbstractDcsController):
 
     def delete_pod(self, name):
         try:
-            self._api.delete_namespaced_pod(name, self._namespace, self._client.V1DeleteOptions())
-        except:
+            self._api.delete_namespaced_pod(name, self._namespace, body=self._client.V1DeleteOptions())
+        except Exception:
             pass
         while True:
             try:
                 self._api.read_namespaced_pod(name, self._namespace)
-            except:
+            except Exception:
                 break
 
     def query(self, key, scope='batman'):
@@ -476,22 +455,23 @@ 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)
-                if key == 'leader':
+                ep = scope + {'leader': '', 'history': '-config', 'initialize': '-config'}.get(key, '-' + key)
+                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:
@@ -511,18 +491,22 @@ 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):
@@ -549,7 +533,8 @@ class ExhibitorController(ZooKeeperController):
 
 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
@@ -582,9 +567,8 @@ 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',
-                        'get_watchdog', 'database_is_running', 'checkpoint_hang', 'patroni_hang',
-                        'terminate_backends', 'backup']:
+        if func not in ['stop', 'query', 'write_label', 'read_label', 'check_role_has_changed_to',
+                        'add_tag_to_config', 'get_watchdog', 'patroni_hang', 'backup']:
             raise AttributeError("PatroniPoolController instance has no attribute '{0}'".format(func))
 
         def wrapper(name, *args, **kwargs):
@@ -598,7 +582,7 @@ 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)
@@ -611,7 +595,7 @@ 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': {
@@ -624,8 +608,9 @@ class PatroniPoolController(object):
             'postgresql': {
                 'parameters': {
                     'archive_mode': 'on',
-                    'archive_command': 'mkdir -p {0} && test ! -f {0}/%f && cp %p {0}/%f'.format(
-                            os.path.join(self._output_dir, 'wal_archive'))
+                    'archive_command': (self.ARCHIVE_RESTORE_SCRIPT + ' --mode archive ' +
+                                        '--dirname {} --filename %f --pathname %p').format(
+                                        os.path.join(self.patroni_path, 'data', 'wal_archive'))
                 },
                 'authentication': {
                     'superuser': {'password': 'zalando1'},
@@ -641,11 +626,14 @@ class PatroniPoolController(object):
             'bootstrap': {
                 'method': 'backup_restore',
                 'backup_restore': {
-                    'command': 'features/backup_restore.sh --sourcedir=' + os.path.join(self._output_dir, 'basebackup'),
+                    'command': (sys.executable + ' features/backup_restore.py --sourcedir=' +
+                                os.path.join(self.patroni_path, 'data', 'basebackup')),
                     'recovery_conf': {
                         'recovery_target_action': 'promote',
                         'recovery_target_timeline': 'latest',
-                        'restore_command': 'cp {0}/wal_archive/%f %p'.format(self._output_dir)
+                        'restore_command': (self.ARCHIVE_RESTORE_SCRIPT + ' --mode restore ' +
+                                            '--dirname {} --filename %f --pathname %p').format(
+                                            os.path.join(self.patroni_path, 'data', 'wal_archive'))
                     }
                 }
             },
@@ -799,8 +787,8 @@ def before_all(context):
 
 def after_all(context):
     context.dcs_ctl.stop()
-    subprocess.call(['coverage', 'combine'])
-    subprocess.call(['coverage', 'report'])
+    subprocess.call([sys.executable, '-m', 'coverage', 'combine'])
+    subprocess.call([sys.executable, '-m', 'coverage', 'report'])
 
 
 def before_feature(context, feature):
@@ -813,3 +801,5 @@ def after_feature(context, feature):
     context.pctl.stop_all()
     shutil.rmtree(os.path.join(context.pctl.patroni_path, 'data'))
     context.dcs_ctl.cleanup_service_tree()
+    if feature.status == 'failed':
+        shutil.copytree(context.pctl.output_dir, context.pctl.output_dir + '_failed')

features/patroni_api.feature

@@ -8,6 +8,10 @@ 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
 	Then I receive a response code 503
 	When I run patronictl.py reinit batman postgres0 --force
@@ -26,10 +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
-	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
+	Given 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
 
@@ -43,7 +44,21 @@ Scenario: check dynamic configuration change via DCS
 	And I receive a response loop_wait 2
 	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 {'tag': 'new_value'}
+	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
@@ -71,6 +86,7 @@ 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
@@ -92,6 +108,7 @@ 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
@@ -100,16 +117,3 @@ 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

@@ -1,16 +1,48 @@
 Feature: standby cluster
+  Scenario: check permanent logical slots are preserved 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, "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
+    And I sleep for 3 seconds
+    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
+    Then "members/postgres0" key in DCS has state=running after 10 seconds
+    And replication works from postgres1 to postgres0 after 15 seconds
+    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 postgres0 without slots sync
-    And I create a replication slot postgres1 on postgres0
-    And I start postgres1 in a standby cluster batman1 as a clone of postgres0
+    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
+    When I issue a GET request to http://127.0.0.1:8009/master
+    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_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
 
   Scenario: check failover
     When I kill postgres1
-	Then postgres2 is replicating from postgres0 after 20 seconds
+    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 503
+    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
+    And replication works from postgres0 to postgres2 after 15 seconds
+    And there is a postgres2_cb.log with "on_start replica batman1\non_role_change standby_leader batman1" in postgres2 data directory

features/steps/cascading_replication.py

@@ -9,10 +9,11 @@ def start_patroni_with_a_name_value_tag(context, name, tag_name, tag_value):
     return context.pctl.start(name, custom_config={'tags': {tag_name: tag_value}})
 
 
-@then('There is a label with "{content:w}" in {name:w} data directory')
-def check_label(context, content, name):
-    label = context.pctl.read_label(name)
-    assert label == content, "{0} is not equal to {1}".format(label, content)
+@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)
+    label = label.replace('\n', '\\n')
+    assert content in label, "{0} doesn't contain {1}".format(label, content)
 
 
 @step('I create label with "{content:w}" in {name:w} data directory')
@@ -33,3 +34,17 @@ def check_member(context, name, key, value, time_limit):
             pass
         time.sleep(1)
     assert False, "{0} does not have {1}={2} in dcs after {3} seconds".format(name, key, 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

@@ -1,18 +1,19 @@
-import base64
 import json
 import os
 import parse
-import requests
 import shlex
 import subprocess
+import sys
 import time
 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|\.|:|/)+')
@@ -44,9 +45,9 @@ def sleep_for_n_seconds(context, value):
 
 
 def _set_response(context, response):
-    context.status_code = response.status_code
-    data = response.content.decode('utf-8')
-    ct = response.headers.get('content-type', '')
+    context.status_code = response.status
+    data = response.data.decode('utf-8')
+    ct = response.getheader('content-type', '')
     if ct.startswith('application/json') or\
             ct.startswith('text/yaml') or\
             ct.startswith('text/x-yaml') or\
@@ -62,13 +63,7 @@ def _set_response(context, response):
 
 @step('I issue a GET request to {url:url}')
 def do_get(context, url):
-    try:
-        r = requests.get(url)
-    except requests.exceptions.RequestException:
-        context.status_code = None
-        context.response = None
-    else:
-        _set_response(context, r)
+    do_request(context, 'GET', url, None)
 
 
 @step('I issue an empty POST request to {url:url}')
@@ -78,24 +73,18 @@ 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 {}
-    headers = {'Authorization': 'Basic ' + base64.b64encode('username:password'.encode('utf-8')).decode('utf-8'),
-               'Content-Type': 'application/json'}
+    data = data and json.loads(data)
     try:
-        if request_method == 'PATCH':
-            r = requests.patch(url, headers=headers, json=data)
-        else:
-            r = requests.post(url, headers=headers, json=data)
-    except requests.exceptions.RequestException:
-        context.status_code = None
-        context.response = None
+        r = request_executor.request(request_method, url, data)
+    except Exception:
+        context.status_code = context.response = None
     else:
         _set_response(context, r)
 
 
 @step('I run {cmd}')
 def do_run(context, cmd):
-    cmd = ['coverage', 'run', '--source=patroni', '-p'] + shlex.split(cmd)
+    cmd = [sys.executable, '-m', 'coverage', 'run', '--source=patroni', '-p'] + shlex.split(cmd)
     try:
         # XXX: Dirty hack! We need to take name/passwd from the config!
         env = os.environ.copy()
@@ -148,8 +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)
-        if (value in r.content.decode('utf-8')) != negate:
+        r = request_executor.request('GET', url)
+        if (value in r.data.decode('utf-8')) != negate:
             break
         time.sleep(1)
     else:

features/steps/standby_cluster.py

@@ -1,3 +1,5 @@
+import os
+import sys
 import time
 
 from behave import step
@@ -8,19 +10,16 @@ SELECT * FROM pg_catalog.pg_stat_replication
 WHERE application_name = '{0}'
 """
 
-create_replication_slot_query = """
-SELECT pg_create_physical_replication_slot('{0}')
-"""
+executable = sys.executable if os.name != 'nt' else sys.executable.replace('\\', '/')
+callback = executable + " features/callback2.py "
 
 
-@step('I start {name:w} without slots sync')
-def start_patroni_without_slots_sync(context, name):
+@step('I start {name:w} with callback configured')
+def start_patroni_with_callbacks(context, name):
     return context.pctl.start(name, custom_config={
-        "bootstrap": {
-            "dcs": {
-                "postgresql": {
-                    "use_slots": False
-                }
+        "postgresql": {
+            "callbacks": {
+                "on_role_change": executable + " features/callback.py"
             }
         }
     })
@@ -29,25 +28,41 @@ def start_patroni_without_slots_sync(context, name):
 @step('I start {name:w} in a cluster {cluster_name:w}')
 def start_patroni(context, name, cluster_name):
     return context.pctl.start(name, custom_config={
-        "scope": cluster_name
+        "scope": cluster_name,
+        "postgresql": {
+            "callbacks": {c: callback + name for c in ('on_start', 'on_stop', 'on_restart', 'on_role_change')},
+            "backup_restore": {
+                "command": (executable + " features/backup_restore.py --sourcedir=" +
+                            os.path.join(context.pctl.patroni_path, 'data', 'basebackup'))}
+        }
     })
 
 
 @step('I start {name:w} in a standby cluster {cluster_name:w} as a clone of {name2:w}')
-def start_patroni_stanby_cluster(context, name, cluster_name, name2):
+def start_patroni_standby_cluster(context, name, cluster_name, name2):
+    # we need to remove patroni.dynamic.json in order to "bootstrap" standby cluster with existing PGDATA
+    os.unlink(os.path.join(context.pctl._processes[name]._data_dir, 'patroni.dynamic.json'))
     port = context.pctl._processes[name2]._connkwargs.get('port')
-    return context.pctl.start(name, custom_config={
+    context.pctl._processes[name].update_config({
         "scope": cluster_name,
         "bootstrap": {
             "dcs": {
+                "ttl": 20,
+                "loop_wait": 2,
+                "retry_timeout": 5,
                 "standby_cluster": {
                     "host": "localhost",
                     "port": port,
-                    "primary_slot_name": "postgres1",
+                    "primary_slot_name": "pm_1",
+                    "create_replica_methods": ["backup_restore", "basebackup"]
                 }
             }
+        },
+        "postgresql": {
+            "callbacks": {c: callback + name for c in ('on_start', 'on_stop', 'on_restart', 'on_role_change')}
         }
     })
+    return context.pctl.start(name)
 
 
 @step('{pg_name1:w} is replicating from {pg_name2:w} after {timeout:d} seconds')
@@ -62,17 +77,8 @@ def check_replication_status(context, pg_name1, pg_name2, timeout):
         )
 
         if cur and len(cur.fetchall()) != 0:
-            return True
+            break
 
         time.sleep(1)
-
-    return False
-
-
-@step('I create a replication slot {slot_name:w} on {pg_name:w}')
-def create_replication_slot(context, slot_name, pg_name):
-    return context.pctl.query(
-        pg_name,
-        create_replication_slot_query.format(slot_name),
-        fail_ok=True
-    )
+    else:
+        assert False, "{0} is not replicating from {1} after {2} seconds".format(pg_name1, pg_name2, timeout)

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/Dockerfile

@@ -1,32 +1,34 @@
-FROM postgres:9.6
+FROM postgres:11
 MAINTAINER Alexander Kukushkin <[email protected]>
 
 RUN export DEBIAN_FRONTEND=noninteractive \
     && echo 'APT::Install-Recommends "0";\nAPT::Install-Suggests "0";' > /etc/apt/apt.conf.d/01norecommend \
     && apt-get update -y \
     && apt-get upgrade -y \
-    && apt-get install -y git curl jq python-psycopg2 python-yaml python-requests python-six python-pysocks \
-        python-dateutil python-pip python-prettytable python-wheel python-psutil python locales \
-
+    && apt-cache depends patroni | sed -n -e 's/.* Depends: \(python3-.\+\)$/\1/p' \
+            | grep -Ev '^python3-(sphinx|etcd|consul|kazoo|kubernetes)' \
+            | xargs apt-get install -y vim-tiny curl jq locales git python3-pip python3-wheel \
     ## Make sure we have a en_US.UTF-8 locale available
     && localedef -i en_US -c -f UTF-8 -A /usr/share/locale/locale.alias en_US.UTF-8 \
-
-    && pip install setuptools pip --upgrade \
-    && pip install 'git+https://github.com/zalando/patroni.git#egg=patroni[kubernetes]' \
-
-    && mkdir -p /home/postgres \
-    && chown postgres:postgres /home/postgres \
-
+    && pip3 install setuptools \
+    && pip3 install 'git+https://github.com/zalando/patroni.git#egg=patroni[kubernetes]' \
+    && PGHOME=/home/postgres \
+    && mkdir -p $PGHOME \
+    && chown postgres $PGHOME \
+    && sed -i "s|/var/lib/postgresql.*|$PGHOME:/bin/bash|" /etc/passwd \
+    # Set permissions for OpenShift
+    && chmod 775 $PGHOME \
+    && chmod 664 /etc/passwd \
     # Clean up
-    && apt-get remove -y git python-pip python-setuptools \
+    && apt-get remove -y git python3-pip python3-wheel \
     && apt-get autoremove -y \
     && apt-get clean -y \
     && rm -rf /var/lib/apt/lists/* /root/.cache
 
-ADD entrypoint.sh callback.py /
+ADD entrypoint.sh /
 
 EXPOSE 5432 8008
-ENV LC_ALL=en_US.UTF-8 LANG=en_US.UTF-8
+ENV LC_ALL=en_US.UTF-8 LANG=en_US.UTF-8 EDITOR=/usr/bin/editor
 USER postgres
 WORKDIR /home/postgres
 CMD ["/bin/bash", "/entrypoint.sh"]

kubernetes/callback.py

@@ -1,62 +0,0 @@
-#!/usr/bin/env python
-
-import logging
-import os
-import socket
-import sys
-import time
-
-from kubernetes import client as k8s_client, config as k8s_config
-from urllib3.exceptions import HTTPError
-from six.moves.http_client import HTTPException
-
-logger = logging.getLogger(__name__)
-
-
-class CoreV1Api(k8s_client.CoreV1Api):
-
-    def retry(func):
-        def wrapped(*args, **kwargs):
-            count = 0
-            while True:
-                try:
-                    return func(*args, **kwargs)
-                except (HTTPException, HTTPError, socket.error, socket.timeout):
-                    if count >= 10:
-                        raise
-                    logger.info('Throttling API requests...')
-                    time.sleep(2 ** count * 0.5)
-                    count += 1
-        return wrapped
-
-    @retry
-    def patch_namespaced_endpoints(self, *args, **kwargs):
-        return super(CoreV1Api, self).patch_namespaced_endpoints(*args, **kwargs)
-
-
-def patch_master_endpoint(api, namespace, cluster):
-    addresses = [k8s_client.V1EndpointAddress(ip=os.environ['POD_IP'])]
-    ports = [k8s_client.V1EndpointPort(port=5432)]
-    subsets = [k8s_client.V1EndpointSubset(addresses=addresses, ports=ports)]
-    body = k8s_client.V1Endpoints(subsets=subsets)
-    return api.patch_namespaced_endpoints(cluster, namespace, body)
-
-
-def main():
-    logging.basicConfig(format='%(asctime)s %(levelname)s: %(message)s', level=logging.INFO)
-    if len(sys.argv) != 4 or sys.argv[1] not in ('on_start', 'on_stop', 'on_role_change'):
-        sys.exit('Usage: %s <action> <role> <cluster_name>', sys.argv[0])
-
-    action, role, cluster = sys.argv[1:4]
-
-    k8s_config.load_incluster_config()
-    k8s_api = CoreV1Api()
-
-    namespace = os.environ['KUBERNETES_NAMESPACE']
-
-    if role == 'master' and action in ('on_start', 'on_role_change'):
-        patch_master_endpoint(k8s_api, namespace, cluster)
-
-
-if __name__ == '__main__':
-    main()

kubernetes/entrypoint.sh

@@ -1,5 +1,12 @@
 #!/bin/bash
 
+if [[ $UID -ge 10000 ]]; then
+    GID=$(id -g)
+    sed -e "s/^postgres:x:[^:]*:[^:]*:/postgres:x:$UID:$GID:/" /etc/passwd > /tmp/passwd
+    cat /tmp/passwd > /etc/passwd
+    rm /tmp/passwd
+fi
+
 cat > /home/postgres/patroni.yml <<__EOF__
 bootstrap:
   dcs:
@@ -13,24 +20,20 @@ bootstrap:
   - data-checksums
   pg_hba:
   - host all all 0.0.0.0/0 md5
-  - host replication ${PATRONI_REPLICATION_USERNAME} ${POD_IP}/16    md5
+  - host replication ${PATRONI_REPLICATION_USERNAME} ${PATRONI_KUBERNETES_POD_IP}/16 md5
 restapi:
-  connect_address: '${POD_IP}:8008'
+  connect_address: '${PATRONI_KUBERNETES_POD_IP}:8008'
 postgresql:
-  connect_address: '${POD_IP}:5432'
+  connect_address: '${PATRONI_KUBERNETES_POD_IP}:5432'
   authentication:
     superuser:
       password: '${PATRONI_SUPERUSER_PASSWORD}'
     replication:
       password: '${PATRONI_REPLICATION_PASSWORD}'
-  callbacks:
-    on_start: /callback.py
-    on_stop: /callback.py
-    on_role_change: /callback.py
 __EOF__
 
 unset PATRONI_SUPERUSER_PASSWORD PATRONI_REPLICATION_PASSWORD
 export KUBERNETES_NAMESPACE=$PATRONI_KUBERNETES_NAMESPACE
 export POD_NAME=$PATRONI_NAME
 
-exec /usr/bin/python /usr/local/bin/patroni /home/postgres/patroni.yml
+exec /usr/bin/python3 /usr/local/bin/patroni /home/postgres/patroni.yml

kubernetes/openshift-example/README.md

@@ -0,0 +1,49 @@
+# Patroni OpenShift Configuration
+Patroni can be run in OpenShift. Based on the kubernetes configuration, the Dockerfile and Entrypoint has been modified to support the dynamic UID/GID configuration that is applied in OpenShift. This can be run under the standard `restricted` SCC. 
+
+# Examples
+
+## Create test project
+
+```
+oc new-project patroni-test
+```
+
+## Build the image
+
+Note: Update the references when merged upstream. 
+Note: If deploying as a template for multiple users, the following commands should be performed in a shared namespace like `openshift`. 
+
+```
+oc import-image postgres:10 --confirm -n openshift
+oc new-build https://github.com/zalando/patroni --context-dir=kubernetes -n openshift
+```
+
+## Deploy the Image 
+Two configuration templates exist in [templates](templates) directory: 
+- Patroni Ephemeral
+- Patroni Persistent
+
+The only difference is whether or not the statefulset requests persistent storage. 
+
+## Create the Template
+Install the template into the `openshift` namespace if this should be shared across projects: 
+
+```
+oc create -f templates/template_patroni_ephemeral.yml -n openshift
+```
+
+Then, from your own project: 
+
+```
+oc new-app patroni-pgsql-ephemeral
+```
+
+Once the pods are running, two configmaps should be available: 
+
+```
+$ oc get configmap
+NAME                DATA      AGE
+patroniocp-config   0         1m
+patroniocp-leader   0         1m
+```

kubernetes/openshift-example/templates/template_patroni_ephemeral.yml

@@ -0,0 +1,287 @@
+apiVersion: v1
+kind: Template
+metadata:
+  name: patroni-pgsql-ephemeral
+  annotations:
+    description: |-
+      Patroni Postgresql database cluster, without 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 (Ephemeral)
+    openshift.io/long-description: This template deploys a a patroni postgresql HA cluster without persistent storage.
+    tags: postgresql
+objects:
+- apiVersion: v1
+  kind: Service
+  metadata:
+    creationTimestamp: null
+    labels:
+      application: ${APPLICATION_NAME}
+      cluster-name: ${PATRONI_CLUSTER_NAME}
+    name: ${PATRONI_CLUSTER_NAME}
+  spec:
+    ports:
+    - port: 5432
+      protocol: TCP
+      targetPort: 5432
+    sessionAffinity: None
+    type: ClusterIP
+  status:
+    loadBalancer: {}
+- apiVersion: v1
+  kind: Service
+  metadata:
+    creationTimestamp: null
+    labels:
+      application: ${APPLICATION_NAME}
+      cluster-name: ${PATRONI_CLUSTER_NAME}
+    name: ${PATRONI_MASTER_SERVICE_NAME}
+  spec:
+    ports:
+    - port: 5432
+      protocol: TCP
+      targetPort: 5432
+    selector:
+      application: ${APPLICATION_NAME}
+      cluster-name: ${PATRONI_CLUSTER_NAME}
+      role: master
+    sessionAffinity: None
+    type: ClusterIP
+  status:
+    loadBalancer: {}
+- apiVersion: v1
+  kind: Secret
+  metadata:
+    name: ${PATRONI_CLUSTER_NAME}
+    labels:
+      application: ${APPLICATION_NAME}
+      cluster-name: ${PATRONI_CLUSTER_NAME}
+  stringData:
+    superuser-password: ${PATRONI_SUPERUSER_PASSWORD}
+    replication-password: ${PATRONI_REPLICATION_PASSWORD}
+- apiVersion: v1
+  kind: Service
+  metadata:
+    creationTimestamp: null
+    labels:
+      application: ${APPLICATION_NAME}
+      cluster-name: ${PATRONI_CLUSTER_NAME}
+    name: ${PATRONI_REPLICA_SERVICE_NAME}
+  spec:
+    ports:
+    - port: 5432
+      protocol: TCP
+      targetPort: 5432
+    selector:
+      application: ${APPLICATION_NAME}
+      cluster-name: ${PATRONI_CLUSTER_NAME}
+      role: replica
+    sessionAffinity: None
+    type: ClusterIP
+  status:
+    loadBalancer: {}
+- apiVersion: apps/v1
+  kind: StatefulSet
+  metadata:
+    creationTimestamp: null
+    generation: 3
+    labels:
+      application: ${APPLICATION_NAME}
+      cluster-name: ${PATRONI_CLUSTER_NAME}
+    name: ${APPLICATION_NAME}
+  spec:
+    podManagementPolicy: OrderedReady
+    replicas: 3
+    revisionHistoryLimit: 10
+    selector:
+      matchLabels:
+        application: ${APPLICATION_NAME}
+        cluster-name: ${PATRONI_CLUSTER_NAME}
+    serviceName: ${APPLICATION_NAME}
+    template:
+      metadata:
+        creationTimestamp: null
+        labels:
+          application: ${APPLICATION_NAME}
+          cluster-name: ${PATRONI_CLUSTER_NAME}
+      spec:
+        containers:
+        - env:
+          - name: PATRONI_KUBERNETES_POD_IP
+            valueFrom:
+              fieldRef:
+                apiVersion: v1
+                fieldPath: status.podIP
+          - name: PATRONI_KUBERNETES_NAMESPACE
+            valueFrom:
+              fieldRef:
+                apiVersion: v1
+                fieldPath: metadata.namespace
+          - name: PATRONI_KUBERNETES_LABELS
+            value: '{application: ${APPLICATION_NAME}, cluster-name: ${PATRONI_CLUSTER_NAME}}'
+          - name: PATRONI_SUPERUSER_USERNAME
+            value: ${PATRONI_SUPERUSER_USERNAME}
+          - name: PATRONI_SUPERUSER_PASSWORD
+            valueFrom:
+              secretKeyRef:
+                key: superuser-password
+                name: ${PATRONI_CLUSTER_NAME}
+          - name: PATRONI_REPLICATION_USERNAME
+            value: ${PATRONI_REPLICATION_USERNAME}
+          - name: PATRONI_REPLICATION_PASSWORD
+            valueFrom:
+              secretKeyRef:
+                key: replication-password
+                name: ${PATRONI_CLUSTER_NAME}
+          - name: PATRONI_SCOPE
+            value: ${PATRONI_CLUSTER_NAME}
+          - name: PATRONI_NAME
+            valueFrom:
+              fieldRef:
+                apiVersion: v1
+                fieldPath: metadata.name
+          - name: PATRONI_POSTGRESQL_DATA_DIR
+            value: /home/postgres/pgdata/pgroot/data
+          - name: PATRONI_POSTGRESQL_PGPASS
+            value: /tmp/pgpass
+          - name: PATRONI_POSTGRESQL_LISTEN
+            value: 0.0.0.0:5432
+          - name: PATRONI_RESTAPI_LISTEN
+            value: 0.0.0.0:8008
+          image: docker-registry.default.svc:5000/${NAMESPACE}/patroni:latest
+          imagePullPolicy: IfNotPresent
+          name: ${APPLICATION_NAME}
+          ports:
+          - containerPort: 8008
+            protocol: TCP
+          - containerPort: 5432
+            protocol: TCP
+          resources: {}
+          terminationMessagePath: /dev/termination-log
+          terminationMessagePolicy: File
+          volumeMounts:
+          - mountPath: /home/postgres/pgdata
+            name: pgdata
+        dnsPolicy: ClusterFirst
+        restartPolicy: Always
+        schedulerName: default-scheduler
+        securityContext: {}
+        serviceAccount: ${SERVICE_ACCOUNT}
+        serviceAccountName: ${SERVICE_ACCOUNT}
+        terminationGracePeriodSeconds: 0
+        volumes:
+        - name: pgdata
+          emptyDir: {}
+    updateStrategy:
+      type: OnDelete
+- apiVersion: v1
+  kind: Endpoints
+  metadata:
+    name: ${APPLICATION_NAME}
+    labels:
+      application: ${APPLICATION_NAME}
+      cluster-name: ${PATRONI_CLUSTER_NAME}
+  subsets: []
+- apiVersion: v1
+  kind: ServiceAccount
+  metadata:
+    name: ${SERVICE_ACCOUNT}
+- apiVersion: rbac.authorization.k8s.io/v1
+  kind: Role
+  metadata:
+    name: ${SERVICE_ACCOUNT}
+  rules:
+  - apiGroups:
+    - ""
+    resources:
+    - configmaps
+    verbs:
+    - create
+    - get
+    - list
+    - patch
+    - update
+    - watch
+    # delete is required only for 'patronictl remove'
+    - delete
+  - apiGroups:
+    - ""
+    resources:
+    - endpoints
+    verbs:
+    - get
+    - patch
+    - update
+    # the following three privileges are necessary only when using endpoints
+    - create
+    - list
+    - watch
+    # delete is required only for for 'patronictl remove'
+    - delete
+  - apiGroups:
+    - ""
+    resources:
+    - pods
+    verbs:
+    - get
+    - list
+    - patch
+    - update
+    - watch
+- apiVersion: rbac.authorization.k8s.io/v1
+  kind: RoleBinding
+  metadata:
+    name: ${SERVICE_ACCOUNT}
+  roleRef:
+    apiGroup: rbac.authorization.k8s.io
+    kind: Role
+    name: ${SERVICE_ACCOUNT}
+  subjects:
+  - kind: ServiceAccount
+    name: ${SERVICE_ACCOUNT}
+parameters:
+- description: The name of the application for labelling all artifacts.
+  displayName: Application Name
+  name: APPLICATION_NAME
+  value: patroni-ephemeral
+- description: The name of the patroni-pgsql cluster.
+  displayName: Cluster Name
+  name: PATRONI_CLUSTER_NAME
+  value: patroni-ephemeral
+- description: The name of the OpenShift Service exposed for the patroni-ephemeral-master container.
+  displayName: Master service name.  
+  name: PATRONI_MASTER_SERVICE_NAME
+  value: patroni-ephemeral-master
+- description: The name of the OpenShift Service exposed for the patroni-ephemeral-replica containers.
+  displayName: Replica service name. 
+  name: PATRONI_REPLICA_SERVICE_NAME
+  value: patroni-ephemeral-replica
+- description: Maximum amount of memory the container can use.
+  displayName: Memory Limit
+  name: MEMORY_LIMIT
+  value: 512Mi
+- description: The OpenShift Namespace where the patroni and postgresql ImageStream resides.
+  displayName: ImageStream Namespace
+  name: NAMESPACE
+  value: openshift
+- description: Username of the superuser account for initialization.
+  displayName: Superuser Username
+  name: PATRONI_SUPERUSER_USERNAME
+  value: postgres
+- description: Password of the superuser account for initialization.
+  displayName: Superuser Passsword
+  name: PATRONI_SUPERUSER_PASSWORD
+  value: postgres
+- description: Username of the replication account for initialization.
+  displayName: Replication Username
+  name: PATRONI_REPLICATION_USERNAME
+  value: postgres
+- description: Password of the replication account for initialization.
+  displayName: Repication Passsword
+  name: PATRONI_REPLICATION_PASSWORD
+  value: postgres
+- description: Service account name used for pods and rolebindings to form a cluster in the project. 
+  displayName: Service Account
+  name: SERVICE_ACCOUNT
+  value: patroniocp

kubernetes/openshift-example/templates/template_patroni_persistent.yaml

@@ -0,0 +1,317 @@
+apiVersion: v1
+kind: Template
+metadata:
+  name: patroni-pgsql-persistent
+  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.
+    tags: postgresql
+objects:
+- apiVersion: v1
+  kind: Service
+  metadata:
+    creationTimestamp: null
+    labels:
+      application: ${APPLICATION_NAME}
+      cluster-name: ${PATRONI_CLUSTER_NAME}
+    name: ${PATRONI_CLUSTER_NAME}
+  spec:
+    ports:
+    - port: 5432
+      protocol: TCP
+      targetPort: 5432
+    sessionAffinity: None
+    type: ClusterIP
+  status:
+    loadBalancer: {}
+- apiVersion: v1
+  kind: Service
+  metadata:
+    creationTimestamp: null
+    labels:
+      application: ${APPLICATION_NAME}
+      cluster-name: ${PATRONI_CLUSTER_NAME}
+    name: ${PATRONI_MASTER_SERVICE_NAME}
+  spec:
+    ports:
+    - port: 5432
+      protocol: TCP
+      targetPort: 5432
+    selector:
+      application: ${APPLICATION_NAME}
+      cluster-name: ${PATRONI_CLUSTER_NAME}
+      role: master
+    sessionAffinity: None
+    type: ClusterIP
+  status:
+    loadBalancer: {}
+- apiVersion: v1
+  kind: Secret
+  metadata:
+    name: ${PATRONI_CLUSTER_NAME}
+    labels:
+      application: ${APPLICATION_NAME}
+      cluster-name: ${PATRONI_CLUSTER_NAME}
+  stringData:
+    superuser-password: ${PATRONI_SUPERUSER_PASSWORD}
+    replication-password: ${PATRONI_REPLICATION_PASSWORD}
+- apiVersion: v1
+  kind: Service
+  metadata:
+    creationTimestamp: null
+    labels:
+      application: ${APPLICATION_NAME}
+      cluster-name: ${PATRONI_CLUSTER_NAME}
+    name: ${PATRONI_REPLICA_SERVICE_NAME}
+  spec:
+    ports:
+    - port: 5432
+      protocol: TCP
+      targetPort: 5432
+    selector:
+      application: ${APPLICATION_NAME}
+      cluster-name: ${PATRONI_CLUSTER_NAME}
+      role: replica
+    sessionAffinity: None
+    type: ClusterIP
+  status:
+    loadBalancer: {}
+- apiVersion: apps/v1
+  kind: StatefulSet
+  metadata:
+    creationTimestamp: null
+    generation: 3
+    labels:
+      application: ${APPLICATION_NAME}
+      cluster-name: ${PATRONI_CLUSTER_NAME}
+    name: ${APPLICATION_NAME}
+  spec:
+    podManagementPolicy: OrderedReady
+    replicas: 3
+    revisionHistoryLimit: 10
+    selector:
+      matchLabels:
+        application: ${APPLICATION_NAME}
+        cluster-name: ${PATRONI_CLUSTER_NAME}
+    serviceName: ${APPLICATION_NAME}
+    template:
+      metadata:
+        creationTimestamp: null
+        labels:
+          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
+            valueFrom:
+              fieldRef:
+                apiVersion: v1
+                fieldPath: status.podIP
+          - name: PATRONI_KUBERNETES_NAMESPACE
+            valueFrom:
+              fieldRef:
+                apiVersion: v1
+                fieldPath: metadata.namespace
+          - name: PATRONI_KUBERNETES_LABELS
+            value: '{application: ${APPLICATION_NAME}, cluster-name: ${PATRONI_CLUSTER_NAME}}'
+          - name: PATRONI_SUPERUSER_USERNAME
+            value: ${PATRONI_SUPERUSER_USERNAME}
+          - name: PATRONI_SUPERUSER_PASSWORD
+            valueFrom:
+              secretKeyRef:
+                key: superuser-password
+                name: ${PATRONI_CLUSTER_NAME}
+          - name: PATRONI_REPLICATION_USERNAME
+            value: ${PATRONI_REPLICATION_USERNAME}
+          - name: PATRONI_REPLICATION_PASSWORD
+            valueFrom:
+              secretKeyRef:
+                key: replication-password
+                name: ${PATRONI_CLUSTER_NAME}
+          - name: PATRONI_SCOPE
+            value: ${PATRONI_CLUSTER_NAME}
+          - name: PATRONI_NAME
+            valueFrom:
+              fieldRef:
+                apiVersion: v1
+                fieldPath: metadata.name
+          - name: PATRONI_POSTGRESQL_DATA_DIR
+            value: /home/postgres/pgdata/pgroot/data
+          - name: PATRONI_POSTGRESQL_PGPASS
+            value: /tmp/pgpass
+          - name: PATRONI_POSTGRESQL_LISTEN
+            value: 0.0.0.0:5432
+          - name: PATRONI_RESTAPI_LISTEN
+            value: 0.0.0.0:8008
+          image: docker-registry.default.svc:5000/${NAMESPACE}/patroni:latest
+          imagePullPolicy: IfNotPresent
+          name: ${APPLICATION_NAME}
+          ports:
+          - containerPort: 8008
+            protocol: TCP
+          - containerPort: 5432
+            protocol: TCP
+          resources: {}
+          terminationMessagePath: /dev/termination-log
+          terminationMessagePolicy: File
+          volumeMounts:
+          - mountPath: /home/postgres/pgdata
+            name: ${APPLICATION_NAME}
+        dnsPolicy: ClusterFirst
+        restartPolicy: Always
+        schedulerName: default-scheduler
+        securityContext: {}
+        serviceAccount: ${SERVICE_ACCOUNT}
+        serviceAccountName: ${SERVICE_ACCOUNT}
+        terminationGracePeriodSeconds: 0
+        volumes:
+        - name: ${APPLICATION_NAME}
+          persistentVolumeClaim: 
+            claimName: ${APPLICATION_NAME}
+    volumeClaimTemplates:
+    - metadata:
+        labels:
+          application: ${APPLICATION_NAME}
+        name: ${APPLICATION_NAME}
+      spec:
+        accessModes:
+        - ReadWriteOnce
+        resources:
+          requests:
+            storage: ${PVC_SIZE}
+    updateStrategy:
+      type: OnDelete
+- apiVersion: v1
+  kind: Endpoints
+  metadata:
+    name: ${APPLICATION_NAME}
+    labels:
+      application: ${APPLICATION_NAME}
+      cluster-name: ${PATRONI_CLUSTER_NAME}
+  subsets: []
+- apiVersion: v1
+  kind: ServiceAccount
+  metadata:
+    name: ${SERVICE_ACCOUNT}
+- apiVersion: rbac.authorization.k8s.io/v1
+  kind: Role
+  metadata:
+    name: ${SERVICE_ACCOUNT}
+  rules:
+  - apiGroups:
+    - ""
+    resources:
+    - configmaps
+    verbs:
+    - create
+    - get
+    - list
+    - patch
+    - update
+    - watch
+    # delete is required only for 'patronictl remove'
+    - delete
+  - apiGroups:
+    - ""
+    resources:
+    - endpoints
+    verbs:
+    - get
+    - patch
+    - update
+    # the following three privileges are necessary only when using endpoints
+    - create
+    - list
+    - watch
+    # delete is required only for for 'patronictl remove'
+    - delete
+  - apiGroups:
+    - ""
+    resources:
+    - pods
+    verbs:
+    - get
+    - list
+    - patch
+    - update
+    - watch
+- apiVersion: rbac.authorization.k8s.io/v1
+  kind: RoleBinding
+  metadata:
+    name: ${SERVICE_ACCOUNT}
+  roleRef:
+    apiGroup: rbac.authorization.k8s.io
+    kind: Role
+    name: ${SERVICE_ACCOUNT}
+  subjects:
+  - kind: ServiceAccount
+    name: ${SERVICE_ACCOUNT}
+parameters:
+- description: The name of the application for labelling all artifacts.
+  displayName: Application Name
+  name: APPLICATION_NAME
+  value: patroni-persistent
+- description: The name of the patroni-pgsql cluster.
+  displayName: Cluster Name
+  name: PATRONI_CLUSTER_NAME
+  value: patroni-persistent
+- description: The name of the OpenShift Service exposed for the patroni-persistent-master container.
+  displayName: Master service name.  
+  name: PATRONI_MASTER_SERVICE_NAME
+  value: patroni-persistent-master
+- description: The name of the OpenShift Service exposed for the patroni-persistent-replica containers.
+  displayName: Replica service name. 
+  name: PATRONI_REPLICA_SERVICE_NAME
+  value: patroni-persistent-replica
+- description: Maximum amount of memory the container can use.
+  displayName: Memory Limit
+  name: MEMORY_LIMIT
+  value: 512Mi
+- description: The OpenShift Namespace where the patroni and postgresql ImageStream resides.
+  displayName: ImageStream Namespace
+  name: NAMESPACE
+  value: openshift
+- description: Username of the superuser account for initialization.
+  displayName: Superuser Username
+  name: PATRONI_SUPERUSER_USERNAME
+  value: postgres
+- description: Password of the superuser account for initialization.
+  displayName: Superuser Passsword
+  name: PATRONI_SUPERUSER_PASSWORD
+  value: postgres
+- description: Username of the replication account for initialization.
+  displayName: Replication Username
+  name: PATRONI_REPLICATION_USERNAME
+  value: postgres
+- description: Password of the replication account for initialization.
+  displayName: Repication Passsword
+  name: PATRONI_REPLICATION_PASSWORD
+  value: postgres
+- description: Service account name used for pods and rolebindings to form a cluster in the project. 
+  displayName: Service Account
+  name: SERVICE_ACCOUNT
+  value: patroni-persistent
+- description: The size of the persistent volume to create. 
+  displayName: Persistent Volume Size
+  name: PVC_SIZE
+  value: 5Gi

kubernetes/openshift-example/test/Jenkinsfile

@@ -0,0 +1,43 @@
+pipeline {
+  agent any
+  stages {
+    stage ('Deploy test pod'){
+      when {
+          expression {
+            openshift.withCluster() {
+              openshift.withProject() {
+                return !openshift.selector( "dc", "pgbench" ).exists()
+              }
+            }
+          }
+        }
+      steps {
+        script {
+            openshift.withCluster() {
+                openshift.withProject() {
+                  def pgbench = openshift.newApp( "https://github.com/stewartshea/docker-pgbench/", "--name=pgbench", "-e PGPASSWORD=postgres", "-e PGUSER=postgres", "-e PGHOST=patroni-persistent-master", "-e PGDATABASE=postgres", "-e TEST_CLIENT_COUNT=20", "-e TEST_DURATION=120" )
+                  def pgbenchdc = openshift.selector( "dc", "pgbench" )
+                  timeout(5) { 
+                    pgbenchdc.rollout().status()
+                    }
+                  }
+                }
+            }
+        }
+    }
+    stage ('Run benchmark Test'){
+      steps {
+          sh '''
+          oc exec $(oc get pods -l app=pgbench | grep Running | awk '{print $1}')  ./test.sh
+          '''
+      }
+    }
+    stage ('Clean up pgtest pod'){
+      steps {
+          sh '''
+          oc delete all -l app=pgbench
+          '''
+      }
+    }
+  }
+}

kubernetes/openshift-example/test/README.md

@@ -0,0 +1,2 @@
+# Jenkins Test
+This pipeline test will create a separate deployment config for a pgbench pod and execute a test against the patroni cluster. This is a sample and should be customized. 

kubernetes/patroni_k8s.yaml

@@ -1,3 +1,15 @@
+# headless service to avoid deletion of patronidemo-config endpoint
+apiVersion: v1
+kind: Service
+metadata:
+  name: patronidemo-config
+  labels:
+    application: patroni
+    cluster-name: patronidemo
+spec:
+  clusterIP: None
+
+---
 apiVersion: apps/v1beta1
 kind: StatefulSet
 metadata:
@@ -28,7 +40,7 @@ spec:
         - mountPath: /home/postgres/pgdata
           name: pgdata
         env:
-        - name: POD_IP
+        - name: PATRONI_KUBERNETES_POD_IP
           valueFrom:
             fieldRef:
               fieldPath: status.podIP
@@ -36,6 +48,8 @@ spec:
           valueFrom:
             fieldRef:
               fieldPath: metadata.namespace
+        - name: PATRONI_KUBERNETES_USE_ENDPOINTS
+          value: 'true'
         - name: PATRONI_KUBERNETES_LABELS
           value: '{application: patroni, cluster-name: patronidemo}'
         - name: PATRONI_SUPERUSER_USERNAME
@@ -145,8 +159,9 @@ rules:
   - patch
   - update
   - watch
-  # delete is required only for 'patronictl remove'
+  # delete and deletecollection are required only for 'patronictl remove'
   - delete
+  - deletecollection
 - apiGroups:
   - ""
   resources:
@@ -159,8 +174,9 @@ rules:
   - create
   - list
   - watch
-  # delete is required only for for 'patronictl remove'
+  # delete and deletecollection are required only for for 'patronictl remove'
   - delete
+  - deletecollection
 - apiGroups:
   - ""
   resources:
@@ -171,6 +187,16 @@ rules:
   - patch
   - update
   - watch
+# The following privilege is only necessary for creation of headless service
+# for patronidemo-config endpoint, in order to prevent cleaning it up by the
+# k8s master. You can avoid giving this privilege by explicitly creating the
+# service like it is done in this manifest (lines 2..10)
+- apiGroups:
+  - ""
+  resources:
+  - services
+  verbs:
+  - create
 
 ---
 apiVersion: rbac.authorization.k8s.io/v1

patroni/__init__.py

@@ -4,30 +4,38 @@ import signal
 import sys
 import time
 
+from patroni.version import __version__
+
 logger = logging.getLogger(__name__)
 
+PATRONI_ENV_PREFIX = 'PATRONI_'
+KUBERNETES_ENV_PREFIX = 'KUBERNETES_'
+
 
 class Patroni(object):
 
-    def __init__(self):
+    def __init__(self, conf):
         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()
 
         self.version = __version__
-        self.config = Config()
+        self.logger = PatroniLogger()
+        self.config = conf
+        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.tags = self.get_tags()
@@ -49,6 +57,7 @@ class Patroni(object):
                 break
             except DCSError:
                 logger.warning('Can not get cluster from dcs')
+                time.sleep(5)
 
     def get_tags(self):
         return {tag: value for tag, value in self.config.get('tags', {}).items()
@@ -62,13 +71,16 @@ class Patroni(object):
     def nosync(self):
         return bool(self.tags.get('nosync', False))
 
-    def reload_config(self):
+    def reload_config(self, sighup=False):
         try:
             self.tags = self.get_tags()
-            self.dcs.reload_config(self.config)
+            self.logger.reload_config(self.config.get('log', {}))
             self.watchdog.reload_config(self.config)
+            if sighup:
+                self.request.reload_config(self.config)
             self.api.reload_config(self.config['restapi'])
-            self.postgresql.reload_config(self.config['postgresql'])
+            self.postgresql.reload_config(self.config['postgresql'], sighup)
+            self.dcs.reload_config(self.config)
         except Exception:
             logger.exception('Failed to reload config_file=%s', self.config.config_file)
 
@@ -80,9 +92,10 @@ class Patroni(object):
         self._received_sighup = True
 
     def sigterm_handler(self, *args):
-        if not self._received_sigterm:
-            self._received_sigterm = True
-            sys.exit()
+        with self._sigterm_lock:
+            if not self._received_sigterm:
+                self._received_sigterm = True
+                sys.exit()
 
     @property
     def noloadbalance(self):
@@ -101,15 +114,23 @@ class Patroni(object):
         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.logger.start()
         self.next_run = time.time()
 
-        while not self._received_sigterm:
+        while not self.received_sigterm:
             if self._received_sighup:
                 self._received_sighup = False
                 if self.config.reload_local_configuration():
-                    self.reload_config()
+                    self.reload_config(True)
+                else:
+                    self.postgresql.config.reload_config(self.config['postgresql'], True)
 
             logger.info(self.ha.run_cycle())
 
@@ -123,28 +144,57 @@ class Patroni(object):
             self.schedule_next_run()
 
     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.ha.shutdown()
+        except Exception:
+            logger.exception('Exception during Ha.shutdown')
+        self.logger.shutdown()
 
 
 def patroni_main():
-    logformat = os.environ.get('PATRONI_LOGFORMAT', '%(asctime)s %(levelname)s: %(message)s')
-    loglevel = os.environ.get('PATRONI_LOGLEVEL', 'INFO')
-    requests_loglevel = os.environ.get('PATRONI_REQUESTS_LOGLEVEL', 'WARNING')
-    logging.basicConfig(format=logformat, level=loglevel)
-    logging.getLogger('requests').setLevel(requests_loglevel)
+    import argparse
 
-    patroni = Patroni()
+    from multiprocessing import freeze_support
+    from patroni.config import Config, ConfigParseError
+    from patroni.validator import schema
+
+    freeze_support()
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--version', action='version', version='%(prog)s {0}'.format(__version__))
+    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 args.validate_config:
+            conf = Config(args.configfile, validator=schema)
+            sys.exit()
+        else:
+            conf = Config(args.configfile)
+    except ConfigParseError as e:
+        if e.value:
+            print(e.value)
+        parser.print_help()
+        sys.exit(1)
+    patroni = Patroni(conf)
     try:
         patroni.run()
     except KeyboardInterrupt:
@@ -153,34 +203,38 @@ def patroni_main():
         patroni.shutdown()
 
 
-def pg_ctl_start(args):
-    import subprocess
-    if os.name != 'nt':
-        os.setsid()
-    postmaster = subprocess.Popen(args)
-    print(postmaster.pid)
+def fatal(string, *args):
+    sys.stderr.write('FATAL: ' + string.format(*args) + '\n')
+    sys.exit(1)
 
 
-def call_self(args, **kwargs):
-    """This function executes Patroni once again with provided arguments.
+def check_psycopg2():
+    min_psycopg2 = (2, 5, 4)
+    min_psycopg2_str = '.'.join(map(str, min_psycopg2))
 
-    :args: list of arguments to call Patroni with.
-    :returns: `Popen` object"""
+    def parse_version(version):
+        for e in version.split('.'):
+            try:
+                yield int(e)
+            except ValueError:
+                break
 
-    exe = [sys.executable]
-    if not getattr(sys, 'frozen', False):  # Binary distribution?
-        exe.append(sys.argv[0])
-
-    import subprocess
-    return subprocess.Popen(exe + args, **kwargs)
+    try:
+        import psycopg2
+        version_str = psycopg2.__version__.split(' ')[0]
+        version = tuple(parse_version(version_str))
+        if version < min_psycopg2:
+            fatal('Patroni requires psycopg2>={0}, but only {1} is available', min_psycopg2_str, version_str)
+    except ImportError:
+        fatal('Patroni requires psycopg2>={0} or psycopg2-binary', min_psycopg2_str)
 
 
 def main():
     if os.getpid() != 1:
-        if len(sys.argv) > 5 and sys.argv[1] == 'pg_ctl_start':
-            return pg_ctl_start(sys.argv[2:])
+        check_psycopg2()
         return patroni_main()
 
+    # Patroni started with PID=1, it looks like we are in the container
     pid = 0
 
     # Looks like we are in a docker, so we will act like init
@@ -199,16 +253,18 @@ def main():
         if pid:
             os.kill(pid, signo)
 
-    signal.signal(signal.SIGCHLD, sigchld_handler)
     if os.name != 'nt':
+        signal.signal(signal.SIGCHLD, sigchld_handler)
         signal.signal(signal.SIGHUP, passtochild)
         signal.signal(signal.SIGQUIT, passtochild)
+        signal.signal(signal.SIGUSR1, passtochild)
+        signal.signal(signal.SIGUSR2, passtochild)
     signal.signal(signal.SIGINT, passtochild)
-    signal.signal(signal.SIGUSR1, passtochild)
-    signal.signal(signal.SIGUSR2, passtochild)
     signal.signal(signal.SIGABRT, passtochild)
     signal.signal(signal.SIGTERM, passtochild)
 
-    patroni = call_self(sys.argv[1:])
+    import multiprocessing
+    patroni = multiprocessing.Process(target=patroni_main)
+    patroni.start()
     pid = patroni.pid
-    patroni.wait()
+    patroni.join()

patroni/api.py

@@ -3,13 +3,17 @@ import json
 import logging
 import psycopg2
 import time
+import traceback
 import dateutil.parser
 import datetime
 import os
+import six
+import socket
 
-from patroni.postgresql import PostgresConnectionException, PostgresException, Postgresql
+from patroni.exceptions import PostgresConnectionException, PostgresException
+from patroni.postgresql.misc import postgres_version_to_int
 from patroni.utils import deep_compare, parse_bool, patch_config, Retry, \
-    RetryFailedError, parse_int, split_host_port, tzutc
+    RetryFailedError, parse_int, split_host_port, tzutc, uri, cluster_as_json
 from six.moves.BaseHTTPServer import BaseHTTPRequestHandler, HTTPServer
 from six.moves.socketserver import ThreadingMixIn
 from threading import Thread
@@ -17,20 +21,6 @@ from threading import Thread
 logger = logging.getLogger(__name__)
 
 
-def check_auth(func):
-    """Decorator function to check authorization header.
-
-    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_response(self, status_code, body, content_type='text/html', headers=None):
@@ -46,14 +36,20 @@ class RestApiHandler(BaseHTTPRequestHandler):
     def _write_json_response(self, status_code, response):
         self._write_response(status_code, json.dumps(response), content_type='application/json')
 
-    def send_auth_request(self, body):
-        headers = {'WWW-Authenticate': 'Basic realm="' + self.server.patroni.__class__.__name__ + '"'}
-        self._write_response(401, body, headers=headers)
+    def check_auth(func):
+        """Decorator function to check authorization header or client certificates
 
-    def check_auth_header(self):
-        auth_header = self.headers.get('Authorization')
-        status = self.server.check_auth_header(auth_header)
-        return not status or self.send_auth_request(status)
+        Usage example:
+        @check_auth
+        def do_PUT_foo():
+            pass
+        """
+
+        def wrapper(self, *args, **kwargs):
+            if self.server.check_auth(self):
+                return func(self, *args, **kwargs)
+
+        return wrapper
 
     def _write_status_response(self, status_code, response):
         patroni = self.server.patroni
@@ -73,26 +69,54 @@ class RestApiHandler(BaseHTTPRequestHandler):
             response['watchdog_failed'] = True
         if patroni.ha.is_paused():
             response['pause'] = True
+        qsize = patroni.logger.queue_size
+        if qsize > patroni.logger.NORMAL_LOG_QUEUE_SIZE:
+            response['logger_queue_size'] = qsize
+            lost = patroni.logger.records_lost
+            if lost:
+                response['logger_records_lost'] = lost
         self._write_json_response(status_code, response)
 
     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
 
-        replica_status_code = 200 if not patroni.noloadbalance and response.get('role') == 'replica' else 503
+        replica_status_code = 200 if not patroni.noloadbalance and \
+            response.get('role') == 'replica' and response.get('state') == 'running' else 503
+
+        if not cluster and patroni.ha.is_paused():
+            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():
+            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:
+            primary_status_code = standby_leader_status_code = 503
+
         status_code = 503
 
-        if patroni.config.is_standby_cluster and ('standby_leader' in path or 'standby-leader' in path):
-            status_code = 200 if patroni.ha.is_leader() else 503
-        elif 'master' in path or 'leader' in path or 'primary' in path:
-            status_code = 200 if patroni.ha.is_leader() else 503
+        if 'standby_leader' in path or 'standby-leader' in path:
+            status_code = standby_leader_status_code
+        elif 'master' in path or 'leader' in path or 'primary' in path or 'read-write' in path:
+            status_code = primary_status_code
         elif 'replica' in path:
             status_code = replica_status_code
+        elif 'read-only' in path:
+            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
@@ -103,10 +127,14 @@ class RestApiHandler(BaseHTTPRequestHandler):
 
         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.wfile.write('{0} {1} {2}\r\n'.format(self.protocol_version, status_code, message).encode('utf-8'))
+            self.wfile.write('{0} {1} {2}\r\n\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)
 
@@ -114,6 +142,14 @@ class RestApiHandler(BaseHTTPRequestHandler):
         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:
@@ -161,18 +197,8 @@ class RestApiHandler(BaseHTTPRequestHandler):
 
     @check_auth
     def do_POST_reload(self):
-        try:
-            if self.server.patroni.config.reload_local_configuration(True):
-                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)
+        self.server.patroni.sighup_handler()
+        self._write_response(202, 'reload scheduled')
 
     @staticmethod
     def parse_schedule(schedule, action):
@@ -224,7 +250,7 @@ class RestApiHandler(BaseHTTPRequestHandler):
                     break
             elif k == 'postgres_version':
                 try:
-                    Postgresql.postgres_version_to_int(request[k])
+                    postgres_version_to_int(request[k])
                 except PostgresException as e:
                     status_code = 400
                     data = e.value
@@ -266,6 +292,20 @@ class RestApiHandler(BaseHTTPRequestHandler):
             code = 404
         self._write_response(code, data)
 
+    @check_auth
+    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_auth
     def do_POST_reinitialize(self):
         request = self._read_json_content(body_is_optional=True)
@@ -401,71 +441,69 @@ 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 ("
-                    "SELECT usename, application_name, client_addr, state, sync_state, sync_priority"
-                    " FROM pg_stat_replication) SELECT"
-                    " to_char(pg_postmaster_start_time(), 'YYYY-MM-DD HH24:MI:SS.MS TZ'),"
-                    " CASE WHEN pg_is_in_recovery() THEN 0"
-                    " ELSE ('x' || SUBSTR(pg_{0}file_name(pg_current_{0}_{1}()), 1, 8))::bit(32)::int END,"
-                    " CASE WHEN pg_is_in_recovery() THEN 0"
-                    " ELSE pg_{0}_{1}_diff(pg_current_{0}_{1}(), '0/0')::bigint END,"
-                    " pg_{0}_{1}_diff(COALESCE(pg_last_{0}_receive_{1}(), pg_last_{0}_replay_{1}()), '0/0')::bigint,"
-                    " pg_{0}_{1}_diff(pg_last_{0}_replay_{1}(), '0/0')::bigint,"
-                    " to_char(pg_last_xact_replay_timestamp(), 'YYYY-MM-DD HH24:MI:SS.MS TZ'),"
-                    " pg_is_in_recovery() AND pg_is_{0}_replay_paused(),"
-                    " (SELECT array_to_json(array_agg(row_to_json(ri))) FROM replication_info ri)")
+            stmt = ("SELECT " + postgresql.POSTMASTER_START_TIME + ", " + postgresql.TL_LSN + ","
+                    " pg_catalog.to_char(pg_catalog.pg_last_xact_replay_timestamp(), 'YYYY-MM-DD HH24:MI:SS.MS TZ'),"
+                    " pg_catalog.array_to_json(pg_catalog.array_agg(pg_catalog.row_to_json(ri))) "
+                    "FROM (SELECT (SELECT rolname FROM pg_authid WHERE oid = usesysid) AS usename,"
+                    " application_name, client_addr, w.state, sync_state, sync_priority"
+                    " FROM pg_catalog.pg_stat_get_wal_senders() w, pg_catalog.pg_stat_get_activity(pid)) AS ri")
 
-            row = self.query(stmt.format(self.server.patroni.postgresql.wal_name,
-                                         self.server.patroni.postgresql.lsn_name), retry=retry)[0]
+            row = self.query(stmt.format(postgresql.wal_name, 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],
-                    'replayed_location': row[4],
-                    'replayed_timestamp': row[5],
-                    'paused': row[6]} if row[1] == 0 else {
+                    'received_location': row[4] or row[3],
+                    'replayed_location': row[3],
+                    'replayed_timestamp': row[6],
+                    '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'] = 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 log_message(self, fmt, *args):
         logger.debug("API thread: %s - - [%s] %s", self.client_address[0], self.log_date_time_string(), fmt % args)
 
 
 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.__set_config_parameters(config)
+        self.__ssl_options = None
+        self.reload_config(config)
         self.daemon = True
 
     def query(self, sql, *params):
@@ -496,47 +534,119 @@ class RestApiServer(ThreadingMixIn, HTTPServer, Thread):
             if not auth_header.startswith('Basic ') or not self.check_basic_auth_key(auth_header[6:]):
                 return 'not authenticated'
 
+    def check_auth(self, rh):
+        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 __get_ssl_options(config):
-        return {option: config[option] for option in ['certfile', 'keyfile'] if option in config}
+    def __has_dual_stack():
+        if hasattr(socket, 'AF_INET6') and hasattr(socket, 'IPPROTO_IPV6') and hasattr(socket, 'IPV6_V6ONLY'):
+            sock = None
+            try:
+                sock = socket.socket(socket.AF_INET6, socket.SOCK_STREAM)
+                sock.setsockopt(socket.IPPROTO_IPV6, socket.IPV6_V6ONLY, False)
+                return True
+            except socket.error as e:
+                logger.debug('Error when working with ipv6 socket: %s', e)
+            finally:
+                if sock:
+                    sock.close()
+        return False
 
-    def __set_config_parameters(self, config):
-        self.__auth_key = base64.b64encode(config['auth'].encode('utf-8')).decode('utf-8') if 'auth' in config else None
-        self.connection_string = '{0}://{1}/patroni'.format(self.__protocol,
-                                                            config.get('connect_address') or self.__listen)
+    def __httpserver_init(self, host, port):
+        dual_stack = self.__has_dual_stack()
+        if host in ('', '*'):
+            host = None
 
-    def __initialize(self, config):
+        info = socket.getaddrinfo(host, port, socket.AF_UNSPEC, socket.SOCK_STREAM, 0, socket.AI_PASSIVE)
+        # in case dual stack is not supported we want IPv4 to be preferred over IPv6
+        info.sort(key=lambda x: x[0] == socket.AF_INET, reverse=not dual_stack)
+
+        self.address_family = info[0][0]
         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.__ssl_options = self.__get_ssl_options(config)
+        self.__listen = listen
+        self.__ssl_options = ssl_options
 
-        HTTPServer.__init__(self, (host, port), RestApiHandler)
+        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'.
-        if self.__ssl_options.get('certfile'):
+        self.__protocol = 'https' if ssl_options.get('certfile') else 'http'
+        if self.__protocol == 'https':
             import ssl
-            self.socket = ssl.wrap_socket(self.socket, server_side=True, **self.__ssl_options)
-            self.__protocol = 'https'
-        return True
+            ctx = ssl.create_default_context(ssl.Purpose.CLIENT_AUTH, cafile=ssl_options.get('cafile'))
+            ctx.load_cert_chain(certfile=ssl_options['certfile'], keyfile=ssl_options.get('keyfile'))
+            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.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)
+        if hasattr(sock, 'context'):  # SSLSocket, we want to do the deferred handshake from a thread
+            newsock = (sock, newsock)
+        return newsock, addr
 
     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)) \
-                and self.__initialize(config):
-            self.start()
-        self.__set_config_parameters(config)
+        ssl_options = {n: config[n] for n in ('certfile', 'keyfile', 'cafile') if n in config}
+
+        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:
+            self.__initialize(config['listen'], ssl_options)
+
+        self.__auth_key = base64.b64encode(config['auth'].encode('utf-8')).decode('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):
+        address, port = client_address
+        logger.warning('Exception happened during processing of request from {}:{}'.format(address, port))
+        logger.warning(traceback.format_exc())

patroni/async_executor.py

@@ -52,8 +52,8 @@ class CriticalTask(object):
 
 class AsyncExecutor(object):
 
-    def __init__(self, state_handler, ha_wakeup):
-        self.state_handler = state_handler
+    def __init__(self, cancellable, ha_wakeup):
+        self._cancellable = cancellable
         self._ha_wakeup = ha_wakeup
         self._thread_lock = RLock()
         self._scheduled_action = None
@@ -92,7 +92,7 @@ class AsyncExecutor(object):
                     return
                 self._finish_event.clear()
 
-            self.state_handler.reset_is_cancelled()
+            self._cancellable.reset_is_cancelled()
             # if the func returned something (not None) - wake up main HA loop
             wakeup = func(*args) if args else func()
             return wakeup
@@ -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:
@@ -118,7 +124,7 @@ class AsyncExecutor(object):
                 logger.warning('Cancelling long running task %s', self._scheduled_action)
             self._is_cancelled = True
 
-        self.state_handler.cancel()
+        self._cancellable.cancel()
         self._finish_event.wait()
 
         with self:

patroni/callback_executor.py

@@ -1,37 +0,0 @@
-import logging
-import subprocess
-from threading import Event, Lock, Thread
-
-logger = logging.getLogger(__name__)
-
-
-class CallbackExecutor(Thread):
-
-    def __init__(self):
-        super(CallbackExecutor, self).__init__()
-        self.daemon = True
-        self._lock = Lock()
-        self._cmd = None
-        self._process = None
-        self._callback_event = Event()
-        self.start()
-
-    def call(self, cmd):
-        with self._lock:
-            if self._process and self._process.poll() is None:
-                self._process.kill()
-                logger.warning('Killed the old callback process because it was still running: %s', self._cmd)
-        self._cmd = cmd
-        self._callback_event.set()
-
-    def run(self):
-        while True:
-            self._callback_event.wait()
-            self._callback_event.clear()
-            with self._lock:
-                try:
-                    self._process = subprocess.Popen(self._cmd, close_fds=True)
-                except Exception:
-                    logger.exception('Failed to execute %s',  self._cmd)
-                    continue
-            self._process.wait()

patroni/config.py

@@ -2,20 +2,36 @@ import json
 import logging
 import os
 import shutil
-import six
-import sys
 import tempfile
 import yaml
 
 from collections import defaultdict
 from copy import deepcopy
-from patroni.dcs import ClusterConfig, is_standby_cluster
-from patroni.postgresql import Postgresql
+from patroni import PATRONI_ENV_PREFIX
+from patroni.exceptions import ConfigParseError
+from patroni.dcs import ClusterConfig
+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',
+    'sslrootcert',
+    'sslcrl',
+    'gssencmode',
+    'channel_binding'
+)
+
+
+def default_validator(conf):
+    if not conf:
+        return "Config is empty."
+
 
 class Config(object):
     """
@@ -37,14 +53,15 @@ 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,
+        'check_timeline': False,
         'master_start_timeout': 300,
+        'master_stop_timeout': 0,
         'synchronous_mode': False,
         'synchronous_mode_strict': False,
         'standby_cluster': {
@@ -59,34 +76,33 @@ class Config(object):
         'postgresql': {
             'bin_dir': '',
             'use_slots': True,
-            'parameters': CaseInsensitiveDict({p: v[0] for p, v in Postgresql.CMDLINE_OPTIONS.items()})
+            'parameters': CaseInsensitiveDict({p: v[0] for p, v in ConfigHandler.CMDLINE_OPTIONS.items()})
         },
         '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.isfile(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:
-                print('Usage: {0} config.yml'.format(sys.argv[0]))
-                print('\tPatroni may also read the configuration from the {0} environment variable'.
-                      format(self.PATRONI_CONFIG_VARIABLE))
-                sys.exit(1)
+        if validator:
+            error = validator(self._local_configuration)
+            if error:
+                raise ConfigParseError(error)
 
         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
@@ -99,10 +115,6 @@ class Config(object):
     def dynamic_configuration(self):
         return deepcopy(self._dynamic_configuration)
 
-    @property
-    def is_standby_cluster(self):
-        return is_standby_cluster(self._dynamic_configuration.get('standby_cluster'))
-
     def check_mode(self, mode):
         return bool(parse_bool(self._dynamic_configuration.get(mode)))
 
@@ -162,31 +174,25 @@ 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):
-        ret = {}
-        for name, value in (parameters or {}).items():
-            if name not in Postgresql.CMDLINE_OPTIONS or not is_local and Postgresql.CMDLINE_OPTIONS[name][1](value):
-                ret[name] = value
-        return ret
+        return {name: value for name, value in (parameters or {}).items()
+                if name not in ConfigHandler.CMDLINE_OPTIONS or
+                not is_local and ConfigHandler.CMDLINE_OPTIONS[name][1](value)}
 
     def _safe_copy_dynamic_configuration(self, dynamic_configuration):
         config = deepcopy(self.__DEFAULT_CONFIG)
@@ -199,12 +205,9 @@ class Config(object):
                     elif name not in ('connect_address', 'listen', 'data_dir', 'pgpass', 'authentication'):
                         config['postgresql'][name] = deepcopy(value)
             elif name == 'standby_cluster':
-                allowed_keys = self.__DEFAULT_CONFIG['standby_cluster'].keys()
-                expected = {
-                    k: v for k, v in (value or {}).items()
-                    if (k in allowed_keys and isinstance(v, six.string_types))
-                }
-                config['standby_cluster'].update(expected)
+                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
                 if name in ('synchronous_mode', 'synchronous_mode_strict'):
                     config[name] = value
@@ -217,25 +220,52 @@ 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)
             if value:
                 ret[param] = value
 
+        def _fix_log_env(name, oldname):
+            value = _popenv(oldname)
+            name = PATRONI_ENV_PREFIX + 'LOG_' + name.upper()
+            if value and name not in os.environ:
+                os.environ[name] = value
+
+        for name, oldname in (('level', 'loglevel'), ('format', 'logformat'), ('dateformat', 'log_datefmt')):
+            _fix_log_env(name, oldname)
+
         def _set_section_values(section, params):
             for param in params:
                 value = _popenv(section + '_' + param)
                 if value:
                     ret[section][param] = value
 
-        _set_section_values('restapi', ['listen', 'connect_address', 'certfile', 'keyfile'])
-        _set_section_values('postgresql', ['listen', 'connect_address', 'data_dir', 'pgpass', 'bin_dir'])
+        _set_section_values('restapi', ['listen', 'connect_address', 'certfile', 'keyfile', 'cafile', 'verify_client'])
+        _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', 'traceback_level', 'format', 'dateformat', 'max_queue_size',
+                                    'dir', 'file_size', 'file_num', 'loggers'])
 
-        def _get_auth(name):
+        def _parse_dict(value):
+            if not value.strip().startswith('{'):
+                value = '{{{0}}}'.format(value)
+            try:
+                return yaml.safe_load(value)
+            except Exception:
+                logger.exception('Exception when parsing dict %s', value)
+                return None
+
+        value = ret.get('log', {}).pop('loggers', None)
+        if value:
+            value = _parse_dict(value)
+            if value:
+                ret['log']['loggers'] = value
+
+        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
@@ -246,16 +276,14 @@ class Config(object):
             ret['restapi']['authentication'] = restapi_auth
 
         authentication = {}
-        for user_type in ('replication', 'superuser'):
-            entry = _get_auth(user_type)
+        for user_type in ('replication', 'superuser', 'rewind'):
+            entry = _get_auth(user_type, _AUTH_ALLOWED_PARAMETERS)
             if entry:
                 authentication[user_type] = entry
 
         if authentication:
             ret['postgresql']['authentication'] = authentication
 
-        users = {}
-
         def _parse_list(value):
             if not (value.strip().startswith('-') or '[' in value):
                 value = '[{0}]'.format(value)
@@ -266,38 +294,41 @@ class Config(object):
                 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 name and suffix:
-                    # PATRONI_(ETCD|CONSUL|ZOOKEEPER|EXHIBITOR|...)_(HOSTS?|PORT|..)
-                    if suffix in ('HOST', 'HOSTS', 'PORT', 'SRV', 'URL', 'PROXY', 'CACERT', 'CERT',
-                                  'KEY', 'VERIFY', 'TOKEN', 'CHECKS', 'DC', 'NAMESPACE', 'CONTEXT',
-                                  'USE_ENDPOINTS', 'SCOPE_LABEL', 'ROLE_LABEL', 'POD_IP', 'PORTS', 'LABELS'):
-                        value = os.environ.pop(param)
-                        if suffix == 'PORT':
-                            value = value and parse_int(value)
-                        elif suffix in ('HOSTS', 'PORTS', 'CHECKS'):
-                            value = value and _parse_list(value)
-                        elif suffix == 'LABELS':
-                            if not value.strip().startswith('{'):
-                                value = '{{{0}}}'.format(value)
-                            try:
-                                value = yaml.safe_load(value)
-                            except Exception:
-                                logger.exception('Exception when parsing dict %s', value)
-                                value = None
-                        if value:
-                            ret[name.lower()][suffix.lower()] = value
-                    # PATRONI_<username>_PASSWORD=<password>, PATRONI_<username>_OPTIONS=<option1,option2,...>
-                    # CREATE USER "<username>" WITH <OPTIONS> PASSWORD '<password>'
-                    elif suffix == 'PASSWORD':
-                        password = os.environ.pop(param)
-                        if password:
-                            users[name] = {'password': password}
-                            options = os.environ.pop(param[:-9] + '_OPTIONS', None)
-                            options = options and _parse_list(options)
-                            if options:
-                                users[name]['options'] = options
+                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:
+                    value = os.environ.pop(param)
+                    if suffix == 'PORT':
+                        value = value and parse_int(value)
+                    elif suffix in ('HOSTS', 'PORTS', 'CHECKS'):
+                        value = value and _parse_list(value)
+                    elif suffix == 'LABELS':
+                        value = _parse_dict(value)
+                    elif suffix in ('USE_PROXIES', 'REGISTER_SERVICE'):
+                        value = parse_bool(value)
+                    if value:
+                        ret[name.lower()][suffix.lower()] = value
+        if 'etcd' in ret:
+            ret['etcd'].update(_get_auth('etcd'))
+
+        users = {}
+        for param in list(os.environ.keys()):
+            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>'
+                if name and suffix == 'PASSWORD':
+                    password = os.environ.pop(param)
+                    if password:
+                        users[name] = {'password': password}
+                        options = os.environ.pop(param[:-9] + '_OPTIONS', None)
+                        options = options and _parse_list(options)
+                        if options:
+                            users[name]['options'] = options
         if users:
             ret['bootstrap']['users'] = users
 
@@ -316,7 +347,7 @@ class Config(object):
                 config[name] = deepcopy(value) if value else {}
 
         # 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
@@ -335,6 +366,11 @@ 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']
@@ -344,7 +380,7 @@ class Config(object):
             'scope',
             'retry_timeout',
             'synchronous_mode',
-            'maximum_lag_on_failover'
+            'synchronous_mode_strict',
         )
 
         pg_config.update({p: config[p] for p in updated_fields if p in config})

patroni/ctl.py

@@ -1,12 +1,13 @@
+# -*- coding: utf-8 -*-
 '''
 Patroni Control
 '''
 
-import base64
 import click
 import codecs
 import datetime
 import dateutil.parser
+import dateutil.tz
 import cdiff
 import copy
 import difflib
@@ -14,27 +15,26 @@ import io
 import json
 import logging
 import os
-import psycopg2
 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 patroni.dcs import get_dcs as _get_dcs
 from patroni.exceptions import PatroniException
 from patroni.postgresql import Postgresql
-from patroni.utils import patch_config, polling_loop
+from patroni.postgresql.misc import postgres_version_to_int
+from patroni.utils import cluster_as_json, patch_config, polling_loop
+from patroni.request import PatroniRequest
 from patroni.version import __version__
-from prettytable import PrettyTable
 from six.moves.urllib_parse import urlparse
-from six import text_type
+from texttable import Texttable
 
 CONFIG_DIR_PATH = click.get_app_dir('patroni')
 CONFIG_FILE_PATH = os.path.join(CONFIG_DIR_PATH, 'patronictl.yaml')
@@ -48,14 +48,52 @@ class PatroniCtlException(ClickException):
     pass
 
 
+class PatronictlPrettyTable(Texttable):
+
+    def __init__(self, header=None):
+        Texttable.__init__(self, 0)
+        self.__table_header = header
+        self.__hline_num = 0
+
+        if sys.platform != 'win32':
+            self._char_horiz = u'─'
+            self._char_vert = u'│'
+            self._hline_header = self._hline
+        self._char_header = self._char_horiz
+
+    def _is_first_hline(self):
+        return self.__hline_num == 0
+
+    def _is_last_hline(self):
+        return self.__hline_num > (len(self._rows) if self._has_hlines() else int(bool(self._header)))
+
+    def _hline(self):
+        if sys.platform == 'win32':
+            left = right = self._char_corner
+        elif self._is_first_hline():
+            left, self._char_corner, right = u'┌', u'┬', u'┐'
+        elif not self._is_last_hline():
+            left, self._char_corner, right = u'├', u'┼', u'┤'
+        else:
+            left, self._char_corner, right = u'└', u'┴', u'┘'
+
+        line = self._build_hline()
+
+        if self._is_first_hline() and self.__table_header:
+            left += self.__table_header
+
+        self.__hline_num += 1
+        return left + line[len(left):-2] + right + '\n'
+
+
 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 == '':
@@ -63,23 +101,22 @@ def parse_dcs(dcs):
     elif scheme not in DCS_DEFAULTS:
         raise PatroniCtlException('Unknown dcs scheme: {}'.format(scheme))
 
-    dcs_info = DCS_DEFAULTS[scheme]
-    return yaml.load(dcs_info['template'].format(host=parsed.hostname or 'localhost', port=port or dcs_info['port']))
+    default = DCS_DEFAULTS[scheme]
+    return yaml.safe_load(default['template'].format(host=parsed.hostname or 'localhost', port=port or default['port']))
 
 
 def load_config(path, dcs):
-    logging.debug('Loading configuration from file %s', path)
-    config = {}
-    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
+    from patroni.config import Config
+
+    if not (os.path.exists(path) and os.access(path, os.R_OK)):
+        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(path, validator=None).copy()
 
     dcs = parse_dcs(dcs) or parse_dcs(config.get('dcs_api')) or {}
     if dcs:
@@ -97,7 +134,7 @@ 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')
@@ -107,12 +144,16 @@ option_insecure = click.option('-k', '--insecure', is_flag=True, help='Allow con
 
 
 @click.group()
-@click.option('--config-file', '-c', help='Configuration file', default=CONFIG_FILE_PATH)
+@click.option('--config-file', '-c', help='Configuration file',
+              envvar='PATRONICTL_CONFIG_FILE', default=CONFIG_FILE_PATH)
 @click.option('--dcs', '-d', help='Use this DCS', envvar='DCS')
 @option_insecure
 @click.pass_context
 def ctl(ctx, config_file, dcs, insecure):
-    logging.basicConfig(format='%(asctime)s - %(levelname)s - %(message)s', level=os.environ.get('LOGLEVEL', 'WARNING'))
+    level = 'WARNING'
+    for name in ('LOGLEVEL', 'PATRONI_LOGLEVEL', 'PATRONI_LOG_LEVEL'):
+        level = os.environ.get(name, level)
+    logging.basicConfig(format='%(asctime)s - %(levelname)s - %(message)s', level=level)
     logging.captureWarnings(True)  # Capture eventual SSL warning
     ctx.obj = load_config(config_file, dcs)
     # backward compatibility for configuration file where ctl section is not define
@@ -128,63 +169,45 @@ def get_dcs(config, scope):
         raise PatroniCtlException(str(e))
 
 
-def auth_header(config):
-    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):
+def request_patroni(member, method='GET', endpoint=None, data=None):
     ctx = click.get_current_context()  # the current click context
-    headers = headers or {}
-    url_parts = urlparse(member.api_url)
-    logging.debug(url_parts)
-    if 'Content-Type' not in headers:
-        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)
+    request_executor = ctx.obj.get('__request_patroni')
+    if not request_executor:
+        request_executor = ctx.obj['__request_patroni'] = PatroniRequest(ctx.obj)
+    return request_executor(member, method, endpoint, data)
 
 
-def print_output(columns, rows=None, alignment=None, fmt='pretty', header=True, delimiter='\t'):
-    rows = rows or []
-    if fmt == 'pretty':
-        t = PrettyTable(columns)
-        for k, v in (alignment or {}).items():
-            t.align[k] = v
-        for r in rows:
-            t.add_row(r)
-        click.echo(t)
-        return
+def print_output(columns, rows, alignment=None, fmt='pretty', header=None, delimiter='\t'):
+    if fmt in {'json', 'yaml', 'yml'}:
+        elements = [{k: v for k, v in zip(columns, r) if not header or str(v)} for r in rows]
+        func = json.dumps if fmt == 'json' else format_config_for_editing
+        click.echo(func(elements))
+    elif fmt in {'pretty', 'tsv', 'topology'}:
+        list_cluster = bool(header and columns and columns[0] == 'Cluster')
+        if list_cluster and 'Tags' in columns:  # we want to format member tags as YAML
+            i = columns.index('Tags')
+            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']:
-        elements = [dict(zip(columns, r)) for r in rows]
-        if fmt == 'json':
-            click.echo(json.dumps(elements))
-        elif fmt in ('yaml', 'yml'):
-            click.echo(yaml.safe_dump(elements, encoding=None, default_flow_style=False, allow_unicode=True, width=200))
-
-    if fmt == 'tsv':
-        if columns is not None and header:
-            click.echo(delimiter.join(columns) + '\n')
-
-        for r in rows:
-            c = [str(c) for c in r]
-            click.echo(delimiter.join(c))
+        if fmt == 'tsv':
+            for r in ([columns] if columns else []) + rows:
+                click.echo(delimiter.join(map(str, r)))
+        else:
+            table = PatronictlPrettyTable(header)
+            if not any(any(isinstance(c, six.string_types) and '\n' in c for c in r) for r in rows):
+                table.set_deco(Texttable.VLINES | Texttable.BORDER | Texttable.HEADER)
+            if rows:
+                if columns:
+                    table.header(columns)
+                    table.set_cols_align([(alignment or {}).get(c, 'l') for c in columns])
+                table.add_rows(rows, header=False)
+            else:
+                table.add_rows([columns], header=False)
+            click.echo(table.draw())
 
 
 def watching(w, watch, max_count=None, clear=True):
@@ -234,6 +257,15 @@ 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:
@@ -246,13 +278,14 @@ def get_cursor(cluster, connect_parameters, role='master', member=None):
     else:
         params.pop('database')
 
+    import psycopg2
     conn = psycopg2.connect(**params)
     conn.autocommit = True
     cursor = conn.cursor()
     if role == 'any':
         return cursor
 
-    cursor.execute('SELECT pg_is_in_recovery()')
+    cursor.execute('SELECT pg_catalog.pg_is_in_recovery()')
     in_recovery = cursor.fetchone()[0]
 
     if in_recovery and role == 'replica' or not in_recovery and role == 'master':
@@ -263,10 +296,12 @@ def get_cursor(cluster, connect_parameters, role='master', member=None):
     return None
 
 
-def get_members(cluster, cluster_name, member_names, role, force, action):
+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:
@@ -286,12 +321,25 @@ def get_members(cluster, cluster_name, member_names, role, force, action):
         if member_name not in candidates:
             raise PatroniCtlException('{0} is not a member of cluster'.format(member_name))
 
-    if not force:
-        confirm = click.confirm('Are you sure you want to {0} members {1}?'.format(action, ', '.join(member_names)))
-        if not confirm:
-            raise PatroniCtlException('Aborted {0}'.format(action))
+    members = [candidates[n] for n in member_names]
+    if ask_confirmation:
+        confirm_members_action(members, force, action)
+    return members
 
-    return [candidates[n] for n in member_names]
+
+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([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([m.name for m in members])))
+            if not confirm:
+                raise PatroniCtlException('Aborted {0}'.format(action))
 
 
 @ctl.command('dsn', help='Generate a dsn for the provided member, defaults to a dsn of the master')
@@ -317,8 +365,7 @@ 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, json)', default='tsv')
+@click.option('--format', 'fmt', help='Output format (pretty, tsv, json, yaml)', 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)
@@ -375,11 +422,12 @@ def query(
         if cursor is None:
             cluster = dcs.get_cluster()
 
-        output, cursor = query_member(cluster, cursor, member, role, command, connect_parameters)
-        print_output(None, output, fmt=fmt, delimiter=delimiter)
+        output, header = query_member(cluster, cursor, member, role, command, connect_parameters)
+        print_output(header, output, fmt=fmt, delimiter=delimiter)
 
 
 def query_member(cluster, cursor, member, role, command, connect_parameters):
+    import psycopg2
     try:
         if cursor is None:
             cursor = get_cursor(cluster, connect_parameters, role=role, member=member)
@@ -392,15 +440,8 @@ def query_member(cluster, cursor, member, role, command, connect_parameters):
             logging.debug(message)
             return [[timestamp(0), message]], None
 
-        cursor.execute('SELECT 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:
@@ -440,9 +481,9 @@ 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:
@@ -455,7 +496,7 @@ 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))
@@ -476,18 +517,17 @@ 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))
-        if r.status_code == 200:
+        r = request_patroni(member, 'post', 'reload')
+        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'))
             )
 
 
@@ -509,7 +549,15 @@ 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')
+    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)
+
     if p_any:
         random.shuffle(members)
         members = members[:1]
@@ -524,16 +572,12 @@ def restart(obj, cluster_name, member_names, force, role, p_any, scheduled, vers
 
     if version:
         try:
-            Postgresql.postgres_version_to_int(version)
+            postgres_version_to_int(version)
         except PatroniException as e:
             raise PatroniCtlException(e.value)
 
         content['postgres_version'] = version
 
-    if scheduled is None and not force:
-        scheduled = click.prompt('When should the restart take place (e.g. 2015-10-01T14:30) ', type=str, default='now')
-
-    scheduled_at = parse_scheduled(scheduled)
     if scheduled_at:
         if cluster.is_paused():
             raise PatroniCtlException("Can't schedule restart in the paused state")
@@ -545,19 +589,19 @@ 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))
-        if r.status_code == 200:
+        r = request_patroni(member, 'post', 'restart', content)
+        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'))
             )
 
 
@@ -565,20 +609,39 @@ 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))
-            if not check_response(r, member.name, 'reinitialize') and r.text.endswith(' already in progress') \
+            r = request_patroni(member, 'post', 'reinitialize', body)
+            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):
@@ -625,10 +688,12 @@ def _do_failover_or_switchover(obj, action, cluster_name, master, candidate, for
         raise PatroniCtlException('Member {0} does not exist in cluster {1}'.format(candidate, cluster_name))
 
     scheduled_at_str = None
+    scheduled_at = None
 
     if action == 'switchover':
         if scheduled is None and not force:
-            scheduled = click.prompt('When should the switchover take place (e.g. 2015-10-01T14:30) ',
+            next_hour = (datetime.datetime.now() + datetime.timedelta(hours=1)).strftime('%Y-%m-%dT%H:%M')
+            scheduled = click.prompt('When should the switchover take place (e.g. ' + next_hour + ' ) ',
                                      type=str, default='now')
 
         scheduled_at = parse_scheduled(scheduled)
@@ -647,27 +712,32 @@ def _do_failover_or_switchover(obj, action, cluster_name, master, candidate, for
 
     if not force:
         demote_msg = ', demoting current master ' + master if master else ''
-
-        if not click.confirm('Are you sure you want to {0} cluster {1}{2}?'.format(action, cluster_name, demote_msg)):
-            raise PatroniCtlException('Aborting ' + action)
+        if scheduled_at_str:
+            if not click.confirm('Are you sure you want to schedule {0} of cluster {1} at {2}{3}?'
+                                 .format(action, cluster_name, scheduled_at_str, demote_msg)):
+                raise PatroniCtlException('Aborting scheduled ' + action)
+        else:
+            if not click.confirm('Are you sure you want to {0} cluster {1}{2}?'
+                                 .format(action, cluster_name, demote_msg)):
+                raise PatroniCtlException('Aborting ' + action)
 
     r = None
     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:
-            r = request_patroni(member, 'post', 'failover', failover_value, auth_header(obj))
+        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)
 
-        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)
@@ -701,75 +771,85 @@ 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
-    if cluster.leader:
-        leader_name = cluster.leader.name
+    initialize = {None: 'uninitialized', '': 'initializing'}.get(cluster.initialize, cluster.initialize)
+    cluster = cluster_as_json(cluster)
 
-    xlog_location_cluster = cluster.last_leader_operation or 0
+    columns = ['Cluster', 'Member', 'Host', 'Role', 'State', 'TL', 'Lag in MB']
+    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)
 
-    # Mainly for consistent pretty printing and watching we sort the output
-    cluster.members.sort(key=lambda x: x.name)
+    # Show Host as 'host:port' if somebody is running on non-standard port or two nodes are running on the same host
+    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)
 
-    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)
-
-    for m in cluster.members:
+    sort = topology_sort if fmt == 'topology' else iter
+    for m in sort(cluster['members']):
         logging.debug(m)
 
-        role = ''
-        if m.name == leader_name:
-            role = 'Leader'
-        elif m.name == cluster.sync.sync_standby:
-            role = 'Sync standby'
+        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 '')
 
-        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)
+        if append_port and m['host'] and m.get('port'):
+            m['host'] = ':'.join([m['host'], str(m['port'])])
 
-        row = [name, m.name, m.conn_kwargs()['host'], role, m.data.get('state', ''), lag]
+        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
 
-        if extended or has_pending_restarts:
-            row.append('*' if m.data.get('pending_restart') else '')
+        rows.append([m.get(n.lower().replace(' ', '_'), '') for n in columns])
 
-        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'])
+    print_output(columns, rows, {'Member': 'l', 'Lag in MB': 'r', 'TL': 'r', 'Tags': 'l'},
+                 fmt, ' Cluster: {0} ({1}) '.format(name, initialize))
 
-            row.append(value)
-
-        rows.append(row)
-
-    columns = ['Cluster', 'Member', 'Host', 'Role', 'State', 'Lag in MB']
-    alignment = {'Lag in MB': 'r'}
-
-    if extended or has_pending_restarts:
-        columns.append('Pending restart')
-
-    if extended or has_scheduled_restarts:
-        columns.append('Scheduled restart')
-
-    print_output(columns, rows, alignment, fmt)
+    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:
-        info = 'Switchover scheduled at: ' + cluster.failover.scheduled_at.isoformat()
-        if cluster.failover.leader:
-            info += '\n                    from: ' + cluster.failover.leader
-        if cluster.failover.candidate:
-            info += '\n                      to: ' + cluster.failover.candidate
+    if 'scheduled_switchover' in cluster:
+        info = 'Switchover scheduled at: ' + cluster['scheduled_switchover']['at']
+        for name in ('from', 'to'):
+            if name in cluster['scheduled_switchover']:
+                info += '\n{0:>24}: {1}'.format(name, cluster['scheduled_switchover'][name])
         service_info.append(info)
 
     if service_info:
@@ -802,6 +882,16 @@ 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]
 
@@ -869,25 +959,45 @@ 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')
-    for member in members:
-        if target == 'restart':
+    if target == 'restart':
+        for member in get_members(cluster, cluster_name, member_names, role, force, 'flush'):
             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):
@@ -914,27 +1024,22 @@ 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 = []
-    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:
+    for member in get_all_members_leader_first(cluster):
         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')
@@ -981,7 +1086,7 @@ 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))
 
@@ -989,7 +1094,7 @@ 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:
@@ -1002,12 +1107,12 @@ def show_diff(before_editing, after_editing):
             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):
@@ -1071,15 +1176,40 @@ 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
 
     :param before_editing: human representation before editing
     :returns tuple of human readable and parsed datastructure after changes
     """
+
     editor_cmd = os.environ.get('EDITOR')
     if not editor_cmd:
-        raise PatroniCtlException('EDITOR environment variable is not set')
+        for editor in ('editor', 'vi'):
+            editor_cmd = find_executable(editor)
+            if editor_cmd:
+                logging.debug('Setting fallback editor_cmd=%s', editor)
+                break
+    if not editor_cmd:
+        raise PatroniCtlException('EDITOR environment variable is not set. editor or vi are not available')
 
     with temporary_file(contents=before_editing.encode('utf-8'),
                         suffix='.yaml',
@@ -1173,8 +1303,8 @@ 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')
-                    data = response.json()
+                    response = request_patroni(m)
+                    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 ""
@@ -1183,6 +1313,19 @@ 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/dcs/__init__.py

@@ -9,17 +9,36 @@ import pkgutil
 import re
 import six
 import sys
+import time
 
-from collections import namedtuple
+from collections import defaultdict, namedtuple
+from copy import deepcopy
 from patroni.exceptions import PatroniException
-from patroni.utils import parse_bool
+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
 
+slot_name_re = re.compile('^[a-z0-9_]{1,63}$')
 logger = logging.getLogger(__name__)
 
 
+def slot_name_from_member_name(member_name):
+    """Translate member name to valid PostgreSQL slot name.
+
+    PostgreSQL replication slot names must be valid PostgreSQL names. This function maps the wider space of
+    member names to valid PostgreSQL names. Names are lowercased, dashes and periods common in hostnames
+    are replaced with underscores, other characters are encoded as their unicode codepoint. Name is truncated
+    to 64 characters. Multiple different member names may map to a single slot name."""
+
+    def replace_char(match):
+        c = match.group(0)
+        return '_' if c in '-.' else "u{:04d}".format(ord(c))
+
+    slot_name = re.sub('[^a-z0-9_]', replace_char, member_name.lower())
+    return slot_name[0:63]
+
+
 def parse_connection_string(value):
     """Original Governor stores connection strings for each cluster members if a following format:
         postgres://{username}:{password}@{connect_address}/postgres
@@ -47,33 +66,44 @@ def dcs_modules():
     module_prefix = __package__ + '.'
 
     if getattr(sys, 'frozen', False):
-        importer = pkgutil.get_importer(dcs_dirname)
-        return [module for module in list(importer.toc) if module.startswith(module_prefix) and module.count('.') == 2]
+        toc = set()
+        for importer in pkgutil.iter_importers(dcs_dirname):
+            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()
-    for module_name in dcs_modules():
-        try:
-            module = importlib.import_module(module_name)
-            for name in filter(lambda name: not name.startswith('__'), dir(module)):  # iterate through module content
-                item = getattr(module, name)
-                name = name.lower()
-                # try to find implementation of AbstractDCS interface, class name must match with module_name
-                if inspect.isclass(item) and issubclass(item, AbstractDCS) and __package__ + '.' + name == module_name:
-                    available_implementations.add(name)
-                    if name in config:  # which has configuration section in the config file
+    modules = dcs_modules()
+
+    for module_name in modules:
+        name = module_name.split('.')[-1]
+        if name in config:  # we will try to import only modules which have configuration section in the config file
+            try:
+                module = importlib.import_module(module_name)
+                for key, item in module.__dict__.items():  # iterate through the module content
+                    # try to find implementation of AbstractDCS interface, class name must match with module_name
+                    if key.lower() == name and inspect.isclass(item) and issubclass(item, AbstractDCS):
                         # 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 PatroniException("""Can not find suitable configuration of distributed configuration store
-Available implementations: """ + ', '.join(available_implementations))
+Available implementations: """ + ', '.join(sorted(set(available_implementations))))
 
 
 class Member(namedtuple('Member', 'index,name,session,data')):
@@ -110,30 +140,30 @@ 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 = 'postgresql://{host}:{port}'.format(
-                host=conn_kwargs.get('host'),
-                port=conn_kwargs.get('port'),
-            )
+            conn_url = uri('postgresql', (conn_kwargs.get('host'), conn_kwargs.get('port', 5432)))
             self.data['conn_url'] = conn_url
             return conn_url
 
     def conn_kwargs(self, auth=None):
         defaults = {
-            "host": "",
-            "port": "",
-            "database": ""
+            "host": None,
+            "port": None,
+            "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,
@@ -141,11 +171,11 @@ 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']
-            if 'password' in auth:
-                ret['password'] = auth['password']
+                ret['user'] = ret.pop('username')
         return ret
 
     @property
@@ -189,13 +219,12 @@ class RemoteMember(Member):
                 'create_replica_methods',
                 'restore_command',
                 'archive_cleanup_command',
-                'recovery_min_apply_delay')
+                'recovery_min_apply_delay',
+                'no_replication_slot')
 
     def __getattr__(self, name):
-        if name not in RemoteMember.allowed_keys():
-            return
-
-        return self.data.get(name)
+        if name in RemoteMember.allowed_keys():
+            return self.data.get(name)
 
 
 class Leader(namedtuple('Leader', 'index,session,member')):
@@ -217,9 +246,27 @@ 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.data.get('version')
+        if version:
+            try:
+                # 1.5.6 is the last version which doesn't expose checkpoint_after_promote: false
+                if tuple(map(int, version.split('.'))) > (1, 5, 6):
+                    return self.data['role'] == 'master' and 'checkpoint_after_promote' not in self.data
+            except Exception:
+                logger.debug('Failed to parse Patroni version %s', version)
 
 
 class Failover(namedtuple('Failover', 'index,leader,candidate,scheduled_at')):
@@ -275,14 +322,28 @@ class ClusterConfig(namedtuple('ClusterConfig', 'index,data,modify_index')):
     def from_node(index, data, modify_index=None):
         """
         >>> ClusterConfig.from_node(1, '{') is None
-        True
+        False
         """
 
         try:
             data = json.loads(data)
         except (TypeError, ValueError):
-            return None
-        return ClusterConfig(index, data, modify_index or index)
+            data = None
+            modify_index = 0
+        if not isinstance(data, dict):
+            data = {}
+        return ClusterConfig(index, data, index if modify_index is None else modify_index)
+
+    @property
+    def permanent_slots(self):
+        return isinstance(self.data, dict) and (
+                self.data.get('permanent_replication_slots') or
+                self.data.get('permanent_slots') or self.data.get('slots')
+        ) or {}
+
+    @property
+    def max_timelines_history(self):
+        return self.data.get('max_timelines_history', 0)
 
 
 class SyncState(namedtuple('SyncState', 'index,leader,sync_standby')):
@@ -341,7 +402,7 @@ class SyncState(namedtuple('SyncState', 'index,leader,sync_standby')):
         return name is not None and name in (self.leader, self.sync_standby)
 
 
-class TimelineHistory(namedtuple('TimelineHistory', 'index,lines')):
+class TimelineHistory(namedtuple('TimelineHistory', 'index,value,lines')):
     """Object representing timeline history file"""
 
     @staticmethod
@@ -357,7 +418,7 @@ class TimelineHistory(namedtuple('TimelineHistory', 'index,lines')):
             lines = None
         if not isinstance(lines, list):
             lines = []
-        return TimelineHistory(index, lines)
+        return TimelineHistory(index, value, lines)
 
 
 class Cluster(namedtuple('Cluster', 'initialize,config,leader,last_leader_operation,members,failover,sync,history')):
@@ -398,8 +459,85 @@ class Cluster(namedtuple('Cluster', 'initialize,config,leader,last_leader_operat
     def is_synchronous_mode(self):
         return self.check_mode('synchronous_mode')
 
-    def is_standby_cluster(self):
-        return is_standby_cluster(self.config and self.config.data.get('standby_cluster'))
+    def get_replication_slots(self, my_name, role):
+        # 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.config and self.config.data.get('postgresql', {}).get('use_slots', True)
+        if role in ('master', 'standby_leader'):
+            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 == my_name or
+                             not self.has_member(m.replicatefrom))]
+            permanent_slots = (self.config and self.config.permanent_slots or {}).copy()
+        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 use_slots and
+                            m.replicatefrom == my_name and m.name != self.leader.name]
+            permanent_slots = {}
+
+        slots = {slot_name_from_member_name(name): {'type': 'physical'} for name in slot_members}
+
+        if len(slots) < len(slot_members):
+            # Find which names are conflicting for a nicer error message
+            slot_conflicts = defaultdict(list)
+            for name in slot_members:
+                slot_conflicts[slot_name_from_member_name(name)].append(name)
+            logger.error("Following cluster members share a replication slot name: %s",
+                         "; ".join("{} map to {}".format(", ".join(v), k)
+                                   for k, v in slot_conflicts.items() if len(v) > 1))
+
+        # "merge" replication slots for members with permanent_replication_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
+
+            value = deepcopy(value) if value else {'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':
+                    if name != my_name:  # Don't try to create permanent physical replication slot for yourself
+                        slots[name] = value
+                    continue
+                elif value['type'] == 'logical' and value.get('database') and value.get('plugin'):
+                    if 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])
+
+        return slots
+
+    def has_permanent_logical_slots(self, name):
+        slots = self.get_replication_slots(name, 'master').values()
+        return any(v for v in slots if v.get("type") == "logical")
+
+    @property
+    def timeline(self):
+        """
+        >>> Cluster(0, 0, 0, 0, 0, 0, 0, 0).timeline
+        0
+        >>> Cluster(0, 0, 0, 0, 0, 0, 0, TimelineHistory.from_node(1, '[]')).timeline
+        1
+        >>> Cluster(0, 0, 0, 0, 0, 0, 0, TimelineHistory.from_node(1, '[["a"]]')).timeline
+        0
+        """
+        if self.history:
+            if self.history.lines:
+                try:
+                    return int(self.history.lines[-1][0]) + 1
+                except Exception:
+                    logger.error('Failed to parse cluster history from DCS: %s', self.history.lines)
+            elif self.history.value == '[]':
+                return 1
+        return 0
 
 
 @six.add_metaclass(abc.ABCMeta)
@@ -426,6 +564,7 @@ class AbstractDCS(object):
 
         self._ctl = bool(config.get('patronictl', False))
         self._cluster = None
+        self._cluster_valid_till = 0
         self._cluster_thread_lock = Lock()
         self._last_leader_operation = ''
         self.event = Event()
@@ -473,6 +612,10 @@ class AbstractDCS(object):
     def set_ttl(self, ttl):
         """Set the new ttl value for leader key"""
 
+    @abc.abstractmethod
+    def ttl(self):
+        """Get new ttl value"""
+
     @abc.abstractmethod
     def set_retry_timeout(self, retry_timeout):
         """Set the new value for retry_timeout"""
@@ -499,23 +642,32 @@ 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:
+            self.reset_cluster()
+            raise
+
         with self._cluster_thread_lock:
-            try:
-                self._load_cluster()
-            except Exception:
-                self._cluster = None
-                raise
-            return self._cluster
+            self._cluster = cluster
+            self._cluster_valid_till = time.time() + self.ttl
+            return cluster
 
     @property
     def cluster(self):
         with self._cluster_thread_lock:
-            return self._cluster
+            return self._cluster if self._cluster_valid_till > time.time() else None
 
     def reset_cluster(self):
         with self._cluster_thread_lock:
             self._cluster = None
+            self._cluster_valid_till = 0
 
     @abc.abstractmethod
     def _write_leader_optime(self, last_operation):
@@ -537,7 +689,7 @@ 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):
+    def update_leader(self, last_operation, access_is_restricted=False):
         """Update leader key (or session) ttl and optime/leader
 
         :param last_operation: absolute xlog location in bytes
@@ -581,7 +733,7 @@ class AbstractDCS(object):
         """Create or update `/config` key"""
 
     @abc.abstractmethod
-    def touch_member(self, data, ttl=None, permanent=False):
+    def touch_member(self, data, permanent=False):
         """Update member key in DCS.
         This method should create or update key with the name = '/members/' + `~self._name`
         and value = data in a given DCS.
@@ -611,10 +763,19 @@ class AbstractDCS(object):
         otherwise it should return `!False`"""
 
     @abc.abstractmethod
-    def delete_leader(self):
-        """Voluntarily remove leader key from DCS
+    def _delete_leader(self):
+        """Remove leader key from DCS.
         This method should remove leader key if current instance is the leader"""
 
+    def delete_leader(self, last_operation=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_operation: latest checkpoint location in bytes"""
+
+        if last_operation:
+            self.write_leader_optime(last_operation)
+        return self._delete_leader()
+
     @abc.abstractmethod
     def cancel_initialization(self):
         """ Removes the initialize key for a cluster """
@@ -654,14 +815,3 @@ class AbstractDCS(object):
 
         self.event.wait(timeout)
         return self.event.isSet()
-
-
-def is_standby_cluster(config):
-    """ Check whether or not provided configuration describes a standby cluster.
-        Config can be both patroni config or cluster.config.data
-    """
-    return isinstance(config, dict) and (
-        config.get('host') or
-        config.get('port') or
-        config.get('restore_command')
-    )

patroni/dcs/consul.py

@@ -9,13 +9,14 @@ import time
 import urllib3
 
 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
 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__)
 
 
@@ -27,16 +28,20 @@ class ConsulInternalError(ConsulException):
     """An internal Consul server error occurred"""
 
 
-class InvalidSessionTTL(ConsulInternalError):
+class InvalidSessionTTL(ConsulException):
     """Session TTL is too small or too big"""
 
 
+class InvalidSession(ConsulException):
+    """invalid session"""
+
+
 class HTTPClient(object):
 
     def __init__(self, host='127.0.0.1', port=8500, token=None, scheme='http', verify=True, cert=None, ca_cert=None):
         self.token = token
         self._read_timeout = 10
-        self.base_uri = '{0}://{1}:{2}'.format(scheme, host, port)
+        self.base_uri = uri(scheme, (host, port))
         kwargs = {}
         if cert:
             if isinstance(cert, tuple):
@@ -48,9 +53,8 @@ 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
-        self.http = urllib3.PoolManager(num_pools=10, **kwargs)
+        kwargs['cert_reqs'] = ssl.CERT_REQUIRED if verify or ca_cert else ssl.CERT_NONE
+        self.http = urllib3.PoolManager(num_pools=10, maxsize=10, **kwargs)
         self._ttl = None
 
     def set_read_timeout(self, timeout):
@@ -72,6 +76,8 @@ class HTTPClient(object):
             msg = '{0} {1}'.format(response.status, data)
             if data.startswith('Invalid Session TTL'):
                 raise InvalidSessionTTL(msg)
+            elif data.startswith('invalid session'):
+                raise InvalidSession(msg)
             else:
                 raise ConsulInternalError(msg)
         return base.Response(response.status, response.headers, data)
@@ -96,12 +102,18 @@ class HTTPClient(object):
                 params = {k: v for k, v in params}
             kwargs = {'retries': 0, 'preload_content': False, 'body': data}
             if method == 'get' and isinstance(params, dict) and 'index' in params:
-                kwargs['timeout'] = (float(params['wait'][:-1]) if 'wait' in params else 300) + 1
+                timeout = float(params['wait'][:-1]) if 'wait' in params else 300
+                # 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.
+                kwargs['timeout'] = timeout + max(timeout/15.0, 1)
             else:
                 kwargs['timeout'] = self._read_timeout
             token = params.pop('token', self.token) if isinstance(params, dict) else self.token
+            kwargs['headers'] = urllib3.make_headers(user_agent=USER_AGENT)
             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
 
@@ -111,7 +123,7 @@ class ConsulClient(base.Consul):
     def __init__(self, *args, **kwargs):
         self._cert = kwargs.pop('cert', None)
         self._ca_cert = kwargs.pop('ca_cert', None)
-        self._token = kwargs.get('token')
+        self.token = kwargs.get('token')
         super(ConsulClient, self).__init__(*args, **kwargs)
 
     def connect(self, *args, **kwargs):
@@ -120,10 +132,15 @@ class ConsulClient(base.Consul):
             kwargs['cert'] = self._cert
         if self._ca_cert:
             kwargs['ca_cert'] = self._ca_cert
-        if self._token:
-            kwargs['token'] = self._token
+        if self.token:
+            kwargs['token'] = self.token
         return HTTPClient(**kwargs)
 
+    def reload_config(self, config):
+        self.http.token = self.token = config.get('token')
+        self.consistency = config.get('consistency', 'default')
+        self.dc = config.get('dc')
+
 
 def catch_consul_errors(func):
     def wrapper(*args, **kwargs):
@@ -186,7 +203,7 @@ class Consul(AbstractDCS):
         if config.get('key') and config.get('cert'):
             config['cert'] = (config['cert'], config['key'])
 
-        config_keys = ('host', 'port', 'token', 'scheme', 'cert', 'ca_cert', 'dc')
+        config_keys = ('host', 'port', 'token', 'scheme', 'cert', 'ca_cert', 'dc', 'consistency')
         kwargs = {p: config.get(p) for p in config_keys if config.get(p)}
 
         verify = config.get('verify')
@@ -199,7 +216,7 @@ 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_name = service_name_from_scope_name(self._scope)
@@ -221,11 +238,19 @@ class Consul(AbstractDCS):
                 logger.info('waiting on consul')
                 time.sleep(5)
 
+    def reload_config(self, config):
+        super(Consul, self).reload_config(config)
+        self._client.reload_config(config.get('consul', {}))
+
     def set_ttl(self, ttl):
         if self._client.http.set_ttl(ttl/2.0):  # Consul multiplies the TTL by 2x
             self._session = None
             self.__do_not_watch = True
 
+    @property
+    def ttl(self):
+        return self._client.http.ttl
+
     def set_retry_timeout(self, retry_timeout):
         self._retry.deadline = retry_timeout
         self._client.http.set_read_timeout(retry_timeout)
@@ -288,7 +313,7 @@ class Consul(AbstractDCS):
             nodes = {}
             for node in results:
                 node['Value'] = (node['Value'] or b'').decode('utf-8')
-                nodes[os.path.relpath(node['Key'], path).replace('\\', '/')] = node
+                nodes[node['Key'][len(path):].lstrip('/')] = node
 
             # get initialize flag
             initialize = nodes.get(self._INITIALIZE)
@@ -331,24 +356,26 @@ class Consul(AbstractDCS):
             sync = nodes.get(self._SYNC)
             sync = SyncState.from_node(sync and sync['ModifyIndex'], sync and sync['Value'])
 
-            self._cluster = Cluster(initialize, config, leader, last_leader_operation, members, failover, sync, history)
+            return Cluster(initialize, config, leader, last_leader_operation, members, failover, sync, history)
         except NotFound:
-            self._cluster = Cluster(None, None, None, None, [], None, None, None)
+            return Cluster(None, None, None, None, [], None, None, None)
         except Exception:
             logger.exception('get_cluster')
             raise ConsulError('Consul is not responding properly')
 
-    def touch_member(self, data, ttl=None, permanent=False):
+    @catch_consul_errors
+    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):
-            try:
-                self._client.kv.delete(self.member_path)
-                create_member = True
-            except Exception:
-                return False
+            self._client.kv.delete(self.member_path)
+            create_member = True
 
         if not create_member and member and deep_compare(data, member.data):
             return True
@@ -359,6 +386,9 @@ class Consul(AbstractDCS):
             if self._register_service:
                 self.update_service(not create_member and member and member.data or {}, data)
             return True
+        except InvalidSession:
+            self._session = None
+            logger.error('Our session disappeared from Consul, can not "touch_member"')
         except Exception:
             logger.exception('touch_member')
         return False
@@ -382,7 +412,8 @@ class Consul(AbstractDCS):
         api_parts = urlparse(data['api_url'])
         api_parts = api_parts._replace(path='/{0}'.format(role))
         conn_parts = urlparse(data['conn_url'])
-        check = base.Check.http(api_parts.geturl(), self._service_check_interval, deregister=self._client.http.ttl * 10)
+        check = base.Check.http(api_parts.geturl(), self._service_check_interval,
+                                deregister='{0}s'.format(self._client.http.ttl * 10))
         params = {
             'service_id': '{0}/{1}'.format(self._scope, self._name),
             'address': conn_parts.hostname,
@@ -416,14 +447,21 @@ class Consul(AbstractDCS):
             return self._update_service(new_data)
 
     @catch_consul_errors
-    def _do_attempt_to_acquire_leader(self, kwargs):
-        return self.retry(self._client.kv.put, self.leader_path, self._name, **kwargs)
+    def _do_attempt_to_acquire_leader(self, permanent):
+        try:
+            kwargs = {} if permanent else {'acquire': self._session}
+            return self.retry(self._client.kv.put, self.leader_path, self._name, **kwargs)
+        except InvalidSession:
+            self._session = None
+            logger.error('Our session disappeared from Consul. Will try to get a new one and retry attempt')
+            self.refresh_session()
+            return self.retry(self._client.kv.put, self.leader_path, self._name, acquire=self._session)
 
     def attempt_to_acquire_leader(self, permanent=False):
         if not self._session and not permanent:
             self.refresh_session()
 
-        ret = self._do_attempt_to_acquire_leader({} if permanent else {'acquire': self._session})
+        ret = self._do_attempt_to_acquire_leader(permanent)
         if not ret:
             logger.info('Could not take out TTL lock')
 
@@ -469,7 +507,7 @@ 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)
@@ -483,6 +521,7 @@ class Consul(AbstractDCS):
         return self.retry(self._client.kv.delete, self.sync_path, cas=index)
 
     def watch(self, leader_index, timeout):
+        self._last_session_refresh = 0
         if self.__do_not_watch:
             self.__do_not_watch = False
             return True

patroni/dcs/etcd.py

@@ -5,28 +5,29 @@ 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 patroni.exceptions import DCSError
-from patroni.utils import Retry, RetryFailedError, split_host_port
-from urllib3.exceptions import HTTPError, ReadTimeoutError
-from requests.exceptions import RequestException
+from urllib3 import Timeout
+from urllib3.exceptions import HTTPError, ReadTimeoutError, ProtocolError
 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__)
 
 
-def uri(protocol, host, port, endpoint=''):
-    return '{0}://{1}:{2}{3}'.format(protocol, host, port, endpoint)
+class EtcdRaftInternal(etcd.EtcdException):
+    """Raft Internal Error"""
 
 
 class EtcdError(DCSError):
@@ -69,12 +70,15 @@ 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:
-            logger.warning('failed to resolve host %s', host)
+        except Exception as e:
+            logger.warning('failed to resolve host %s: %s', host, e)
             return []
 
 
@@ -87,30 +91,68 @@ class Client(etcd.Client):
         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)
+        # 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 = not self._use_proxies
+        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):
+        """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."""
+
+        per_node_timeout = timeout = float(timeout or self.read_timeout)
+
+        max_retries = 4 - min(etcd_nodes, 3)
+        per_node_retries = 1
+        min_timeout = 1.0
+
+        while etcd_nodes > 0:
+            per_node_timeout = float(timeout) / 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 _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 _build_request_parameters(self, etcd_nodes, timeout=None):
         kwargs = {'headers': self._get_headers(), 'redirect': self.allow_redirect}
 
-        # calculate the number of retries and timeout *per node*
-        # 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:
-        #     kwargs.update({'retries': 0, 'timeout': float(self.read_timeout)/etcd_nodes})
-        # elif etcd_nodes > 1:
-        #     kwargs.update({'retries': 1, 'timeout': self.read_timeout/2.0/etcd_nodes})
-        # else:
-        #     kwargs.update({'retries': 2, 'timeout': self.read_timeout/3.0})
-        kwargs['timeout'] = self.read_timeout/float(kwargs['retries'] + 1)/etcd_nodes
+        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
 
+    @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
@@ -123,56 +165,64 @@ 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._build_request_parameters(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 = [n.strip() for n in self._handle_server_response(response).data.decode('utf-8').split(',')]
+                response = self.http.request(self._MGET, base_uri + self.version_prefix + '/machines', **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)
-                random.shuffle(machines)
-                for url in 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
+                if machines:
+                    random.shuffle(machines)
+                    self._update_dns_cache(self._dns_resolver.resolve_async, machines)
+                    return machines
             except Exception as e:
-                # We can't get the list of machines, if one server is in the
-                # machines cache, try on it
-                logger.error("Failed to get list of machines from %s%s: %r", self._base_uri, self.version_prefix, e)
-                if self._machines_cache:
-                    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 []
+                self.http.clear()
+                logger.error("Failed to get list of machines from %s%s: %r", base_uri, self.version_prefix, e)
+
+        raise etcd.EtcdConnectionFailed('No more machines in the cluster')
 
     def set_read_timeout(self, timeout):
         self._read_timeout = timeout
 
-    def _do_http_request(self, request_executor, method, url, fields=None, **kwargs):
-        try:
-            response = request_executor(method, url, fields=fields, **kwargs)
-            response.data.decode('utf-8')
-            self._check_cluster_id(response)
-        except (HTTPError, HTTPException, socket.error, socket.timeout) as e:
-            if (isinstance(fields, dict) and fields.get("wait") == "true" and
-                    isinstance(e, ReadTimeoutError)):
-                logger.debug("Watch timed out.")
-                raise etcd.EtcdWatchTimedOut("Watch timed out: {0}".format(e), cause=e)
-            logger.error("Request to server %s failed: %r", self._base_uri, e)
-            logger.info("Reconnection allowed, looking for another server.")
-            self._base_uri = self._next_server(cause=e)
-            response = False
-        return response
+    def _do_http_request(self, retry, machines_cache, request_executor, method, path, fields=None, **kwargs):
+        some_request_failed = False
+        for i, base_uri in enumerate(machines_cache):
+            if i > 0:
+                logger.info("Retrying on %s", base_uri)
+            try:
+                response = request_executor(method, base_uri + path, fields=fields, **kwargs)
+                response.data.decode('utf-8')
+                self._check_cluster_id(response)
+                if some_request_failed:
+                    self.set_base_uri(base_uri)
+                    self._refresh_machines_cache()
+                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')
 
     def api_execute(self, path, method, params=None, timeout=None):
         if not path.startswith('/'):
             raise ValueError('Path does not start with /')
 
+        retry = params.pop('retry', None) if isinstance(params, dict) else None
         kwargs = {'fields': params, 'preload_content': False}
 
         if method in [self._MGET, self._MDELETE]:
@@ -186,38 +236,37 @@ class Client(etcd.Client):
         # Update machines_cache if previous attempt of update has failed
         if self._update_machines_cache:
             self._load_machines_cache()
-        elif time.time() - self._machines_cache_updated > self._machines_cache_ttl:
-            self._machines_cache = self.machines
-            if self._base_uri in self._machines_cache:
-                self._machines_cache.remove(self._base_uri)
-            self._machines_cache_updated = time.time()
+        elif not self._use_proxies and time.time() - self._machines_cache_updated > self._machines_cache_ttl:
+            self._refresh_machines_cache()
 
-        kwargs.update(self._build_request_parameters())
+        machines_cache = self.machines_cache
+        etcd_nodes = len(machines_cache)
+        kwargs.update(self._build_request_parameters(etcd_nodes, timeout))
 
-        if timeout is not None:
-            kwargs.update({'retries': 0, 'timeout': timeout})
-
-        response = False
-
-        try:
-            some_request_failed = False
-            while not response:
-                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 and not self._use_proxies:
-                self._machines_cache = self.machines
-                if self._base_uri in self._machines_cache:
-                    self._machines_cache.remove(self._base_uri)
-        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:
+        while True:
+            try:
+                response = self._do_http_request(retry, machines_cache, request_executor, method, path, **kwargs)
+                return self._handle_server_response(response)
+            except etcd.EtcdWatchTimedOut:
                 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):
@@ -236,15 +285,15 @@ class Client(etcd.Client):
             protocol = 'https' if '-ssl' in r else 'http'
             endpoint = '/members' if '-server' in r else ''
             for host, port in self.get_srv_record('_etcd{0}._tcp.{1}'.format(r, srv)):
-                url = uri(protocol, host, port, endpoint)
+                url = uri(protocol, (host, port), endpoint)
                 if endpoint:
                     try:
-                        response = requests.get(url, timeout=self.read_timeout, verify=False)
-                        if response.ok:
-                            for member in response.json():
+                        response = requests_get(url, timeout=self.read_timeout, verify=False)
+                        if response.status < 400:
+                            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)
@@ -258,15 +307,32 @@ class Client(etcd.Client):
     def _get_machines_cache_from_dns(self, host, port):
         """One host might be resolved into multiple ip addresses. We will make list out of it"""
         if self.protocol == 'http':
-            ret = []
-            for af, _, _, _, sa in self._dns_resolver.resolve(host, port):
-                host, port = sa[:2]
-                if af == socket.AF_INET6:
-                    host = '[{0}]'.format(host)
-                ret.append(uri(self.protocol, host, port))
+            ret = map(lambda res: uri(self.protocol, res[-1][:2]), self._dns_resolver.resolve(host, port))
             if ret:
                 return list(set(ret))
-        return [uri(self.protocol, host, port)]
+        return [uri(self.protocol, (host, port))]
+
+    def _get_machines_cache_from_config(self):
+        if 'proxy' in self._config:
+            return [uri(self.protocol, (self._config['host'], self._config['port']))]
+
+        machines_cache = []
+        if 'srv' in self._config:
+            machines_cache = self._get_machines_cache_from_srv(self._config['srv'])
+
+        if not machines_cache and 'hosts' in self._config:
+            machines_cache = list(self._config['hosts'])
+
+        if not machines_cache and 'host' in self._config:
+            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.
@@ -279,34 +345,49 @@ 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')
 
-        if self._use_proxies:
-            self._machines_cache = [uri(self.protocol, self._config['host'], self._config['port'])]
-        else:
-            self._machines_cache = []
-
-            if 'srv' in self._config:
-                self._machines_cache = self._get_machines_cache_from_srv(self._config['srv'])
-
-            if not self._machines_cache and 'hosts' in self._config:
-                self._machines_cache = list(self._config['hosts'])
-
-            if not self._machines_cache and 'host' in self._config:
-                self._machines_cache = self._get_machines_cache_from_dns(self._config['host'], self._config['port'])
-
+        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
-        self._base_uri = self._next_server()
-        self._machines_cache = self.machines
+        # enforce resolving dns name,they might get new ips
+        self._update_dns_cache(self._dns_resolver.remove, machines_cache)
 
-        if self._base_uri in self._machines_cache:
-            self._machines_cache.remove(self._base_uri)
+        # 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 set_base_uri(self, value):
+        logger.info('Selected new etcd server %s', value)
+        self._base_uri = value
+
 
 class Etcd(AbstractDCS):
 
@@ -314,15 +395,15 @@ class Etcd(AbstractDCS):
         super(Etcd, self).__init__(config)
         self._ttl = int(config.get('ttl') or 30)
         self._retry = Retry(deadline=config['retry_timeout'], max_delay=1, max_tries=-1,
-                            retry_exceptions=(etcd.EtcdLeaderElectionInProgress,
-                                              etcd.EtcdWatcherCleared,
-                                              etcd.EtcdEventIndexCleared))
+                            retry_exceptions=(etcd.EtcdLeaderElectionInProgress, EtcdRaftInternal))
         self._client = self.get_etcd_client(config)
         self.__do_not_watch = False
         self._has_failed = False
 
     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:
@@ -370,7 +451,7 @@ 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
@@ -426,6 +507,8 @@ class Etcd(AbstractDCS):
         while not client:
             try:
                 client = Client(config, dns_resolver)
+                if 'use_proxies' in config and not client.machines:
+                    raise etcd.EtcdException
             except etcd.EtcdException:
                 logger.info('waiting on etcd')
                 time.sleep(5)
@@ -437,6 +520,10 @@ class Etcd(AbstractDCS):
         self._ttl = ttl
         self._client.set_machines_cache_ttl(ttl*10)
 
+    @property
+    def ttl(self):
+        return self._ttl
+
     def set_retry_timeout(self, retry_timeout):
         self._retry.deadline = retry_timeout
         self._client.set_read_timeout(retry_timeout)
@@ -446,9 +533,10 @@ class Etcd(AbstractDCS):
         return Member.from_node(node.modifiedIndex, os.path.basename(node.key), node.ttl, node.value)
 
     def _load_cluster(self):
+        cluster = None
         try:
             result = self.retry(self._client.read, self.client_path(''), recursive=True)
-            nodes = {os.path.relpath(node.key, result.key).replace('\\', '/'): node for node in result.leaves}
+            nodes = {node.key[len(result.key):].lstrip('/'): node for node in result.leaves}
 
             # get initialize flag
             initialize = nodes.get(self._INITIALIZE)
@@ -486,21 +574,22 @@ class Etcd(AbstractDCS):
             sync = nodes.get(self._SYNC)
             sync = SyncState.from_node(sync and sync.modifiedIndex, sync and sync.value)
 
-            self._cluster = Cluster(initialize, config, leader, last_leader_operation, members, failover, sync, history)
+            cluster = Cluster(initialize, config, leader, last_leader_operation, members, failover, sync, history)
         except etcd.EtcdKeyNotFound:
-            self._cluster = Cluster(None, None, None, None, [], None, None, None)
+            cluster = Cluster(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
+        return cluster
 
     @catch_etcd_errors
-    def touch_member(self, data, ttl=None, permanent=False):
+    def touch_member(self, data, permanent=False):
         data = json.dumps(data, separators=(',', ':'))
-        return self._client.set(self.member_path, data, None if permanent else ttl or self._ttl)
+        return self._client.set(self.member_path, data, None if permanent else self._ttl)
 
     @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:
@@ -529,14 +618,14 @@ class Etcd(AbstractDCS):
 
     @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
@@ -575,7 +664,6 @@ class Etcd(AbstractDCS):
                     # 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
@@ -590,3 +678,6 @@ class Etcd(AbstractDCS):
             return super(Etcd, self).watch(None, timeout)
         finally:
             self.event.clear()
+
+
+etcd.EtcdError.error_exceptions[300] = EtcdRaftInternal

patroni/dcs/exhibitor.py

@@ -1,10 +1,11 @@
+import json
 import logging
 import random
-import requests
 import time
 
 from patroni.dcs.zookeeper import ZooKeeper
-from requests.exceptions import RequestException
+from patroni.request import get as requests_get
+from patroni.utils import uri
 
 logger = logging.getLogger(__name__)
 
@@ -46,12 +47,11 @@ class ExhibitorEnsembleProvider(object):
     def _query_exhibitors(self, exhibitors):
         random.shuffle(exhibitors)
         for host in exhibitors:
-            uri = 'http://{0}:{1}{2}'.format(host, self._exhibitor_port, self._uri_path)
             try:
-                response = requests.get(uri, timeout=self.TIMEOUT)
-                return response.json()
-            except RequestException:
-                pass
+                response = requests_get(uri('http', (host, self._exhibitor_port), self._uri_path), timeout=self.TIMEOUT)
+                return json.loads(response.data.decode('utf-8'))
+            except Exception:
+                logging.debug('Request to %s failed', host)
         return None
 
     @property

patroni/dcs/kubernetes.py

@@ -8,11 +8,14 @@ import sys
 import time
 
 from kubernetes import client as k8s_client, config as k8s_config, watch as k8s_watch
-from patroni.dcs import AbstractDCS, ClusterConfig, Cluster, Failover, Leader, Member, SyncState, TimelineHistory
-from patroni.exceptions import DCSError
-from patroni.utils import deep_compare, tzutc, Retry, RetryFailedError
+from urllib3 import Timeout
 from urllib3.exceptions import HTTPError
 from six.moves.http_client import HTTPException
+from threading import Condition, Lock, Thread
+
+from . import AbstractDCS, Cluster, ClusterConfig, Failover, Leader, Member, SyncState, TimelineHistory
+from ..exceptions import DCSError
+from ..utils import deep_compare, Retry, RetryFailedError, tzutc, USER_AGENT
 
 logger = logging.getLogger(__name__)
 
@@ -28,16 +31,39 @@ class KubernetesRetriableException(k8s_client.rest.ApiException):
         self.body = orig.body
         self.headers = orig.headers
 
+    @property
+    def sleeptime(self):
+        try:
+            return int(self.headers['retry-after'])
+        except Exception:
+            return None
+
 
 class CoreV1ApiProxy(object):
 
     def __init__(self, use_endpoints=False):
         self._api = k8s_client.CoreV1Api()
+        self._api.api_client.user_agent = USER_AGENT
+        self._api.api_client.rest_client.pool_manager.connection_pool_kw['maxsize'] = 10
         self._request_timeout = None
-        self._use_endpoints = use_endpoints
+        self._use_endpoints = bool(use_endpoints)
 
-    def set_timeout(self, timeout):
-        self._request_timeout = (1, timeout / 3.0)
+    def configure_timeouts(self, loop_wait, retry_timeout, ttl):
+        # Normally every loop_wait seconds we should have receive something from the socket.
+        # If we didn't received anything after the loop_wait + retry_timeout it is a time
+        # to start worrying (send keepalive messages). Finally, the connection should be
+        # considered as dead if we received nothing from the socket after the ttl seconds.
+        cnt = 3
+        idle = int(loop_wait + retry_timeout)
+        intvl = max(1, int(float(ttl - idle) / cnt))
+        self._api.api_client.rest_client.pool_manager.connection_pool_kw['socket_options'] = [
+            (socket.SOL_SOCKET, socket.SO_KEEPALIVE, 1),
+            (socket.IPPROTO_TCP, socket.TCP_KEEPIDLE, idle),
+            (socket.IPPROTO_TCP, socket.TCP_KEEPINTVL, intvl),
+            (socket.IPPROTO_TCP, socket.TCP_KEEPCNT, cnt),
+            (socket.IPPROTO_TCP, 18, int(ttl * 1000))  # TCP_USER_TIMEOUT
+        ]
+        self._request_timeout = (1, retry_timeout / 3.0)
 
     def __getattr__(self, func):
         if func.endswith('_kind'):
@@ -49,11 +75,15 @@ class CoreV1ApiProxy(object):
             try:
                 return getattr(self._api, func)(*args, **kwargs)
             except k8s_client.rest.ApiException as e:
-                if e.status in (502, 503, 504):  # XXX
+                if e.status in (500, 503, 504) or e.headers and 'retry-after' in e.headers:  # XXX
                     raise KubernetesRetriableException(e)
                 raise
         return wrapper
 
+    @property
+    def use_endpoints(self):
+        return self._use_endpoints
+
 
 def catch_kubernetes_errors(func):
     def wrapper(*args, **kwargs):
@@ -70,6 +100,124 @@ def catch_kubernetes_errors(func):
     return wrapper
 
 
+class ObjectCache(Thread):
+
+    def __init__(self, dcs, func, retry, condition, name=None):
+        Thread.__init__(self)
+        self.daemon = True
+        self._api_client = k8s_client.ApiClient()
+        self._dcs = dcs
+        self._func = func
+        self._retry = retry
+        self._condition = condition
+        self._name = name  # name of this pod
+        self._is_ready = False
+        self._object_cache = {}
+        self._object_cache_lock = Lock()
+        self._annotations_map = {self._dcs.leader_path: self._dcs._LEADER, self._dcs.config_path: self._dcs._CONFIG}
+        self.start()
+
+    def _list(self):
+        try:
+            return self._func(_request_timeout=(self._retry.deadline, Timeout.DEFAULT_TIMEOUT))
+        except Exception:
+            time.sleep(1)
+            raise
+
+    def _watch(self, resource_version):
+        return self._func(_request_timeout=(self._retry.deadline, Timeout.DEFAULT_TIMEOUT),
+                          _preload_content=False, watch=True, resource_version=resource_version)
+
+    def set(self, name, value):
+        with self._object_cache_lock:
+            old_value = self._object_cache.get(name)
+            ret = not old_value or int(old_value.metadata.resource_version) < int(value.metadata.resource_version)
+            if ret:
+                self._object_cache[name] = value
+        return ret, old_value
+
+    def delete(self, name, resource_version):
+        with self._object_cache_lock:
+            old_value = self._object_cache.get(name)
+            ret = old_value and int(old_value.metadata.resource_version) < int(resource_version)
+            if ret:
+                del self._object_cache[name]
+        return not old_value or ret, old_value
+
+    def copy(self):
+        with self._object_cache_lock:
+            return self._object_cache.copy()
+
+    def get(self, name):
+        with self._object_cache_lock:
+            return self._object_cache.get(name)
+
+    def _build_cache(self):
+        objects = self._list()
+        return_type = 'V1' + objects.kind[:-4]
+        with self._object_cache_lock:
+            self._object_cache = {item.metadata.name: item for item in objects.items}
+        with self._condition:
+            self._is_ready = True
+            self._condition.notify()
+
+        response = self._watch(objects.metadata.resource_version)
+        try:
+            for line in k8s_watch.watch.iter_resp_lines(response):
+                event = json.loads(line)
+                obj = event['object']
+                if obj.get('code') == 410:
+                    break
+
+                ev_type = event['type']
+                name = obj['metadata']['name']
+
+                if ev_type in ('ADDED', 'MODIFIED'):
+                    obj = k8s_watch.watch.SimpleNamespace(data=json.dumps(obj))
+                    obj = self._api_client.deserialize(obj, return_type)
+                    success, old_value = self.set(name, obj)
+                    if success:
+                        new_value = (obj.metadata.annotations or {}).get(self._annotations_map.get(name))
+                elif ev_type == 'DELETED':
+                    success, old_value = self.delete(name, obj['metadata']['resourceVersion'])
+                    new_value = None
+                else:
+                    logger.warning('Unexpected event type: %s', ev_type)
+                    continue
+
+                if success and return_type != 'V1Pod':
+                    if old_value:
+                        old_value = (old_value.metadata.annotations or {}).get(self._annotations_map.get(name))
+
+                    value_changed = old_value != new_value and \
+                        (name != self._dcs.config_path or old_value is not None and new_value is not None)
+
+                    if value_changed:
+                        logger.debug('%s changed from %s to %s', name, old_value, new_value)
+
+                    # Do not wake up HA loop if we run as leader and received leader object update event
+                    if value_changed or name == self._dcs.leader_path and self._name != new_value:
+                        self._dcs.event.set()
+        finally:
+            with self._condition:
+                self._is_ready = False
+            response.close()
+            response.release_conn()
+
+    def run(self):
+        while True:
+            try:
+                self._build_cache()
+            except Exception as e:
+                with self._condition:
+                    self._is_ready = False
+                logger.error('ObjectCache.run %r', e)
+
+    def is_ready(self):
+        """Must be called only when holding the lock on `_condition`"""
+        return self._is_ready
+
+
 class Kubernetes(AbstractDCS):
 
     def __init__(self, config):
@@ -89,25 +237,33 @@ class Kubernetes(AbstractDCS):
         except k8s_config.ConfigException:
             k8s_config.load_kube_config(context=config.get('context', 'local'))
 
-        self.__subsets = None
-        use_endpoints = config.get('use_endpoints') and (config.get('patronictl') or 'pod_ip' in config)
-        if use_endpoints:
-            addresses = [k8s_client.V1EndpointAddress(ip=config['pod_ip'])]
-            ports = []
-            for p in config.get('ports', [{}]):
-                port = {'port': int(p.get('port', '5432'))}
-                port.update({n: p[n] for n in ('name', 'protocol') if p.get(n)})
-                ports.append(k8s_client.V1EndpointPort(**port))
-            self.__subsets = [k8s_client.V1EndpointSubset(addresses=addresses, ports=ports)]
-        self._api = CoreV1ApiProxy(use_endpoints)
-        self.set_retry_timeout(config['retry_timeout'])
-        self.set_ttl(config.get('ttl') or 30)
+        self.__my_pod = None
+        self.__ips = [] if config.get('patronictl') else [config.get('pod_ip')]
+        self.__ports = []
+        for p in config.get('ports', [{}]):
+            port = {'port': int(p.get('port', '5432'))}
+            port.update({n: p[n] for n in ('name', 'protocol') if p.get(n)})
+            self.__ports.append(k8s_client.V1EndpointPort(**port))
+
+        self._api = CoreV1ApiProxy(config.get('use_endpoints'))
+        self._should_create_config_service = self._api.use_endpoints
+        self.reload_config(config)
+        # leader_observed_record, leader_resource_version, and leader_observed_time are used only for leader race!
         self._leader_observed_record = {}
         self._leader_observed_time = None
         self._leader_resource_version = None
-        self._leader_observed_subsets = []
         self.__do_not_watch = False
 
+        self._condition = Condition()
+
+        pods_func = functools.partial(self._api.list_namespaced_pod, self._namespace,
+                                      label_selector=self._label_selector)
+        self._pods = ObjectCache(self, pods_func, self._retry, self._condition)
+
+        kinds_func = functools.partial(self._api.list_namespaced_kind, self._namespace,
+                                       label_selector=self._label_selector)
+        self._kinds = ObjectCache(self, kinds_func, self._retry, self._condition, self._name)
+
     def retry(self, *args, **kwargs):
         return self._retry.copy()(*args, **kwargs)
 
@@ -116,16 +272,23 @@ class Kubernetes(AbstractDCS):
 
     @property
     def leader_path(self):
-        return self._base_path[1:] if self.__subsets else super(Kubernetes, self).leader_path
+        return self._base_path[1:] if self._api.use_endpoints else super(Kubernetes, self).leader_path
 
     def set_ttl(self, ttl):
         ttl = int(ttl)
         self.__do_not_watch = self._ttl != ttl
         self._ttl = ttl
 
+    @property
+    def ttl(self):
+        return self._ttl
+
     def set_retry_timeout(self, retry_timeout):
         self._retry.deadline = retry_timeout
-        self._api.set_timeout(retry_timeout)
+
+    def reload_config(self, config):
+        super(Kubernetes, self).reload_config(config)
+        self._api.configure_timeouts(self.loop_wait, self._retry.deadline, self.ttl)
 
     @staticmethod
     def member(pod):
@@ -134,14 +297,21 @@ class Kubernetes(AbstractDCS):
         member.data['pod_labels'] = pod.metadata.labels
         return member
 
+    def _wait_caches(self):
+        stop_time = time.time() + self._retry.deadline
+        while not (self._pods.is_ready() and self._kinds.is_ready()):
+            timeout = stop_time - time.time()
+            if timeout <= 0:
+                raise RetryFailedError('Exceeded retry deadline')
+            self._condition.wait(timeout)
+
     def _load_cluster(self):
         try:
-            # get list of members
-            response = self.retry(self._api.list_namespaced_pod, self._namespace, label_selector=self._label_selector)
-            members = [self.member(pod) for pod in response.items]
+            with self._condition:
+                self._wait_caches()
 
-            response = self.retry(self._api.list_namespaced_kind, self._namespace, label_selector=self._label_selector)
-            nodes = {item.metadata.name: item for item in response.items}
+                members = [self.member(pod) for pod in self._pods.copy().values()]
+                nodes = self._kinds.copy()
 
             config = nodes.get(self.config_path)
             metadata = config and config.metadata
@@ -152,7 +322,8 @@ class Kubernetes(AbstractDCS):
 
             # get global dynamic configuration
             config = ClusterConfig.from_node(metadata and metadata.resource_version,
-                                             annotations.get(self._CONFIG) or '{}')
+                                             annotations.get(self._CONFIG) or '{}',
+                                             metadata.resource_version if self._CONFIG in annotations else 0)
 
             # get timeline history
             history = TimelineHistory.from_node(metadata and metadata.resource_version,
@@ -161,7 +332,6 @@ class Kubernetes(AbstractDCS):
             leader = nodes.get(self.leader_path)
             metadata = leader and leader.metadata
             self._leader_resource_version = metadata.resource_version if metadata else None
-            self._leader_observed_subsets = leader.subsets if self.__subsets and leader and leader.subsets else []
             annotations = metadata and metadata.annotations or {}
 
             # get last leader operation
@@ -187,19 +357,20 @@ class Kubernetes(AbstractDCS):
             if metadata:
                 member = Member(-1, leader, None, {})
                 member = ([m for m in members if m.name == leader] or [member])[0]
-                leader = Leader(response.metadata.resource_version, None, member)
+                leader = Leader(metadata.resource_version, None, member)
 
             # failover key
             failover = nodes.get(self.failover_path)
             metadata = failover and failover.metadata
-            failover = Failover.from_node(metadata and metadata.resource_version, metadata and metadata.annotations)
+            failover = Failover.from_node(metadata and metadata.resource_version,
+                                          metadata and (metadata.annotations or {}).copy())
 
             # get synchronization state
             sync = nodes.get(self.sync_path)
             metadata = sync and sync.metadata
             sync = SyncState.from_node(metadata and metadata.resource_version,  metadata and metadata.annotations)
 
-            self._cluster = Cluster(initialize, config, leader, last_leader_operation, members, failover, sync, history)
+            return Cluster(initialize, config, leader, last_leader_operation, members, failover, sync, history)
         except Exception:
             logger.exception('get_cluster')
             raise KubernetesError('Kubernetes API is not responding properly')
@@ -209,50 +380,71 @@ class Kubernetes(AbstractDCS):
         return p1.name == p2.name and p1.port == p2.port and (p1.protocol or 'TCP') == (p2.protocol or 'TCP')
 
     @staticmethod
-    def subsets_changed(last_observed_subsets, subsets):
+    def subsets_changed(last_observed_subsets, ip, ports):
         """
-        >>> Kubernetes.subsets_changed([], [])
-        False
-        >>> Kubernetes.subsets_changed([], [k8s_client.V1EndpointSubset()])
+        >>> Kubernetes.subsets_changed([], None, [])
         True
-        >>> s1 = [k8s_client.V1EndpointSubset(addresses=[k8s_client.V1EndpointAddress(ip='1.2.3.4')])]
-        >>> s2 = [k8s_client.V1EndpointSubset(addresses=[k8s_client.V1EndpointAddress(ip='1.2.3.5')])]
-        >>> Kubernetes.subsets_changed(s1, s2)
+        >>> ip = '1.2.3.4'
+        >>> a = [k8s_client.V1EndpointAddress(ip=ip)]
+        >>> s = [k8s_client.V1EndpointSubset(addresses=a)]
+        >>> Kubernetes.subsets_changed(s, '1.2.3.5', [])
         True
-        >>> a = [k8s_client.V1EndpointAddress(ip='1.2.3.4')]
-        >>> s1 = [k8s_client.V1EndpointSubset(addresses=a, ports=[k8s_client.V1EndpointPort(protocol='TCP', port=1)])]
-        >>> s2 = [k8s_client.V1EndpointSubset(addresses=a, ports=[k8s_client.V1EndpointPort(port=5432)])]
-        >>> Kubernetes.subsets_changed(s1, s2)
+        >>> s = [k8s_client.V1EndpointSubset(addresses=a, ports=[k8s_client.V1EndpointPort(protocol='TCP', port=1)])]
+        >>> Kubernetes.subsets_changed(s, '1.2.3.4', [k8s_client.V1EndpointPort(port=5432)])
         True
         >>> p1 = k8s_client.V1EndpointPort(name='port1', port=1)
         >>> p2 = k8s_client.V1EndpointPort(name='port2', port=2)
         >>> p3 = k8s_client.V1EndpointPort(name='port3', port=3)
-        >>> s1 = [k8s_client.V1EndpointSubset(addresses=a, ports=[p1, p2])]
-        >>> s2 = [k8s_client.V1EndpointSubset(addresses=a, ports=[p2, p3])]
-        >>> Kubernetes.subsets_changed(s1, s2)
+        >>> s = [k8s_client.V1EndpointSubset(addresses=a, ports=[p1, p2])]
+        >>> Kubernetes.subsets_changed(s, ip, [p2, p3])
         True
         >>> s2 = [k8s_client.V1EndpointSubset(addresses=a, ports=[p2, p1])]
-        >>> Kubernetes.subsets_changed(s1, s2)
+        >>> Kubernetes.subsets_changed(s, ip, [p2, p1])
         False
         """
-        if len(last_observed_subsets) != len(subsets):
+
+        if len(last_observed_subsets) != 1:
             return True
-        if subsets == []:
-            return False
         if len(last_observed_subsets[0].addresses or []) != 1 or \
-                last_observed_subsets[0].addresses[0].ip != subsets[0].addresses[0].ip or \
-                len(last_observed_subsets[0].ports) != len(subsets[0].ports):
+                last_observed_subsets[0].addresses[0].ip != ip or \
+                len(last_observed_subsets[0].ports) != len(ports):
             return True
-        if len(subsets[0].ports) == 1:
-            return not Kubernetes.compare_ports(last_observed_subsets[0].ports[0], subsets[0].ports[0])
+        if len(ports) == 1:
+            return not Kubernetes.compare_ports(last_observed_subsets[0].ports[0], ports[0])
         observed_ports = {p.name: p for p in last_observed_subsets[0].ports}
-        for p in subsets[0].ports:
+        for p in ports:
             if p.name not in observed_ports or not Kubernetes.compare_ports(p, observed_ports.pop(p.name)):
                 return True
         return False
 
-    @catch_kubernetes_errors
-    def patch_or_create(self, name, annotations, resource_version=None, patch=False, retry=True, subsets=None):
+    def __target_ref(self, leader_ip, latest_subsets, pod):
+        # we want to re-use existing target_ref if possible
+        for subset in latest_subsets:
+            for address in subset.addresses or []:
+                if address.ip == leader_ip and address.target_ref and address.target_ref.name == self._name:
+                    return address.target_ref
+        return k8s_client.V1ObjectReference(kind='Pod', uid=pod.metadata.uid, namespace=self._namespace,
+                                            name=self._name, resource_version=pod.metadata.resource_version)
+
+    def _map_subsets(self, endpoints, ips):
+        leader = self._kinds.get(self.leader_path)
+        latest_subsets = leader and leader.subsets or []
+        if not ips:
+            # We want to have subsets empty
+            if latest_subsets:
+                endpoints['subsets'] = []
+            return
+
+        pod = self._pods.get(self._name)
+        leader_ip = ips[0] or pod and pod.status.pod_ip
+        # don't touch subsets if our (leader) ip is unknown or subsets is valid
+        if leader_ip and self.subsets_changed(latest_subsets, leader_ip, self.__ports):
+            kwargs = {'hostname': pod.spec.hostname, 'node_name': pod.spec.node_name,
+                      'target_ref': self.__target_ref(leader_ip, latest_subsets, pod)} if pod else {}
+            address = k8s_client.V1EndpointAddress(ip=leader_ip, **kwargs)
+            endpoints['subsets'] = [k8s_client.V1EndpointSubset(addresses=[address], ports=self.__ports)]
+
+    def _patch_or_create(self, name, annotations, resource_version=None, patch=False, retry=None, ips=None):
         metadata = {'namespace': self._namespace, 'name': name, 'labels': self._labels, 'annotations': annotations}
         if patch or resource_version:
             if resource_version is not None:
@@ -264,14 +456,40 @@ class Kubernetes(AbstractDCS):
             metadata['annotations'] = {k: v for k, v in metadata['annotations'].items() if v is not None}
 
         metadata = k8s_client.V1ObjectMeta(**metadata)
-        if subsets is not None and self.__subsets:
+        if ips is not None and self._api.use_endpoints:
             endpoints = {'metadata': metadata}
-            if self.subsets_changed(self._leader_observed_subsets, subsets):
-                endpoints['subsets'] = subsets
+            self._map_subsets(endpoints, ips)
             body = k8s_client.V1Endpoints(**endpoints)
         else:
             body = k8s_client.V1ConfigMap(metadata=metadata)
-        return self.retry(func, self._namespace, body) if retry else func(self._namespace, body)
+        ret = retry(func, self._namespace, body) if retry else func(self._namespace, body)
+        if ret:
+            self._kinds.set(name, ret)
+        return ret
+
+    @catch_kubernetes_errors
+    def patch_or_create(self, name, annotations, resource_version=None, patch=False, retry=True, ips=None):
+        if retry is True:
+            retry = self.retry
+        return self._patch_or_create(name, annotations, resource_version, patch, retry, ips)
+
+    def patch_or_create_config(self, annotations, resource_version=None, patch=False, retry=True):
+        # SCOPE-config endpoint requires corresponding service otherwise it might be "cleaned" by k8s master
+        if self._api.use_endpoints and not patch and not resource_version:
+            self._should_create_config_service = True
+            self._create_config_service()
+        return self.patch_or_create(self.config_path, annotations, resource_version, patch, retry)
+
+    def _create_config_service(self):
+        metadata = k8s_client.V1ObjectMeta(namespace=self._namespace, name=self.config_path, labels=self._labels)
+        body = k8s_client.V1Service(metadata=metadata, spec=k8s_client.V1ServiceSpec(cluster_ip='None'))
+        try:
+            if not self._api.create_namespaced_service(self._namespace, body):
+                return
+        except Exception as e:
+            if not isinstance(e, k8s_client.rest.ApiException) or e.status != 409:  # Service already exists
+                return logger.exception('create_config_service failed')
+        self._should_create_config_service = False
 
     def _write_leader_optime(self, last_operation):
         """Unused"""
@@ -279,18 +497,58 @@ class Kubernetes(AbstractDCS):
     def _update_leader(self):
         """Unused"""
 
-    def update_leader(self, last_operation):
+    def _update_leader_with_retry(self, annotations, resource_version, ips):
+        retry = self._retry.copy()
+
+        def _retry(*args, **kwargs):
+            return retry(*args, **kwargs)
+
+        try:
+            return self._patch_or_create(self.leader_path, annotations, resource_version, ips=ips, retry=_retry)
+        except k8s_client.rest.ApiException as e:
+            if e.status == 409:
+                logger.warning('Concurrent update of %s', self.leader_path)
+            else:
+                logger.exception('Permission denied' if e.status == 403 else 'Unexpected error from Kubernetes API')
+                return False
+        except RetryFailedError:
+            return False
+
+        deadline = retry.stoptime - time.time()
+        if deadline < 2:
+            return False
+
+        retry.sleep_func(1)  # Give a chance for ObjectCache to receive the latest version
+
+        kind = self._kinds.get(self.leader_path)
+        kind_annotations = kind and kind.metadata.annotations or {}
+        kind_resource_version = kind and kind.metadata.resource_version
+
+        # There is different leader or resource_version in cache didn't change
+        if kind and (kind_annotations.get(self._LEADER) != self._name or kind_resource_version == resource_version):
+            return False
+
+        retry.deadline = deadline - 1  # Update deadline and retry
+        return self.patch_or_create(self.leader_path, annotations, kind_resource_version, ips=ips, retry=_retry)
+
+    def update_leader(self, last_operation, access_is_restricted=False):
+        kind = self._kinds.get(self.leader_path)
+        kind_annotations = kind and kind.metadata.annotations or {}
+
+        if kind and kind_annotations.get(self._LEADER) != self._name:
+            return False
+
         now = datetime.datetime.now(tzutc).isoformat()
+        leader_observed_record = kind_annotations or self._leader_observed_record
         annotations = {self._LEADER: self._name, 'ttl': str(self._ttl), 'renewTime': now,
-                       'acquireTime': self._leader_observed_record.get('acquireTime') or now,
-                       'transitions': self._leader_observed_record.get('transitions') or '0'}
+                       'acquireTime': leader_observed_record.get('acquireTime') or now,
+                       'transitions': leader_observed_record.get('transitions') or '0'}
         if last_operation:
             annotations[self._OPTIME] = last_operation
 
-        ret = self.patch_or_create(self.leader_path, annotations, self._leader_resource_version, subsets=self.__subsets)
-        if ret:
-            self._leader_resource_version = ret.metadata.resource_version
-        return ret
+        resource_version = kind and kind.metadata.resource_version
+        ips = [] if access_is_restricted else self.__ips
+        return self._update_leader_with_retry(annotations, resource_version, ips)
 
     def attempt_to_acquire_leader(self, permanent=False):
         now = datetime.datetime.now(tzutc).isoformat()
@@ -307,10 +565,9 @@ class Kubernetes(AbstractDCS):
             else:
                 annotations['acquireTime'] = self._leader_observed_record.get('acquireTime') or now
             annotations['transitions'] = str(transitions)
-        ret = self.patch_or_create(self.leader_path, annotations, self._leader_resource_version, subsets=self.__subsets)
-        if ret:
-            self._leader_resource_version = ret.metadata.resource_version
-        else:
+        ips = [] if self._api.use_endpoints else None
+        ret = self.patch_or_create(self.leader_path, annotations, self._leader_resource_version, ips=ips)
+        if not ret:
             logger.info('Could not take out TTL lock')
         return ret
 
@@ -321,19 +578,24 @@ class Kubernetes(AbstractDCS):
         """Unused"""
 
     def manual_failover(self, leader, candidate, scheduled_at=None, index=None):
-        annotations = {'leader': leader or None, 'member': candidate or None, 'scheduled_at': scheduled_at}
+        annotations = {'leader': leader or None, 'member': candidate or None,
+                       'scheduled_at': scheduled_at and scheduled_at.isoformat()}
         patch = bool(self.cluster and isinstance(self.cluster.failover, Failover) and self.cluster.failover.index)
         return self.patch_or_create(self.failover_path, annotations, index, bool(index or patch), False)
 
+    @property
+    def _config_resource_version(self):
+        config = self._kinds.get(self.config_path)
+        return config and config.metadata.resource_version
+
     def set_config_value(self, value, index=None):
-        patch = bool(index or self.cluster and self.cluster.config and self.cluster.config.index)
-        return self.patch_or_create(self.config_path, {self._CONFIG: value}, index, patch, False)
+        return self.patch_or_create_config({self._CONFIG: value}, index, bool(self._config_resource_version), False)
 
     @catch_kubernetes_errors
-    def touch_member(self, data, ttl=None, permanent=False):
+    def touch_member(self, data, permanent=False):
         cluster = self.cluster
         if cluster and cluster.leader and cluster.leader.name == self._name:
-            role = 'master'
+            role = 'promoted' if data['role'] in ('replica', 'promoted') else 'master'
         elif data['state'] == 'running' and data['role'] != 'master':
             role = data['role']
         else:
@@ -348,28 +610,38 @@ class Kubernetes(AbstractDCS):
                         'annotations': {'status': json.dumps(data, separators=(',', ':'))}}
             body = k8s_client.V1Pod(metadata=k8s_client.V1ObjectMeta(**metadata))
             ret = self._api.patch_namespaced_pod(self._name, self._namespace, body)
+            if ret:
+                self._pods.set(self._name, ret)
+        if self._should_create_config_service:
+            self._create_config_service()
         return ret
 
     def initialize(self, create_new=True, sysid=""):
         cluster = self.cluster
         resource_version = cluster.config.index if cluster and cluster.config and cluster.config.index else None
-        return self.patch_or_create(self.config_path, {self._INITIALIZE: sysid}, resource_version)
+        return self.patch_or_create_config({self._INITIALIZE: sysid}, resource_version)
 
-    def delete_leader(self):
-        if self.cluster and isinstance(self.cluster.leader, Leader) and self.cluster.leader.name == self._name:
-            self.patch_or_create(self.leader_path, {self._LEADER: None}, self._leader_resource_version, True, False, [])
+    def _delete_leader(self):
+        """Unused"""
+
+    def delete_leader(self, last_operation=None):
+        kind = self._kinds.get(self.leader_path)
+        if kind and (kind.metadata.annotations or {}).get(self._LEADER) == self._name:
+            annotations = {self._LEADER: None}
+            if last_operation:
+                annotations[self._OPTIME] = last_operation
+            self.patch_or_create(self.leader_path, annotations, kind.metadata.resource_version, True, False, [])
             self.reset_cluster()
 
     def cancel_initialization(self):
-        self.patch_or_create(self.config_path, {self._INITIALIZE: None}, self.cluster.config.index, True)
+        self.patch_or_create_config({self._INITIALIZE: None}, self._config_resource_version, True)
 
     @catch_kubernetes_errors
     def delete_cluster(self):
         self.retry(self._api.delete_collection_namespaced_kind, self._namespace, label_selector=self._label_selector)
 
     def set_history_value(self, value):
-        patch = bool(self.cluster and self.cluster.config and self.cluster.config.index)
-        return self.patch_or_create(self.config_path, {self._HISTORY: value}, None, patch, False)
+        return self.patch_or_create_config({self._HISTORY: value}, None, bool(self._config_resource_version), False)
 
     def set_sync_state_value(self, value, index=None):
         """Unused"""
@@ -385,25 +657,7 @@ class Kubernetes(AbstractDCS):
             self.__do_not_watch = False
             return True
 
-        if leader_index:
-            end_time = time.time() + timeout
-            w = k8s_watch.Watch()
-            while timeout >= 1:
-                try:
-                    for event in w.stream(self._api.list_namespaced_kind, self._namespace,
-                                          resource_version=leader_index, timeout_seconds=int(timeout + 0.5),
-                                          field_selector='metadata.name=' + self.leader_path,
-                                          _request_timeout=(1, timeout + 1)):
-                        return event['raw_object'].get('metadata', {}).get('resourceVersion') != leader_index
-                    return False
-                except KeyboardInterrupt:
-                    raise
-                except Exception:
-                    logger.exception('watch')
-
-                timeout = end_time - time.time()
-
         try:
-            return super(Kubernetes, self).watch(None, timeout)
+            return super(Kubernetes, self).watch(None, timeout + 0.5)
         finally:
             self.event.clear()

patroni/dcs/zookeeper.py

@@ -1,5 +1,6 @@
 import json
 import logging
+import select
 import time
 
 from kazoo.client import KazooClient, KazooState, KazooRetry
@@ -37,12 +38,21 @@ class PatroniSequentialThreadingHandler(SequentialThreadingHandler):
                      `connect_timeout` (negotiated session timeout) as the second element."""
 
         args = list(args)
-        if len(args) == 1:
+        if len(args) == 0:  # kazoo 2.6.0 slightly changed the way how it calls create_connection method
+            kwargs['timeout'] = max(self._connect_timeout, kwargs.get('timeout', self._connect_timeout*10)/10.0)
+        elif len(args) == 1:
             args.append(self._connect_timeout)
         else:
             args[1] = max(self._connect_timeout, args[1]/10.0)
         return super(PatroniSequentialThreadingHandler, self).create_connection(*args, **kwargs)
 
+    def select(self, *args, **kwargs):
+        """Python3 raises `ValueError` if socket is closed, because fd == -1"""
+        try:
+            return super(PatroniSequentialThreadingHandler, self).select(*args, **kwargs)
+        except ValueError as e:
+            raise select.error(9, str(e))
+
 
 class ZooKeeper(AbstractDCS):
 
@@ -66,7 +76,7 @@ class ZooKeeper(AbstractDCS):
 
         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
@@ -78,7 +88,7 @@ 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):
@@ -116,6 +126,10 @@ class ZooKeeper(AbstractDCS):
             self._client.restart()
             return True
 
+    @property
+    def ttl(self):
+        return self._client._session_timeout
+
     def set_retry_timeout(self, retry_timeout):
         retry = self._client.retry if isinstance(self._client.retry, KazooRetry) else self._client._retry
         retry.deadline = retry_timeout
@@ -196,16 +210,21 @@ class ZooKeeper(AbstractDCS):
         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])
 
-        self._cluster = Cluster(initialize, config, leader, last_leader_operation, members, failover, sync, history)
+        return Cluster(initialize, config, leader, last_leader_operation, members, failover, sync, history)
 
     def _load_cluster(self):
-        if self._fetch_cluster or self._cluster is None:
+        cluster = self.cluster
+        if self._fetch_cluster or cluster is None:
             try:
-                self._client.retry(self._inner_load_cluster)
+                cluster = self._client.retry(self._inner_load_cluster)
             except Exception:
                 logger.exception('get_cluster')
                 self.cluster_watcher(None)
                 raise ZooKeeperError('ZooKeeper in not responding properly')
+        return cluster
+
+    def _bypass_caches(self):
+        self._fetch_cluster = True
 
     def _create(self, path, value, retry=False, ephemeral=False):
         try:
@@ -254,11 +273,14 @@ class ZooKeeper(AbstractDCS):
         return self._create(self.initialize_path, sysid, retry=True) if create_new \
             else self._client.retry(self._client.set, self.initialize_path, sysid)
 
-    def touch_member(self, data, ttl=None, permanent=False):
+    def touch_member(self, data, permanent=False):
         cluster = self.cluster
         member = cluster and cluster.get_member(self._name, fallback_to_leader=False)
         encoded_data = json.dumps(data, separators=(',', ':')).encode('utf-8')
-        if member and self._client.client_id is not None and member.session != self._client.client_id[0]:
+        if member and (self._client.client_id is not None and member.session != self._client.client_id[0] or
+                       not (deep_compare(member.data.get('tags', {}), data.get('tags', {})) and
+                            member.data.get('version') == data.get('version') and
+                            member.data.get('checkpoint_after_promote') == data.get('checkpoint_after_promote'))):
             try:
                 self._client.delete_async(self.member_path).get(timeout=1)
             except NoNodeError:
@@ -296,7 +318,7 @@ class ZooKeeper(AbstractDCS):
     def _update_leader(self):
         return True
 
-    def delete_leader(self):
+    def _delete_leader(self):
         self._client.restart()
         return True
 

patroni/exceptions.py

@@ -27,3 +27,7 @@ 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
@@ -12,33 +11,38 @@ 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.postgresql import ACTION_ON_START
-from patroni.utils import polling_loop, tzutc
+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, is_standby_cluster as _is_standby_cluster, parse_int
 from patroni.dcs import RemoteMember
 from threading import RLock
 
 logger = logging.getLogger(__name__)
 
 
-class _MemberStatus(namedtuple('_MemberStatus', 'member,reachable,in_recovery,wal_position,tags,watchdog_failed')):
+class _MemberStatus(namedtuple('_MemberStatus', ['member', 'reachable', 'in_recovery', 'timeline',
+                                                 'wal_position', 'tags', 'watchdog_failed'])):
     """Node status distilled from API response:
 
         member - dcs.Member object of the node
         reachable - `!False` if the node is not reachable or is not responding with correct JSON
         in_recovery - `!True` if pg_is_in_recovery() == true
-        wal_position - value of `replayed_location` or `location` from JSON, dependin on its role.
+        timeline - timeline value from JSON
+        wal_position - maximum value of `replayed_location` or `received_location` from JSON
         tags - dictionary with values of different tags (i.e. nofailover)
         watchdog_failed - indicates that watchdog is required by configuration but not available or failed
     """
     @classmethod
     def from_api_response(cls, member, json):
         is_master = json['role'] == 'master'
+        timeline = json.get('timeline', 0)
         wal = not is_master and max(json['xlog'].get('received_location', 0), json['xlog'].get('replayed_location', 0))
-        return cls(member, True, not is_master, wal, json.get('tags', {}), json.get('watchdog_failed', False))
+        return cls(member, True, not is_master, timeline, wal, json.get('tags', {}), json.get('watchdog_failed', False))
 
     @classmethod
     def unknown(cls, member):
-        return cls(member, False, None, 0, {}, False)
+        return cls(member, False, None, 0, 0, {}, False)
 
     def failover_limitation(self):
         """Returns reason why this node can't promote or None if everything is ok."""
@@ -56,18 +60,20 @@ class Ha(object):
     def __init__(self, patroni):
         self.patroni = patroni
         self.state_handler = patroni.postgresql
+        self._rewind = Rewind(self.state_handler)
         self.dcs = patroni.dcs
         self.cluster = None
         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._crash_recovery_executed = False
         self._start_timeout = None
-        self._async_executor = AsyncExecutor(self.state_handler, self.wakeup)
+        self._async_executor = AsyncExecutor(self.state_handler.cancellable, self.wakeup)
         self.watchdog = patroni.watchdog
 
         # Each member publishes various pieces of information to the DCS using touch_member. This lock protects
@@ -88,16 +94,38 @@ 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')
 
+    def check_timeline(self):
+        return self.check_mode('check_timeline')
+
+    def get_standby_cluster_config(self):
+        if self.cluster and self.cluster.config and self.cluster.config.modify_index:
+            config = self.cluster.config.data
+        else:
+            config = self.patroni.config.dynamic_configuration
+        return config.get('standby_cluster')
+
+    def is_standby_cluster(self):
+        return _is_standby_cluster(self.get_standby_cluster_config())
+
     def is_leader(self):
         with self._is_leader_lock:
-            return self._is_leader
+            return self._is_leader > time.time() and not self._leader_access_is_restricted
 
     def set_is_leader(self, value):
         with self._is_leader_lock:
-            self._is_leader = value
+            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()
@@ -107,12 +135,13 @@ class Ha(object):
             self.old_cluster = cluster
         self.cluster = cluster
 
-        if self.cluster.is_unlocked() or self.cluster.leader.name != self.state_handler.name:
+        if not self.has_lock(False):
             self.set_is_leader(False)
 
         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
@@ -124,15 +153,16 @@ class Ha(object):
                 last_operation = self.state_handler.last_operation()
             except Exception:
                 logger.exception('Exception when called state_handler.last_operation()')
-        ret = self.dcs.update_leader(last_operation)
+        ret = self.dcs.update_leader(last_operation, self._leader_access_is_restricted)
         self.set_is_leader(ret)
         if ret:
             self.watchdog.keepalive()
         return ret
 
-    def has_lock(self):
+    def has_lock(self, info=True):
         lock_owner = self.cluster.leader and self.cluster.leader.name
-        logger.info('Lock owner: %s; I am %s', lock_owner, self.state_handler.name)
+        if info:
+            logger.info('Lock owner: %s; I am %s', lock_owner, self.state_handler.name)
         return lock_owner == self.state_handler.name
 
     def get_effective_tags(self):
@@ -149,8 +179,15 @@ class Ha(object):
                 'conn_url': self.state_handler.connection_string,
                 'api_url': self.patroni.api.connection_string,
                 'state': self.state_handler.state,
-                'role': self.state_handler.role
+                'role': self.state_handler.role,
+                'version': self.patroni.version
             }
+
+            # following two lines are mainly necessary for consul, to avoid creation of master service
+            if data['role'] == 'master' and not self.is_leader():
+                data['role'] = 'promoted'
+            if self.is_leader() and not self._rewind.checkpoint_after_promote():
+                data['checkpoint_after_promote'] = False
             tags = self.get_effective_tags()
             if tags:
                 data['tags'] = tags
@@ -159,10 +196,22 @@ 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:
@@ -178,7 +227,11 @@ class Ha(object):
             return self.dcs.touch_member(data)
 
     def clone(self, clone_member=None, msg='(without leader)'):
-        if self.state_handler.clone(clone_member):
+        if self.is_standby_cluster() and not isinstance(clone_member, RemoteMember):
+            clone_member = self.get_remote_member(clone_member)
+
+        self._rewind.reset_state()
+        if self.state_handler.bootstrap.clone(clone_member):
             logger.info('bootstrapped %s', msg)
             cluster = self.dcs.get_cluster()
             node_to_follow = self._get_node_to_follow(cluster)
@@ -192,9 +245,8 @@ 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))
-            self._async_executor.run_async(self.clone, args=(clone_member, msg))
-            return 'trying to bootstrap {0}'.format(msg)
+            ret = self._async_executor.try_run_async('bootstrap {0}'.format(msg), self.clone, args=(clone_member, msg))
+            return ret or '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:
@@ -202,34 +254,29 @@ class Ha(object):
                 self.state_handler.bootstrapping = True
                 self._post_bootstrap_task = CriticalTask()
 
-                if self.patroni.config.is_standby_cluster:
-                    self._async_executor.schedule('bootstrap_standby_leader')
-                    self._async_executor.run_async(self.bootstrap_standby_leader)
-                    return 'trying to bootstrap a new standby leader'
+                if self.is_standby_cluster():
+                    ret = self._async_executor.try_run_async('bootstrap_standby_leader', self.bootstrap_standby_leader)
+                    return ret or 'trying to bootstrap a new standby leader'
                 else:
-                    self._async_executor.schedule('bootstrap')
-                    self._async_executor.run_async(
-                        self.state_handler.bootstrap,
-                        args=(self.patroni.config['bootstrap'],)
-                    )
-                    return 'trying to bootstrap a new cluster'
+                    ret = self._async_executor.try_run_async('bootstrap', self.state_handler.bootstrap.bootstrap,
+                                                             args=(self.patroni.config['bootstrap'],))
+                    return ret or 'trying to bootstrap a new cluster'
             else:
                 return 'failed to acquire initialize lock'
         else:
-            if self.state_handler.can_create_replica_without_replication_connection():
+            create_replica_methods = self.get_standby_cluster_config().get('create_replica_methods', []) \
+                                     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)
-                self._async_executor.run_async(self.clone)
-                return 'trying to ' + msg
-            return 'waiting for leader to bootstrap'
+                return self._async_executor.try_run_async(msg, self.clone) or 'trying to ' + msg
+            return 'waiting for {0}leader to bootstrap'.format('standby_' if self.is_standby_cluster() else '')
 
     def bootstrap_standby_leader(self):
         """ If we found 'standby' key in the configuration, we need to bootstrap
             not a real master, but a 'standby leader', that will take base backup
             from a remote master and start follow it.
         """
-        patroni_config = self.patroni.config.dynamic_configuration
-        clone_source = self.get_remote_master(patroni_config)
+        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)
@@ -238,11 +285,19 @@ class Ha(object):
 
         return result
 
-    def _handle_rewind(self):
-        if self.state_handler.rewind_needed_and_possible(self.cluster.leader):
-            self._async_executor.schedule('running pg_rewind from ' + self.cluster.leader.name)
-            self._async_executor.run_async(self.state_handler.rewind, (self.cluster.leader,))
-            return True
+    def _handle_rewind_or_reinitialize(self):
+        leader = self.get_remote_master() if self.is_standby_cluster() else self.cluster.leader
+        if not self._rewind.rewind_or_reinitialize_needed_and_possible(leader):
+            return None
+
+        if self._rewind.can_rewind:
+            msg = 'running pg_rewind from ' + leader.name
+            return self._async_executor.try_run_async(msg, self._rewind.execute, args=(leader,)) or msg
+
+        # remove_data_directory_on_diverged_timelines is set
+        if not self.is_standby_cluster():
+            msg = 'reinitializing due to diverged timelines'
+            return self._async_executor.try_run_async(msg, self._do_reinitialize, args=(self.cluster,)) or msg
 
     def recover(self):
         # Postgres is not running and we will restart in standby mode. Watchdog is not needed until we promote.
@@ -264,47 +319,66 @@ class Ha(object):
             timeout = None
 
         data = self.state_handler.controldata()
-        if data.get('Database cluster state') in ('in production', 'shutting down', 'in crash recovery') and \
-                not self._crash_recovery_executed and (self.cluster.is_unlocked() or self.state_handler.can_rewind):
+        logger.info('pg_controldata:\n%s\n', '\n'.join('  {0}: {1}'.format(k, v) for k, v in data.items()))
+        if data.get('Database cluster state') in ('in production', 'shutting down', 'in crash recovery') \
+                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._async_executor.run_async(self.state_handler.fix_cluster_state)
-            return self._async_executor.scheduled_action
+            msg = 'doing crash recovery in a single user mode'
+            return self._async_executor.try_run_async(msg, self._rewind.ensure_clean_shutdown) or msg
 
         self.load_cluster_from_dcs()
 
-        if self.has_lock():
+        role = 'replica'
+        if self.is_standby_cluster() or not self.has_lock():
+            if not self._rewind.executed:
+                self._rewind.trigger_check_diverged_lsn()
+            msg = self._handle_rewind_or_reinitialize()
+            if msg:
+                return msg
+
+            if self.has_lock():  # in standby cluster
+                msg = "starting as a standby leader because i had the session lock"
+                role = 'standby_leader'
+                node_to_follow = self._get_node_to_follow(self.cluster)
+            elif self.is_standby_cluster() and self.cluster.is_unlocked():
+                msg = "trying to follow a remote master because standby cluster is unhealthy"
+                node_to_follow = self.get_remote_master()
+            else:
+                msg = "starting as a secondary"
+                node_to_follow = self._get_node_to_follow(self.cluster)
+        elif self.has_lock():
             msg = "starting as readonly because i had the session lock"
             node_to_follow = None
-        else:
-            if not self.state_handler.rewind_executed:
-                self.state_handler.trigger_check_diverged_lsn()
-            if self._handle_rewind():
-                return self._async_executor.scheduled_action
-            msg = "starting as a secondary"
-            node_to_follow = self._get_node_to_follow(self.cluster)
 
-        self.recovering = True
-
-        self._async_executor.schedule('restarting after failure')
-        self._async_executor.run_async(self.state_handler.follow, (node_to_follow, timeout))
+        if self._async_executor.try_run_async('restarting after failure', self.state_handler.follow,
+                                              args=(node_to_follow, role, timeout)) is None:
+            self.recovering = True
         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.
-        is_leader = self.cluster.leader and self.state_handler.name == self.cluster.leader.name
-
-        if self.cluster.is_standby_cluster() and is_leader:
-            node_to_follow = self.get_remote_master(cluster.config.data)
+        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
 
-        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:
@@ -315,22 +389,45 @@ class Ha(object):
         node_to_follow = self._get_node_to_follow(self.cluster)
 
         if self.is_paused():
-            if not (self.state_handler.need_rewind and self.state_handler.can_rewind) or self.cluster.is_unlocked():
-                self.state_handler.set_role('master' if is_leader else 'replica')
+            if not (self._rewind.is_needed and self._rewind.can_rewind_or_reinitialize_allowed)\
+                    or self.cluster.is_unlocked():
                 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':
+                    self.state_handler.set_role('replica')
+
+                if not node_to_follow:
                     return 'no action'
         elif is_leader:
             self.demote('immediate-nolock')
             return demote_reason
 
-        if self._handle_rewind():
-            return self._async_executor.scheduled_action
+        if self.is_standby_cluster() and self._leader_timeline and \
+                self.state_handler.get_history(self._leader_timeline + 1):
+            self._rewind.trigger_check_diverged_lsn()
 
-        if not self.state_handler.check_recovery_conf(node_to_follow):
-            self._async_executor.schedule('changing primary_conninfo and restarting')
-            self._async_executor.run_async(self.state_handler.follow, (node_to_follow,))
+        msg = self._handle_rewind_or_reinitialize()
+        if msg:
+            return msg
+
+        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'
+        else:
+            change_required, restart_required = self.state_handler.config.check_recovery_conf(node_to_follow)
+            if change_required:
+                if restart_required:
+                    self._async_executor.try_run_async('changing primary_conninfo and restarting',
+                                                       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
 
@@ -365,14 +462,17 @@ class Ha(object):
                     logger.warning("No standbys available!")
 
                 logger.info("Assigning synchronous standby status to %s", picked)
-                self.state_handler.set_synchronous_standby(picked)
+                self.state_handler.config.set_synchronous_standby(picked)
 
                 if picked and 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)
                 if allow_promote:
-                    cluster = self.dcs.get_cluster()
+                    try:
+                        cluster = self.dcs.get_cluster()
+                    except DCSError:
+                        return logger.warning("Could not get cluster state from DCS during process_sync_replication()")
                     if cluster.sync.leader and cluster.sync.leader != self.state_handler.name:
                         logger.info("Synchronous replication key updated by someone else")
                         return
@@ -383,7 +483,7 @@ class Ha(object):
         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.set_synchronous_standby(None)
+            self.state_handler.config.set_synchronous_standby(None)
 
     def is_sync_standby(self, cluster):
         return cluster.leader and cluster.sync.leader == cluster.leader.name \
@@ -433,9 +533,10 @@ 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 \
@@ -445,9 +546,7 @@ class Ha(object):
                 self.dcs.set_history_value(json.dumps(history, separators=(',', ':')))
 
     def enforce_follow_remote_master(self, message):
-        self.state_handler.set_role('standby_leader')
         demote_reason = 'cannot be a real master in standby cluster'
-
         return self.follow(demote_reason, message)
 
     def enforce_master_role(self, message, promote_message):
@@ -477,22 +576,29 @@ 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.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 None)
             if self.state_handler.role != 'master':
-                self._async_executor.schedule('promote')
-                self._async_executor.run_async(self.state_handler.promote, args=(self.dcs.loop_wait,))
+                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.try_run_async('promote', self.state_handler.promote,
+                                                   args=(self.dcs.loop_wait, on_success,
+                                                         self._leader_access_is_restricted))
             return promote_message
 
-    @staticmethod
-    def fetch_node_status(member):
+    def fetch_node_status(self, 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)
-            logger.info('Got response from %s %s: %s', member.name, member.api_url, response.content)
-            return _MemberStatus.from_api_response(member, response.json())
+            response = self.patroni.request(member, timeout=2, retries=0)
+            data = response.data.decode('utf-8')
+            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)
@@ -511,15 +617,24 @@ class Ha(object):
         :returns True when node is lagging
         """
         lag = (self.cluster.last_leader_operation or 0) - wal_position
-        return lag > self.state_handler.config.get('maximum_lag_on_failover', 0)
+        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()
+        # We don't call `last_operation()` here because it returns a string
+        _, my_wal_position, _ = self.state_handler.timeline_wal_position()
         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
 
+        if not self.is_standby_cluster() and self.check_timeline():
+            cluster_timeline = self.cluster.timeline
+            my_timeline = self.state_handler.replica_cached_timeline(cluster_timeline)
+            if my_timeline < cluster_timeline:
+                logger.info('My timeline %s is behind last known cluster timeline %s', my_timeline, cluster_timeline)
+                return False
+
         # Prepare list of nodes to run check against
         members = [m for m in members if m.name != self.state_handler.name and not m.nofailover and m.api_url]
 
@@ -530,11 +645,17 @@ class Ha(object):
                         logger.warning('Master (%s) is still alive', st.member.name)
                         return False
                     if my_wal_position < st.wal_position:
-                        return False
+                        logger.info('Wal position of %s is ahead of my wal position', st.member.name)
+                        # In synchronous mode the former leader might be still accessible and even be ahead of us.
+                        # We should not disqualify himself from the leader race in such a situation.
+                        if not self.is_synchronous_mode() or st.member.name != self.cluster.sync.leader:
+                            return False
+                        logger.info('Ignoring the former leader being ahead of us')
         return True
 
     def is_failover_possible(self, members):
         ret = False
+        cluster_timeline = self.cluster.timeline
         members = [m for m in members if m.name != self.state_handler.name and not m.nofailover and m.api_url]
         if members:
             for st in self.fetch_nodes_statuses(members):
@@ -543,6 +664,9 @@ class Ha(object):
                     logger.info('Member %s is %s', st.member.name, not_allowed_reason)
                 elif self.is_lagging(st.wal_position):
                     logger.info('Member %s exceeds maximum replication lag', st.member.name)
+                elif self.check_timeline() and (not st.timeline or st.timeline < cluster_timeline):
+                    logger.info('Timeline %s of member %s is behind the cluster timeline %s',
+                                st.timeline, st.member.name, cluster_timeline)
                 else:
                     ret = True
         else:
@@ -634,10 +758,14 @@ class Ha(object):
 
         return self._is_healthiest_node(members.values())
 
-    def release_leader_key_voluntarily(self):
-        self.dcs.delete_leader()
-        self.touch_member()
+    def _delete_leader(self, last_operation=None):
+        self.set_is_leader(False)
+        self.dcs.delete_leader(last_operation)
         self.dcs.reset_cluster()
+
+    def release_leader_key_voluntarily(self, last_operation=None):
+        self._delete_leader(last_operation)
+        self.touch_member()
         logger.info("Leader key released")
 
     def demote(self, mode):
@@ -658,14 +786,17 @@ class Ha(object):
             'immediate-nolock': dict(stop='immediate', checkpoint=False, release=False, offline=False, async_req=True),
         }[mode]
 
-        self.state_handler.trigger_check_diverged_lsn()
+        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']:
-            self.release_leader_key_voluntarily()
+            checkpoint_location = self.state_handler.latest_checkpoint_location() if mode == 'graceful' else None
+            with self._async_executor:
+                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
@@ -677,12 +808,11 @@ 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.run_async(self.state_handler.follow, (node_to_follow,))
+            self._async_executor.try_run_async('starting after demotion', self.state_handler.follow, (node_to_follow,))
         else:
             if self.is_synchronous_mode():
-                self.state_handler.set_synchronous_standby(None)
-            if self.state_handler.rewind_needed_and_possible(leader):
+                self.state_handler.config.set_synchronous_standby(None)
+            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)
 
@@ -751,9 +881,8 @@ 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')
-                        self._async_executor.run_async(self.demote, ('graceful',))
-                        return 'manual failover: demoting myself'
+                        ret = self._async_executor.try_run_async('manual failover: demote', self.demote, ('graceful',))
+                        return ret or 'manual failover: demoting myself'
                     else:
                         logger.warning('manual failover: no healthy members found, failover is not possible')
             else:
@@ -780,11 +909,11 @@ class Ha(object):
                         self.dcs.manual_failover('', '')
                 self.load_cluster_from_dcs()
 
-                if self.cluster.is_standby_cluster():
+                if self.is_standby_cluster():
                     # standby leader disappeared, and this is a healthiest
                     # replica, so it should become a new standby leader.
                     # This imply that we need to start following a remote master
-                    msg = 'promoted self to a standby leader because i had the session lock'
+                    msg = 'promoted self to a standby leader by acquiring session lock'
                     return self.enforce_follow_remote_master(msg)
                 else:
                     return self.enforce_master_role(
@@ -798,7 +927,7 @@ class Ha(object):
             # when we are doing manual failover there is no guaranty that new leader is ahead of any other node
             # node tagged as nofailover can be ahead of the new leader either, but it is always excluded from elections
             if bool(self.cluster.failover) or self.patroni.nofailover:
-                self.state_handler.trigger_check_diverged_lsn()
+                self._rewind.trigger_check_diverged_lsn()
                 time.sleep(2)  # Give a time to somebody to take the leader lock
 
             if self.patroni.nofailover:
@@ -813,20 +942,29 @@ class Ha(object):
                 if self.cluster.failover and self.cluster.failover.candidate == self.state_handler.name:
                     return 'waiting to become master after promote...'
 
-                self.dcs.delete_leader()
-                self.dcs.reset_cluster()
+                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
 
-                if self.cluster.is_standby_cluster():
+                # check if the node is ready to be used by pg_rewind
+                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 the standby leader with the lock' \
+                          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(
@@ -845,6 +983,9 @@ class Ha(object):
                     return 'not promoting because failed to update leader lock in DCS'
         else:
             logger.info('does not have lock')
+        if self.is_standby_cluster():
+            return self.follow('cannot be a real master in standby cluster',
+                               'no action.  i am a secondary and i am following a standby 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 and i am following a leader', refresh=False)
 
@@ -881,8 +1022,7 @@ class Ha(object):
         if role and role != self.state_handler.role:
             reason_to_cancel = "host role mismatch"
 
-        if (postgres_version and
-           self.state_handler.postgres_version_to_int(postgres_version) <= int(self.state_handler.server_version)):
+        if postgres_version and postgres_version_to_int(postgres_version) <= int(self.state_handler.server_version):
             reason_to_cancel = "postgres version mismatch"
 
         if pending_restart and not self.state_handler.pending_restart:
@@ -961,8 +1101,9 @@ class Ha(object):
                 return (False, 'restart failed')
 
     def _do_reinitialize(self, cluster):
-        self.state_handler.stop('immediate')
-        self.state_handler.remove_data_directory()
+        self.state_handler.stop('immediate', stop_timeout=self.patroni.config['retry_timeout'])
+        # Commented redundant data directory cleanup here
+        # self.state_handler.remove_data_directory()
 
         clone_member = self.cluster.get_clone_member(self.state_handler.name)
         member_role = 'leader' if clone_member == self.cluster.leader else 'replica'
@@ -975,7 +1116,7 @@ class Ha(object):
             if self.cluster.is_unlocked():
                 return 'Cluster has no leader, can not reinitialize'
 
-            if self.cluster.leader.name == self.state_handler.name:
+            if self.has_lock(False):
                 return 'I am the leader, can not reinitialize'
 
         if force:
@@ -1020,18 +1161,20 @@ class Ha(object):
         if not self.state_handler.is_running():
             self.watchdog.disable()
             if self.has_lock():
-                self.state_handler.set_role('demoted')
-                self.dcs.delete_leader()
-                self.dcs.reset_cluster()
+                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
+        if self._rewind.executed and not self._rewind.failed:
+            self._rewind.reset_state()
         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')
 
@@ -1045,16 +1188,17 @@ class Ha(object):
                 return 'waiting for end of recovery after bootstrap'
 
             self.state_handler.set_role('master')
-            self._async_executor.schedule('post_bootstrap')
-            self._async_executor.run_async(self.state_handler.post_bootstrap,
-                                           args=(self.patroni.config['bootstrap'], self._post_bootstrap_task))
-            return 'running post_bootstrap'
+            ret = self._async_executor.try_run_async('post_bootstrap', self.state_handler.bootstrap.post_bootstrap,
+                                                     args=(self.patroni.config['bootstrap'], self._post_bootstrap_task))
+            return ret or '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.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.state_handler.slots_handler.sync_replication_slots(self.cluster)
         self.dcs.take_leader()
         self.set_is_leader(True)
         self.state_handler.call_nowait(ACTION_ON_START)
@@ -1143,7 +1287,7 @@ class Ha(object):
             if self.state_handler.bootstrapping:
                 return self.post_bootstrap()
 
-            if self.recovering and not self.state_handler.need_rewind:
+            if self.recovering and not self._rewind.is_needed:
                 self.recovering = False
                 # Check if we tried to recover and failed
                 msg = self.post_recover()
@@ -1153,7 +1297,7 @@ class Ha(object):
             # 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()
 
@@ -1163,40 +1307,41 @@ class Ha(object):
                     return 'released leader key voluntarily as data dir empty and currently leader'
 
                 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:
-                    logger.fatal("system ID mismatch, node %s belongs to a different cluster: %s != %s",
-                                 self.state_handler.name, self.cluster.initialize, self.state_handler.sysid)
-                    sys.exit(1)
+                if self.sysid_valid(self.cluster.initialize):
+                    if self.cluster.initialize != data_sysid:
+                        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():
                     if self.has_lock():
-                        self.dcs.delete_leader()
-                        self.dcs.reset_cluster()
+                        self._delete_leader()
                         return 'removed leader lock because postgres is not running'
                     # Normally we don't start Postgres in a paused state. We make an exception for the demoted primary
                     # that needs to be started after it had been stopped by demote. When there is no need to call rewind
                     # the demote code follows through to starting Postgres right away, however, in the rewind case
                     # it returns from demote and reaches this point to start PostgreSQL again after rewind. In that
                     # case it makes no sense to continue to recover() unless rewind has finished successfully.
-                    elif (self.state_handler.rewind_failed or
-                          not (self.state_handler.need_rewind and self.state_handler.can_rewind)):
+                    elif self._rewind.failed or not self._rewind.executed and not \
+                            (self._rewind.is_needed and self._rewind.can_rewind_or_reinitialize_allowed):
                         return 'postgres is not running'
 
                 # try to start dead postgres
@@ -1213,11 +1358,11 @@ class Ha(object):
                 # 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)
                     if not self.state_handler.cb_called:
                         if not self.state_handler.is_leader():
-                            self.state_handler.trigger_check_diverged_lsn()
+                            self._rewind.trigger_check_diverged_lsn()
                         self.state_handler.call_nowait(ACTION_ON_START)
-                    self.state_handler.sync_replication_slots(self.cluster)
         except DCSError:
             dcs_failed = True
             logger.error('Error communicating with DCS')
@@ -1233,8 +1378,12 @@ class Ha(object):
 
     def run_cycle(self):
         with self._async_executor:
-            info = self._run_cycle()
-            return (self.is_paused() and 'PAUSE: ' or '') + info
+            try:
+                info = self._run_cycle()
+                return (self.is_paused() and 'PAUSE: ' or '') + info
+            except Exception:
+                logger.exception('Unexpected exception')
+                return 'Unexpected exception raised, please report it as a BUG'
 
     def shutdown(self):
         if self.is_paused():
@@ -1246,10 +1395,12 @@ class Ha(object):
             # 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():
-                    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
@@ -1259,13 +1410,11 @@ class Ha(object):
                              (" Leaving watchdog running." if self.watchdog.is_running else ""))
 
     def watch(self, timeout):
-        cluster = self.cluster
         # watch on leader key changes if the postgres is running and leader is known and current node is not lock owner
-        if not self._async_executor.busy and cluster and cluster.leader \
-                and cluster.leader.name != self.state_handler.name:
-            leader_index = cluster.leader.index
-        else:
+        if self._async_executor.busy or self.cluster.is_unlocked() or self.has_lock(False):
             leader_index = None
+        else:
+            leader_index = self.cluster.leader.index
 
         return self.dcs.watch(leader_index, timeout)
 
@@ -1275,26 +1424,24 @@ class Ha(object):
         This usually happens on the master or if the node is running async action"""
         self.dcs.event.set()
 
-    def get_remote_master(self, config):
+    def get_remote_member(self, member=None):
         """ In case of standby cluster this will tel us from which remote
             master to stream. Config can be both patroni config or
             cluster.config.data
         """
-        config = config or (self.config is not None and self.config.data)
+        cluster_params = self.get_standby_cluster_config()
 
-        if config and config.get('standby_cluster'):
-            cluster_params = config.get('standby_cluster')
-            unique_name = 'remote_master:{}'.format(uuid.uuid1())
-            data = {
-                'conn_kwargs': {
-                    "host": cluster_params.get('host'),
-                    "port": cluster_params.get('port'),
-                },
-                'no_replication_slot': 'primary_slot_name' not in cluster_params,
-            }
-            data.update({
-                k: v for k, v in cluster_params.items()
-                if k in RemoteMember.allowed_keys()
-            })
+        if cluster_params:
+            name = member.name if member else 'remote_master:{}'.format(uuid.uuid1())
 
-            return RemoteMember(unique_name, data)
+            data = {k: v for k, v in cluster_params.items() if k in RemoteMember.allowed_keys()}
+            data['no_replication_slot'] = 'primary_slot_name' not in cluster_params
+            conn_kwargs = member.conn_kwargs() if member else \
+                {k: cluster_params[k] for k in ('host', 'port') if k in cluster_params}
+            if conn_kwargs:
+                data['conn_kwargs'] = conn_kwargs
+
+            return RemoteMember(name, data)
+
+    def get_remote_master(self):
+        return self.get_remote_member()

patroni/log.py

@@ -0,0 +1,194 @@
+import logging
+import os
+import sys
+
+from copy import deepcopy
+from logging.handlers import RotatingFileHandler
+from patroni.utils import deep_compare
+from six.moves.queue import Queue, Full
+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):
+        logging.Handler.__init__(self)
+        self.queue = Queue()
+        self._records_lost = 0
+
+    def _put_record(self, record):
+        self.format(record)
+        record.msg = record.message
+        record.args = None
+        record.exc_info = None
+        self.queue.put_nowait(record)
+
+    def _try_to_report_lost_records(self):
+        if self._records_lost:
+            try:
+                record = _LOGGER.makeRecord(_LOGGER.name, logging.WARNING, __file__, 0,
+                                            'QueueHandler has lost %s log records',
+                                            (self._records_lost,), None, 'emit')
+                self._put_record(record)
+                self._records_lost = 0
+            except Exception:
+                pass
+
+    def emit(self, record):
+        try:
+            self._put_record(record)
+            self._try_to_report_lost_records()
+        except Exception:
+            self._records_lost += 1
+
+    @property
+    def records_lost(self):
+        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
+    DEFAULT_MAX_QUEUE_SIZE = 1000
+    LOGGING_BROKEN_EXIT_CODE = 5
+
+    def __init__(self):
+        super(PatroniLogger, self).__init__()
+        self._queue_handler = QueueHandler()
+        self._root_logger = logging.getLogger()
+        self._config = None
+        self.log_handler = None
+        self.log_handler_lock = Lock()
+        self._old_handlers = []
+        self.reload_config({'level': 'DEBUG'})
+        # 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 {})
+        for name, logger in self._root_logger.manager.loggerDict.items():
+            if not isinstance(logger, logging.PlaceHolder):
+                level = loggers.pop(name, logging.NOTSET)
+                logger.setLevel(level)
+
+        for name, level in loggers.items():
+            logger = self._root_logger.manager.getLogger(name)
+            logger.setLevel(level)
+
+    def reload_config(self, config):
+        if self._config is None or not deep_compare(self._config, config):
+            with self._queue_handler.queue.mutex:
+                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):
+                    new_handler = RotatingFileHandler(os.path.join(config['dir'], __name__))
+                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):
+                    new_handler = logging.StreamHandler()
+                handler = new_handler or self.log_handler
+
+            oldlogformat = (self._config or {}).get('format', PatroniLogger.DEFAULT_FORMAT)
+            logformat = config.get('format', PatroniLogger.DEFAULT_FORMAT)
+
+            olddateformat = (self._config or {}).get('dateformat') or None
+            dateformat = config.get('dateformat') or None  # Convert empty string to `None`
+
+            if oldlogformat != logformat or olddateformat != dateformat or new_handler:
+                handler.setFormatter(logging.Formatter(logformat, dateformat))
+
+            if new_handler:
+                with self.log_handler_lock:
+                    if self.log_handler:
+                        self._old_handlers.append(self.log_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:
+                if not self._old_handlers:
+                    break
+                handler = self._old_handlers.pop()
+            try:
+                handler.close()
+            except Exception:
+                _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)
+
+        while True:
+            self._close_old_handlers()
+
+            record = self._queue_handler.queue.get(True)
+            if record is None:
+                break
+
+            self.log_handler.handle(record)
+            self._queue_handler.queue.task_done()
+
+    def shutdown(self):
+        try:
+            self._queue_handler.queue.put_nowait(None)
+        except Full:  # Queue is full.
+            # It seems that logging is not working, exiting with non-standard exit-code is the best we can do.
+            sys.exit(self.LOGGING_BROKEN_EXIT_CODE)
+        self.join()
+        logging.shutdown()
+
+    @property
+    def queue_size(self):
+        return self._queue_handler.queue.qsize()
+
+    @property
+    def records_lost(self):
+        return self._queue_handler.records_lost

patroni/postgresql/__init__.py

@@ -0,0 +1,937 @@
+import logging
+import os
+import psycopg2
+import shlex
+import shutil
+import subprocess
+import time
+
+from contextlib import contextmanager
+from copy import deepcopy
+from dateutil import tz
+from datetime import datetime
+from patroni.postgresql.callback_executor import CallbackExecutor
+from patroni.postgresql.bootstrap import Bootstrap
+from patroni.postgresql.cancellable import CancellableSubprocess
+from patroni.postgresql.config import ConfigHandler, mtime
+from patroni.postgresql.connection import Connection, get_connection_cursor
+from patroni.postgresql.misc import parse_history, parse_lsn, 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, data_directory_is_empty, parse_int
+from threading import current_thread, Lock
+from psutil import TimeoutExpired
+
+
+logger = logging.getLogger(__name__)
+
+ACTION_ON_START = "on_start"
+ACTION_ON_STOP = "on_stop"
+ACTION_ON_RESTART = "on_restart"
+ACTION_ON_RELOAD = "on_reload"
+ACTION_ON_ROLE_CHANGE = "on_role_change"
+ACTION_NOOP = "noop"
+
+STATE_RUNNING = 'running'
+STATE_REJECT = 'rejecting connections'
+STATE_NO_RESPONSE = 'not responding'
+STATE_UNKNOWN = 'unknown'
+
+STOP_POLLING_INTERVAL = 1
+
+
+@contextmanager
+def null_context():
+    yield
+
+
+class Postgresql(object):
+
+    POSTMASTER_START_TIME = "pg_catalog.to_char(pg_catalog.pg_postmaster_start_time(), 'YYYY-MM-DD HH24:MI:SS.MS TZ')"
+    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()
+        self.set_state('stopped')
+
+        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)
+        self.bootstrapping = False
+        self.__thread_ident = current_thread().ident
+
+        self.slots_handler = SlotsHandler(self)
+
+        self._callback_executor = CallbackExecutor()
+        self.__cb_called = False
+        self.__cb_pending = None
+
+        self.cancellable = CancellableSubprocess()
+
+        self._sysid = None
+        self.retry = Retry(max_tries=-1, deadline=config['retry_timeout']/2.0, max_delay=1,
+                           retry_exceptions=PostgresConnectionException)
+
+        # Retry 'pg_is_in_recovery()' only once
+        self._is_leader_retry = Retry(max_tries=1, deadline=config['retry_timeout']/2.0, max_delay=1,
+                                      retry_exceptions=PostgresConnectionException)
+
+        self._role_lock = Lock()
+        self.set_role(self.get_postgres_role_from_data_directory())
+        self._state_entry_timestamp = None
+
+        self._cluster_info_state = {}
+        self._cached_replica_timeline = None
+
+        # Last known running process
+        self._postmaster_proc = None
+
+        if self.is_running():
+            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
+            hba_saved = self.config.replace_pg_hba()
+            ident_saved = self.config.replace_pg_ident()
+            if hba_saved or ident_saved:
+                self.reload()
+        elif self.role == 'master':
+            self.set_role('demoted')
+
+    @property
+    def create_replica_methods(self):
+        return self.config.get('create_replica_methods', []) or self.config.get('create_replica_method', [])
+
+    @property
+    def major_version(self):
+        return self._major_version
+
+    @property
+    def database(self):
+        return self._database
+
+    @property
+    def data_dir(self):
+        return self._data_dir
+
+    @property
+    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'
+
+    @property
+    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 = (", CASE WHEN latest_end_lsn IS NULL THEN NULL ELSE received_tli END,"
+                     " slot_name, conninfo FROM pg_catalog.pg_stat_get_wal_receiver()")
+            if self.role == 'standby_leader':
+                extra = "timeline_id" + extra + ", pg_catalog.pg_control_checkpoint()"
+            else:
+                extra = "0" + extra
+        else:
+            extra = "0, 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)
+
+    def get_major_version(self):
+        if self._version_file_exists():
+            try:
+                with open(self._version_file) as f:
+                    return postgres_major_version_to_int(f.read().strip())
+            except Exception:
+                logger.exception('Failed to read PG_VERSION from %s', self._data_dir)
+        return 0
+
+    def pgcommand(self, cmd):
+        """Returns path to the specified PostgreSQL command"""
+        return os.path.join(self._bin_dir, cmd)
+
+    def pg_ctl(self, cmd, *args, **kwargs):
+        """Builds and executes pg_ctl command
+
+        :returns: `!True` when return_code == 0, otherwise `!False`"""
+
+        pg_ctl = [self.pgcommand('pg_ctl'), cmd]
+        return subprocess.call(pg_ctl + ['-D', self._data_dir] + list(args), **kwargs) == 0
+
+    def pg_isready(self):
+        """Runs pg_isready to see if PostgreSQL is accepting connections.
+
+        :returns: 'ok' if PostgreSQL is up, 'reject' if starting up, 'no_resopnse' if not up."""
+
+        r = self.config.local_connect_kwargs
+        cmd = [self.pgcommand('pg_isready'), '-p', r['port'], '-d', self._database]
+
+        # Host is not set if we are connecting via default unix socket
+        if 'host' in r:
+            cmd.extend(['-h', r['host']])
+
+        # We only need the username because pg_isready does not try to authenticate
+        if 'user' in r:
+            cmd.extend(['-U', r['user']])
+
+        ret = subprocess.call(cmd)
+        return_codes = {0: STATE_RUNNING,
+                        1: STATE_REJECT,
+                        2: STATE_NO_RESPONSE,
+                        3: STATE_UNKNOWN}
+        return return_codes.get(ret, STATE_UNKNOWN)
+
+    def reload_config(self, config, sighup=False):
+        self.config.reload_config(config, sighup)
+        self._is_leader_retry.deadline = self.retry.deadline = config['retry_timeout']/2.0
+
+    @property
+    def pending_restart(self):
+        return self._pending_restart
+
+    def set_pending_restart(self, value):
+        self._pending_restart = value
+
+    @property
+    def sysid(self):
+        if not self._sysid and not self.bootstrapping:
+            data = self.controldata()
+            self._sysid = data.get('Database system identifier', "")
+        return self._sysid
+
+    def get_postgres_role_from_data_directory(self):
+        if self.data_directory_empty() or not self.controldata():
+            return 'uninitialized'
+        elif self.config.recovery_conf_exists():
+            return 'replica'
+        else:
+            return 'master'
+
+    @property
+    def server_version(self):
+        return self._connection.server_version
+
+    def connection(self):
+        return self._connection.get()
+
+    def set_connection_kwargs(self, kwargs):
+        self._connection.set_conn_kwargs(kwargs)
+
+    def _query(self, sql, *params):
+        """We are always using the same cursor, therefore this method is not thread-safe!!!
+        You can call it from different threads only if you are holding explicit `AsyncExecutor` lock,
+        because the main thread is always holding this lock when running HA cycle."""
+        cursor = None
+        try:
+            cursor = self._connection.cursor()
+            cursor.execute(sql, params)
+            return cursor
+        except psycopg2.Error as e:
+            if cursor and cursor.connection.closed == 0:
+                # When connected via unix socket, psycopg2 can't recoginze 'connection lost'
+                # and leaves `_cursor_holder.connection.closed == 0`, but psycopg2.OperationalError
+                # is still raised (what is correct). It doesn't make sense to continiue with existing
+                # connection and we will close it, to avoid its reuse by the `cursor` method.
+                if isinstance(e, psycopg2.OperationalError):
+                    self._connection.close()
+                else:
+                    raise e
+            if self.state == 'restarting':
+                raise RetryFailedError('cluster is being restarted')
+            raise PostgresConnectionException('connection problems')
+
+    def query(self, sql, *args, **kwargs):
+        if not kwargs.get('retry', True):
+            return self._query(sql, *args)
+        try:
+            return self.retry(self._query, sql, *args)
+        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):
+        if self.pg_control_exists():
+            return False
+        return data_directory_is_empty(self._data_dir)
+
+    def replica_method_options(self, method):
+        return deepcopy(self.config.get(method, {}))
+
+    def replica_method_can_work_without_replication_connection(self, method):
+        return method != 'basebackup' and self.replica_method_options(method).get('no_master')
+
+    def can_create_replica_without_replication_connection(self, replica_methods=None):
+        """ go through the replication methods to see if there are ones
+            that does not require a working replication connection.
+        """
+        if replica_methods is None:
+            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):
+        self._cluster_info_state = {}
+
+    def _cluster_info_state_get(self, name):
+        if not self._cluster_info_state:
+            try:
+                result = self._is_leader_retry(self._query, self.cluster_info_query).fetchone()
+                self._cluster_info_state = dict(zip(['timeline', 'wal_position', 'replayed_location',
+                                                     'received_location', 'replay_paused', 'pg_control_timeline',
+                                                     'received_tli', 'slot_name', 'conninfo'], result))
+            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:
+                    self.set_state('starting')
+
+        if 'error' in self._cluster_info_state:
+            raise PostgresConnectionException(self._cluster_info_state['error'])
+
+        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 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 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 latest_checkpoint_location(self):
+        """Returns checkpoint location for the cleanly shut down primary"""
+
+        data = self.controldata()
+        lsn = data.get('Latest checkpoint location')
+        if data.get('Database cluster state') == 'shut down' and lsn:
+            try:
+                return str(parse_lsn(lsn))
+            except (IndexError, ValueError) as e:
+                logger.error('Exception when parsing lsn %s: %r', lsn, e)
+
+    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."""
+        if self._postmaster_proc:
+            if self._postmaster_proc.is_running():
+                return self._postmaster_proc
+            self._postmaster_proc = None
+
+        # we noticed that postgres was restarted, force syncing of replication
+        self.slots_handler.schedule()
+
+        self._postmaster_proc = PostmasterProcess.from_pidfile(self._data_dir)
+        return self._postmaster_proc
+
+    @property
+    def cb_called(self):
+        return self.__cb_called
+
+    def call_nowait(self, cb_name):
+        """ pick a callback command and call it without waiting for it to finish """
+        if self.bootstrapping:
+            return
+        if cb_name in (ACTION_ON_START, ACTION_ON_STOP, ACTION_ON_RESTART, ACTION_ON_ROLE_CHANGE):
+            self.__cb_called = True
+
+        if self.callback and cb_name in self.callback:
+            cmd = self.callback[cb_name]
+            try:
+                cmd = shlex.split(self.callback[cb_name]) + [cb_name, self.role, self.scope]
+                self._callback_executor.call(cmd)
+            except Exception:
+                logger.exception('callback %s %s %s %s failed', cmd, cb_name, self.role, self.scope)
+
+    @property
+    def role(self):
+        with self._role_lock:
+            return self._role
+
+    def set_role(self, value):
+        with self._role_lock:
+            self._role = value
+
+    @property
+    def state(self):
+        with self._state_lock:
+            return self._state
+
+    def set_state(self, value):
+        with self._state_lock:
+            self._state = value
+            self._state_entry_timestamp = time.time()
+
+    def time_in_state(self):
+        return time.time() - self._state_entry_timestamp
+
+    def is_starting(self):
+        return self.state == 'starting'
+
+    def wait_for_port_open(self, postmaster, timeout):
+        """Waits until PostgreSQL opens ports."""
+        for _ in polling_loop(timeout):
+            if self.cancellable.is_cancelled:
+                return False
+
+            if not postmaster.is_running():
+                logger.error('postmaster is not running')
+                self.set_state('start failed')
+                return False
+
+            isready = self.pg_isready()
+            if isready != STATE_NO_RESPONSE:
+                if isready not in [STATE_REJECT, STATE_RUNNING]:
+                    logger.warning("Can't determine PostgreSQL startup status, assuming running")
+                return True
+
+        logger.warning("Timed out waiting for PostgreSQL to start")
+        return False
+
+    def start(self, timeout=None, task=None, block_callbacks=False, role=None):
+        """Start PostgreSQL
+
+        Waits for postmaster to open ports or terminate so pg_isready can be used to check startup completion
+        or failure.
+
+        :returns: True if start was initiated and postmaster ports are open, False if start failed"""
+        # make sure we close all connections established against
+        # the former node, otherwise, we might get a stalled one
+        # after kill -9, which would report incorrect data to
+        # patroni.
+        self._connection.close()
+
+        if self.is_running():
+            logger.error('Cannot start PostgreSQL because one is already running.')
+            self.set_state('starting')
+            return True
+
+        if not block_callbacks:
+            self.__cb_pending = ACTION_ON_START
+
+        self.set_role(role or self.get_postgres_role_from_data_directory())
+
+        self.set_state('starting')
+        self._pending_restart = False
+
+        try:
+            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 self.cancellable.is_cancelled:
+            return False
+
+        with task or null_context():
+            if task and task.is_cancelled:
+                logger.info("PostgreSQL start cancelled.")
+                return False
+
+            self._postmaster_proc = PostmasterProcess.start(self.pgcommand('postgres'),
+                                                            self._data_dir,
+                                                            self.config.postgresql_conf,
+                                                            options)
+
+            if task:
+                task.complete(self._postmaster_proc)
+
+        start_timeout = timeout
+        if not start_timeout:
+            try:
+                start_timeout = float(self.config.get('pg_ctl_timeout', 60))
+            except ValueError:
+                start_timeout = 60
+
+        # We want postmaster to open ports before we continue
+        if not self._postmaster_proc or not self.wait_for_port_open(self._postmaster_proc, start_timeout):
+            return False
+
+        ret = self.wait_for_startup(start_timeout)
+        if ret is not None:
+            return ret
+        elif timeout is not None:
+            return False
+        else:
+            return 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")
+                if check_not_is_in_recovery:
+                    cur.execute('SELECT pg_catalog.pg_is_in_recovery()')
+                    if cur.fetchone()[0]:
+                        return 'is_in_recovery=true'
+                return cur.execute('CHECKPOINT')
+        except psycopg2.Error:
+            logger.exception('Exception during CHECKPOINT')
+            return 'not accessible or not healty'
+
+    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
+        commit to users. Currently this means we wait for user backends to close. But in the future alternate mechanisms
+        could be added.
+
+        :param on_safepoint: This callback is called when no user backends are running.
+        """
+        if checkpoint is None:
+            checkpoint = False if mode == 'immediate' else True
+
+        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
+            if not block_callbacks:
+                self.set_state('stopped')
+                if pg_signaled:
+                    self.call_nowait(ACTION_ON_STOP)
+        else:
+            logger.warning('pg_ctl stop failed')
+            self.set_state('stop failed')
+        return success
+
+    def _do_stop(self, mode, block_callbacks, checkpoint, on_safepoint, stop_timeout):
+        postmaster = self.is_running()
+        if not postmaster:
+            if on_safepoint:
+                on_safepoint()
+            return True, False
+
+        if checkpoint and not self.is_starting():
+            self.checkpoint(timeout=stop_timeout)
+
+        if not block_callbacks:
+            self.set_state('stopping')
+
+        # Send signal to postmaster to stop
+        success = postmaster.signal_stop(mode, self.pgcommand('pg_ctl'))
+        if success is not None:
+            if success and on_safepoint:
+                on_safepoint()
+            return success, True
+
+        # We can skip safepoint detection if we don't have a callback
+        if on_safepoint:
+            # Wait for our connection to terminate so we can be sure that no new connections are being initiated
+            self._wait_for_connection_close(postmaster)
+            postmaster.wait_for_user_backends_to_close()
+            on_safepoint()
+
+        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
+
+    def terminate_postmaster(self, postmaster, mode, stop_timeout):
+        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', self.pgcommand('pg_ctl'))
+        postmaster.wait()
+
+    def _wait_for_connection_close(self, postmaster):
+        try:
+            with self.connection().cursor() as cur:
+                while postmaster.is_running():  # Need a timeout here?
+                    cur.execute("SELECT 1")
+                    time.sleep(STOP_POLLING_INTERVAL)
+        except psycopg2.Error:
+            pass
+
+    def reload(self, block_callbacks=False):
+        ret = self.pg_ctl('reload')
+        if ret and not block_callbacks:
+            self.call_nowait(ACTION_ON_RELOAD)
+        return ret
+
+    def check_for_startup(self):
+        """Checks PostgreSQL status and returns if PostgreSQL is in the middle of startup."""
+        return self.is_starting() and not self.check_startup_state_changed()
+
+    def check_startup_state_changed(self):
+        """Checks if PostgreSQL has completed starting up or failed or still starting.
+
+        Should only be called when state == 'starting'
+
+        :returns: True if state was changed from 'starting'
+        """
+        ready = self.pg_isready()
+
+        if ready == STATE_REJECT:
+            return False
+        elif ready == STATE_NO_RESPONSE:
+            ret = not self.is_running()
+            if ret:
+                self.set_state('start failed')
+                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.
+                # Let the main loop of run cycle clean up the mess.
+                logger.warning("%s status returned from pg_isready",
+                               "Unknown" if ready == STATE_UNKNOWN else "Invalid")
+            self.set_state('running')
+            self.slots_handler.schedule()
+            self.config.save_configuration_files(True)
+            # TODO: __cb_pending can be None here after PostgreSQL restarts on its own. Do we want to call the callback?
+            # Previously we didn't even notice.
+            action = self.__cb_pending or ACTION_ON_START
+            self.call_nowait(action)
+            self.__cb_pending = None
+
+            return True
+
+    def wait_for_startup(self, timeout=None):
+        """Waits for PostgreSQL startup to complete or fail.
+
+        :returns: True if start was successful, False otherwise"""
+        if not self.is_starting():
+            # Should not happen
+            logger.warning("wait_for_startup() called when not in starting state")
+
+        while not self.check_startup_state_changed():
+            if self.cancellable.is_cancelled or timeout and self.time_in_state() > timeout:
+                return None
+            time.sleep(1)
+
+        return self.state == 'running'
+
+    def restart(self, timeout=None, task=None, block_callbacks=False, role=None):
+        """Restarts PostgreSQL.
+
+        When timeout parameter is set the call will block either until PostgreSQL has started, failed to start or
+        timeout arrives.
+
+        :returns: True when restart was successful and timeout did not expire when waiting.
+        """
+        self.set_state('restarting')
+        if not block_callbacks:
+            self.__cb_pending = ACTION_ON_RESTART
+        ret = self.stop(block_callbacks=True) and self.start(timeout, task, True, role)
+        if not ret and not self.is_starting():
+            self.set_state('restart failed ({0})'.format(self.state))
+        return ret
+
+    def is_healthy(self):
+        if not self.is_running():
+            logger.warning('Postgresql is not running.')
+            return False
+        return True
+
+    def get_guc_value(self, name):
+        cmd = [self.pgcommand('postgres'), 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 """
+        # Don't try to call pg_controldata during backup restore
+        if self._version_file_exists() and self.state != 'creating replica':
+            try:
+                env = os.environ.copy()
+                env.update(LANG='C', LC_ALL='C')
+                data = subprocess.check_output([self.pgcommand('pg_controldata'), self._data_dir], env=env)
+                if data:
+                    data = filter(lambda e: ':' in e, data.decode('utf-8').splitlines())
+                    # pg_controldata output depends on major version. Some of parameters are prefixed by 'Current '
+                    return {k.replace('Current ', '', 1): v.strip() for k, v in map(lambda e: e.split(':', 1), data)}
+            except subprocess.CalledProcessError:
+                logger.exception("Error when calling pg_controldata")
+        return {}
+
+    @contextmanager
+    def get_replication_connection_cursor(self, host='localhost', port=5432, **kwargs):
+        conn_kwargs = self.config.replication.copy()
+        conn_kwargs.update(host=host, port=int(port) if port else None, user=conn_kwargs.pop('username'),
+                           connect_timeout=3, replication=1, options='-c statement_timeout=2000')
+        with get_connection_cursor(**conn_kwargs) as cur:
+            yield cur
+
+    def get_replica_timeline(self):
+        try:
+            with self.get_replication_connection_cursor(**self.config.local_replication_address) as cur:
+                cur.execute('IDENTIFY_SYSTEM')
+                return cur.fetchone()[1]
+        except Exception:
+            logger.exception('Can not fetch local timeline and lsn from replication connection')
+
+    def replica_cached_timeline(self, master_timeline):
+        if not self._cached_replica_timeline or not master_timeline or self._cached_replica_timeline != master_timeline:
+            self._cached_replica_timeline = self.get_replica_timeline()
+        return self._cached_replica_timeline
+
+    def get_master_timeline(self):
+        return self._cluster_info_state_get('timeline')
+
+    def get_history(self, timeline):
+        history_path = os.path.join(self.wal_dir, '{0:08X}.history'.format(timeline))
+        history_mtime = mtime(history_path)
+        if history_mtime:
+            try:
+                with open(history_path, 'r') as f:
+                    history = f.read()
+                history = list(parse_history(history))
+                if history[-1][0] == timeline - 1:
+                    history_mtime = datetime.fromtimestamp(history_mtime).replace(tzinfo=tz.tzlocal())
+                    history[-1].append(history_mtime.isoformat())
+                return history
+            except Exception:
+                logger.exception('Failed to read and parse %s', (history_path,))
+
+    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():
+            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)
+
+        if change_role:
+            # TODO: postpone this until start completes, or maybe do even earlier
+            self.call_nowait(ACTION_ON_ROLE_CHANGE)
+        return True
+
+    def _wait_promote(self, wait_seconds):
+        for _ in polling_loop(wait_seconds):
+            data = self.controldata()
+            if data.get('Database cluster state') == 'in production':
+                return True
+
+    def promote(self, wait_seconds, on_success=None, access_is_restricted=False):
+        if self.role == 'master':
+            return True
+        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)
+            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:
+            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]
+
+        wal_position = self._wal_position(timeline, wal_position, received_location, replayed_location)
+        return (timeline, wal_position, pg_control_timeline)
+
+    def postmaster_start_time(self):
+        try:
+            query = "SELECT " + self.POSTMASTER_START_TIME
+            if current_thread().ident == self.__thread_ident:
+                return self.query(query).fetchone()[0]
+            with self.connection().cursor() as cursor:
+                cursor.execute(query)
+                return cursor.fetchone()[0]
+        except psycopg2.Error:
+            return None
+
+    def last_operation(self):
+        return str(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 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'))
+                logger.info('renaming data directory to %s', new_name)
+                os.rename(self._data_dir, new_name)
+            except OSError:
+                logger.exception("Could not rename data directory %s", self._data_dir)
+
+    def remove_data_directory(self):
+        self.set_role('uninitialized')
+        logger.info('Removing data directory: %s', self._data_dir)
+        try:
+            if os.path.islink(self._data_dir):
+                os.unlink(self._data_dir)
+            elif not os.path.exists(self._data_dir):
+                return
+            elif os.path.isfile(self._data_dir):
+                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
+                # should clean the target
+                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)
+                        logger.info('Removing WAL directory: %s', pg_wal_realpath)
+                        shutil.rmtree(pg_wal_realpath)
+                # Remove user defined tablespace directory
+                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)
+                            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):
+        """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.
+
+        :returns tuple of candidate name or None, and bool showing if the member is the active synchronous standby.
+        """
+        current = cluster.sync.sync_standby
+        current = current.lower() if current else current
+        members = {m.name.lower(): m for m in cluster.members}
+        candidates = []
+        # Pick candidates based on who has flushed WAL farthest.
+        # TODO: for synchronous_commit = remote_write we actually want to order on write_location
+        for app_name, state, sync_state in self.query(
+                "SELECT pg_catalog.lower(application_name), state, sync_state"
+                " FROM pg_catalog.pg_stat_replication"
+                " ORDER BY flush_{0} DESC".format(self.lsn_name)):
+            member = members.get(app_name)
+            if state != 'streaming' or not member or member.tags.get('nosync', False):
+                continue
+            if sync_state == 'sync':
+                return member.name, True
+            if sync_state == 'potential' and app_name == current:
+                # Prefer current even if not the best one any more to avoid indecisivness and spurious swaps.
+                return cluster.sync.sync_standby, False
+            if sync_state in ('async', 'potential'):
+                candidates.append(member.name)
+
+        if candidates:
+            return candidates[0], False
+        return None, False
+
+    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
+            2. get new 'Database system identifier' to make sure that it wasn't changed
+        """
+        self.slots_handler.schedule()
+        self._sysid = None

patroni/postgresql/bootstrap.py

@@ -0,0 +1,379 @@
+import logging
+import os
+import shlex
+import tempfile
+import time
+
+from patroni.dcs import RemoteMember
+from patroni.utils import deep_compare
+from six import string_types
+
+logger = logging.getLogger(__name__)
+
+
+class Bootstrap(object):
+
+    def __init__(self, postgresql):
+        self._postgresql = postgresql
+        self._running_custom_bootstrap = False
+
+    @property
+    def running_custom_bootstrap(self):
+        return self._running_custom_bootstrap
+
+    @property
+    def keep_existing_recovery_conf(self):
+        return self._running_custom_bootstrap and self._keep_existing_recovery_conf
+
+    @staticmethod
+    def process_user_options(tool, options, not_allowed_options, error_handler):
+        user_options = []
+
+        def option_is_allowed(name):
+            ret = name not in not_allowed_options
+            if not ret:
+                error_handler('{0} option for {1} is not allowed'.format(name, tool))
+            return ret
+
+        if isinstance(options, dict):
+            for k, v in options.items():
+                if k and v:
+                    user_options.append('--{0}={1}'.format(k, v))
+        elif isinstance(options, list):
+            for opt in options:
+                if isinstance(opt, string_types) and option_is_allowed(opt):
+                    user_options.append('--{0}'.format(opt))
+                elif isinstance(opt, dict):
+                    keys = list(opt.keys())
+                    if len(keys) != 1 or not isinstance(opt[keys[0]], string_types) or not option_is_allowed(keys[0]):
+                        error_handler('Error when parsing {0} key-value option {1}: only one key-value is allowed'
+                                      ' and value should be a string'.format(tool, opt[keys[0]]))
+                    user_options.append('--{0}={1}'.format(keys[0], opt[keys[0]]))
+                else:
+                    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))
+        return user_options
+
+    def _initdb(self, config):
+        self._postgresql.set_state('initalizing new cluster')
+        not_allowed_options = ('pgdata', 'nosync', 'pwfile', 'sync-only', 'version')
+
+        def error_handler(e):
+            raise Exception(e)
+
+        options = self.process_user_options('initdb', config or [], not_allowed_options, error_handler)
+        pwfile = None
+
+        if self._postgresql.config.superuser:
+            if 'username' in self._postgresql.config.superuser:
+                options.append('--username={0}'.format(self._postgresql.config.superuser['username']))
+            if 'password' in self._postgresql.config.superuser:
+                (fd, pwfile) = tempfile.mkstemp()
+                os.write(fd, self._postgresql.config.superuser['password'].encode('utf-8'))
+                os.close(fd)
+                options.append('--pwfile={0}'.format(pwfile))
+        options = ['-o', ' '.join(options)] if options else []
+
+        ret = self._postgresql.pg_ctl('initdb', *options)
+        if pwfile:
+            os.remove(pwfile)
+        if ret:
+            self._postgresql.configure_server_parameters()
+        else:
+            self._postgresql.set_state('initdb failed')
+        return ret
+
+    def _post_restore(self):
+        self._postgresql.config.restore_configuration_files()
+        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.get('recovery_conf', {}).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]
+        try:
+            logger.info('Running custom bootstrap script: %s', config['command'])
+            if self._postgresql.cancellable.call(shlex.split(config['command']) + params) != 0:
+                self._postgresql.set_state('custom bootstrap failed')
+                return False
+        except Exception:
+            logger.exception('Exception during custom bootstrap')
+            return False
+        self._post_restore()
+
+        if 'recovery_conf' in config:
+            self._postgresql.config.write_recovery_conf(config['recovery_conf'])
+        elif not self.keep_existing_recovery_conf:
+            self._postgresql.config.remove_recovery_conf()
+        return True
+
+    def call_post_bootstrap(self, config):
+        """
+        runs a script after initdb or custom bootstrap script is called and waits until completion.
+        """
+        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' not in r:
+                # 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
+
+            env = self._postgresql.config.write_pgpass(r) if 'password' in r else None
+            env['PGOPTIONS'] = '-c synchronous_commit=local'
+
+            try:
+                ret = self._postgresql.cancellable.call(shlex.split(cmd) + [connstring], env=env)
+            except OSError:
+                logger.error('post_init script %s failed', cmd)
+                return False
+            if ret != 0:
+                logger.error('post_init script %s returned non-zero code %d', cmd, ret)
+                return False
+        return True
+
+    def create_replica(self, clone_member):
+        """
+            create the replica according to the replica_method
+            defined by the user.  this is a list, so we need to
+            loop through all methods the user supplies
+        """
+
+        self._postgresql.set_state('creating replica')
+        self._postgresql.schedule_sanity_checks_after_pause()
+
+        is_remote_master = isinstance(clone_member, RemoteMember)
+
+        # get list of replica methods either from clone member or from
+        # the config. If there is no configuration key, or no value is
+        # specified, use basebackup
+        replica_methods = (clone_member.create_replica_methods if is_remote_master
+                           else self._postgresql.create_replica_methods) or ['basebackup']
+
+        if clone_member and clone_member.conn_url:
+            r = clone_member.conn_kwargs(self._postgresql.config.replication)
+            # add the credentials to connect to the replica origin to pgpass.
+            env = self._postgresql.config.write_pgpass(r)
+            connstring = self._postgresql.config.format_dsn(r, True)
+        else:
+            connstring = ''
+            env = os.environ.copy()
+            # if we don't have any source, leave only replica methods that work without it
+            replica_methods = [r for r in replica_methods
+                               if self._postgresql.replica_method_can_work_without_replication_connection(r)]
+
+        # go through them in priority order
+        ret = 1
+        for replica_method in replica_methods:
+            if self._postgresql.cancellable.is_cancelled:
+                break
+
+            method_config = self._postgresql.replica_method_options(replica_method)
+
+            # if the method is basebackup, then use the built-in
+            if replica_method == "basebackup":
+                ret = self.basebackup(connstring, env, method_config)
+                if ret == 0:
+                    logger.info("replica has been created using basebackup")
+                    # if basebackup succeeds, exit with success
+                    break
+            else:
+                if not self._postgresql.data_directory_empty():
+                    if method_config.get('keep_data', False):
+                        logger.info('Leaving data directory uncleaned')
+                    else:
+                        self._postgresql.remove_data_directory()
+
+                cmd = replica_method
+                # user-defined method; check for configuration
+                # not required, actually
+                if method_config:
+                    # look to see if the user has supplied a full command path
+                    # if not, use the method name as the command
+                    cmd = method_config.pop('command', cmd)
+
+                # add the default parameters
+                if not method_config.get('no_params', False):
+                    method_config.update({"scope": self._postgresql.scope,
+                                          "role": "replica",
+                                          "datadir": self._postgresql.data_dir,
+                                          "connstring": connstring})
+                else:
+                    for param in ('no_params', 'no_master', 'keep_data'):
+                        method_config.pop(param, None)
+                params = ["--{0}={1}".format(arg, val) for arg, val in method_config.items()]
+                try:
+                    # call script with the full set of parameters
+                    ret = self._postgresql.cancellable.call(shlex.split(cmd) + params, env=env)
+                    # if we succeeded, stop
+                    if ret == 0:
+                        logger.info('replica has been created using %s', replica_method)
+                        break
+                    else:
+                        logger.error('Error creating replica using method %s: %s exited with code=%s',
+                                     replica_method, cmd, ret)
+                except Exception:
+                    logger.exception('Error creating replica using method %s', replica_method)
+                    ret = 1
+
+        self._postgresql.set_state('stopped')
+        return ret
+
+    def basebackup(self, conn_url, env, options):
+        # creates a replica data dir using pg_basebackup.
+        # this is the default, built-in create_replica_methods
+        # tries twice, then returns failure (as 1)
+        # uses "stream" as the xlog-method to avoid sync issues
+        # supports additional user-supplied options, those are not validated
+        maxfailures = 2
+        ret = 1
+        not_allowed_options = ('pgdata', 'format', 'wal-method', 'xlog-method', 'gzip',
+                               'version', 'compress', 'dbname', 'host', 'port', 'username', 'password')
+        user_options = self.process_user_options('basebackup', options, not_allowed_options, logger.error)
+
+        for bbfailures in range(0, maxfailures):
+            if self._postgresql.cancellable.is_cancelled:
+                break
+            if not self._postgresql.data_directory_empty():
+                self._postgresql.remove_data_directory()
+            try:
+                ret = self._postgresql.cancellable.call([self._postgresql.pgcommand('pg_basebackup'),
+                                                         '--pgdata=' + self._postgresql.data_dir, '-X', 'stream',
+                                                         '--dbname=' + conn_url] + user_options, env=env)
+                if ret == 0:
+                    break
+                else:
+                    logger.error('Error when fetching backup: pg_basebackup exited with code=%s', ret)
+
+            except Exception as e:
+                logger.error('Error when fetching backup with pg_basebackup: %s', e)
+
+            if bbfailures < maxfailures - 1:
+                logger.warning('Trying again in 5 seconds')
+                time.sleep(5)
+
+        return ret
+
+    def clone(self, clone_member):
+        """
+             - initialize the replica from an existing member (master or replica)
+             - initialize the replica using the replica creation method that
+               works without the replication connection (i.e. restore from on-disk
+               base backup)
+        """
+
+        ret = self.create_replica(clone_member) == 0
+        if ret:
+            self._post_restore()
+        return ret
+
+    def bootstrap(self, config):
+        """ Initialize a new node from scratch and start it. """
+        pg_hba = config.get('pg_hba', [])
+        method = config.get('method') or 'initdb'
+        if method != 'initdb' and method in config and 'command' in config[method]:
+            self._keep_existing_recovery_conf = config[method].get('keep_existing_recovery_conf')
+            self._running_custom_bootstrap = True
+            do_initialize = self._custom_bootstrap
+        else:
+            method = 'initdb'
+            do_initialize = self._initdb
+        return do_initialize(config.get(method)) and self._postgresql.config.append_pg_hba(pg_hba) \
+            and self._postgresql.config.save_configuration_files() and self._postgresql.start()
+
+    def create_or_update_role(self, name, password, options):
+        options = list(map(str.upper, options))
+        if 'NOLOGIN' not in options and 'LOGIN' not in options:
+            options.append('LOGIN')
+
+        params = [name]
+        if password:
+            options.extend(['PASSWORD', '%s'])
+            params.extend([password, password])
+
+        sql = """DO $$
+BEGIN
+    SET local synchronous_commit = 'local';
+    PERFORM * FROM pg_authid WHERE rolname = %s;
+    IF FOUND THEN
+        ALTER ROLE "{0}" WITH {1};
+    ELSE
+        CREATE ROLE "{0}" WITH {1};
+    END IF;
+END;$$""".format(name, ' '.join(options))
+        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:
+            postgresql = self._postgresql
+            superuser = postgresql.config.superuser
+            if 'username' in superuser and 'password' in superuser:
+                self.create_or_update_role(superuser['username'], superuser['password'], ['SUPERUSER'])
+
+            task.complete(self.call_post_bootstrap(config))
+            if task.result:
+                replication = postgresql.config.replication
+                self.create_or_update_role(replication['username'], replication.get('password'), ['REPLICATION'])
+
+                rewind = postgresql.config.rewind_credentials
+                if not deep_compare(rewind, superuser):
+                    self.create_or_update_role(rewind['username'], rewind.get('password'), [])
+                    for f in ('pg_ls_dir(text, boolean, boolean)', 'pg_stat_file(text, boolean)',
+                              'pg_read_binary_file(text)', 'pg_read_binary_file(text, bigint, bigint, boolean)'):
+                        sql = """DO $$
+BEGIN
+    SET local synchronous_commit = 'local';
+    GRANT EXECUTE ON function pg_catalog.{0} TO "{1}";
+END;$$""".format(f, rewind['username'])
+                        postgresql.query(sql)
+
+                for name, value in (config.get('users') or {}).items():
+                    if all(name != a.get('username') for a in (superuser, replication, rewind)):
+                        self.create_or_update_role(name, value.get('password'), value.get('options', []))
+
+                # We were doing a custom bootstrap instead of running initdb, therefore we opened trust
+                # access from certain addresses to be able to reach cluster and change password
+                if self._running_custom_bootstrap:
+                    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)
+                        postgresql.config.restore_configuration_files()
+                    postgresql.config.write_postgresql_conf()
+                    postgresql.config.replace_pg_ident()
+
+                    # at this point there should be no recovery.conf
+                    postgresql.config.remove_recovery_conf()
+
+                    if postgresql.config.hba_file:
+                        postgresql.restart()
+                    else:
+                        postgresql.config.replace_pg_hba()
+                        if postgresql.pending_restart:
+                            postgresql.restart()
+                        else:
+                            postgresql.reload()
+                            time.sleep(1)  # give a time to postgres to "reload" configuration files
+                            postgresql.connection().close()  # close connection to reconnect with a new password
+        except Exception:
+            logger.exception('post_bootstrap')
+            task.complete(False)
+        return task.result

patroni/postgresql/callback_executor.py

@@ -0,0 +1,36 @@
+import logging
+
+from patroni.postgresql.cancellable import CancellableExecutor
+from threading import Condition, Thread
+
+logger = logging.getLogger(__name__)
+
+
+class CallbackExecutor(CancellableExecutor, Thread):
+
+    def __init__(self):
+        CancellableExecutor.__init__(self)
+        Thread.__init__(self)
+        self.daemon = True
+        self._cmd = None
+        self._condition = Condition()
+        self.start()
+
+    def call(self, cmd):
+        self._kill_process()
+        with self._condition:
+            self._cmd = cmd
+            self._condition.notify()
+
+    def run(self):
+        while True:
+            with self._condition:
+                if self._cmd is None:
+                    self._condition.wait()
+                cmd, self._cmd = self._cmd, None
+
+            with self._lock:
+                if not self._start_process(cmd, close_fds=True):
+                    continue
+            self._process.wait()
+            self._kill_children()

patroni/postgresql/cancellable.py

@@ -0,0 +1,125 @@
+import logging
+import psutil
+import subprocess
+
+from patroni.exceptions import PostgresException
+from patroni.utils import polling_loop
+from threading import Lock
+
+logger = logging.getLogger(__name__)
+
+
+class CancellableExecutor(object):
+
+    def __init__(self):
+        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 childrens
+                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 = kwargs.pop('communicate', None)
+        if isinstance(communicate, dict):
+            input_data = communicate.get('input')
+            if input_data:
+                if input_data[-1] != '\n':
+                    input_data += '\n'
+                input_data = input_data.encode('utf-8')
+            kwargs['stdin'] = subprocess.PIPE
+            kwargs['stdout'] = subprocess.PIPE
+            kwargs['stderr'] = subprocess.PIPE
+
+        try:
+            with self._lock:
+                if self._is_cancelled:
+                    raise PostgresException('cancelled')
+
+                self._is_cancelled = False
+                started = self._start_process(*args, **kwargs)
+
+            if started:
+                if isinstance(communicate, dict):
+                    communicate['stdout'], communicate['stderr'] = self._process.communicate(input_data)
+                return self._process.wait()
+        finally:
+            with self._lock:
+                self._process = None
+            self._kill_children()
+
+    def reset_is_cancelled(self):
+        with self._lock:
+            self._is_cancelled = False
+
+    @property
+    def is_cancelled(self):
+        with self._lock:
+            return self._is_cancelled
+
+    def cancel(self):
+        with self._lock:
+            self._is_cancelled = True
+            if self._process is None or not self._process.is_running():
+                return
+            self._process.terminate()
+
+        for _ in polling_loop(10):
+            with self._lock:
+                if self._process is None or not self._process.is_running():
+                    return
+
+        self._kill_process()

patroni/postgresql/connection.py

@@ -0,0 +1,46 @@
+import logging
+import psycopg2
+
+from contextlib import contextmanager
+from threading import Lock
+
+logger = logging.getLogger(__name__)
+
+
+class Connection(object):
+
+    def __init__(self):
+        self._lock = Lock()
+        self._connection = None
+        self._cursor_holder = None
+
+    def set_conn_kwargs(self, conn_kwargs):
+        self._conn_kwargs = conn_kwargs
+
+    def get(self):
+        with self._lock:
+            if not self._connection or self._connection.closed != 0:
+                self._connection = psycopg2.connect(**self._conn_kwargs)
+                self._connection.autocommit = True
+                self.server_version = self._connection.server_version
+        return self._connection
+
+    def cursor(self):
+        if not self._cursor_holder or self._cursor_holder.closed or self._cursor_holder.connection.closed != 0:
+            logger.info("establishing a new patroni connection to the postgres cluster")
+            self._cursor_holder = self.get().cursor()
+        return self._cursor_holder
+
+    def close(self):
+        if self._connection and self._connection.closed == 0:
+            self._connection.close()
+            logger.info("closed patroni connection to the postgresql cluster")
+        self._cursor_holder = self._connection = None
+
+
+@contextmanager
+def get_connection_cursor(**kwargs):
+    with psycopg2.connect(**kwargs) as conn:
+        conn.autocommit = True
+        with conn.cursor() as cur:
+            yield cur

patroni/postgresql/misc.py

@@ -0,0 +1,70 @@
+import logging
+
+from patroni.exceptions import PostgresException
+
+logger = logging.getLogger(__name__)
+
+
+def postgres_version_to_int(pg_version):
+    """Convert the server_version to integer
+
+    >>> postgres_version_to_int('9.5.3')
+    90503
+    >>> postgres_version_to_int('9.3.13')
+    90313
+    >>> postgres_version_to_int('10.1')
+    100001
+    >>> postgres_version_to_int('10')  # doctest: +IGNORE_EXCEPTION_DETAIL
+    Traceback (most recent call last):
+        ...
+    PostgresException: 'Invalid PostgreSQL version format: X.Y or X.Y.Z is accepted: 10'
+    >>> postgres_version_to_int('9.6')  # doctest: +IGNORE_EXCEPTION_DETAIL
+    Traceback (most recent call last):
+        ...
+    PostgresException: 'Invalid PostgreSQL version format: X.Y or X.Y.Z is accepted: 9.6'
+    >>> postgres_version_to_int('a.b.c')  # doctest: +IGNORE_EXCEPTION_DETAIL
+    Traceback (most recent call last):
+        ...
+    PostgresException: 'Invalid PostgreSQL version: a.b.c'
+    """
+
+    try:
+        components = list(map(int, pg_version.split('.')))
+    except ValueError:
+        raise PostgresException('Invalid PostgreSQL version: {0}'.format(pg_version))
+
+    if len(components) < 2 or len(components) == 2 and components[0] < 10 or len(components) > 3:
+        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
+        components.insert(1, 0)
+
+    return int(''.join('{0:02d}'.format(c) for c in components))
+
+
+def postgres_major_version_to_int(pg_version):
+    """
+    >>> postgres_major_version_to_int('10')
+    100000
+    >>> postgres_major_version_to_int('9.6')
+    90600
+    """
+    return postgres_version_to_int(pg_version + '.0')
+
+
+def parse_lsn(lsn):
+    t = lsn.split('/')
+    return int(t[0], 16) * 0x100000000 + int(t[1], 16)
+
+
+def parse_history(data):
+    for line in data.split('\n'):
+        values = line.strip().split('\t')
+        if len(values) == 3:
+            try:
+                values[0] = int(values[0])
+                values[1] = parse_lsn(values[1])
+                yield values
+            except (IndexError, ValueError):
+                logger.exception('Exception when parsing timeline history line "%s"', values)

patroni/postgresql/postmaster.py

@@ -1,21 +1,43 @@
 import logging
+import multiprocessing
 import os
 import psutil
 import re
 import signal
 import subprocess
+import sys
 
-from patroni import call_self
+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,
-    'fast': signal.SIGINT,
-    'immediate': signal.SIGQUIT if os.name != 'nt' else signal.SIGABRT,
+    'smart': 'TERM',
+    'fast': 'INT',
+    'immediate': 'QUIT',
 }
 
 
+def pg_ctl_start(conn, cmdline, env):
+    if os.name != 'nt':
+        os.setsid()
+    try:
+        postmaster = subprocess.Popen(cmdline, close_fds=True, env=env)
+        conn.send(postmaster.pid)
+    except Exception:
+        logger.exception('Failed to execute %s', cmdline)
+        conn.send(None)
+    conn.close()
+
+
 class PostmasterProcess(psutil.Process):
 
     def __init__(self, pid):
@@ -83,7 +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
@@ -91,8 +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:
@@ -101,29 +161,42 @@ 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 .. 9.6
-        aux_proc_re = re.compile("(?:postgres:)( .*:)? (?:""(?:startup|logger|checkpointer|writer|wal writer|"
-                                 "autovacuum launcher|autovacuum worker|stats collector|wal receiver|archiver|"
-                                 "wal sender) process|bgworker: )")
+        # These regexps are cross checked against versions PostgreSQL 9.1 .. 11
+        aux_proc_re = re.compile("(?:postgres:)( .*:)? (?:(?:archiver|startup|autovacuum launcher|autovacuum worker|"
+                                 "checkpointer|logger|stats collector|wal receiver|wal writer|writer)(?: process  )?|"
+                                 "walreceiver|wal sender process|walsender|walwriter|background writer|"
+                                 "logical replication launcher|logical replication worker for|bgworker:) ")
 
         try:
-            user_backends = []
-            user_backends_cmdlines = []
-            for child in self.children():
-                try:
-                    cmdline = child.cmdline()[0]
-                    if not aux_proc_re.match(cmdline):
-                        user_backends.append(child)
-                        user_backends_cmdlines.append(cmdline)
-                except psutil.NoSuchProcess:
-                    pass
-            if user_backends:
-                logger.debug('Waiting for user backends %s to close', ', '.join(user_backends_cmdlines))
-                psutil.wait_procs(user_backends)
-            logger.debug("Backends closed")
+            children = self.children()
         except psutil.Error:
-            logger.exception('wait_for_user_backends_to_close')
+            return logger.debug('Failed to get list of postmaster children')
+
+        user_backends = []
+        user_backends_cmdlines = []
+        for child in children:
+            try:
+                cmdline = child.cmdline()[0]
+                if not aux_proc_re.match(cmdline):
+                    user_backends.append(child)
+                    user_backends_cmdlines.append(cmdline)
+            except psutil.NoSuchProcess:
+                pass
+        if user_backends:
+            logger.debug('Waiting for user backends %s to close', ', '.join(user_backends_cmdlines))
+            psutil.wait_procs(user_backends)
+        logger.debug("Backends closed")
 
     @staticmethod
     def start(pgcommand, data_dir, conf, options):
@@ -139,7 +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():
@@ -156,10 +230,14 @@ class PostmasterProcess(psutil.Process):
             pass
         cmdline = [pgcommand, '-D', data_dir, '--config-file={}'.format(conf)] + options
         logger.debug("Starting postgres: %s", " ".join(cmdline))
-        proc = call_self(['pg_ctl_start'] + cmdline, close_fds=(os.name != 'nt'),
-                         stdout=subprocess.PIPE, stderr=subprocess.STDOUT, env=env)
-        pid = int(proc.stdout.readline().strip())
-        proc.wait()
+        ctx = multiprocessing.get_context('spawn') if sys.version_info >= (3, 4) else multiprocessing
+        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()
+        if pid is None:
+            return
         logger.info('postmaster pid=%s', pid)
 
         # TODO: In an extremely unlikely case, the process could have exited and the pid reassigned. The start

patroni/postgresql/rewind.py

@@ -0,0 +1,430 @@
+import logging
+import os
+import six
+import subprocess
+
+from threading import Lock, Thread
+
+from .connection import get_connection_cursor
+from .misc import parse_history, parse_lsn
+from ..async_executor import CriticalTask
+from ..dcs import Leader
+
+logger = logging.getLogger(__name__)
+
+REWIND_STATUS = type('Enum', (), {'INITIAL': 0, 'CHECKPOINT': 1, 'CHECK': 2, 'NEED': 3,
+                                  'NOT_NEED': 4, 'SUCCESS': 5, 'FAILED': 6})
+
+
+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)
+
+
+class Rewind(object):
+
+    def __init__(self, postgresql):
+        self._postgresql = postgresql
+        self._checkpoint_task_lock = Lock()
+        self.reset_state()
+
+    @staticmethod
+    def configuration_allows_rewind(data):
+        return data.get('wal_log_hints setting', 'off') == 'on' or data.get('Data page checksum version', '0') != '0'
+
+    @property
+    def can_rewind(self):
+        """ check if pg_rewind executable is there and that pg_controldata indicates
+            we have either wal_log_hints or checksums turned on
+        """
+        # low-hanging fruit: check if pg_rewind configuration is there
+        if not self._postgresql.config.get('use_pg_rewind'):
+            return False
+
+        cmd = [self._postgresql.pgcommand('pg_rewind'), '--help']
+        try:
+            ret = subprocess.call(cmd, stdout=open(os.devnull, 'w'), stderr=subprocess.STDOUT)
+            if ret != 0:  # pg_rewind is not there, close up the shop and go home
+                return False
+        except OSError:
+            return False
+        return self.configuration_allows_rewind(self._postgresql.controldata())
+
+    @property
+    def can_rewind_or_reinitialize_allowed(self):
+        return self._postgresql.config.get('remove_data_directory_on_diverged_timelines') or self.can_rewind
+
+    def trigger_check_diverged_lsn(self):
+        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):
+        if not kwargs.get('database'):
+            kwargs['database'] = self._postgresql.database
+        try:
+            with get_connection_cursor(connect_timeout=3, options='-c statement_timeout=2000', **kwargs) as cur:
+                cur.execute('SELECT pg_catalog.pg_is_in_recovery()')
+                if not cur.fetchone()[0]:
+                    return True
+                logger.info('Leader is still in_recovery and therefore can\'t be used for rewind')
+        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 wil 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."""
+
+        cmd = self._postgresql.pgcommand('pg_{0}dump'.format(self._postgresql.wal_name))
+        lsn8 = format_lsn(lsn, True)
+        lsn = format_lsn(lsn)
+        env = os.environ.copy()
+        env.update(LANG='C', LC_ALL='C', PGDATA=self._postgresql.data_dir)
+        try:
+            waldump = subprocess.Popen([cmd, '-t', str(timeline), '-s', lsn, '-n', '2'],
+                                       stdout=subprocess.PIPE, stderr=subprocess.PIPE, env=env)
+            out, err = waldump.communicate()
+            waldump.wait()
+        except Exception as e:
+            logger.error('Failed to execute `%s -t %s -s %s -n 2`: %r', cmd, timeline, lsn, e)
+        else:
+            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 `%s -t %s -s %s -n 2` output', cmd, timeline, lsn)
+            logger.error(' stdout=%s', '\n'.join(out))
+            logger.error(' stderr=%s', '\n'.join(err))
+
+        return 0
+
+    def _get_local_timeline_lsn_from_controldata(self):
+        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')
+
+        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 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
+            in_recovery, timeline, lsn = self._get_local_timeline_lsn_from_controldata()
+
+        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 _check_timeline_and_lsn(self, leader):
+        in_recovery, local_timeline, local_lsn = self._get_local_timeline_lsn()
+        if local_timeline is None or local_lsn is None:
+            return
+
+        if isinstance(leader, Leader):
+            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)):
+            return
+
+        history = need_rewind = None
+        try:
+            with self._postgresql.get_replication_connection_cursor(**leader.conn_kwargs()) as cur:
+                cur.execute('IDENTIFY_SYSTEM')
+                master_timeline = cur.fetchone()[1]
+                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')
+                    logger.debug('master: history=%s', history)
+        except Exception:
+            return logger.exception('Exception when working with master via replication connection')
+
+        if history is not None:
+            history = list(parse_history(history))
+            for i, (parent_timeline, switchpoint, _) in enumerate(history):
+                if parent_timeline == local_timeline:
+                    # We don't need to rewind when:
+                    # 1. for replica: replayed location is not ahead of switchpoint
+                    # 2. for the former primary: end of checkpoint record is the same as switchpoint
+                    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
+
+    def rewind_or_reinitialize_needed_and_possible(self, leader):
+        if leader and leader.name != self._postgresql.name and leader.conn_url and self._state == REWIND_STATUS.CHECK:
+            self._check_timeline_and_lsn(leader)
+        return leader and leader.conn_url and self._state == REWIND_STATUS.NEED
+
+    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():
+            with self._checkpoint_task_lock:
+                if self._checkpoint_task:
+                    with self._checkpoint_task:
+                        if self._checkpoint_task.result:
+                            self._state = REWIND_STATUS.CHECKPOINT
+                        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(cmd, shell=True) == 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:
+                    waldir, wal_filename = os.path.split(line[b:e])
+                    if waldir.endswith(os.path.sep + '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.config.write_pgpass(r)
+        env.update(LANG='C', LC_ALL='C', PGOPTIONS='-c statement_timeout=0')
+        dsn = self._postgresql.config.format_dsn(r, True)
+        logger.info('running pg_rewind from %s', dsn)
+
+        restore_command = self._postgresql.config.get('recovery_conf', {}).get('restore_command') \
+            if self._postgresql.major_version < 120000 else self._postgresql.get_guc_value('restore_command')
+
+        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)
+
+        # 1. make sure that we are really trying to rewind from the master
+        # 2. make sure that pg_control contains the new timeline by:
+        #   running a checkpoint or
+        #   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))
+            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):
+            return
+
+        if self.pg_rewind(r):
+            self._state = REWIND_STATUS.SUCCESS
+        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)
+
+            for name in ('remove_data_directory_on_rewind_failure', 'remove_data_directory_on_diverged_timelines'):
+                if self._postgresql.config.get(name):
+                    logger.warning('%s is set. removing...', name)
+                    self._postgresql.remove_data_directory()
+                    self._state = REWIND_STATUS.INITIAL
+                    break
+            else:
+                self._state = REWIND_STATUS.FAILED
+        return False
+
+    def reset_state(self):
+        self._state = REWIND_STATUS.INITIAL
+        with self._checkpoint_task_lock:
+            self._checkpoint_task = None
+
+    @property
+    def is_needed(self):
+        return self._state in (REWIND_STATUS.CHECK, REWIND_STATUS.NEED)
+
+    @property
+    def executed(self):
+        return self._state > REWIND_STATUS.NOT_NEED
+
+    @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()
+        return self.single_user_mode(options=opts) == 0 or None

patroni/postgresql/slots.py

@@ -0,0 +1,103 @@
+import logging
+
+from patroni.postgresql.connection import get_connection_cursor
+from collections import defaultdict
+
+logger = logging.getLogger(__name__)
+
+
+def compare_slots(s1, s2):
+    return s1['type'] == s2['type'] and (s1['type'] == 'physical' or
+                                         s1['database'] == s2['database'] and s1['plugin'] == s2['plugin'])
+
+
+class SlotsHandler(object):
+
+    def __init__(self, postgresql):
+        self._postgresql = postgresql
+        self._replication_slots = {}  # already existing replication slots
+        self.schedule()
+
+    def _query(self, sql, *params):
+        return self._postgresql.query(sql, *params, retry=False)
+
+    def load_replication_slots(self):
+        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')
+            for r in cursor:
+                value = {'type': r[1]}
+                if r[1] == 'logical':
+                    value.update({'plugin': r[2], 'database': r[3]})
+                replication_slots[r[0]] = value
+            self._replication_slots = replication_slots
+            self._schedule_load_slots = False
+
+    def drop_replication_slot(self, name):
+        cursor = self._query(('SELECT pg_catalog.pg_drop_replication_slot(%s) WHERE EXISTS (SELECT 1 ' +
+                              'FROM pg_catalog.pg_replication_slots WHERE slot_name = %s AND NOT active)'), name, name)
+        # 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):
+        if self._postgresql.major_version >= 90400:
+            try:
+                self.load_replication_slots()
+
+                slots = cluster.get_replication_slots(self._postgresql.name, self._postgresql.role)
+
+                # drop old replication slots which are not presented in desired 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 ''
+
+                logical_slots = defaultdict(dict)
+                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 not self.drop_replication_slot(name):
+                            logger.error("Failed to drop replication slot '%s'", name)
+                            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
+
+    def schedule(self, value=None):
+        if value is None:
+            value = self._postgresql.major_version >= 90400
+        self._schedule_load_slots = value

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)
-        except RequestException:
+            r = requests_get('http://169.254.169.254/latest/dynamic/instance-identity/document', timeout=2.1)
+        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

@@ -224,13 +224,12 @@ class WALERestore(object):
                             lsn_name = 'location'
                         con.autocommit = True
                         with con.cursor() as cur:
-                            cur.execute("""SELECT CASE WHEN pg_is_in_recovery()
-                                                       THEN GREATEST(
-                                                                pg_{0}_{1}_diff(COALESCE(
-                                                                  pg_last_{0}_receive_{1}(), '0/0'), %s)::bigint,
-                                                                pg_{0}_{1}_diff(pg_last_{0}_replay_{1}(), %s)::bigint)
-                                                       ELSE pg_{0}_{1}_diff(pg_current_{0}_{1}(), %s)::bigint
-                                                   END""".format(wal_name, lsn_name),
+                            cur.execute(("SELECT CASE WHEN pg_catalog.pg_is_in_recovery()"
+                                         " THEN GREATEST(pg_catalog.pg_{0}_{1}_diff(COALESCE("
+                                         "pg_last_{0}_receive_{1}(), '0/0'), %s)::bigint, "
+                                         "pg_catalog.pg_{0}_{1}_diff(pg_catalog.pg_last_{0}_replay_{1}(), %s)::bigint)"
+                                         " ELSE pg_catalog.pg_{0}_{1}_diff(pg_catalog.pg_current_{0}_{1}(), %s)::bigint"
+                                         " END").format(wal_name, lsn_name),
                                         (backup_start_lsn, backup_start_lsn, backup_start_lsn))
 
                             diff_in_bytes = int(cur.fetchone()[0])

patroni/utils.py

@@ -1,11 +1,26 @@
+import logging
+import os
+import platform
 import random
+import re
+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]+')
+DBL_RE = re.compile(r'^[-+]?[0-9]*\.?[0-9]+([eE][-+]?[0-9]+)?')
+
 
 def deep_compare(obj1, obj2):
     """
@@ -88,42 +103,77 @@ def strtol(value, strict=True):
     True
     >>> strtol(' d ') == (None, 'd')
     True
+    >>> strtol(' 1 d ') == (1, ' d')
+    True
     >>> strtol('9s', False) == (9, 's')
     True
     >>> strtol(' s ', False) == (1, 's')
     True
     """
     value = str(value).strip()
-    ln = len(value)
-    i = 0
-    # skip sign:
-    if i < ln and value[i] in ('-', '+'):
-        i += 1
+    for regex, base in ((HEX_RE, 16), (OCT_RE, 8), (DEC_RE, 10)):
+        match = regex.match(value)
+        if match:
+            end = match.end()
+            return int(value[:end], base), value[end:]
+    return (None if strict else 1), value
 
-    # we always expect to get digit in the beginning
-    if i < ln and value[i].isdigit():
-        if value[i] == '0':
-            i += 1
-            if i < ln and value[i] in ('x', 'X'):  # '0' followed by 'x': HEX
-                base = 16
-                i += 1
-            else:  # just starts with '0': OCT
-                base = 8
-        else:  # any other digit: DEC
-            base = 10
 
-        ret = None
-        while i <= ln:
-            try:  # try to find maximally long number
-                i += 1  # by giving to `int` longer and longer strings
-                ret = int(value[:i], base)
-            except ValueError:  # until we will not get an exception or end of the string
-                i -= 1
-                break
-        if ret is not None:  # yay! there is a number in the beginning of the string
-            return ret, value[i:].strip()  # return the number and the "rest"
+def strtod(value):
+    """As most as possible close equivalent of strtod(3) function used by postgres to parse parameter values.
+    >>> strtod(' A ') == (None, 'A')
+    True
+    """
+    value = str(value).strip()
+    match = DBL_RE.match(value)
+    if match:
+        end = match.end()
+        return float(value[:end]), value[end:]
+    return None, value
 
-    return (None if strict else 1), value.strip()
+
+def rint(value):
+    """
+    >>> rint(0.5) == 0
+    True
+    >>> rint(0.501) == 1
+    True
+    >>> rint(1.5) == 2
+    True
+    """
+
+    ret = round(value)
+    return 2.0 * round(value / 2.0) if abs(ret - value) == 0.5 else ret
+
+
+def convert_to_base_unit(value, unit, base_unit):
+    convert = {
+        'B': {'B': 1, 'kB': 1024, 'MB': 1024 * 1024, 'GB': 1024 * 1024 * 1024, 'TB': 1024 * 1024 * 1024 * 1024},
+        'kB': {'B': 1.0 / 1024, 'kB': 1, 'MB': 1024, 'GB': 1024 * 1024, 'TB': 1024 * 1024 * 1024},
+        'MB': {'B': 1.0 / (1024 * 1024), 'kB': 1.0 / 1024, 'MB': 1, 'GB': 1024, 'TB': 1024 * 1024},
+        'ms': {'us': 1.0 / 1000, 'ms': 1, 's': 1000, 'min': 1000 * 60, 'h': 1000 * 60 * 60, 'd': 1000 * 60 * 60 * 24},
+        's': {'us': 1.0 / (1000 * 1000), 'ms': 1.0 / 1000, 's': 1, 'min': 60, 'h': 60 * 60, 'd': 60 * 60 * 24},
+        'min': {'us': 1.0 / (1000 * 1000 * 60), 'ms': 1.0 / (1000 * 60), 's': 1.0 / 60, 'min': 1, 'h': 60, 'd': 60 * 24}
+    }
+
+    round_order = {
+        'TB': 'GB', 'GB': 'MB', 'MB': 'kB', 'kB': 'B',
+        'd': 'h', 'h': 'min', 'min': 's', 's': 'ms', 'ms': 'us'
+    }
+
+    if base_unit and base_unit not in convert:
+        base_value, base_unit = strtol(base_unit, False)
+    else:
+        base_value = 1
+
+    if base_unit in convert and unit in convert[base_unit]:
+        value *= convert[base_unit][unit] / float(base_value)
+
+        if unit in round_order:
+            multiplier = convert[base_unit][round_order[unit]]
+            value = rint(value / float(multiplier)) * multiplier
+
+        return value
 
 
 def parse_int(value, base_unit=None):
@@ -136,56 +186,71 @@ def parse_int(value, base_unit=None):
     True
     >>> parse_int('1000 ms', 's') == 1
     True
-    >>> parse_int('1GB', 'MB') is None
+    >>> parse_int('1TB', 'GB') is None
     True
     >>> parse_int(0) == 0
     True
+    >>> parse_int('6GB', '16MB') == 384
+    True
+    >>> parse_int('4097.4kB', 'kB') == 4097
+    True
+    >>> parse_int('4097.5kB', 'kB') == 4098
+    True
     """
 
-    convert = {
-        'kB': {'kB': 1, 'MB': 1024, 'GB': 1024 * 1024, 'TB': 1024 * 1024 * 1024},
-        'ms': {'ms': 1, 's': 1000, 'min': 1000 * 60, 'h': 1000 * 60 * 60, 'd': 1000 * 60 * 60 * 24},
-        's': {'ms': -1000, 's': 1, 'min': 60, 'h': 60 * 60, 'd': 60 * 60 * 24},
-        'min': {'ms': -1000 * 60, 's': -60, 'min': 1, 'h': 60, 'd': 60 * 24}
-    }
+    val, unit = strtol(value)
+    if val is None and unit.startswith('.') or unit and unit[0] in ('.', 'e', 'E'):
+        val, unit = strtod(value)
 
-    value, unit = strtol(value)
-    if value is not None:
+    if val is not None:
+        unit = unit.strip()
         if not unit:
-            return value
+            return int(rint(val))
 
-        if base_unit and base_unit not in convert:
-            base_value, base_unit = strtol(base_unit, False)
-        else:
-            base_value = 1
-        if base_unit in convert and unit in convert[base_unit]:
-            multiplier = convert[base_unit][unit]
-            if multiplier < 0:
-                value /= -multiplier
-            else:
-                value *= multiplier
-            return int(value/base_value)
+        val = convert_to_base_unit(val, unit, base_unit)
+        if val is not None:
+            return int(rint(val))
+
+
+def parse_real(value, base_unit=None):
+    """
+    >>> parse_real(' +0.0005 ') == 0.0005
+    True
+    >>> parse_real('0.0005ms', 'ms') == 0.0
+    True
+    >>> parse_real('0.00051ms', 'ms') == 0.001
+    True
+    """
+    val, unit = strtod(value)
+
+    if val is not None:
+        unit = unit.strip()
+        if not unit:
+            return val
+
+        return convert_to_base_unit(val, unit, base_unit)
 
 
 def compare_values(vartype, unit, old_value, new_value):
     """
     >>> compare_values('enum', None, 'remote_write', 'REMOTE_WRITE')
     True
-    >>> compare_values('real', None, '1.23', 1.23)
+    >>> compare_values('real', None, '1e-06', 0.000001)
     True
     """
 
-    # if the integer or bool new_value is not correct this function will return False
-    if vartype == 'bool':
-        old_value = parse_bool(old_value)
-        new_value = parse_bool(new_value)
-    elif vartype == 'integer':
-        old_value = parse_int(old_value)
-        new_value = parse_int(new_value, unit)
-    elif vartype == 'enum':
-        return str(old_value).lower() == str(new_value).lower()
-    else:  # ('string', 'real')
-        return str(old_value) == str(new_value)
+    converters = {
+        'bool': lambda v1, v2: parse_bool(v1),
+        'integer': parse_int,
+        'real': parse_real,
+        'enum': lambda v1, v2: str(v1).lower(),
+        'string': lambda v1, v2: str(v1)
+    }
+
+    convert = converters.get(vartype) or converters['string']
+    old_value = convert(old_value, None)
+    new_value = convert(new_value, unit)
+
     return old_value is not None and new_value is not None and old_value == new_value
 
 
@@ -237,6 +302,17 @@ 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`
 
@@ -252,18 +328,20 @@ class Retry(object):
                 if self.deadline is not None and self._cur_stoptime is None:
                     self._cur_stoptime = time.time() + self.deadline
                 return func(*args, **kwargs)
-            except self.retry_exceptions:
+            except self.retry_exceptions as e:
                 # Note: max_tries == -1 means infinite tries.
                 if self._attempts == self.max_tries:
+                    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")
-                else:
-                    self.sleep_func(sleeptime)
-                self._cur_delay = min(self._cur_delay * self.backoff, self.max_delay)
+                logger.debug('Retry got exception: %s', e)
+                self.sleep_func(sleeptime)
+                self.update_delay()
 
 
 def polling_loop(timeout, interval=1):
@@ -279,5 +357,99 @@ 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])
+
+
+def uri(proto, netloc, path='', user=None):
+    host, port = netloc if isinstance(netloc, (list, tuple)) else split_host_port(netloc, 0)
+    if host and ':' in host and host[0] != '[' and host[-1] != ']':
+        host = '[{0}]'.format(host)
+    port = ':{0}'.format(port) if port else ''
+    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 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
+    xlog_location_cluster = cluster.last_leader_operation 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 == cluster.sync.sync_standby:
+            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:
+            xlog_location = m.data.get('xlog_location')
+            if xlog_location is None:
+                member['lag'] = 'unknown'
+            elif xlog_location_cluster >= xlog_location:
+                member['lag'] = xlog_location_cluster - xlog_location
+            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)
+            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))

patroni/validator.py

@@ -0,0 +1,379 @@
+#!/usr/bin/env python3
+import os
+import socket
+import re
+import subprocess
+
+from patroni.utils import split_host_port, data_directory_is_empty
+from patroni.ctl import find_executable
+from patroni.dcs import dcs_modules
+from patroni.exceptions import ConfigParseError
+from six import string_types
+
+
+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 comma_separated_host_port(string):
+    assert all([validate_host_port(s.strip()) for s in string.split(",")]), "didn't pass the validation"
+    return True
+
+
+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()]
+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
+
+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": {
+          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})
+         },
+      "exhibitor": {
+          "hosts": [str],
+          "port": lambda i: assert_(int(i) <= 65535),
+          Optional("pool_interval"): int
+          },
+      "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__ = '1.5.0'
+__version__ = '1.6.5'

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

@@ -1,6 +1,5 @@
 import collections
 import ctypes
-import fcntl
 import os
 import platform
 from patroni.watchdog.base import WatchdogBase, WatchdogError
@@ -17,11 +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
@@ -162,7 +161,9 @@ class LinuxWatchdogDevice(WatchdogBase):
         Raises OSError or IOError (Python 2) when the ioctl fails."""
         if self._fd is None:
             raise WatchdogError("Watchdog device is closed")
-        fcntl.ioctl(self._fd, func, arg, True)
+        if os.name != 'nt':
+            import fcntl
+            fcntl.ioctl(self._fd, func, arg, True)
 
     def get_support(self):
         if self._support_cache is None:
@@ -217,7 +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
 

postgres0.yml

@@ -17,7 +17,18 @@ 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
 
 bootstrap:
   # this section will be written into Etcd:/<namespace>/<scope>/config after initializing new cluster
@@ -55,6 +66,9 @@ bootstrap:
   - data-checksums
 
   pg_hba:  # Add following lines to pg_hba.conf after running 'initdb'
+  # For kerberos gss based connectivity (discard @.*$)
+  #- host replication replicator 127.0.0.1/32 gss include_realm=0
+  #- host all all 0.0.0.0/0 gss include_realm=0
   - host replication replicator 127.0.0.1/32 md5
   - host all all 0.0.0.0/0 md5
 #  - hostssl all all 0.0.0.0/0 md5
@@ -84,7 +98,15 @@ postgresql:
     superuser:
       username: postgres
       password: zalando
+    rewind:  # Has no effect on postgres 10 and lower
+      username: rewind_user
+      password: rewind_password
+  # Server side kerberos spn
+#  krbsrvname: postgres
   parameters:
+    # Fully qualified kerberos ticket file for the running user
+    # same as KRB5CCNAME used by the GSS
+#   krb_server_keyfile: /var/spool/keytabs/postgres
     unix_socket_directories: '.'
 
 #watchdog:

postgres1.yml

@@ -17,7 +17,18 @@ 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
 
 bootstrap:
   # this section will be written into Etcd:/<namespace>/<scope>/config after initializing new cluster
@@ -49,6 +60,9 @@ bootstrap:
   - data-checksums
 
   pg_hba:  # Add following lines to pg_hba.conf after running 'initdb'
+  # For kerberos gss based connectivity (discard @.*$)
+  #- host replication replicator 127.0.0.1/32 gss include_realm=0
+  #- host all all 0.0.0.0/0 gss include_realm=0
   - host replication replicator 127.0.0.1/32 md5
   - host all all 0.0.0.0/0 md5
 #  - hostssl all all 0.0.0.0/0 md5
@@ -78,7 +92,15 @@ postgresql:
     superuser:
       username: postgres
       password: zalando
+    rewind:  # Has no effect on postgres 10 and lower
+      username: rewind_user
+      password: rewind_password
+  # Server side kerberos spn
+#  krbsrvname: postgres
   parameters:
+    # Fully qualified kerberos ticket file for the running user
+    # same as KRB5CCNAME used by the GSS
+#    krb_server_keyfile: /var/spool/keytabs/postgres
     unix_socket_directories: '.'
   basebackup:
       - verbose

postgres2.yml

@@ -17,7 +17,18 @@ 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
 
 bootstrap:
   # this section will be written into Etcd:/<namespace>/<scope>/config after initializing new cluster
@@ -49,6 +60,9 @@ bootstrap:
   - data-checksums
 
   pg_hba:  # Add following lines to pg_hba.conf after running 'initdb'
+  # For kerberos gss based connectivity (discard @.*$)
+  #- host replication replicator 127.0.0.1/32 gss include_realm=0
+  #- host all all 0.0.0.0/0 gss include_realm=0
   - host replication replicator 127.0.0.1/32 md5
   - host all all 0.0.0.0/0 md5
 #  - hostssl all all 0.0.0.0/0 md5
@@ -75,7 +89,15 @@ postgresql:
     superuser:
       username: postgres
       password: zalando
+    rewind:  # Has no effect on postgres 10 and lower
+      username: rewind_user
+      password: rewind_password
+  # Server side kerberos spn
+#  krbsrvname: postgres
   parameters:
+    # Fully qualified kerberos ticket file for the running user
+    # same as KRB5CCNAME used by the GSS
+#    krb_server_keyfile: /var/spool/keytabs/postgres
     unix_socket_directories: '.'
 tags:
     nofailover: false

requirements.txt

@@ -1,16 +1,13 @@
 urllib3>=1.19.1,!=1.21
 boto
-psycopg2>=2.5.4
 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
+texttable
 python-dateutil
-psutil
+psutil>=2.0.0
 cdiff
-kubernetes>=2.0.0,<=7.0.0,!=4.0.*,!=5.0.*
+kubernetes>=2.0.0,<=10.0.1,!=4.0.*,!=5.0.*

setup.py

@@ -8,27 +8,12 @@ import inspect
 import os
 import sys
 
-from setuptools.command.test import test as TestCommand
-from setuptools import find_packages, setup
-
-if sys.version_info < (2, 7, 0):
-    sys.stderr.write('FATAL: patroni needs to be run with Python 2.7+\n')
-    sys.exit(1)
+from setuptools import Command, find_packages, setup
 
 __location__ = os.path.join(os.getcwd(), os.path.dirname(inspect.getfile(inspect.currentframe())))
 
-
-def read_version(package):
-    data = {}
-    with open(os.path.join(package, 'version.py'), 'r') as fd:
-        exec(fd.read(), data)
-    return data['__version__']
-
-
 NAME = 'patroni'
 MAIN_PACKAGE = NAME
-SCRIPTS = 'scripts'
-VERSION = read_version(MAIN_PACKAGE)
 DESCRIPTION = 'PostgreSQL High-Available orchestrator and CLI'
 LICENSE = 'The MIT License'
 URL = 'https://github.com/zalando/patroni'
@@ -37,9 +22,10 @@ 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'], 'consul': ['python-consul'],
+                  'exhibitor': ['kazoo'], 'zookeeper': ['kazoo'], 'kubernetes': ['kubernetes']}
 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
@@ -69,82 +55,123 @@ CONSOLE_SCRIPTS = ['patroni = patroni: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=',
-                    None, 'Generate junit html report'), ('junitxml=', None, 'Generate xml of test results')]
+    user_options = []
 
     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'):
-            os.environ['SYSTEMROOT'] = '/'
-        errno = pytest.main(**params)
+
+        args = ['--verbose', 'tests', '--doctest-modules', MAIN_PACKAGE] +\
+            ['-s' if logging.getLogger().getEffectiveLevel() < silence else '--capture=fd']
+        if self.cov:
+            args += self.cov
+
+        errno = pytest.main(args=args)
         sys.exit(errno)
 
-
-def get_install_requirements(path):
-    content = open(os.path.join(__location__, path)).read()
-    return [req for req in content.split('\n') if req != '']
+    def run(self):
+        from pkg_resources import evaluate_marker
+        requirements = self.distribution.install_requires + ['mock>=2.0.0', 'pytest-cov', 'pytest'] +\
+            [v for k, v in self.distribution.extras_require.items() if not k.startswith(':') or evaluate_marker(k[1:])]
+        self.distribution.fetch_build_eggs(requirements)
+        self.run_tests()
 
 
 def read(fname):
-    return open(os.path.join(__location__, fname)).read()
+    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}
-
-    # Some helper variables
-    version = os.getenv('GO_PIPELINE_LABEL', VERSION)
+    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 get_install_requirements('requirements.txt'):
+    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, v in EXTRAS_REQUIRE.items():
             if r.startswith(v[0]):
-                extras_require[e] = [r]
+                EXTRAS_REQUIRE[e] = [r]
                 extra = True
         if not extra:
             install_requires.append(r)
 
-    command_options = {'test': {'test_suite': ('setup.py', 'tests')}}
-    if JUNIT_XML:
-        command_options['test']['junitxml'] = 'setup.py', 'junit.xml'
+    command_options = {'test': {}}
     if COVERAGE_XML:
         command_options['test']['cov_xml'] = 'setup.py', True
     if COVERAGE_HTML:
@@ -161,17 +188,28 @@ 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

@@ -0,0 +1,205 @@
+import os
+import shutil
+import unittest
+
+from mock import Mock, patch
+
+import psycopg2
+import urllib3
+
+from patroni.dcs import Leader, Member
+from patroni.postgresql import Postgresql
+from patroni.postgresql.config import ConfigHandler
+from patroni.utils import RetryFailedError
+
+
+class SleepException(Exception):
+    pass
+
+
+class MockResponse(object):
+
+    def __init__(self, status_code=200):
+        self.status_code = status_code
+        self.content = '{}'
+
+    @property
+    def data(self):
+        return self.content.encode('utf-8')
+
+    @property
+    def status(self):
+        return self.status_code
+
+    @staticmethod
+    def getheader(*args):
+        return ''
+
+
+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 urllib3.exceptions.HTTPError()
+    elif ':8011/patroni' in url:
+        response.content = '{"role": "replica", "xlog": {"received_location": 0}, "tags": {}}'
+    elif url.endswith('/members'):
+        response.content = '[{}]' if url.startswith('http://error') else members
+    elif url.startswith('http://exhibitor'):
+        response.content = '{"servers":["127.0.0.1","127.0.0.2","127.0.0.3"],"port":2181}'
+    elif url.endswith(':8011/reinitialize'):
+        data = kwargs.get('data', '')
+        if ' false}' in data:
+            response.status_code = 503
+            response.content = 'restarting after failure already in progress'
+    else:
+        response.status_code = 404
+    return response
+
+
+class MockPostmaster(object):
+    def __init__(self, is_running=True, is_single_master=False):
+        self.is_running = Mock(return_value=is_running)
+        self.is_single_master = Mock(return_value=is_single_master)
+        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):
+
+    def __init__(self, connection):
+        self.connection = connection
+        self.closed = False
+        self.rowcount = 0
+        self.results = []
+        self.description = [Mock()]
+
+    def execute(self, sql, *params):
+        if sql.startswith('blabla'):
+            raise psycopg2.ProgrammingError()
+        elif sql == 'CHECKPOINT' or sql.startswith('SELECT pg_catalog.pg_create_'):
+            raise psycopg2.OperationalError()
+        elif sql.startswith('RetryFailedError'):
+            raise RetryFailedError('retry')
+        elif sql.startswith('SELECT slot_name'):
+            self.results = [('blabla', 'physical'), ('foobar', 'physical'), ('ls', 'logical', 'a', 'b')]
+        elif sql.startswith('SELECT CASE WHEN pg_catalog.pg_is_in_recovery()'):
+            self.results = [(1, 2, 1, 0, False, 1, 1, None, None)]
+        elif sql.startswith('SELECT pg_catalog.pg_is_in_recovery()'):
+            self.results = [(False, 2)]
+        elif sql.startswith('SELECT pg_catalog.to_char'):
+            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)]
+        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', 3, '0/402EEC0', '')]
+        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'
+                                 b'2\t0/402DD98\tno recovery target specified\n\n'
+                                 b'3\t0/403DD98\tno recovery target specified\n')]
+        else:
+            self.results = [(None, None, None, None, None, None, None, None, None, None)]
+
+    def fetchone(self):
+        return self.results[0]
+
+    def fetchall(self):
+        return self.results
+
+    def __iter__(self):
+        for i in self.results:
+            yield i
+
+    def __enter__(self):
+        return self
+
+    def __exit__(self, *args):
+        pass
+
+
+class MockConnect(object):
+
+    server_version = 99999
+    autocommit = False
+    closed = 0
+
+    def cursor(self):
+        return MockCursor(self)
+
+    def __enter__(self):
+        return self
+
+    def __exit__(self, *args):
+        pass
+
+    @staticmethod
+    def close():
+        pass
+
+
+def psycopg2_connect(*args, **kwargs):
+    return MockConnect()
+
+
+class PostgresInit(unittest.TestCase):
+    _PARAMETERS = {'wal_level': 'hot_standby', 'max_replication_slots': 5, 'f.oo': 'bar',
+                   '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',
+                   'stats_temp_directory': '/tmp'}
+
+    @patch('psycopg2.connect', psycopg2_connect)
+    @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 = 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(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'}},
+                             'remove_data_directory_on_rewind_failure': True,
+                             'use_pg_rewind': True, 'pg_ctl_timeout': 'bla',
+                             'parameters': self._PARAMETERS,
+                             'recovery_conf': {'foo': 'bar'},
+                             'pg_hba': ['host all all 0.0.0.0/0 md5'],
+                             'pg_ident': ['krb realm postgres'],
+                             'callbacks': {'on_start': 'true', 'on_stop': 'true', 'on_reload': 'true',
+                                           'on_restart': 'true', 'on_role_change': 'true'}})
+
+
+class BaseTestPostgresql(PostgresInit):
+
+    def setUp(self):
+        super(BaseTestPostgresql, self).setUp()
+
+        if not os.path.exists(self.p.data_dir):
+            os.makedirs(self.p.data_dir)
+
+        self.leadermem = Member(0, 'leader', 28, {'conn_url': 'postgres://replicator:[email protected]:5435/postgres'})
+        self.leader = Leader(-1, 28, self.leadermem)
+        self.other = Member(0, 'test-1', 28, {'conn_url': 'postgres://replicator:[email protected]:5433/postgres',
+                                              'tags': {'replicatefrom': 'leader'}})
+        self.me = Member(0, 'test0', 28, {'conn_url': 'postgres://replicator:[email protected]:5434/postgres'})
+
+    def tearDown(self):
+        if os.path.exists(self.p.data_dir):
+            shutil.rmtree(self.p.data_dir)

tests/test_api.py

@@ -2,6 +2,7 @@ import datetime
 import json
 import psycopg2
 import unittest
+import socket
 
 from mock import Mock, PropertyMock, patch
 from patroni.api import RestApiHandler, RestApiServer
@@ -10,7 +11,8 @@ from patroni.ha import _MemberStatus
 from patroni.utils import tzutc
 from six import BytesIO as IO
 from six.moves import BaseHTTPServer
-from test_postgresql import psycopg2_connect, MockCursor
+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)
@@ -28,6 +30,8 @@ class MockPostgresql(object):
     pending_restart = True
     wal_name = 'wal'
     lsn_name = 'lsn'
+    POSTMASTER_START_TIME = 'pg_catalog.to_char(pg_catalog.pg_postmaster_start_time'
+    TL_LSN = 'CASE WHEN pg_catalog.pg_is_in_recovery()'
 
     @staticmethod
     def connection():
@@ -60,7 +64,7 @@ class MockHa(object):
         return 'reinitialize'
 
     @staticmethod
-    def restart():
+    def restart(*args, **kwargs):
         return (True, '')
 
     @staticmethod
@@ -73,7 +77,7 @@ class MockHa(object):
 
     @staticmethod
     def fetch_nodes_statuses(members):
-        return [_MemberStatus(None, True, None, None, {}, False)]
+        return [_MemberStatus(None, True, None, 0, None, {}, False)]
 
     @staticmethod
     def schedule_future_restart(data):
@@ -95,6 +99,17 @@ class MockHa(object):
     def is_paused():
         return True
 
+    @staticmethod
+    def is_standby_cluster():
+        return False
+
+
+class MockLogger(object):
+
+    NORMAL_LOG_QUEUE_SIZE = 2
+    queue_size = 3
+    records_lost = 1
+
 
 class MockPatroni(object):
 
@@ -102,6 +117,7 @@ class MockPatroni(object):
     config = Mock()
     postgresql = ha.state_handler
     dcs = Mock()
+    logger = MockLogger()
     tags = {}
     version = '0.00'
     noloadbalance = PropertyMock(return_value=False)
@@ -130,25 +146,29 @@ class MockRestApiServer(RestApiServer):
     def __init__(self, Handler, request, config=None):
         self.socket = 0
         self.serve_forever = Mock()
-        BaseHTTPServer.HTTPServer.__init__ = 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'}
         super(MockRestApiServer, self).__init__(MockPatroni(), config)
         Handler(MockRequest(request), ('0.0.0.0', 8080), self)
 
 
-@patch('ssl.wrap_socket', Mock(return_value=0))
+@patch('ssl.SSLContext.load_cert_chain', Mock())
+@patch('ssl.SSLContext.wrap_socket', Mock(return_value=0))
+@patch.object(BaseHTTPServer.HTTPServer, '__init__', Mock())
 class TestRestApiHandler(unittest.TestCase):
 
     _authorization = '\nAuthorization: Basic dGVzdDp0ZXN0'
 
     def test_do_GET(self):
         MockRestApiServer(RestApiHandler, 'GET /replica')
+        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')
         MockPatroni.dcs.cluster.sync.sync_standby = MockPostgresql.name
         MockPatroni.dcs.cluster.is_synchronous_mode = Mock(return_value=True)
@@ -159,6 +179,8 @@ class TestRestApiHandler(unittest.TestCase):
             MockRestApiServer(RestApiHandler, 'GET /asynchronous')
         MockPatroni.ha.is_leader = Mock(return_value=True)
         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')
@@ -167,8 +189,8 @@ class TestRestApiHandler(unittest.TestCase):
         self.assertIsNotNone(MockRestApiServer(RestApiHandler, 'GET /master'))
         with patch.object(RestApiServer, 'query', Mock(return_value=[('', 1, '', '', '', '', False, '')])):
             self.assertIsNotNone(MockRestApiServer(RestApiHandler, 'GET /patroni'))
-        MockPatroni.config.is_standby_cluster = PropertyMock(return_value=True)
-        MockRestApiServer(RestApiHandler, 'GET /standby_leader')
+        with patch.object(MockHa, 'is_standby_cluster', Mock(return_value=True)):
+            MockRestApiServer(RestApiHandler, 'GET /standby_leader')
 
     def test_do_OPTIONS(self):
         self.assertIsNotNone(MockRestApiServer(RestApiHandler, 'OPTIONS / HTTP/1.0'))
@@ -181,6 +203,17 @@ 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 = {}
@@ -217,11 +250,8 @@ 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')
@@ -281,6 +311,15 @@ 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
@@ -387,13 +426,46 @@ class TestRestApiHandler(unittest.TestCase):
         MockRestApiServer(RestApiHandler, post + '37\n\n{"candidate":"2","scheduled_at": "1"}')
 
 
-@patch('ssl.wrap_socket', Mock(return_value=0))
 class TestRestApiServer(unittest.TestCase):
 
+    @patch('ssl.SSLContext.load_cert_chain', 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'})
+
+    @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, srv.reload_config, bad_config)
-        self.assertRaises(ValueError, srv.reload_config, {})
-        srv.reload_config({'listen': '127.0.0.2:8008'})
+        self.assertRaises(ValueError, self.srv.reload_config, bad_config)
+        self.assertRaises(ValueError, self.srv.reload_config, {})
+        with patch.object(socket.socket, 'setsockopt', Mock(side_effect=socket.error)):
+            self.srv.reload_config({'listen': ':8008'})
+
+    def test_check_auth(self):
+        mock_rh = Mock()
+        mock_rh.request.getpeercert.return_value = None
+        self.assertIsNot(self.srv.check_auth(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', Mock(return_value=(1, '2')))
+    def test_get_request(self):
+        self.srv.socket = Mock()
+        self.assertEqual(self.srv.get_request(), ((self.srv.socket, 1), '2'))

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

@@ -0,0 +1,227 @@
+import os
+
+from mock import Mock, PropertyMock, patch
+
+from patroni.async_executor import CriticalTask
+from patroni.postgresql import Postgresql
+from patroni.postgresql.bootstrap import Bootstrap
+from patroni.postgresql.cancellable import CancellableSubprocess
+from patroni.postgresql.config import ConfigHandler
+
+from . import psycopg2_connect, BaseTestPostgresql
+
+
+@patch('subprocess.call', Mock(return_value=0))
+@patch('psycopg2.connect', psycopg2_connect)
+@patch('os.rename', Mock())
+class TestBootstrap(BaseTestPostgresql):
+
+    @patch('patroni.postgresql.CallbackExecutor', Mock())
+    def setUp(self):
+        super(TestBootstrap, self).setUp()
+        self.b = self.p.bootstrap
+
+    @patch('time.sleep', Mock())
+    @patch.object(CancellableSubprocess, 'call')
+    @patch.object(Postgresql, 'remove_data_directory', Mock(return_value=True))
+    @patch.object(Postgresql, 'data_directory_empty', Mock(return_value=False))
+    @patch.object(Bootstrap, '_post_restore', Mock(side_effect=OSError))
+    def test_create_replica(self, mock_cancellable_subprocess_call):
+        self.p.config._config['create_replica_methods'] = ['pgBackRest']
+        self.p.config._config['pgBackRest'] = {'command': 'pgBackRest', 'keep_data': True, 'no_params': True}
+        mock_cancellable_subprocess_call.return_value = 0
+        self.assertEqual(self.b.create_replica(self.leader), 0)
+
+        self.p.config._config['create_replica_methods'] = ['basebackup']
+        self.p.config._config['basebackup'] = [{'max_rate': '100M'}, 'no-sync']
+        self.assertEqual(self.b.create_replica(self.leader), 0)
+
+        self.p.config._config['basebackup'] = [{'max_rate': '100M', 'compress': '9'}]
+        with patch('patroni.postgresql.bootstrap.logger.error', new_callable=Mock()) as mock_logger:
+            self.b.create_replica(self.leader)
+            mock_logger.assert_called_once()
+            self.assertTrue("only one key-value is allowed and value should be a string" in mock_logger.call_args[0][0],
+                            "not matching {0}".format(mock_logger.call_args[0][0]))
+
+        self.p.config._config['basebackup'] = [42]
+        with patch('patroni.postgresql.bootstrap.logger.error', new_callable=Mock()) as mock_logger:
+            self.b.create_replica(self.leader)
+            mock_logger.assert_called_once()
+            self.assertTrue("value should be string value or a single key-value pair" in mock_logger.call_args[0][0],
+                            "not matching {0}".format(mock_logger.call_args[0][0]))
+
+        self.p.config._config['basebackup'] = {"foo": "bar"}
+        self.assertEqual(self.b.create_replica(self.leader), 0)
+
+        self.p.config._config['create_replica_methods'] = ['wale', 'basebackup']
+        del self.p.config._config['basebackup']
+        mock_cancellable_subprocess_call.return_value = 1
+        self.assertEqual(self.b.create_replica(self.leader), 1)
+
+        mock_cancellable_subprocess_call.side_effect = Exception('foo')
+        self.assertEqual(self.b.create_replica(self.leader), 1)
+
+        mock_cancellable_subprocess_call.side_effect = [1, 0]
+        self.assertEqual(self.b.create_replica(self.leader), 0)
+
+        mock_cancellable_subprocess_call.side_effect = [Exception(), 0]
+        self.assertEqual(self.b.create_replica(self.leader), 0)
+
+        self.p.cancellable.cancel()
+        self.assertEqual(self.b.create_replica(self.leader), 1)
+
+    @patch('time.sleep', Mock())
+    @patch.object(CancellableSubprocess, 'call')
+    @patch.object(Postgresql, 'remove_data_directory', Mock(return_value=True))
+    @patch.object(Bootstrap, '_post_restore', Mock(side_effect=OSError))
+    def test_create_replica_old_format(self, mock_cancellable_subprocess_call):
+        """ The same test as before but with old 'create_replica_method'
+            to test backward compatibility
+        """
+        self.p.config._config['create_replica_method'] = ['wale', 'basebackup']
+        self.p.config._config['wale'] = {'command': 'foo'}
+        mock_cancellable_subprocess_call.return_value = 0
+        self.assertEqual(self.b.create_replica(self.leader), 0)
+        del self.p.config._config['wale']
+        self.assertEqual(self.b.create_replica(self.leader), 0)
+
+        self.p.config._config['create_replica_method'] = ['wale']
+        mock_cancellable_subprocess_call.return_value = 1
+        self.assertEqual(self.b.create_replica(self.leader), 1)
+
+    def test_basebackup(self):
+        self.p.cancellable.cancel()
+        self.b.basebackup(None, None, {'foo': 'bar'})
+
+    def test__initdb(self):
+        self.assertRaises(Exception, self.b.bootstrap, {'initdb': [{'pgdata': 'bar'}]})
+        self.assertRaises(Exception, self.b.bootstrap, {'initdb': [{'foo': 'bar', 1: 2}]})
+        self.assertRaises(Exception, self.b.bootstrap, {'initdb': [1]})
+        self.assertRaises(Exception, self.b.bootstrap, {'initdb': 1})
+
+    @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({}))
+
+        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.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()
+            self.assertTrue('host all all 0.0.0.0/0 md5\n' in lines)
+
+        self.p.config._config.pop('pg_hba')
+        config.update({'post_init': '/bin/false',
+                       'pg_hba': ['host replication replicator 127.0.0.1/32 md5',
+                                  'hostssl all all 0.0.0.0/0 md5',
+                                  'host all all 0.0.0.0/0 md5']})
+        self.b.bootstrap(config)
+        with open(os.path.join(self.p.data_dir, 'pg_hba.conf')) as f:
+            lines = f.readlines()
+            self.assertTrue('host replication replicator 127.0.0.1/32 md5\n' in lines)
+
+    @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'}}
+
+        mock_cancellable_subprocess_call.return_value = 1
+        self.assertFalse(self.b.bootstrap(config))
+
+        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()),\
+                patch.object(ConfigHandler, 'restore_configuration_files', Mock()),\
+                patch.object(ConfigHandler, 'write_recovery_conf', Mock()):
+            with self.assertRaises(Exception) as e:
+                self.b.bootstrap(config)
+            self.assertEqual(str(e.exception), '42')
+
+            config['foo']['recovery_conf'] = {'foo': 'bar'}
+
+            with self.assertRaises(Exception) as e:
+                self.b.bootstrap(config)
+            self.assertEqual(str(e.exception), '42')
+
+        mock_cancellable_subprocess_call.side_effect = Exception
+        self.assertFalse(self.b.bootstrap(config))
+
+    @patch('time.sleep', Mock())
+    @patch('os.unlink', Mock())
+    @patch('shutil.copy', Mock())
+    @patch('os.path.isfile', Mock(return_value=True))
+    @patch.object(Bootstrap, 'call_post_bootstrap', Mock(return_value=True))
+    @patch.object(Bootstrap, '_custom_bootstrap', Mock(return_value=True))
+    @patch.object(Postgresql, 'start', Mock(return_value=True))
+    @patch.object(Postgresql, 'get_major_version', Mock(return_value=110000))
+    def test_post_bootstrap(self):
+        config = {'method': 'foo', 'foo': {'command': 'bar'}}
+        self.b.bootstrap(config)
+
+        task = CriticalTask()
+        with patch.object(Bootstrap, 'create_or_update_role', Mock(side_effect=Exception)):
+            self.b.post_bootstrap({}, task)
+            self.assertFalse(task.result)
+
+        self.p.config._config.pop('pg_hba')
+        self.b.post_bootstrap({}, task)
+        self.assertTrue(task.result)
+
+        self.b.bootstrap(config)
+        with patch.object(Postgresql, 'pending_restart', PropertyMock(return_value=True)), \
+                patch.object(Postgresql, 'restart', Mock()) as mock_restart:
+            self.b.post_bootstrap({}, task)
+            mock_restart.assert_called_once()
+
+        self.b.bootstrap(config)
+        self.p.set_state('stopped')
+        self.p.reload_config({'authentication': {'superuser': {'username': 'p', 'password': 'p'},
+                                                 'replication': {'username': 'r', 'password': 'r'},
+                                                 'rewind': {'username': 'rw', 'password': 'rw'}},
+                              'listen': '*', 'retry_timeout': 10, 'parameters': {'wal_level': '', 'hba_file': 'foo'}})
+        with patch.object(Postgresql, 'major_version', PropertyMock(return_value=110000)), \
+                patch.object(Postgresql, 'restart', Mock()) as mock_restart:
+            self.b.post_bootstrap({}, task)
+            mock_restart.assert_called_once()
+
+    @patch.object(CancellableSubprocess, 'call')
+    def test_call_post_bootstrap(self, mock_cancellable_subprocess_call):
+        mock_cancellable_subprocess_call.return_value = 1
+        self.assertFalse(self.b.call_post_bootstrap({'post_init': '/bin/false'}))
+
+        mock_cancellable_subprocess_call.return_value = 0
+        self.p.config.superuser.pop('username')
+        self.assertTrue(self.b.call_post_bootstrap({'post_init': '/bin/false'}))
+        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', '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', 'dbname=postgres port=5432'])
+
+        mock_cancellable_subprocess_call.side_effect = OSError
+        self.assertFalse(self.b.call_post_bootstrap({'post_init': '/bin/false'}))
+
+    @patch('os.path.exists', Mock(return_value=True))
+    @patch('os.unlink', Mock())
+    @patch.object(Bootstrap, 'create_replica', Mock(return_value=0))
+    def test_clone(self):
+        self.b.clone(self.leader)

tests/test_callback_executor.py

@@ -1,24 +1,34 @@
+import psutil
 import unittest
 
 from mock import Mock, patch
-from patroni.callback_executor import CallbackExecutor
+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.poll.return_value = None
+        mock_popen.return_value.children.return_value = []
+        mock_popen.return_value.is_running.return_value = True
 
         ce = CallbackExecutor()
+        ce._kill_children = Mock(side_effect=Exception)
         self.assertIsNone(ce.call([]))
         ce.join()
 
         self.assertIsNone(ce.call([]))
 
+        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])
         self.assertIsNone(ce.call([]))
         ce.join()

tests/test_cancellable.py

@@ -0,0 +1,34 @@
+import psutil
+import unittest
+
+from mock import Mock, patch
+from patroni.exceptions import PostgresException
+from patroni.postgresql.cancellable import CancellableSubprocess
+
+
+class TestCancellableSubprocess(unittest.TestCase):
+
+    def setUp(self):
+        self.c = CancellableSubprocess()
+
+    def test_call(self):
+        self.c.cancel()
+        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.is_running.return_value = True
+        self.c._process.children.side_effect = psutil.Error()
+        self.c._process.suspend.side_effect = psutil.Error()
+        self.c.cancel()
+        self.c._process.is_running.side_effect = [True, False]
+        self.c.cancel()

tests/test_config.py

@@ -1,6 +1,6 @@
 import os
-import unittest
 import sys
+import unittest
 
 from mock import MagicMock, Mock, patch
 from patroni.config import Config
@@ -15,10 +15,7 @@ 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()
-
-    def test_no_config(self):
-        self.assertRaises(SystemExit, Config)
+        self.config = Config(None)
 
     def test_set_dynamic_configuration(self):
         with patch.object(Config, '_build_effective_configuration', Mock(side_effect=Exception)):
@@ -30,6 +27,8 @@ class TestConfig(unittest.TestCase):
             'PATRONI_NAME': 'postgres0',
             'PATRONI_NAMESPACE': '/patroni/',
             'PATRONI_SCOPE': 'batman2',
+            'PATRONI_LOGLEVEL': 'ERROR',
+            'PATRONI_LOG_LOGGERS': 'patroni.postmaster: WARNING, urllib3: DEBUG',
             'PATRONI_RESTAPI_USERNAME': 'username',
             'PATRONI_RESTAPI_PASSWORD': 'password',
             'PATRONI_RESTAPI_LISTEN': '0.0.0.0:8008',
@@ -49,7 +48,8 @@ class TestConfig(unittest.TestCase):
             'PATRONI_ETCD_CERT': '/cert',
             'PATRONI_ETCD_KEY': '/key',
             'PATRONI_CONSUL_HOST': '127.0.0.1:8500',
-            'PATRONI_KUBERNETES_LABELS': 'a:b:c',
+            'PATRONI_CONSUL_REGISTER_SERVICE': 'on',
+            'PATRONI_KUBERNETES_LABELS': 'a: b: c',
             'PATRONI_KUBERNETES_SCOPE_LABEL': 'a',
             'PATRONI_KUBERNETES_PORTS': '[{"name": "postgresql"}]',
             'PATRONI_ZOOKEEPER_HOSTS': "'host1:2181','host2:2181'",
@@ -63,12 +63,10 @@ class TestConfig(unittest.TestCase):
             'PATRONI_admin_PASSWORD': 'admin',
             'PATRONI_admin_OPTIONS': 'createrole,createdb'
         })
-        sys.argv = ['patroni.py', 'postgres0.yml']
-        config = Config()
+        config = Config('postgres0.yml')
         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)
-            self.assertTrue(config.reload_local_configuration(True))
+                config.reload_local_configuration()
             self.assertTrue(config.reload_local_configuration())
             self.assertIsNone(config.reload_local_configuration())
 
@@ -84,3 +82,15 @@ class TestConfig(unittest.TestCase):
             self.config.save_cache()
         with patch('os.fdopen', MagicMock()):
             self.config.save_cache()
+
+    def test_standby_cluster_parameters(self):
+        dynamic_configuration = {
+            'standby_cluster': {
+                'create_replica_methods': ['wal_e', 'basebackup'],
+                'host': 'localhost',
+                'port': 5432
+            }
+        }
+        self.config.set_dynamic_configuration(dynamic_configuration)
+        for name, value in dynamic_configuration['standby_cluster'].items():
+            self.assertEqual(self.config['standby_cluster'][name], value)

tests/test_consul.py

@@ -4,8 +4,8 @@ import unittest
 from consul import ConsulException, NotFound
 from mock import Mock, patch
 from patroni.dcs.consul import AbstractDCS, Cluster, Consul, ConsulInternalError, \
-                                ConsulError, HTTPClient, InvalidSessionTTL
-from test_etcd import SleepException
+                                ConsulError, HTTPClient, InvalidSessionTTL, InvalidSession
+from . import SleepException
 
 
 def kv_get(self, key, **kwargs):
@@ -52,6 +52,8 @@ class TestHTTPClient(unittest.TestCase):
         self.assertRaises(ConsulInternalError, self.client.get, Mock(), '')
         self.client.http.request.return_value.data = b"Invalid Session TTL '3000000000', must be between [10s=24h0m0s]"
         self.assertRaises(InvalidSessionTTL, self.client.get, Mock(), '')
+        self.client.http.request.return_value.data = b"invalid session '16492f43-c2d6-5307-432f-e32d6f7bcbd0'"
+        self.assertRaises(InvalidSession, self.client.get, Mock(), '')
 
     def test_unknown_method(self):
         try:
@@ -81,7 +83,7 @@ class TestConsul(unittest.TestCase):
         self.c = Consul({'ttl': 30, 'scope': 'test', 'name': 'postgresql1', 'host': 'localhost:1', 'retry_timeout': 10,
                          'register_service': True})
         self.c._base_path = '/service/good'
-        self.c._load_cluster()
+        self.c.get_cluster()
 
     @patch('time.sleep', Mock(side_effect=SleepException))
     @patch.object(consul.Consul.Session, 'create', Mock(side_effect=ConsulException))
@@ -110,19 +112,20 @@ class TestConsul(unittest.TestCase):
         self.c._session = 'fd4f44fe-2cac-bba5-a60b-304b51ff39b8'
         self.assertIsInstance(self.c.get_cluster(), Cluster)
 
-    @patch.object(consul.Consul.KV, 'delete', Mock(side_effect=[ConsulException, True, True]))
-    @patch.object(consul.Consul.KV, 'put', Mock(side_effect=[True, ConsulException]))
+    @patch.object(consul.Consul.KV, 'delete', Mock(side_effect=[ConsulException, True, True, True]))
+    @patch.object(consul.Consul.KV, 'put', Mock(side_effect=[True, ConsulException, InvalidSession]))
     def test_touch_member(self):
-        self.c._register_service = True
-        self.c.refresh_session = Mock(return_value=True)
-        self.c.touch_member({'balbla': 'blabla'})
-        self.c.touch_member({'balbla': 'blabla'})
-        self.c.touch_member({'balbla': 'blabla'})
         self.c.refresh_session = Mock(return_value=False)
         self.c.touch_member({'conn_url': 'postgres://replicator:[email protected]:5433/postgres',
                              'api_url': 'http://127.0.0.1:8009/patroni'})
+        self.c._register_service = True
+        self.c.refresh_session = Mock(return_value=True)
+        for _ in range(0, 4):
+            self.c.touch_member({'balbla': 'blabla'})
+        self.c.refresh_session = Mock(side_effect=ConsulError('foo'))
+        self.assertFalse(self.c.touch_member({'balbla': 'blabla'}))
 
-    @patch.object(consul.Consul.KV, 'put', Mock(return_value=False))
+    @patch.object(consul.Consul.KV, 'put', Mock(side_effect=InvalidSession))
     def test_take_leader(self):
         self.c.set_ttl(20)
         self.c.refresh_session = Mock()
@@ -196,3 +199,6 @@ class TestConsul(unittest.TestCase):
         d['state'] = 'running'
         d['role'] = 'bla'
         self.assertIsNone(self.c.update_service({}, d))
+
+    def test_reload_config(self):
+        self.c.reload_config({'consul': {'token': 'foo'}, 'loop_wait': 10, 'ttl': 30, 'retry_timeout': 10})

tests/test_ctl.py

@@ -1,40 +1,38 @@
 import etcd
 import os
-import requests
-import sys
 import unittest
 
 from click.testing import CliRunner
 from datetime import datetime, timedelta
 from mock import patch, Mock
-from patroni.ctl import ctl, store_config, load_config, output_members, request_patroni, get_dcs, parse_dcs, \
-    get_all_members, get_any_member, get_cursor, query_member, configure, PatroniCtlException, apply_config_changes, \
-    format_config_for_editing, show_diff, invoke_editor, format_pg_version
+from patroni.ctl import ctl, store_config, load_config, output_members, get_dcs, parse_dcs, get_all_members, \
+    get_any_member, get_cursor, query_member, configure, PatroniCtlException, apply_config_changes, show_diff, \
+    format_config_for_editing, invoke_editor, format_pg_version, find_executable, print_output, CONFIG_FILE_PATH
 from patroni.dcs.etcd import Client, Failover
 from patroni.utils import tzutc
 from psycopg2 import OperationalError
-from test_etcd import etcd_read, requests_get, socket_getaddrinfo, MockResponse
-from test_ha import get_cluster_initialized_without_leader, get_cluster_initialized_with_leader, \
-    get_cluster_initialized_with_only_leader, get_cluster_not_initialized_without_leader, get_cluster, Member
-from test_postgresql import MockConnect, psycopg2_connect
+from urllib3 import PoolManager
 
-CONFIG_FILE_PATH = './test-ctl.yaml'
+from . import MockConnect, MockCursor, MockResponse, psycopg2_connect
+from .test_etcd import etcd_read, socket_getaddrinfo
+from .test_ha import get_cluster_initialized_without_leader, get_cluster_initialized_with_leader, \
+    get_cluster_initialized_with_only_leader, get_cluster_not_initialized_without_leader, get_cluster, Member
 
 
 def test_rw_config():
     runner = CliRunner()
     with runner.isolated_filesystem():
-        store_config({'etcd': {'host': 'localhost:2379'}}, CONFIG_FILE_PATH + '/dummy')
-        sys.argv = ['patronictl.py', '']
-        load_config(CONFIG_FILE_PATH + '/dummy', None)
-        load_config(CONFIG_FILE_PATH + '/dummy', '0.0.0.0')
-        os.remove(CONFIG_FILE_PATH + '/dummy')
-        os.rmdir(CONFIG_FILE_PATH)
+        load_config(CONFIG_FILE_PATH, None)
+        CONFIG_PATH = './test-ctl.yaml'
+        store_config({'etcd': {'host': 'localhost:2379'}}, CONFIG_PATH + '/dummy')
+        load_config(CONFIG_PATH + '/dummy', '0.0.0.0')
+        os.remove(CONFIG_PATH + '/dummy')
+        os.rmdir(CONFIG_PATH)
 
 
-@patch('patroni.ctl.load_config',
-       Mock(return_value={'scope': 'alpha', 'postgresql': {'data_dir': '.', 'parameters': {}, 'retry_timeout': 5},
-                          'restapi': {'auth': 'u:p', 'listen': ''}, 'etcd': {'host': 'localhost:2379'}}))
+@patch('patroni.ctl.load_config', Mock(return_value={
+    'scope': 'alpha', 'restapi': {'listen': '::', 'certfile': 'a'}, 'etcd': {'host': 'localhost:2379'},
+    'postgresql': {'data_dir': '.', 'pgpass': './pgpass', 'parameters': {}, 'retry_timeout': 5}}))
 class TestCtl(unittest.TestCase):
 
     @patch('socket.getaddrinfo', socket_getaddrinfo)
@@ -44,6 +42,12 @@ class TestCtl(unittest.TestCase):
             self.runner = CliRunner()
             self.e = get_dcs({'etcd': {'ttl': 30, 'host': 'ok:2379', 'retry_timeout': 10}}, 'foo')
 
+    def test_load_config(self):
+        runner = CliRunner()
+        with runner.isolated_filesystem():
+            self.assertRaises(PatroniCtlException, load_config, './non-existing-config-file', None)
+            self.assertRaises(PatroniCtlException, load_config, './non-existing-config-file', None)
+
     @patch('psycopg2.connect', psycopg2_connect)
     def test_get_cursor(self):
         self.assertIsNone(get_cursor(get_cluster_initialized_without_leader(), {}, role='master'))
@@ -68,13 +72,12 @@ class TestCtl(unittest.TestCase):
     def test_output_members(self):
         scheduled_at = datetime.now(tzutc) + timedelta(seconds=600)
         cluster = get_cluster_initialized_with_leader(Failover(1, 'foo', 'bar', scheduled_at))
-        self.assertIsNone(output_members(cluster, name='abc', fmt='pretty'))
-        self.assertIsNone(output_members(cluster, name='abc', fmt='json'))
-        self.assertIsNone(output_members(cluster, name='abc', fmt='yaml'))
-        self.assertIsNone(output_members(cluster, name='abc', fmt='tsv'))
+        del cluster.members[1].data['conn_url']
+        for fmt in ('pretty', 'json', 'yaml', 'tsv', 'topology'):
+            self.assertIsNone(output_members(cluster, name='abc', fmt=fmt))
 
     @patch('patroni.ctl.get_dcs')
-    @patch('patroni.ctl.request_patroni', Mock(return_value=MockResponse()))
+    @patch.object(PoolManager, 'request', Mock(return_value=MockResponse()))
     def test_switchover(self, mock_get_dcs):
         mock_get_dcs.return_value = self.e
         mock_get_dcs.return_value.get_cluster = get_cluster_initialized_with_leader
@@ -89,10 +92,15 @@ class TestCtl(unittest.TestCase):
             result = self.runner.invoke(ctl, ['switchover', 'dummy', '--force', '--scheduled', '2015-01-01T12:00:00'])
             assert result.exit_code == 1
 
-        # Aborting switchover, as we anser NO to the confirmation
+        # Aborting switchover, as we answer NO to the confirmation
         result = self.runner.invoke(ctl, ['switchover', 'dummy'], input='leader\nother\n\nN')
         assert result.exit_code == 1
 
+        # Aborting scheduled switchover, as we answer NO to the confirmation
+        result = self.runner.invoke(ctl, ['switchover', 'dummy', '--scheduled', '2015-01-01T12:00:00+01:00'],
+                                    input='leader\nother\n\nN')
+        assert result.exit_code == 1
+
         # Target and source are equal
         result = self.runner.invoke(ctl, ['switchover', 'dummy'], input='leader\nleader\n\ny')
         assert result.exit_code == 1
@@ -119,18 +127,18 @@ class TestCtl(unittest.TestCase):
         result = self.runner.invoke(ctl, ['switchover', 'dummy'], input='dummy')
         assert result.exit_code == 1
 
-        with patch('patroni.ctl.request_patroni', Mock(side_effect=Exception)):
+        with patch.object(PoolManager, 'request', Mock(side_effect=Exception)):
             # Non-responding patroni
             result = self.runner.invoke(ctl, ['switchover', 'dummy'], input='leader\nother\n2300-01-01T12:23:00\ny')
             assert 'falling back to DCS' in result.output
 
-        with patch('patroni.ctl.request_patroni') as mocked:
-            mocked.return_value.status_code = 500
+        with patch.object(PoolManager, 'request') as mocked:
+            mocked.return_value.status = 500
             result = self.runner.invoke(ctl, ['switchover', 'dummy'], input='leader\nother\n\ny')
             assert 'Switchover failed' in result.output
 
-            mocked.return_value.status_code = 501
-            mocked.return_value.text = 'Server does not support this operation'
+            mocked.return_value.status = 501
+            mocked.return_value.data = b'Server does not support this operation'
             result = self.runner.invoke(ctl, ['switchover', 'dummy'], input='leader\nother\n\ny')
             assert 'Switchover failed' in result.output
 
@@ -145,7 +153,7 @@ class TestCtl(unittest.TestCase):
         assert result.exit_code == 1
 
     @patch('patroni.ctl.get_dcs')
-    @patch('patroni.ctl.request_patroni', Mock(return_value=MockResponse()))
+    @patch.object(PoolManager, 'request', Mock(return_value=MockResponse()))
     def test_failover(self, mock_get_dcs):
         mock_get_dcs.return_value = self.e
         mock_get_dcs.return_value.get_cluster = get_cluster_initialized_with_leader
@@ -153,9 +161,16 @@ class TestCtl(unittest.TestCase):
         result = self.runner.invoke(ctl, ['failover', 'dummy'], input='\n')
         assert 'Failover could be performed only to a specific candidate' in result.output
 
+    @patch('patroni.dcs.dcs_modules', Mock(return_value=['patroni.dcs.dummy', 'patroni.dcs.etcd']))
     def test_get_dcs(self):
         self.assertRaises(PatroniCtlException, get_dcs, {'dummy': {}}, 'dummy')
 
+    @patch('sys.platform', 'win32')
+    @patch('click.echo')
+    def test_print_output(self, mock_click_echo):
+        print_output(['a'], [])
+        mock_click_echo.assert_called_once_with('+---+\n| a |\n+---+')
+
     @patch('psycopg2.connect', psycopg2_connect)
     @patch('patroni.ctl.query_member', Mock(return_value=([['mock column']], None)))
     @patch('patroni.ctl.get_dcs')
@@ -192,24 +207,21 @@ class TestCtl(unittest.TestCase):
 
     def test_query_member(self):
         with patch('patroni.ctl.get_cursor', Mock(return_value=MockConnect().cursor())):
-            rows = query_member(None, None, None, 'master', 'SELECT pg_is_in_recovery()', {})
+            rows = query_member(None, None, None, 'master', 'SELECT pg_catalog.pg_is_in_recovery()', {})
             self.assertTrue('False' in str(rows))
 
-            rows = query_member(None, None, None, 'replica', 'SELECT pg_is_in_recovery()', {})
-            self.assertEqual(rows, (None, None))
-
-            with patch('test_postgresql.MockCursor.execute', Mock(side_effect=OperationalError('bla'))):
-                rows = query_member(None, None, None, 'replica', 'SELECT pg_is_in_recovery()', {})
+            with patch.object(MockCursor, 'execute', Mock(side_effect=OperationalError('bla'))):
+                rows = query_member(None, None, None, 'replica', 'SELECT pg_catalog.pg_is_in_recovery()', {})
 
         with patch('patroni.ctl.get_cursor', Mock(return_value=None)):
-            rows = query_member(None, None, None, None, 'SELECT pg_is_in_recovery()', {})
+            rows = query_member(None, None, None, None, 'SELECT pg_catalog.pg_is_in_recovery()', {})
             self.assertTrue('No connection to' in str(rows))
 
-            rows = query_member(None, None, None, 'replica', 'SELECT pg_is_in_recovery()', {})
+            rows = query_member(None, None, None, 'replica', 'SELECT pg_catalog.pg_is_in_recovery()', {})
             self.assertTrue('No connection to' in str(rows))
 
         with patch('patroni.ctl.get_cursor', Mock(side_effect=OperationalError('bla'))):
-            rows = query_member(None, None, None, 'replica', 'SELECT pg_is_in_recovery()', {})
+            rows = query_member(None, None, None, 'replica', 'SELECT pg_catalog.pg_is_in_recovery()', {})
 
     @patch('patroni.ctl.get_dcs')
     def test_dsn(self, mock_get_dcs):
@@ -225,7 +237,7 @@ class TestCtl(unittest.TestCase):
         result = self.runner.invoke(ctl, ['dsn', 'alpha', '--member', 'dummy'])
         assert result.exit_code == 1
 
-    @patch('requests.post')
+    @patch.object(PoolManager, 'request')
     @patch('patroni.ctl.get_dcs')
     def test_reload(self, mock_get_dcs, mock_post):
         mock_get_dcs.return_value.get_cluster = get_cluster_initialized_with_leader
@@ -233,19 +245,20 @@ class TestCtl(unittest.TestCase):
         result = self.runner.invoke(ctl, ['reload', 'alpha'], input='y')
         assert 'Failed: reload for member' in result.output
 
-        mock_post.return_value.status_code = 200
+        mock_post.return_value.status = 200
         result = self.runner.invoke(ctl, ['reload', 'alpha'], input='y')
         assert 'No changes to apply on member' in result.output
 
-        mock_post.return_value.status_code = 202
+        mock_post.return_value.status = 202
         result = self.runner.invoke(ctl, ['reload', 'alpha'], input='y')
         assert 'Reload request received for member' in result.output
 
-    @patch('requests.post', requests_get)
+    @patch.object(PoolManager, 'request')
     @patch('patroni.ctl.get_dcs')
-    def test_restart_reinit(self, mock_get_dcs):
+    def test_restart_reinit(self, mock_get_dcs, mock_post):
         mock_get_dcs.return_value.get_cluster = get_cluster_initialized_with_leader
-        result = self.runner.invoke(ctl, ['restart', 'alpha'], input='y\n\nnow')
+        mock_post.return_value.status = 503
+        result = self.runner.invoke(ctl, ['restart', 'alpha'], input='now\ny\n')
         assert 'Failed: restart for' in result.output
         assert result.exit_code == 0
 
@@ -257,76 +270,71 @@ class TestCtl(unittest.TestCase):
         assert result.exit_code == 0
 
         # Aborted restart
-        result = self.runner.invoke(ctl, ['restart', 'alpha'], input='N')
+        result = self.runner.invoke(ctl, ['restart', 'alpha'], input='now\nN')
         assert result.exit_code == 1
 
         result = self.runner.invoke(ctl, ['restart', 'alpha', '--pending', '--force'])
         assert result.exit_code == 0
 
+        # Aborted scheduled restart
+        result = self.runner.invoke(ctl, ['restart', 'alpha', '--scheduled', '2019-10-01T14:30'], input='N')
+        assert result.exit_code == 1
+
         # Not a member
-        result = self.runner.invoke(ctl, ['restart', 'alpha', 'dummy', '--any'], input='y')
+        result = self.runner.invoke(ctl, ['restart', 'alpha', 'dummy', '--any'], input='now\ny')
         assert result.exit_code == 1
 
         # Wrong pg version
-        result = self.runner.invoke(ctl, ['restart', 'alpha', '--any', '--pg-version', '9.1'], input='y')
+        result = self.runner.invoke(ctl, ['restart', 'alpha', '--any', '--pg-version', '9.1'], input='now\ny')
         assert 'Error: Invalid PostgreSQL version format' in result.output
         assert result.exit_code == 1
 
         result = self.runner.invoke(ctl, ['restart', 'alpha', '--pending', '--force', '--timeout', '10min'])
         assert result.exit_code == 0
 
-        with patch('requests.delete', Mock(return_value=MockResponse(500))):
-            # normal restart, the schedule is actually parsed, but not validated in patronictl
-            result = self.runner.invoke(ctl, ['restart', 'alpha', 'other', '--force',
-                                              '--scheduled', '2300-10-01T14:30'])
-            assert 'Failed: flush scheduled restart' in result.output
+        # normal restart, the schedule is actually parsed, but not validated in patronictl
+        result = self.runner.invoke(ctl, ['restart', 'alpha', 'other', '--force', '--scheduled', '2300-10-01T14:30'])
+        assert 'Failed: flush scheduled restart' in result.output
 
         with patch('patroni.dcs.Cluster.is_paused', Mock(return_value=True)):
             result = self.runner.invoke(ctl,
                                         ['restart', 'alpha', 'other', '--force', '--scheduled', '2300-10-01T14:30'])
             assert result.exit_code == 1
 
-        with patch('requests.post', Mock(return_value=MockResponse())):
-            # normal restart, the schedule is actually parsed, but not validated in patronictl
-            result = self.runner.invoke(ctl, ['restart', 'alpha', '--pg-version', '42.0.0',
-                                        '--scheduled', '2300-10-01T14:30'], input='y')
-            assert result.exit_code == 0
-
-        with patch('requests.post', Mock(return_value=MockResponse(204))):
-            # get restart with the non-200 return code
-            # normal restart, the schedule is actually parsed, but not validated in patronictl
-            result = self.runner.invoke(ctl, ['restart', 'alpha', '--pg-version', '42.0',
-                                        '--scheduled', '2300-10-01T14:30'], input='y')
-            assert result.exit_code == 0
-
         # force restart with restart already present
-        with patch('patroni.ctl.request_patroni', Mock(return_value=MockResponse(204))):
-            result = self.runner.invoke(ctl, ['restart', 'alpha', 'other', '--force',
-                                              '--scheduled', '2300-10-01T14:30'])
-            assert result.exit_code == 0
+        result = self.runner.invoke(ctl, ['restart', 'alpha', 'other', '--force', '--scheduled', '2300-10-01T14:30'])
+        assert result.exit_code == 0
 
-        with patch('requests.post', Mock(return_value=MockResponse(202))):
-            # get restart with the non-200 return code
-            # normal restart, the schedule is actually parsed, but not validated in patronictl
-            result = self.runner.invoke(
-                ctl, ['restart', 'alpha', '--pg-version', '99.0.0', '--scheduled', '2300-10-01T14:30'], input='y'
-            )
-            assert 'Success: restart scheduled' in result.output
-            assert result.exit_code == 0
+        ctl_args = ['restart', 'alpha', '--pg-version', '99.0', '--scheduled', '2300-10-01T14:30']
+        # normal restart, the schedule is actually parsed, but not validated in patronictl
+        mock_post.return_value.status = 200
+        result = self.runner.invoke(ctl, ctl_args, input='y')
+        assert result.exit_code == 0
 
-        with patch('requests.post', Mock(return_value=MockResponse(409))):
-            # get restart with the non-200 return code
-            # normal restart, the schedule is actually parsed, but not validated in patronictl
-            result = self.runner.invoke(
-                ctl, ['restart', 'alpha', '--pg-version', '99.0.0', '--scheduled', '2300-10-01T14:30'], input='y'
-            )
-            assert 'Failed: another restart is already' in result.output
-            assert result.exit_code == 0
+        # get restart with the non-200 return code
+        # normal restart, the schedule is actually parsed, but not validated in patronictl
+        mock_post.return_value.status = 204
+        result = self.runner.invoke(ctl, ctl_args, input='y')
+        assert result.exit_code == 0
+
+        # get restart with the non-200 return code
+        # normal restart, the schedule is actually parsed, but not validated in patronictl
+        mock_post.return_value.status = 202
+        result = self.runner.invoke(ctl, ctl_args, input='y')
+        assert 'Success: restart scheduled' in result.output
+        assert result.exit_code == 0
+
+        # get restart with the non-200 return code
+        # normal restart, the schedule is actually parsed, but not validated in patronictl
+        mock_post.return_value.status = 409
+        result = self.runner.invoke(ctl, ctl_args, input='y')
+        assert 'Failed: another restart is already' in result.output
+        assert result.exit_code == 0
 
     @patch('patroni.ctl.get_dcs')
     def test_remove(self, mock_get_dcs):
         mock_get_dcs.return_value.get_cluster = get_cluster_initialized_with_leader
-        result = self.runner.invoke(ctl, ['remove', 'alpha'], input='alpha\nslave')
+        result = self.runner.invoke(ctl, ['-k', 'remove', 'alpha'], input='alpha\nslave')
         assert 'Please confirm' in result.output
         assert 'You are about to remove all' in result.output
         # Not typing an exact confirmation
@@ -343,17 +351,6 @@ class TestCtl(unittest.TestCase):
         result = self.runner.invoke(ctl, ['remove', 'alpha'], input='alpha\nYes I am aware\nleader')
         assert result.exit_code == 0
 
-    @patch('requests.post', Mock(side_effect=requests.exceptions.ConnectionError('foo')))
-    @patch('click.get_current_context')
-    def test_request_patroni(self, mock_context):
-        member = get_cluster_initialized_with_leader().leader.member
-
-        mock_context.return_value.obj = {'ctl': {'cacert': 'cert.pem'}}
-        self.assertRaises(requests.exceptions.ConnectionError, request_patroni, member, 'post', 'dummy', {})
-
-        mock_context.return_value.obj = {'ctl': {'insecure': True}}
-        self.assertRaises(requests.exceptions.ConnectionError, request_patroni, member, 'post', 'dummy', {})
-
     def test_ctl(self):
         self.runner.invoke(ctl, ['list'])
 
@@ -427,8 +424,38 @@ class TestCtl(unittest.TestCase):
         assert 'Scheduled restart' in result.output
 
     @patch('patroni.ctl.get_dcs')
-    @patch('requests.delete', Mock(return_value=MockResponse()))
-    def test_flush(self, mock_get_dcs):
+    def test_topology(self, mock_get_dcs):
+        mock_get_dcs.return_value = self.e
+        cluster = get_cluster_initialized_with_leader()
+        cascade_member = Member(0, 'cascade', 28, {'conn_url': 'postgres://replicator:[email protected]:5437/postgres',
+                                                   'api_url': 'http://127.0.0.1:8012/patroni',
+                                                   'state': 'running',
+                                                   'tags': {'replicatefrom': 'other'},
+                                                   })
+        cascade_member_wrong_tags = Member(0, 'wrong_cascade', 28,
+                                           {'conn_url': 'postgres://replicator:[email protected]:5438/postgres',
+                                            'api_url': 'http://127.0.0.1:8013/patroni',
+                                            'state': 'running',
+                                            'tags': {'replicatefrom': 'nonexistinghost'},
+                                            })
+        cluster.members.append(cascade_member)
+        cluster.members.append(cascade_member_wrong_tags)
+        mock_get_dcs.return_value.get_cluster = Mock(return_value=cluster)
+        result = self.runner.invoke(ctl, ['topology', 'dummy'])
+        assert '+\n| leader          | 127.0.0.1:5435 |  Leader |' in result.output
+        assert '|\n| + other         | 127.0.0.1:5436 | Replica |' in result.output
+        assert '|\n|   + cascade     | 127.0.0.1:5437 | Replica |' in result.output
+        assert '|\n| + wrong_cascade | 127.0.0.1:5438 | Replica |' in result.output
+
+        cluster = get_cluster_initialized_without_leader()
+        mock_get_dcs.return_value.get_cluster = Mock(return_value=cluster)
+        result = self.runner.invoke(ctl, ['topology', 'dummy'])
+        assert '+\n| + leader | 127.0.0.1:5435 | Replica |' in result.output
+        assert '|\n| + other  | 127.0.0.1:5436 | Replica |' in result.output
+
+    @patch('patroni.ctl.get_dcs')
+    @patch.object(PoolManager, 'request', Mock(return_value=MockResponse()))
+    def test_flush_restart(self, mock_get_dcs):
         mock_get_dcs.return_value = self.e
         mock_get_dcs.return_value.get_cluster = get_cluster_initialized_with_leader
 
@@ -437,58 +464,78 @@ class TestCtl(unittest.TestCase):
 
         result = self.runner.invoke(ctl, ['flush', 'dummy', 'restart', '--force'])
         assert 'Success: flush scheduled restart' in result.output
-        with patch.object(requests, 'delete', return_value=MockResponse(404)):
+        with patch.object(PoolManager, 'request', return_value=MockResponse(404)):
             result = self.runner.invoke(ctl, ['flush', 'dummy', 'restart', '--force'])
             assert 'Failed: flush scheduled restart' in result.output
 
+    @patch('patroni.ctl.get_dcs')
+    @patch.object(PoolManager, 'request', Mock(return_value=MockResponse()))
+    def test_flush_switchover(self, mock_get_dcs):
+        mock_get_dcs.return_value = self.e
+
+        mock_get_dcs.return_value.get_cluster = get_cluster_initialized_with_leader
+        result = self.runner.invoke(ctl, ['flush', 'dummy', 'switchover'])
+        assert 'No pending scheduled switchover' in result.output
+
+        scheduled_at = datetime.now(tzutc) + timedelta(seconds=600)
+        mock_get_dcs.return_value.get_cluster = Mock(
+                return_value=get_cluster_initialized_with_leader(Failover(1, 'a', 'b', scheduled_at)))
+        result = self.runner.invoke(ctl, ['flush', 'dummy', 'switchover'])
+        assert result.output.startswith('Success: ')
+
+        mock_get_dcs.return_value.manual_failover = Mock()
+        with patch.object(PoolManager, 'request', side_effect=[MockResponse(409), Exception]):
+            result = self.runner.invoke(ctl, ['flush', 'dummy', 'switchover'])
+            assert 'Could not find any accessible member of cluster' in result.output
+
+    @patch.object(PoolManager, 'request')
     @patch('patroni.ctl.get_dcs')
     @patch('patroni.ctl.polling_loop', Mock(return_value=[1]))
-    def test_pause_cluster(self, mock_get_dcs):
+    def test_pause_cluster(self, mock_get_dcs, mock_post):
         mock_get_dcs.return_value = self.e
         mock_get_dcs.return_value.get_cluster = get_cluster_initialized_with_leader
 
-        with patch('requests.patch', Mock(return_value=MockResponse(500))):
-            result = self.runner.invoke(ctl, ['pause', 'dummy'])
-            assert 'Failed' in result.output
+        mock_post.return_value.status = 500
+        result = self.runner.invoke(ctl, ['pause', 'dummy'])
+        assert 'Failed' in result.output
 
-        with patch('requests.patch', Mock(return_value=MockResponse(200))),\
-                patch('patroni.dcs.Cluster.is_paused', Mock(return_value=True)):
+        mock_post.return_value.status = 200
+        with patch('patroni.dcs.Cluster.is_paused', Mock(return_value=True)):
             result = self.runner.invoke(ctl, ['pause', 'dummy'])
             assert 'Cluster is already paused' in result.output
 
-        with patch('requests.patch', Mock(return_value=MockResponse(200))):
-            result = self.runner.invoke(ctl, ['pause', 'dummy', '--wait'])
-            assert "'pause' request sent" in result.output
-            mock_get_dcs.return_value.get_cluster = Mock(side_effect=[get_cluster_initialized_with_leader(),
-                                                                      get_cluster(None, None, [], None, None)])
-            self.runner.invoke(ctl, ['pause', 'dummy', '--wait'])
-            member = Member(1, 'other', 28, {})
-            mock_get_dcs.return_value.get_cluster = Mock(side_effect=[get_cluster_initialized_with_leader(),
-                                                                      get_cluster(None, None, [member], None, None)])
-            self.runner.invoke(ctl, ['pause', 'dummy', '--wait'])
+        result = self.runner.invoke(ctl, ['pause', 'dummy', '--wait'])
+        assert "'pause' request sent" in result.output
+        mock_get_dcs.return_value.get_cluster = Mock(side_effect=[get_cluster_initialized_with_leader(),
+                                                                  get_cluster(None, None, [], None, None)])
+        self.runner.invoke(ctl, ['pause', 'dummy', '--wait'])
+        member = Member(1, 'other', 28, {})
+        mock_get_dcs.return_value.get_cluster = Mock(side_effect=[get_cluster_initialized_with_leader(),
+                                                                  get_cluster(None, None, [member], None, None)])
+        self.runner.invoke(ctl, ['pause', 'dummy', '--wait'])
 
+    @patch.object(PoolManager, 'request')
     @patch('patroni.ctl.get_dcs')
-    def test_resume_cluster(self, mock_get_dcs):
+    def test_resume_cluster(self, mock_get_dcs, mock_post):
         mock_get_dcs.return_value = self.e
         mock_get_dcs.return_value.get_cluster = get_cluster_initialized_with_leader
 
+        mock_post.return_value.status = 200
+        with patch('patroni.dcs.Cluster.is_paused', Mock(return_value=False)):
+            result = self.runner.invoke(ctl, ['resume', 'dummy'])
+            assert 'Cluster is not paused' in result.output
+
         with patch('patroni.dcs.Cluster.is_paused', Mock(return_value=True)):
-            with patch('requests.patch', Mock(return_value=MockResponse(200))):
-                result = self.runner.invoke(ctl, ['resume', 'dummy'])
-                assert 'Success' in result.output
+            result = self.runner.invoke(ctl, ['resume', 'dummy'])
+            assert 'Success' in result.output
 
-            with patch('requests.patch', Mock(return_value=MockResponse(500))):
-                result = self.runner.invoke(ctl, ['resume', 'dummy'])
-                assert 'Failed' in result.output
+            mock_post.return_value.status = 500
+            result = self.runner.invoke(ctl, ['resume', 'dummy'])
+            assert 'Failed' in result.output
 
-            with patch('requests.patch', Mock(return_value=MockResponse(200))),\
-                    patch('patroni.dcs.Cluster.is_paused', Mock(return_value=False)):
-                result = self.runner.invoke(ctl, ['resume', 'dummy'])
-                assert 'Cluster is not paused' in result.output
-
-            with patch('requests.patch', Mock(side_effect=Exception)):
-                result = self.runner.invoke(ctl, ['resume', 'dummy'])
-                assert 'Can not find accessible cluster member' in result.output
+            mock_post.side_effect = Exception
+            result = self.runner.invoke(ctl, ['resume', 'dummy'])
+            assert 'Can not find accessible cluster member' in result.output
 
     def test_apply_config_changes(self):
         config = {"postgresql": {"parameters": {"work_mem": "4MB"}, "use_pg_rewind": True}, "ttl": 30}
@@ -533,10 +580,12 @@ class TestCtl(unittest.TestCase):
         show_diff(b"foo:\n  bar: \xc3\xb6\xc3\xb6\n".decode('utf-8'),
                   b"foo:\n  bar: \xc3\xbc\xc3\xbc\n".decode('utf-8'))
 
-    def test_invoke_editor(self):
-        for e in ('', 'false'):
-            os.environ['EDITOR'] = e
-            self.assertRaises(PatroniCtlException, invoke_editor, 'foo: bar\n', 'test')
+    @patch('subprocess.call', return_value=1)
+    def test_invoke_editor(self, mock_subprocess_call):
+        os.environ.pop('EDITOR', None)
+        for e in ('', '/bin/vi'):
+            with patch('patroni.ctl.find_executable', Mock(return_value=e)):
+                self.assertRaises(PatroniCtlException, invoke_editor, 'foo: bar\n', 'test')
 
     @patch('patroni.ctl.get_dcs')
     def test_show_config(self, mock_get_dcs):
@@ -548,29 +597,64 @@ class TestCtl(unittest.TestCase):
     def test_edit_config(self, mock_get_dcs):
         mock_get_dcs.return_value = self.e
         mock_get_dcs.return_value.get_cluster = get_cluster_initialized_with_leader
+        mock_get_dcs.return_value.set_config_value = Mock(return_value=False)
         os.environ['EDITOR'] = 'true'
         self.runner.invoke(ctl, ['edit-config', 'dummy'])
         self.runner.invoke(ctl, ['edit-config', 'dummy', '-s', 'foo=bar'])
         self.runner.invoke(ctl, ['edit-config', 'dummy', '--replace', 'postgres0.yml'])
         self.runner.invoke(ctl, ['edit-config', 'dummy', '--apply', '-'], input='foo: bar')
         self.runner.invoke(ctl, ['edit-config', 'dummy', '--force', '--apply', '-'], input='foo: bar')
-        mock_get_dcs.return_value.set_config_value = Mock(return_value=True)
+        mock_get_dcs.return_value.set_config_value.return_value = True
         self.runner.invoke(ctl, ['edit-config', 'dummy', '--force', '--apply', '-'], input='foo: bar')
 
     @patch('patroni.ctl.get_dcs')
     def test_version(self, mock_get_dcs):
         mock_get_dcs.return_value = self.e
         mock_get_dcs.return_value.get_cluster = get_cluster_initialized_with_leader
-        with patch('patroni.ctl.request_patroni') as mocked:
+        with patch.object(PoolManager, 'request') as mocked:
             result = self.runner.invoke(ctl, ['version'])
             assert 'patronictl version' in result.output
-            mocked.return_value.json = lambda: {'patroni': {'version': '1.2.3'}, 'server_version': 100001}
+            mocked.return_value.data = b'{"patroni":{"version":"1.2.3"},"server_version": 100001}'
             result = self.runner.invoke(ctl, ['version', 'dummy'])
             assert '1.2.3' in result.output
-        with patch('requests.get', Mock(side_effect=Exception)):
+        with patch.object(PoolManager, 'request', Mock(side_effect=Exception)):
             result = self.runner.invoke(ctl, ['version', 'dummy'])
             assert 'failed to get version' in result.output
 
+    @patch('patroni.ctl.get_dcs')
+    def test_history(self, mock_get_dcs):
+        mock_get_dcs.return_value.get_cluster = Mock()
+        mock_get_dcs.return_value.get_cluster.return_value.history.lines = [[1, 67176, 'no recovery target specified']]
+        result = self.runner.invoke(ctl, ['history'])
+        assert 'Reason' in result.output
+
     def test_format_pg_version(self):
         self.assertEqual(format_pg_version(100001), '10.1')
         self.assertEqual(format_pg_version(90605), '9.6.5')
+
+    @patch('sys.platform', 'win32')
+    def test_find_executable(self):
+        with patch('os.path.isfile', Mock(return_value=True)):
+            self.assertEqual(find_executable('vim'), 'vim.exe')
+        with patch('os.path.isfile', Mock(return_value=False)):
+            self.assertIsNone(find_executable('vim'))
+        with patch('os.path.isfile', Mock(side_effect=[False, True])):
+            self.assertEqual(find_executable('vim', '/'), '/vim.exe')
+
+    @patch('patroni.ctl.get_dcs')
+    def test_get_members(self, mock_get_dcs):
+        mock_get_dcs.return_value = self.e
+        mock_get_dcs.return_value.get_cluster = get_cluster_not_initialized_without_leader
+        result = self.runner.invoke(ctl, ['reinit', 'dummy'])
+        assert "cluster doesn\'t have any members" in result.output
+
+    @patch('time.sleep', Mock())
+    @patch('patroni.ctl.get_dcs')
+    def test_reinit_wait(self, mock_get_dcs):
+        mock_get_dcs.return_value.get_cluster = get_cluster_initialized_with_leader
+        with patch.object(PoolManager, 'request') as mocked:
+            mocked.side_effect = [Mock(data=s, status=200) for s in
+                                  [b"reinitialize", b'{"state":"creating replica"}', b'{"state":"running"}']]
+            result = self.runner.invoke(ctl, ['reinit', 'alpha', 'other', '--wait'], input='y\ny')
+        self.assertIn("Waiting for reinitialize to complete on: other", result.output)
+        self.assertIn("Reinitialize is completed on: other", result.output)

tests/test_etcd.py

@@ -1,7 +1,5 @@
 import etcd
-import json
 import urllib3.util.connection
-import requests
 import socket
 import unittest
 
@@ -9,58 +7,10 @@ from dns.exception import DNSException
 from mock import Mock, patch
 from patroni.dcs.etcd import AbstractDCS, Client, Cluster, Etcd, EtcdError, DnsCachingResolver
 from patroni.exceptions import DCSError
+from patroni.utils import Retry
 from urllib3.exceptions import ReadTimeoutError
 
-
-class MockResponse(object):
-
-    def __init__(self, status_code=200):
-        self.status_code = status_code
-        self.content = '{}'
-        self.ok = True
-
-    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
-
-    @staticmethod
-    def getheader(*args):
-        return ''
-
-
-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()
-    elif ':8011/patroni' in url:
-        response.content = '{"role": "replica", "xlog": {"received_location": 0}, "tags": {}}'
-    elif url.endswith('/members'):
-        response.content = '[{}]' if url.startswith('http://error') else members
-    elif url.startswith('http://exhibitor'):
-        response.content = '{"servers":["127.0.0.1","127.0.0.2","127.0.0.3"],"port":2181}'
-    elif url.endswith(':8011/reinitialize'):
-        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
+from . import SleepException, MockResponse, requests_get
 
 
 def etcd_watch(self, key, index=None, timeout=None, recursive=None):
@@ -122,10 +72,6 @@ def etcd_read(self, key, **kwargs):
     return result
 
 
-class SleepException(Exception):
-    pass
-
-
 def dns_query(name, _):
     if '-server' not in name or '-ssl' in name:
         return []
@@ -148,13 +94,14 @@ def socket_getaddrinfo(*args):
 def http_request(method, url, **kwargs):
     if url == 'http://localhost:2379/timeout':
         raise ReadTimeoutError(None, None, None)
+    ret = MockResponse()
     if url == 'http://localhost:2379/v2/machines':
-        ret = MockResponse()
         ret.content = 'http://localhost:2379,http://localhost:4001'
-        return ret
-    if url == 'http://localhost:2379/':
-        return MockResponse()
-    raise socket.error
+    elif url == 'http://localhost:4001/v2/machines':
+        ret.content = ''
+    elif url != 'http://localhost:2379/':
+        raise socket.error
+    return ret
 
 
 class TestDnsCachingResolver(unittest.TestCase):
@@ -169,12 +116,12 @@ class TestDnsCachingResolver(unittest.TestCase):
 
 @patch('dns.resolver.query', dns_query)
 @patch('socket.getaddrinfo', socket_getaddrinfo)
-@patch('requests.get', requests_get)
+@patch('patroni.dcs.etcd.requests_get', requests_get)
 class TestClient(unittest.TestCase):
 
     @patch('dns.resolver.query', dns_query)
     @patch('socket.getaddrinfo', socket_getaddrinfo)
-    @patch('requests.get', requests_get)
+    @patch('patroni.dcs.etcd.requests_get', requests_get)
     def setUp(self):
         with patch.object(Client, 'machines') as mock_machines:
             mock_machines.__get__ = Mock(return_value=['http://localhost:2379', 'http://localhost:4001'])
@@ -183,12 +130,11 @@ class TestClient(unittest.TestCase):
             self.client.http.request_encode_body = http_request
 
     def test_machines(self):
-        self.client._base_uri = 'http://localhost:4001'
-        self.client._machines_cache = ['http://localhost:2379']
+        self.client._base_uri = 'http://localhost:4002'
+        self.client._machines_cache = ['http://localhost:4002', 'http://localhost:2379']
         self.assertIsNotNone(self.client.machines)
         self.client._base_uri = 'http://localhost:4001'
-        self.client._machines_cache = []
-        self.assertIsNotNone(self.client.machines)
+        self.client._machines_cache = ['http://localhost:4001']
         self.client._update_machines_cache = True
         machines = None
         try:
@@ -199,19 +145,31 @@ class TestClient(unittest.TestCase):
 
     @patch.object(Client, 'machines')
     def test_api_execute(self, mock_machines):
-        mock_machines.__get__ = Mock(return_value=['http://localhost:2379'])
+        mock_machines.__get__ = Mock(return_value=['http://localhost:4001', 'http://localhost:2379'])
         self.assertRaises(ValueError, self.client.api_execute, '', '')
         self.client._base_uri = 'http://localhost:4001'
-        self.client._machines_cache = ['http://localhost:2379']
-        self.client.api_execute('/', 'POST', timeout=0)
-        mock_machines.__get__ = Mock(return_value=['http://localhost:2379'])
+        self.assertRaises(etcd.EtcdException, self.client.api_execute, '/', 'POST', timeout=0)
+        self.client._base_uri = 'http://localhost:4001'
+        rtry = Retry(deadline=10, max_delay=1, max_tries=-1, retry_exceptions=(etcd.EtcdLeaderElectionInProgress,))
+        rtry(self.client.api_execute, '/', 'POST', timeout=0, params={'retry': rtry})
         self.client._machines_cache_updated = 0
         self.client.api_execute('/', 'POST', timeout=0)
         self.client._machines_cache = [self.client._base_uri]
         self.assertRaises(etcd.EtcdWatchTimedOut, self.client.api_execute, '/timeout', 'POST', params={'wait': 'true'})
+        self.assertRaises(etcd.EtcdWatchTimedOut, self.client.api_execute, '/timeout', 'POST', params={'wait': 'true'})
         self.assertRaises(etcd.EtcdException, self.client.api_execute, '/', '')
-        self.client._update_machines_cache = True
-        with patch.object(Client, '_load_machines_cache', Mock(side_effect=etcd.EtcdException)):
+
+        with patch.object(Client, '_calculate_timeouts', Mock(side_effect=[(1, 1, 0), (1, 1, 0), (0, 1, 0)])),\
+                patch.object(Client, '_load_machines_cache', Mock(side_effect=Exception)):
+            self.client.http.request = Mock(side_effect=socket.error)
+            self.assertRaises(etcd.EtcdException, rtry, self.client.api_execute, '/', 'GET', params={'retry': rtry})
+
+        with patch.object(Client, '_calculate_timeouts', Mock(side_effect=[(1, 1, 0), (1, 1, 0), (0, 1, 0)])),\
+                patch.object(Client, '_load_machines_cache', Mock(return_value=True)):
+            self.assertRaises(etcd.EtcdException, rtry, self.client.api_execute, '/', 'GET', params={'retry': rtry})
+
+        with patch.object(Client, '_do_http_request', Mock(side_effect=etcd.EtcdException)):
+            self.client._read_timeout = 0.01
             self.assertRaises(etcd.EtcdException, self.client.api_execute, '/', 'GET')
 
     def test_get_srv_record(self):
@@ -227,8 +185,12 @@ class TestClient(unittest.TestCase):
         self.client._get_machines_cache_from_dns('error', 2379)
 
     @patch.object(Client, 'machines')
-    def test__load_machines_cache(self, mock_machines):
-        mock_machines.__get__ = Mock(return_value=['http://localhost:2379'])
+    def test__refresh_machines_cache(self, mock_machines):
+        mock_machines.__get__ = Mock(side_effect=etcd.EtcdConnectionFailed)
+        self.assertIsNone(self.client._refresh_machines_cache())
+        self.assertRaises(etcd.EtcdException, self.client._refresh_machines_cache, True)
+
+    def test__load_machines_cache(self):
         self.client._config = {}
         self.assertRaises(Exception, self.client._load_machines_cache)
         self.client._config = {'srv': 'blabla'}
@@ -244,7 +206,7 @@ class TestClient(unittest.TestCase):
                           socket_options=[(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)])
 
 
-@patch('requests.get', requests_get)
+@patch('patroni.dcs.etcd.requests_get', requests_get)
 @patch('socket.getaddrinfo', socket_getaddrinfo)
 @patch.object(etcd.Client, 'write', etcd_write)
 @patch.object(etcd.Client, 'read', etcd_read)
@@ -263,17 +225,18 @@ class TestEtcd(unittest.TestCase):
 
     @patch('dns.resolver.query', dns_query)
     def test_get_etcd_client(self):
-        with patch.object(Client, 'machines') as mock_machines:
+        with patch('time.sleep', Mock(side_effect=SleepException)),\
+                patch.object(Client, 'machines') as mock_machines:
             mock_machines.__get__ = Mock(side_effect=etcd.EtcdException)
-            with patch('time.sleep', Mock(side_effect=SleepException)):
-                self.assertRaises(SleepException, self.etcd.get_etcd_client,
-                                  {'discovery_srv': 'test', 'retry_timeout': 10, 'cacert': '1', 'key': '1', 'cert': 1})
-                self.assertRaises(SleepException, self.etcd.get_etcd_client,
-                                  {'url': 'https://test:2379', 'retry_timeout': 10})
-                self.assertRaises(SleepException, self.etcd.get_etcd_client,
-                                  {'proxy': 'https://user:password@test:2379', 'retry_timeout': 10})
-                self.assertRaises(SleepException, self.etcd.get_etcd_client,
-                                  {'hosts': 'foo:4001,bar', 'retry_timeout': 10})
+            self.assertRaises(SleepException, self.etcd.get_etcd_client,
+                              {'discovery_srv': 'test', 'retry_timeout': 10, 'cacert': '1', 'key': '1', 'cert': 1})
+            self.assertRaises(SleepException, self.etcd.get_etcd_client,
+                              {'url': 'https://test:2379', 'retry_timeout': 10})
+            self.assertRaises(SleepException, self.etcd.get_etcd_client,
+                              {'hosts': 'foo:4001,bar', 'retry_timeout': 10})
+            mock_machines.__get__ = Mock(return_value=[])
+            self.assertRaises(SleepException, self.etcd.get_etcd_client,
+                              {'proxy': 'https://user:password@test:2379', 'retry_timeout': 10})
 
     def test_get_cluster(self):
         cluster = self.etcd.get_cluster()

tests/test_exhibitor.py

@@ -1,13 +1,15 @@
 import unittest
+import urllib3
 
 from mock import Mock, patch
 from patroni.dcs.exhibitor import ExhibitorEnsembleProvider, Exhibitor
 from patroni.dcs.zookeeper import ZooKeeperError
-from test_etcd import SleepException, requests_get
-from test_zookeeper import MockKazooClient
+
+from . import SleepException, requests_get
+from .test_zookeeper import MockKazooClient
 
 
-@patch('requests.get', requests_get)
+@patch('patroni.dcs.exhibitor.requests_get', requests_get)
 @patch('time.sleep', Mock(side_effect=SleepException))
 class TestExhibitorEnsembleProvider(unittest.TestCase):
 
@@ -20,7 +22,8 @@ class TestExhibitorEnsembleProvider(unittest.TestCase):
 
 class TestExhibitor(unittest.TestCase):
 
-    @patch('requests.get', requests_get)
+    @patch('urllib3.PoolManager.request', Mock(return_value=urllib3.HTTPResponse(
+        status=200, body=b'{"servers":["127.0.0.1","127.0.0.2","127.0.0.3"],"port":2181}')))
     @patch('patroni.dcs.zookeeper.KazooClient', MockKazooClient)
     def setUp(self):
         self.e = Exhibitor({'hosts': ['localhost', 'exhibitor'], 'port': 8181, 'scope': 'test',

tests/test_ha.py

@@ -1,21 +1,26 @@
 import datetime
 import etcd
 import os
-import unittest
 import sys
 
-from mock import Mock, MagicMock, PropertyMock, patch
-from patroni.async_executor import CriticalTask
+from mock import Mock, MagicMock, PropertyMock, patch, mock_open
 from patroni.config import Config
 from patroni.dcs import Cluster, ClusterConfig, Failover, Leader, Member, get_dcs, SyncState, TimelineHistory
 from patroni.dcs.etcd import Client
 from patroni.exceptions import DCSError, PostgresConnectionException, PatroniException
 from patroni.ha import Ha, _MemberStatus
 from patroni.postgresql import Postgresql
-from patroni.watchdog import Watchdog
+from patroni.postgresql.bootstrap import Bootstrap
+from patroni.postgresql.cancellable import CancellableSubprocess
+from patroni.postgresql.config import ConfigHandler
+from patroni.postgresql.rewind import Rewind
+from patroni.postgresql.slots import SlotsHandler
 from patroni.utils import tzutc
-from test_etcd import socket_getaddrinfo, etcd_read, etcd_write, requests_get
-from test_postgresql import psycopg2_connect, MockPostmaster
+from patroni.watchdog import Watchdog
+from six.moves import builtins
+
+from . import PostgresInit, MockPostmaster, psycopg2_connect, requests_get
+from .test_etcd import socket_getaddrinfo, etcd_read, etcd_write
 
 SYSID = '12345678901'
 
@@ -29,8 +34,10 @@ def false(*args, **kwargs):
 
 
 def get_cluster(initialize, leader, members, failover, sync, cluster_config=None):
-    history = TimelineHistory(1, [(1, 67197376, 'no recovery target specified', datetime.datetime.now().isoformat())])
-    cluster_config = cluster_config or ClusterConfig(1, {1: 2}, 1)
+    t = datetime.datetime.now().isoformat()
+    history = TimelineHistory(1, '[[1,67197376,"no recovery target specified","' + t + '"]]',
+                              [(1, 67197376, 'no recovery target specified', t)])
+    cluster_config = cluster_config or ClusterConfig(1, {'check_timeline': True}, 1)
     return Cluster(initialize, cluster_config, leader, 10, members, failover, sync, history)
 
 
@@ -62,17 +69,6 @@ def get_cluster_initialized_with_only_leader(failover=None, cluster_config=None)
     return get_cluster(True, leader, [leader], failover, None, cluster_config)
 
 
-def get_cluster_not_initialized_standby(failover=None, sync=None):
-    return get_cluster_not_initialized_without_leader(
-        cluster_config=ClusterConfig(1, {
-            "standby_cluster": {
-                "host": "localhost",
-                "port": 5432,
-                "primary_slot_name": "",
-            }}, 1)
-    )
-
-
 def get_standby_cluster_initialized_with_only_leader(failover=None, sync=None):
     return get_cluster_initialized_with_only_leader(
         cluster_config=ClusterConfig(1, {
@@ -84,12 +80,13 @@ def get_standby_cluster_initialized_with_only_leader(failover=None, sync=None):
     )
 
 
-def get_node_status(reachable=True, in_recovery=True, wal_position=10, nofailover=False, watchdog_failed=False):
+def get_node_status(reachable=True, in_recovery=True, timeline=2,
+                    wal_position=10, nofailover=False, watchdog_failed=False):
     def fetch_node_status(e):
         tags = {}
         if nofailover:
             tags['nofailover'] = True
-        return _MemberStatus(e, reachable, in_recovery, wal_position, tags, watchdog_failed)
+        return _MemberStatus(e, reachable, in_recovery, timeline, wal_position, tags, watchdog_failed)
     return fetch_node_status
 
 
@@ -126,7 +123,9 @@ zookeeper:
         # all the extra values that are coming from py.test
         sys.argv = sys.argv[:1]
 
-        self.config = Config()
+        self.config = Config(None)
+        self.config.set_dynamic_configuration({'maximum_lag_on_failover': 5})
+        self.version = '1.5.7'
         self.postgresql = p
         self.dcs = d
         self.api = Mock()
@@ -139,51 +138,55 @@ zookeeper:
         self.scheduled_restart = {'schedule': future_restart_time,
                                   'postmaster_start_time': str(postmaster_start_time)}
         self.watchdog = Watchdog(self.config)
+        self.request = lambda member, **kwargs: requests_get(member.api_url, **kwargs)
 
 
 def run_async(self, func, args=()):
-    return func(*args) if args else func()
+    self.reset_scheduled_action()
+    if args:
+        func(*args)
+    else:
+        func()
 
 
 @patch.object(Postgresql, 'is_running', Mock(return_value=MockPostmaster()))
 @patch.object(Postgresql, 'is_leader', Mock(return_value=True))
-@patch.object(Postgresql, 'timeline_wal_position', Mock(return_value=(1, 10)))
+@patch.object(Postgresql, 'timeline_wal_position', Mock(return_value=(1, 10, 1)))
 @patch.object(Postgresql, '_cluster_info_state_get', Mock(return_value=3))
-@patch.object(Postgresql, 'call_nowait', Mock(return_value=True))
 @patch.object(Postgresql, 'data_directory_empty', Mock(return_value=False))
-@patch.object(Postgresql, 'controldata', Mock(return_value={'Database system identifier': SYSID}))
-@patch.object(Postgresql, 'sync_replication_slots', Mock())
-@patch.object(Postgresql, 'write_pg_hba', Mock())
-@patch.object(Postgresql, 'write_pgpass', Mock(return_value={}))
-@patch.object(Postgresql, 'write_recovery_conf', Mock())
+@patch.object(Postgresql, 'controldata', Mock(return_value={
+    'Database system identifier': SYSID,
+    'Database cluster state': 'shut down',
+    'Latest checkpoint location': '0/12345678'}))
+@patch.object(SlotsHandler, 'sync_replication_slots', Mock())
+@patch.object(ConfigHandler, 'append_pg_hba', Mock())
+@patch.object(ConfigHandler, 'write_pgpass', Mock(return_value={}))
+@patch.object(ConfigHandler, 'write_recovery_conf', Mock())
+@patch.object(ConfigHandler, 'write_postgresql_conf', Mock())
 @patch.object(Postgresql, 'query', Mock())
 @patch.object(Postgresql, 'checkpoint', Mock())
-@patch.object(Postgresql, 'call_nowait', Mock())
-@patch.object(Postgresql, 'cancellable_subprocess_call', Mock(return_value=0))
+@patch.object(CancellableSubprocess, 'call', Mock(return_value=0))
+@patch.object(Postgresql, 'get_replica_timeline', Mock(return_value=2))
+@patch.object(Postgresql, 'get_master_timeline', Mock(return_value=2))
+@patch.object(ConfigHandler, 'restore_configuration_files', Mock())
 @patch.object(etcd.Client, 'write', etcd_write)
 @patch.object(etcd.Client, 'read', etcd_read)
 @patch.object(etcd.Client, 'delete', Mock(side_effect=etcd.EtcdException))
 @patch('patroni.postgresql.polling_loop', Mock(return_value=range(1)))
 @patch('patroni.async_executor.AsyncExecutor.busy', PropertyMock(return_value=False))
 @patch('patroni.async_executor.AsyncExecutor.run_async', run_async)
+@patch('patroni.postgresql.rewind.Thread', Mock())
 @patch('subprocess.call', Mock(return_value=0))
 @patch('time.sleep', Mock())
-class TestHa(unittest.TestCase):
+class TestHa(PostgresInit):
 
     @patch('socket.getaddrinfo', socket_getaddrinfo)
-    @patch('psycopg2.connect', psycopg2_connect)
-    @patch('patroni.dcs.dcs_modules', Mock(return_value=['patroni.dcs.foo', 'patroni.dcs.etcd']))
+    @patch('patroni.dcs.dcs_modules', Mock(return_value=['patroni.dcs.etcd']))
     @patch.object(etcd.Client, 'read', etcd_read)
     def setUp(self):
+        super(TestHa, self).setUp()
         with patch.object(Client, 'machines') as mock_machines:
             mock_machines.__get__ = Mock(return_value=['http://remotehost:2379'])
-            self.p = Postgresql({'name': 'postgresql0', 'scope': 'dummy', 'listen': '127.0.0.1:5432',
-                                 'data_dir': 'data/postgresql0', 'retry_timeout': 10,
-                                 'maximum_lag_on_failover': 5,
-                                 'authentication': {'superuser': {'username': 'foo', 'password': 'bar'},
-                                                    'replication': {'username': '', 'password': ''}},
-                                 'parameters': {'wal_level': 'hot_standby', 'max_replication_slots': 5, 'foo': 'bar',
-                                                'hot_standby': 'on', 'max_wal_senders': 5, 'wal_keep_segments': 8}})
             self.p.set_state('running')
             self.p.set_role('replica')
             self.p.postmaster_start_time = MagicMock(return_value=str(postmaster_start_time))
@@ -199,10 +202,14 @@ class TestHa(unittest.TestCase):
         self.p.last_operation = Mock(side_effect=PostgresConnectionException(''))
         self.assertTrue(self.ha.update_lock(True))
 
+    @patch.object(Postgresql, 'received_timeline', Mock(return_value=None))
     def test_touch_member(self):
-        self.p.timeline_wal_position = Mock(return_value=(0, 1))
+        self.p.timeline_wal_position = Mock(return_value=(0, 1, 0))
         self.p.replica_cached_timeline = Mock(side_effect=Exception)
         self.ha.touch_member()
+        self.p.timeline_wal_position = Mock(return_value=(0, 1, 1))
+        self.p.set_role('standby_leader')
+        self.ha.touch_member()
 
     def test_is_leader(self):
         self.assertFalse(self.ha.is_leader())
@@ -212,48 +219,28 @@ class TestHa(unittest.TestCase):
         self.assertEqual(self.ha.run_cycle(), 'starting as a secondary')
 
     @patch('patroni.dcs.etcd.Etcd.initialize', return_value=True)
-    def test_start_as_standby_leader(self, initialize):
+    def test_bootstrap_as_standby_leader(self, initialize):
         self.p.data_directory_empty = true
-        self.ha.cluster = get_cluster_not_initialized_standby()
+        self.ha.cluster = get_cluster_not_initialized_without_leader(cluster_config=ClusterConfig(0, {}, 0))
         self.ha.cluster.is_unlocked = true
-        self.ha.patroni.config._dynamic_configuration = {"standby_cluster": {
-            "host": "localhost",
-            "port": 5432,
-            "primary_slot_name": "",
-        }}
-        self.assertEqual(
-            self.ha.run_cycle(),
-            'trying to bootstrap a new standby leader'
-        )
+        self.ha.patroni.config._dynamic_configuration = {"standby_cluster": {"port": 5432}}
+        self.assertEqual(self.ha.run_cycle(), 'trying to bootstrap a new standby leader')
+
+    def test_bootstrap_waiting_for_standby_leader(self):
+        self.p.data_directory_empty = true
+        self.ha.cluster = get_cluster_initialized_without_leader()
+        self.ha.cluster.config.data.update({'standby_cluster': {'port': 5432}})
+        self.assertEqual(self.ha.run_cycle(), 'waiting for standby_leader to bootstrap')
 
     @patch.object(Cluster, 'get_clone_member',
-                  Mock(return_value=Member(0, 'test', 1, {'api_url': 'http://127.0.0.1:8011/patroni'})))
-    @patch.object(Postgresql, 'create_replica', Mock(return_value=0))
+                  Mock(return_value=Member(0, 'test', 1, {'api_url': 'http://127.0.0.1:8011/patroni',
+                                                          'conn_url': 'postgres://127.0.0.1:5432/postgres'})))
+    @patch.object(Bootstrap, 'create_replica', Mock(return_value=0))
     def test_start_as_cascade_replica_in_standby_cluster(self):
         self.p.data_directory_empty = true
         self.ha.cluster = get_standby_cluster_initialized_with_only_leader()
         self.ha.cluster.is_unlocked = false
-        self.ha.patroni.config._dynamic_configuration = {"standby_cluster": {
-            "host": "localhost",
-            "port": 5432,
-            "primary_slot_name": "",
-        }}
-        self.assertEqual(
-            self.ha.run_cycle(),
-            "trying to bootstrap from replica 'test'"
-        )
-
-    @patch.object(Postgresql, 'create_replica', Mock(return_value=0))
-    def test_bootstrap_standby_leader(self):
-        self.ha.cluster = get_cluster_not_initialized_standby()
-        self.ha.cluster.is_unlocked = true
-        self.ha.patroni.config._dynamic_configuration = {"standby_cluster": {
-            "host": "localhost",
-            "port": 5432,
-            "primary_slot_name": "",
-        }}
-        self.ha._post_bootstrap_task = CriticalTask()
-        self.assertEqual(self.ha.bootstrap_standby_leader(), True)
+        self.assertEqual(self.ha.run_cycle(), "trying to bootstrap from replica 'test'")
 
     def test_recover_replica_failed(self):
         self.p.controldata = lambda: {'Database cluster state': 'in recovery', 'Database system identifier': SYSID}
@@ -271,18 +258,26 @@ class TestHa(unittest.TestCase):
         self.ha.cluster = get_cluster_initialized_with_leader()
         self.assertEqual(self.ha.run_cycle(), 'starting as readonly because i had the session lock')
 
-    @patch.object(Postgresql, 'fix_cluster_state', Mock())
+    @patch.object(Rewind, 'ensure_clean_shutdown', Mock())
     def test_crash_recovery(self):
         self.p.is_running = false
         self.p.controldata = lambda: {'Database cluster state': 'in production', 'Database system identifier': SYSID}
         self.assertEqual(self.ha.run_cycle(), 'doing crash recovery in a single user mode')
 
-    @patch.object(Postgresql, 'rewind_needed_and_possible', Mock(return_value=True))
+    @patch.object(Rewind, 'rewind_or_reinitialize_needed_and_possible', Mock(return_value=True))
+    @patch.object(Rewind, 'can_rewind', PropertyMock(return_value=True))
     def test_recover_with_rewind(self):
         self.p.is_running = false
         self.ha.cluster = get_cluster_initialized_with_leader()
         self.assertEqual(self.ha.run_cycle(), 'running pg_rewind from leader')
 
+    @patch.object(Rewind, 'rewind_or_reinitialize_needed_and_possible', Mock(return_value=True))
+    @patch.object(Bootstrap, 'create_replica', Mock(return_value=1))
+    def test_recover_with_reinitialize(self):
+        self.p.is_running = false
+        self.ha.cluster = get_cluster_initialized_with_leader()
+        self.assertEqual(self.ha.run_cycle(), 'reinitializing due to diverged timelines')
+
     @patch('sys.exit', return_value=1)
     @patch('patroni.ha.Ha.sysid_valid', MagicMock(return_value=True))
     def test_sysid_no_match(self, exit_mock):
@@ -349,7 +344,7 @@ class TestHa(unittest.TestCase):
             self.assertEqual(self.ha.run_cycle(), 'Not promoting self because watchdog could not be activated')
 
     def test_leader_with_lock(self):
-        self.ha.cluster = get_cluster_not_initialized_without_leader()
+        self.ha.cluster = get_cluster_initialized_with_leader()
         self.ha.cluster.is_unlocked = false
         self.ha.has_lock = true
         self.assertEqual(self.ha.run_cycle(), 'no action.  i am the leader with the lock')
@@ -372,6 +367,7 @@ class TestHa(unittest.TestCase):
         self.p.is_leader = false
         self.assertEqual(self.ha.run_cycle(), 'no action.  i am a secondary and i am following a leader')
         self.ha.patroni.replicatefrom = "foo"
+        self.p.config.check_recovery_conf = Mock(return_value=(True, False))
         self.assertEqual(self.ha.run_cycle(), 'no action.  i am a secondary and i am following a leader')
 
     def test_follow_in_pause(self):
@@ -381,10 +377,11 @@ class TestHa(unittest.TestCase):
         self.p.is_leader = false
         self.assertEqual(self.ha.run_cycle(), 'PAUSE: no action')
 
-    @patch.object(Postgresql, 'rewind_needed_and_possible', Mock(return_value=True))
+    @patch.object(Rewind, 'rewind_or_reinitialize_needed_and_possible', Mock(return_value=True))
+    @patch.object(Rewind, 'can_rewind', PropertyMock(return_value=True))
     def test_follow_triggers_rewind(self):
         self.p.is_leader = false
-        self.p.trigger_check_diverged_lsn()
+        self.ha._rewind.trigger_check_diverged_lsn()
         self.ha.cluster = get_cluster_initialized_with_leader()
         self.assertEqual(self.ha.run_cycle(), 'running pg_rewind from leader')
 
@@ -443,7 +440,7 @@ class TestHa(unittest.TestCase):
 
         self.ha.cluster = get_cluster_initialized_with_leader()
         self.assertIsNone(self.ha.reinitialize(True))
-
+        self.ha._async_executor.schedule('reinitialize')
         self.assertIsNotNone(self.ha.reinitialize())
 
         self.ha.state_handler.name = self.ha.cluster.leader.name
@@ -457,7 +454,7 @@ class TestHa(unittest.TestCase):
         self.p.restart = false
         self.assertEqual(self.ha.restart({}), (False, 'restart failed'))
         self.ha.cluster = get_cluster_initialized_with_leader()
-        self.ha.reinitialize()
+        self.ha._async_executor.schedule('reinitialize')
         self.assertEqual(self.ha.restart({}), (False, 'reinitialize already in progress'))
         with patch.object(self.ha, "restart_matches", return_value=False):
             self.assertEqual(self.ha.restart({'foo': 'bar'}), (False, "restart conditions are not satisfied"))
@@ -465,7 +462,7 @@ class TestHa(unittest.TestCase):
     @patch('os.kill', Mock())
     def test_restart_in_progress(self):
         with patch('patroni.async_executor.AsyncExecutor.busy', PropertyMock(return_value=True)):
-            self.ha.restart({}, run_async=True)
+            self.ha._async_executor.schedule('restart')
             self.assertTrue(self.ha.restart_scheduled())
             self.assertEqual(self.ha.run_cycle(), 'restart in progress')
 
@@ -482,7 +479,6 @@ class TestHa(unittest.TestCase):
                     self.assertEqual(self.ha.run_cycle(), 'lost leader lock during restart')
                     mock_terminate.assert_called()
 
-    @patch('requests.get', requests_get)
     def test_manual_failover_from_leader(self):
         self.ha.fetch_node_status = get_node_status()
         self.ha.has_lock = true
@@ -495,12 +491,14 @@ class TestHa(unittest.TestCase):
         f = Failover(0, self.p.name, '', None)
         self.ha.cluster = get_cluster_initialized_with_leader(f)
         self.assertEqual(self.ha.run_cycle(), 'manual failover: demoting myself')
-        self.p.rewind_needed_and_possible = true
+        self.ha._rewind.rewind_or_reinitialize_needed_and_possible = true
         self.assertEqual(self.ha.run_cycle(), 'manual failover: demoting myself')
         self.ha.fetch_node_status = get_node_status(nofailover=True)
         self.assertEqual(self.ha.run_cycle(), 'no action.  i am the leader with the lock')
         self.ha.fetch_node_status = get_node_status(watchdog_failed=True)
         self.assertEqual(self.ha.run_cycle(), 'no action.  i am the leader with the lock')
+        self.ha.fetch_node_status = get_node_status(timeline=1)
+        self.assertEqual(self.ha.run_cycle(), 'no action.  i am the leader with the lock')
         self.ha.fetch_node_status = get_node_status(wal_position=1)
         self.assertEqual(self.ha.run_cycle(), 'no action.  i am the leader with the lock')
         # manual failover from the previous leader to us won't happen if we hold the nofailover flag
@@ -528,7 +526,6 @@ class TestHa(unittest.TestCase):
         self.ha.cluster = get_cluster_initialized_with_leader(Failover(0, 'blabla', self.p.name, scheduled))
         self.assertEqual('no action.  i am the leader with the lock', self.ha.run_cycle())
 
-    @patch('requests.get', requests_get)
     def test_manual_failover_from_leader_in_pause(self):
         self.ha.has_lock = true
         self.ha.is_paused = true
@@ -538,7 +535,6 @@ class TestHa(unittest.TestCase):
         self.ha.cluster = get_cluster_initialized_with_leader(Failover(0, self.p.name, '', None))
         self.assertEqual('PAUSE: no action.  i am the leader with the lock', self.ha.run_cycle())
 
-    @patch('requests.get', requests_get)
     def test_manual_failover_from_leader_in_synchronous_mode(self):
         self.p.is_leader = true
         self.ha.has_lock = true
@@ -551,7 +547,6 @@ class TestHa(unittest.TestCase):
         self.ha.is_failover_possible = true
         self.assertEqual('manual failover: demoting myself', self.ha.run_cycle())
 
-    @patch('requests.get', requests_get)
     def test_manual_failover_process_no_leader(self):
         self.p.is_leader = false
         self.ha.cluster = get_cluster_initialized_without_leader(failover=Failover(0, '', self.p.name, None))
@@ -602,6 +597,7 @@ class TestHa(unittest.TestCase):
         self.assertFalse(self.ha.is_healthiest_node())
 
     def test__is_healthiest_node(self):
+        self.ha.cluster = get_cluster_initialized_without_leader(sync=('postgresql1', self.p.name))
         self.assertTrue(self.ha._is_healthiest_node(self.ha.old_cluster.members))
         self.p.is_leader = false
         self.ha.fetch_node_status = get_node_status()  # accessible, in_recovery
@@ -610,25 +606,34 @@ class TestHa(unittest.TestCase):
         self.assertFalse(self.ha._is_healthiest_node(self.ha.old_cluster.members))
         self.ha.fetch_node_status = get_node_status(wal_position=11)  # accessible, in_recovery, wal position ahead
         self.assertFalse(self.ha._is_healthiest_node(self.ha.old_cluster.members))
-        with patch('patroni.postgresql.Postgresql.timeline_wal_position', return_value=(1, 1)):
+        # in synchronous_mode consider itself healthy if the former leader is accessible in read-only and ahead of us
+        with patch.object(Ha, 'is_synchronous_mode', Mock(return_value=True)):
+            self.assertTrue(self.ha._is_healthiest_node(self.ha.old_cluster.members))
+        with patch('patroni.postgresql.Postgresql.timeline_wal_position', return_value=(1, 1, 1)):
+            self.assertFalse(self.ha._is_healthiest_node(self.ha.old_cluster.members))
+        with patch('patroni.postgresql.Postgresql.replica_cached_timeline', return_value=1):
             self.assertFalse(self.ha._is_healthiest_node(self.ha.old_cluster.members))
         self.ha.patroni.nofailover = True
         self.assertFalse(self.ha._is_healthiest_node(self.ha.old_cluster.members))
         self.ha.patroni.nofailover = False
 
-    @patch('requests.get', requests_get)
     def test_fetch_node_status(self):
         member = Member(0, 'test', 1, {'api_url': 'http://127.0.0.1:8011/patroni'})
         self.ha.fetch_node_status(member)
         member = Member(0, 'test', 1, {'api_url': 'http://localhost:8011/patroni'})
         self.ha.fetch_node_status(member)
 
+    @patch.object(Rewind, 'pg_rewind', true)
+    @patch.object(Rewind, 'check_leader_is_not_in_recovery', true)
     def test_post_recover(self):
         self.p.is_running = false
         self.ha.has_lock = true
+        self.p.set_role('master')
         self.assertEqual(self.ha.post_recover(), 'removed leader key after trying and failing to start postgres')
         self.ha.has_lock = false
         self.assertEqual(self.ha.post_recover(), 'failed to start postgres')
+        leader = Leader(0, 0, Member(0, 'l', 2, {"version": "1.6", "conn_url": "postgres://a", "role": "master"}))
+        self.ha._rewind.execute(leader)
         self.p.is_running = true
         self.assertIsNone(self.ha.post_recover())
 
@@ -672,7 +677,7 @@ class TestHa(unittest.TestCase):
 
     def test_restart_matches(self):
         self.p._role = 'replica'
-        self.p.server_version = 90500
+        self.p._connection.server_version = 90500
         self.p._pending_restart = True
         self.assertFalse(self.ha.restart_matches("master", "9.5.0", True))
         self.assertFalse(self.ha.restart_matches("replica", "9.4.3", True))
@@ -689,60 +694,63 @@ class TestHa(unittest.TestCase):
         self.ha.cluster = get_cluster_initialized_with_leader(Failover(0, '', self.p.name, None))
         self.assertEqual(self.ha.run_cycle(), 'PAUSE: waiting to become master after promote...')
 
+    @patch('patroni.postgresql.mtime', Mock(return_value=1588316884))
+    @patch.object(builtins, 'open', mock_open(read_data='1\t0/40159C0\tno recovery target specified\n'))
     def test_process_healthy_standby_cluster_as_standby_leader(self):
         self.p.is_leader = false
         self.p.name = 'leader'
-        self.ha.patroni.config._dynamic_configuration = {"standby_cluster": {
-            "host": "localhost",
-            "port": 5432,
-            "primary_slot_name": "",
-        }}
         self.ha.cluster = get_standby_cluster_initialized_with_only_leader()
-        msg = 'no action.  i am the standby leader with the lock'
-        self.assertEqual(self.ha.run_cycle(), msg)
+        self.p.config.check_recovery_conf = Mock(return_value=(False, False))
+        self.ha._leader_timeline = 1
+        self.assertEqual(self.ha.run_cycle(), 'promoted self to a standby leader because i had the session lock')
+        self.assertEqual(self.ha.run_cycle(), 'no action.  i am the standby leader with the lock')
+        self.p.set_role('replica')
+        self.p.config.check_recovery_conf = Mock(return_value=(True, False))
+        self.assertEqual(self.ha.run_cycle(), 'promoted self to a standby leader because i had the session lock')
 
     def test_process_healthy_standby_cluster_as_cascade_replica(self):
         self.p.is_leader = false
         self.p.name = 'replica'
-        self.ha.patroni.config._dynamic_configuration = {"standby_cluster": {
-            "host": "localhost",
-            "port": 5432,
-            "primary_slot_name": "",
-        }}
         self.ha.cluster = get_standby_cluster_initialized_with_only_leader()
-        msg = 'no action.  i am a secondary and i am following a leader'
-        self.assertEqual(self.ha.run_cycle(), msg)
+        self.assertEqual(self.ha.run_cycle(), 'no action.  i am a secondary and i am following a standby leader')
+        with patch.object(Leader, 'conn_url', PropertyMock(return_value='')):
+            self.assertEqual(self.ha.run_cycle(), 'continue following the old known standby leader')
 
-    @patch('patroni.dcs.etcd.Etcd.initialize', return_value=True)
-    def test_process_unhealthy_standby_cluster_as_standby_leader(self, initialize):
+    def test_process_unhealthy_standby_cluster_as_standby_leader(self):
         self.p.is_leader = false
         self.p.name = 'leader'
-        self.ha.patroni.config._dynamic_configuration = {"standby_cluster": {
-            "host": "localhost",
-            "port": 5432,
-            "primary_slot_name": "",
-        }}
         self.ha.cluster = get_standby_cluster_initialized_with_only_leader()
         self.ha.cluster.is_unlocked = true
         self.ha.sysid_valid = true
         self.p._sysid = True
-        msg = 'promoted self to a standby leader because i had the session lock'
-        self.assertEqual(self.ha.run_cycle(), msg)
+        self.assertEqual(self.ha.run_cycle(), 'promoted self to a standby leader by acquiring session lock')
 
-    @patch.object(Postgresql, 'rewind_needed_and_possible', Mock(return_value=True))
-    @patch('patroni.dcs.etcd.Etcd.initialize', return_value=True)
-    def test_process_unhealthy_standby_cluster_as_cascade_replica(self, initialize):
+    @patch.object(Rewind, 'rewind_or_reinitialize_needed_and_possible', Mock(return_value=True))
+    @patch.object(Rewind, 'can_rewind', PropertyMock(return_value=True))
+    def test_process_unhealthy_standby_cluster_as_cascade_replica(self):
         self.p.is_leader = false
         self.p.name = 'replica'
-        self.ha.patroni.config._dynamic_configuration = {"standby_cluster": {
-            "host": "localhost",
-            "port": 5432,
-            "primary_slot_name": "",
-        }}
         self.ha.cluster = get_standby_cluster_initialized_with_only_leader()
         self.ha.is_unlocked = true
-        msg = 'running pg_rewind from leader'
-        self.assertEqual(self.ha.run_cycle(), msg)
+        self.assertTrue(self.ha.run_cycle().startswith('running pg_rewind from remote_master:'))
+
+    def test_recover_unhealthy_leader_in_standby_cluster(self):
+        self.p.is_leader = false
+        self.p.name = 'leader'
+        self.p.is_running = false
+        self.p.follow = false
+        self.ha.cluster = get_standby_cluster_initialized_with_only_leader()
+        self.assertEqual(self.ha.run_cycle(), 'starting as a standby leader because i had the session lock')
+
+    def test_recover_unhealthy_unlocked_standby_cluster(self):
+        self.p.is_leader = false
+        self.p.name = 'leader'
+        self.p.is_running = false
+        self.p.follow = false
+        self.ha.cluster = get_standby_cluster_initialized_with_only_leader()
+        self.ha.cluster.is_unlocked = true
+        self.ha.has_lock = false
+        self.assertEqual(self.ha.run_cycle(), 'trying to follow a remote master because standby cluster is unhealthy')
 
     def test_failed_to_update_lock_in_pause(self):
         self.ha.update_lock = false
@@ -820,6 +828,17 @@ class TestHa(unittest.TestCase):
         self.assertEqual(self.ha.run_cycle(), 'stopped PostgreSQL to fail over after a crash')
         demote.assert_called_once()
 
+    def test_master_stop_timeout(self):
+        self.assertEqual(self.ha.master_stop_timeout(), None)
+        self.ha.patroni.config.set_dynamic_configuration({'master_stop_timeout': 30})
+        with patch.object(Ha, 'is_synchronous_mode', Mock(return_value=True)):
+            self.assertEqual(self.ha.master_stop_timeout(), 30)
+        self.ha.patroni.config.set_dynamic_configuration({'master_stop_timeout': 30})
+        with patch.object(Ha, 'is_synchronous_mode', Mock(return_value=False)):
+            self.assertEqual(self.ha.master_stop_timeout(), None)
+            self.ha.patroni.config.set_dynamic_configuration({'master_stop_timeout': None})
+            self.assertEqual(self.ha.master_stop_timeout(), None)
+
     @patch('patroni.postgresql.Postgresql.follow')
     def test_demote_immediate(self, follow):
         self.ha.has_lock = true
@@ -829,7 +848,7 @@ class TestHa(unittest.TestCase):
 
     def test_process_sync_replication(self):
         self.ha.has_lock = true
-        mock_set_sync = self.p.set_synchronous_standby = Mock()
+        mock_set_sync = self.p.config.set_synchronous_standby = Mock()
         self.p.name = 'leader'
 
         # Test sync key removed when sync mode disabled
@@ -886,11 +905,15 @@ class TestHa(unittest.TestCase):
         self.ha.run_cycle()
         self.assertEqual(self.ha.dcs.write_sync_state.call_count, 2)
 
-        # Test changing sync standby failed due to race
+        # Test updating sync standby key failed due to DCS being not accessible
         self.ha.dcs.write_sync_state = Mock(return_value=True)
+        self.ha.dcs.get_cluster = Mock(side_effect=DCSError('foo'))
+        self.ha.run_cycle()
+
+        # Test changing sync standby failed due to race
         self.ha.dcs.get_cluster = Mock(return_value=get_cluster_initialized_with_leader(sync=('somebodyelse', None)))
         self.ha.run_cycle()
-        self.assertEqual(self.ha.dcs.write_sync_state.call_count, 1)
+        self.assertEqual(self.ha.dcs.write_sync_state.call_count, 2)
 
         # Test sync set to '*' when synchronous_mode_strict is enabled
         mock_set_sync.reset_mock()
@@ -902,7 +925,7 @@ class TestHa(unittest.TestCase):
     def test_sync_replication_become_master(self):
         self.ha.is_synchronous_mode = true
 
-        mock_set_sync = self.p.set_synchronous_standby = Mock()
+        mock_set_sync = self.p.config.set_synchronous_standby = Mock()
         self.p.is_leader = false
         self.p.set_role('replica')
         self.ha.has_lock = true
@@ -994,6 +1017,8 @@ class TestHa(unittest.TestCase):
         self.ha._disable_sync = False
         self.assertEqual(self.ha.get_effective_tags(), {'foo': 'bar'})
 
+    @patch('patroni.postgresql.mtime', Mock(return_value=1588316884))
+    @patch.object(builtins, 'open', Mock(side_effect=Exception))
     def test_restore_cluster_config(self):
         self.ha.cluster.config.data.clear()
         self.ha.has_lock = true
@@ -1025,11 +1050,15 @@ class TestHa(unittest.TestCase):
         # will not say bootstrap from leader as replica can't self elect
         self.assertEqual(self.ha.run_cycle(), "trying to bootstrap from replica 'other'")
 
+    @patch('patroni.postgresql.mtime', Mock(return_value=1588316884))
+    @patch.object(builtins, 'open', mock_open(read_data=('1\t0/40159C0\tno recovery target specified\n\n'
+                                                         '2\t1/40159C0\tno recovery target specified\n')))
     def test_update_cluster_history(self):
-        self.p.get_master_timeline = Mock(return_value=1)
         self.ha.has_lock = true
         self.ha.cluster.is_unlocked = false
-        self.assertEqual(self.ha.run_cycle(), 'no action.  i am the leader with the lock')
+        for tl in (1, 3):
+            self.p.get_master_timeline = Mock(return_value=tl)
+            self.assertEqual(self.ha.run_cycle(), 'no action.  i am the leader with the lock')
 
     @patch('sys.exit', return_value=1)
     def test_abort_join(self, exit_mock):
@@ -1045,3 +1074,17 @@ class TestHa(unittest.TestCase):
         self.assertEqual(self.ha.run_cycle(), 'PAUSE: no action.  i am the leader with the lock')
         self.ha.is_paused = false
         self.assertEqual(self.ha.run_cycle(), 'no action.  i am the leader with the lock')
+
+    @patch('psycopg2.connect', psycopg2_connect)
+    def test_permanent_logical_slots_after_promote(self):
+        config = ClusterConfig(1, {'slots': {'l': {'database': 'postgres', 'plugin': 'test_decoding'}}}, 1)
+        self.ha.cluster = get_cluster_initialized_without_leader(cluster_config=config)
+        self.assertEqual(self.ha.run_cycle(), 'acquired session lock as a leader')
+        self.ha.cluster = get_cluster_initialized_without_leader(leader=True, cluster_config=config)
+        self.ha.has_lock = true
+        self.assertEqual(self.ha.run_cycle(), 'no action.  i am the leader with the lock')
+
+    @patch.object(Cluster, 'has_member', true)
+    def test_run_cycle(self):
+        self.ha.dcs.touch_member = Mock(side_effect=DCSError('foo'))
+        self.assertEqual(self.ha.run_cycle(), 'Unexpected exception raised, please report it as a BUG')