Release 1.5.4 (#931)

* Bump version
* Update release notes
* Make it possible to configure registration of Service in Consul via env variables

.gitignore

@@ -1,11 +1,56 @@
-data/*
-*.pyc
-*.egg/
-*.egg-info/
+*.py[cod]
+
+# vi(m) swap files:
+*.sw?
+
+# C extensions
+*.so
+
+# Packages
 .cache/
+*.egg
+*.eggs
+*.egg-info
+dist
+build
+eggs
+parts
+bin
+var
+sdist
+develop-eggs
+.installed.cfg
+lib
+lib64
+
+# Installer logs
+pip-log.txt
+
+# Unit test / coverage reports
 .coverage
-.eggs/
-build/
+.tox
+nosetests.xml
 coverage.xml
+htmlcov
 junit.xml
+features/output
+dummy
+
+# Translations
+*.mo
+
+# Mr Developer
+.mr.developer.cfg
+.project
+.pydevproject
+
 pgpass
+scm-source.json
+
+# Sphinx-generated documentation
+docs/build/
+docs/source/_static/
+docs/source/_templates/
+
+# Pycharm IDE
+.idea/

.travis.yml

@@ -1,14 +1,153 @@
+sudo: true
+dist: trusty
 language: python
-python:
-  - "2.7"
-  - "3.3"
-  - "3.4"
+env:
+  global:
+  - ETCDVERSION=3.0.17 ZKVERSION=3.4.11 CONSULVERSION=0.7.4
+  - PYVERSIONS="2.7 3.5 3.6"
+  - EXCLUDE_BEHAVE="3.5"
+  - BOTO_CONFIG=/doesnotexist
+matrix:
+  include:
+  - python: "3.5"
+    env: TEST_SUITE="python setup.py"
+  - python: "3.6"
+    env: DCS="etcd" TEST_SUITE="behave"
+  - python: "3.6"
+    env: DCS="exhibitor" TEST_SUITE="behave"
+  - python: "3.6"
+    env: DCS="consul" TEST_SUITE="behave"
+  - python: "3.6"
+    env: DCS="kubernetes" TEST_SUITE="behave"
+branches:
+  only:
+  - master
+  - /^v\d+\.\d+(\.\d+)?$/
+cache:
+  directories:
+  - $HOME/mycache
+before_cache:
+- |
+  rm -fr $HOME/mycache/python*
+  for pv in $PYVERSIONS; do
+      if [[ $TEST_SUITE != "behave" || $pv != $EXCLUDE_BEHAVE ]]; then
+          fpv=$(basename $(readlink $HOME/virtualenv/python${pv}))
+          mv $HOME/virtualenv/${fpv} $HOME/mycache/${fpv}
+      fi
+  done
 install:
-  - if [[ $TRAVIS_PYTHON_VERSION == 2* ]]; then pip install -r requirements-py2.txt --use-mirrors; fi
-  - if [[ $TRAVIS_PYTHON_VERSION == 3* ]]; then pip install -r requirements-py3.txt --use-mirrors; fi
-  - pip install coveralls
+  - |
+    set -e
+
+    if [[ $TEST_SUITE == "behave" ]]; then
+        function get_consul() {
+            CC=~/mycache/consul_${CONSULVERSION}
+            if [[ ! -x $CC ]]; then
+                curl -L https://releases.hashicorp.com/consul/${CONSULVERSION}/consul_${CONSULVERSION}_linux_amd64.zip \
+                    | gunzip > $CC
+                [[ ${PIPESTATUS[0]} == 0 ]] || return 1
+                chmod +x $CC
+            fi
+            ln -s $CC consul
+        }
+
+        function get_etcd() {
+            EC=~/mycache/etcd_${ETCDVERSION}
+            if [[ ! -x $EC ]]; then
+                curl -L https://github.com/coreos/etcd/releases/download/v${ETCDVERSION}/etcd-v${ETCDVERSION}-linux-amd64.tar.gz \
+                    | tar xz -C . --strip=1 --wildcards --no-anchored etcd
+                [[ ${PIPESTATUS[0]} == 0 ]] || return 1
+                mv etcd $EC
+            fi
+            ln -s $EC etcd
+        }
+
+        function get_kubernetes() {
+            wget -O localkube "https://storage.googleapis.com/minikube/k8sReleases/v1.7.0/localkube-linux-amd64"
+            chmod +x localkube
+            sudo nohup ./localkube --logtostderr=true --enable-dns=false > localkube.log 2>&1 &
+
+            echo "Waiting for localkube to start..."
+            if ! timeout 120 sh -c "while ! curl -ks http://127.0.0.1:8080/ >/dev/null; do sleep 1; done"; then
+                sudo cat localkube.log
+                echo "localkube did not start"
+                exit 1
+            fi
+            echo "Check certificate permissions"
+            sudo chmod 644 /var/lib/localkube/certs/*
+            sudo ls -altr /var/lib/localkube/certs/
+
+            echo "Set up .kube/config"
+            mkdir ~/.kube
+            echo -e "apiVersion: v1\nclusters:\n- cluster:\n    certificate-authority: /var/lib/localkube/certs/ca.crt\n    server: https://127.0.0.1:8443\n  name: local\ncontexts:\n- context:\n    cluster: local\n    user: myself\n  name: local\ncurrent-context: local\nkind: Config\npreferences: {}\nusers:\n- name: myself\n  user:\n    client-certificate: /var/lib/localkube/certs/apiserver.crt\n    client-key: /var/lib/localkube/certs/apiserver.key\n" > ~/.kube/config
+        }
+
+        function get_exhibitor() {
+            ZC=~/mycache/zookeeper-${ZKVERSION}
+            if [[ ! -d $ZC ]]; then
+                curl -L http://www.apache.org/dist/zookeeper/zookeeper-${ZKVERSION}/zookeeper-${ZKVERSION}.tar.gz | tar xz
+                [[ ${PIPESTATUS[0]} == 0 ]] || return 1
+                mv zookeeper-${ZKVERSION}/conf/zoo_sample.cfg zookeeper-${ZKVERSION}/conf/zoo.cfg
+                mv zookeeper-${ZKVERSION} $ZC
+            fi
+            $ZC/bin/zkServer.sh start
+            # following lines are 'emulating' exhibitor REST API
+            while true; do
+                echo -e 'HTTP/1.0 200 OK\nContent-Type: application/json\n\n{"servers":["127.0.0.1"],"port":2181}' \
+                    | nc -l 8181 &> /dev/null
+            done&
+        }
+
+        attempt_num=1
+        until get_${DCS}; do
+            [[ $attempt_num -ge 3 ]] && exit 1
+            echo "Attempt $attempt_num failed! Trying again in $attempt_num seconds..."
+            sleep $(( attempt_num++ ))
+        done
+    fi
+
+    for pv in $PYVERSIONS; do
+        if [[ $TEST_SUITE != "behave" || $pv != $EXCLUDE_BEHAVE ]]; then
+            fpv=$(basename $(readlink $HOME/virtualenv/python$pv))
+            if [[ -d ~/mycache/${fpv} ]]; then
+                mv ~/virtualenv/${fpv} ~/virtualenv/${fpv}.bckp
+                mv ~/mycache/${fpv} ~/virtualenv/${fpv}
+            fi
+            source ~/virtualenv/python${pv}/bin/activate
+            # explicitly install all needed python modules to cache them
+            for p in '-r requirements.txt' 'behave codacy-coverage coverage coveralls flake8 mock pytest-cov pytest setuptools'; do
+                pip install $p --upgrade
+            done
+        fi
+    done
 script:
-  - python setup.py test
-  - python setup.py flake8
+  - |
+    for pv in $PYVERSIONS; do
+        if [[ $TEST_SUITE == "behave" && $pv == $EXCLUDE_BEHAVE ]]; then
+            continue
+        fi
+
+        source ~/virtualenv/python${pv}/bin/activate
+
+        if [[ $TEST_SUITE != "behave" ]]; then
+            echo Running unit tests using python${pv}
+            $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 $TEST_SUITE; then
+                # output all log files when tests are failing
+                grep . features/output/*/*postgres?.*
+                exit 1
+            fi
+        fi
+    done
+
+    set +e
 after_success:
+  # before_cache is executed earlier than after_success, so we need to restore one of virtualenv directories
+  - fpv=$(basename $(readlink $HOME/virtualenv/python3.6)) && mv $HOME/mycache/${fpv} $HOME/virtualenv/${fpv}
   - coveralls
+  - if [[ $TEST_SUITE != "behave" ]]; then python-codacy-coverage -r coverage.xml; fi
+  - if [[ $DCS == "exhibitor" ]]; then ~/mycache/zookeeper-${ZKVERSION}/bin/zkServer.sh stop; fi
+  - sudo kill $(jobs -p)

Dockerfile

@@ -1,36 +1,45 @@
 ## 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 ubuntu:14.04
-MAINTAINER Feike Steenbergen <feike.steenberge[email protected]>
+FROM postgres:10
+MAINTAINER Alexander Kukushkin <alexander.kukushki[email protected]>
 
-# We need curl
-RUN apt-get update -y && apt-get install curl -y
+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 \
+    # 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 \
 
-# Add PGDG repositories
-RUN echo "deb http://apt.postgresql.org/pub/repos/apt/ $(lsb_release -cs)-pgdg main" > /etc/apt/sources.list.d/pgdg.list
-RUN curl https://www.postgresql.org/media/keys/ACCC4CF8.asc | apt-key add -
-RUN apt-get update -y
-RUN apt-get upgrade -y
+    ## 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 \
 
-ENV PGVERSION 9.4
-RUN apt-get install python python-yaml python-requests python-boto postgresql-${PGVERSION} python-dnspython python-kazoo python-pip -y
-RUN apt-get install python-dev postgresql-server-dev-${PGVERSION} -y
-RUN pip install python-etcd psycopg2
+    && mkdir -p /home/postgres \
+    && chown postgres:postgres /home/postgres \
 
-ENV PATH /usr/lib/postgresql/${PGVERSION}/bin:$PATH
+    # Clean up
+    && apt-get purge -y libpython2.7-stdlib libpython2.7-minimal \
+    && apt-get autoremove -y \
+    && apt-get clean -y \
+    && rm -rf /var/lib/apt/lists/* /root/.cache
 
-ADD patroni.py /patroni.py
-ADD patroni/ /patroni
+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
 
-ENV ETCDVERSION 2.0.13
-RUN curl -L https://github.com/coreos/etcd/releases/download/v${ETCDVERSION}/etcd-v${ETCDVERSION}-linux-amd64.tar.gz | tar xz -C /bin --strip=1 --wildcards --no-anchored etcd etcdctl
+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
 
-### Setting up a simple script that will serve as an entrypoint
-RUN mkdir /data/ && touch /var/log/etcd.log /var/log/etcd.err /pgpass /patroni/postgres.yml
-RUN chown postgres:postgres -R /patroni/ /data/ /pgpass /var/log/etcd.* /patroni/postgres.yml
-ADD docker/entrypoint.sh /entrypoint.sh
+ADD patronictl.py patroni.py docker/entrypoint.sh /
+ADD patroni /patroni/
+ADD extras/confd /etc/confd
 
-EXPOSE 4001 5432 2380
+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/
 
+EXPOSE 2379 5432 8008
+
+ENV LC_ALL=en_US.UTF-8 LANG=en_US.UTF-8
 ENTRYPOINT ["/bin/bash", "/entrypoint.sh"]
 USER postgres

MAINTAINERS

@@ -0,0 +1,3 @@
+Alexander Kukushkin <[email protected]>
+Feike Steenbergen <[email protected]>
+Oleksii Kliukin <[email protected]>

README.rst

@@ -1,34 +1,111 @@
 |Build Status| |Coverage Status|
 
-Patroni: A Template for PostgreSQL HA with ZooKeeper or etcd
-------------------------------------------------------------
+Patroni: A Template for PostgreSQL HA with ZooKeeper, etcd or Consul
+--------------------------------------------------------------------
 
-Patroni was previously known as Governor.
+You can find a version of this documentation that is searchable and also easier to navigate at `patroni.readthedocs.io <https://patroni.readthedocs.io>`__.
 
-*There are many ways to run high availability with PostgreSQL. Here, we
-present a template for you to create your own customized, high-availability 
-solution using Python and — for maximum accessibility — a distributed 
-configuration store like ZooKeeper or etcd.*
 
-Getting Started
----------------
+There are many ways to run high availability with PostgreSQL; for a list, see the `PostgreSQL Documentation <https://wiki.postgresql.org/wiki/Replication,_Clustering,_and_Connection_Pooling>`__.
+
+Patroni is a template for you to create your own customized, high-availability solution using Python and - for maximum accessibility - a distributed configuration store like `ZooKeeper <https://zookeeper.apache.org/>`__, `etcd <https://github.com/coreos/etcd>`__, `Consul <https://github.com/hashicorp/consul>`__ or `Kubernetes <https://kubernetes.io>`__. Database engineers, DBAs, DevOps engineers, and SREs who are looking to quickly deploy HA PostgreSQL in the datacenter-or anywhere else-will hopefully find it useful.
+
+We call Patroni a "template" because it is far from being a one-size-fits-all or plug-and-play replication system. It will have its own caveats. Use wisely.
+
+**Note to Kubernetes users**: Patroni can run natively on top of Kubernetes. Take a look at the `Kubernetes <https://github.com/zalando/patroni/blob/master/docs/kubernetes.rst>`__ chapter of the Patroni documentation.
+
+.. contents::
+    :local:
+    :depth: 1
+    :backlinks: none
+
+=================
+How Patroni Works
+=================
+
+Patroni originated as a fork of `Governor <https://github.com/compose/governor>`__, the project from Compose. It includes plenty of new features.
+
+For an example of a Docker-based deployment with Patroni, see `Spilo <https://github.com/zalando/spilo>`__, currently in use at Zalando.
+
+For additional background info, see:
+
+* `Elephants on Automatic: HA Clustered PostgreSQL with Helm <https://www.youtube.com/watch?v=CftcVhFMGSY>`_, talk by Josh Berkus and Oleksii Kliukin at KubeCon Berlin 2017
+* `PostgreSQL HA with Kubernetes and Patroni <https://www.youtube.com/watch?v=iruaCgeG7qs>`__, talk by Josh Berkus at KubeCon 2016 (video)
+* `Feb. 2016 Zalando Tech blog post <https://tech.zalando.de/blog/zalandos-patroni-a-template-for-high-availability-postgresql/>`__
+
+==================
+Development Status
+==================
+
+Patroni is in active development and accepts contributions. See our `Contributing <https://github.com/zalando/patroni/blob/master/docs/CONTRIBUTING.rst>`__ section below for more details.
+
+We report new releases information `here <https://github.com/zalando/patroni/releases>`__.
+
+=========
+Community
+=========
+
+There are two places to connect with the Patroni community: `on github <https://github.com/zalando/patroni>`__, via Issues and PRs, and on channel #patroni in the `PostgreSQL Slack <https://postgres-slack.herokuapp.com/>`__.  If you're using Patroni, or just interested, please join us.
+
+===================================
+Technical Requirements/Installation
+===================================
+
+**Pre-requirements for Mac OS**
+
+To install requirements on a Mac, run the following:
+
+::
+
+    brew install postgresql etcd haproxy libyaml python
+
+**General installation for pip**
+
+Patroni can be installed with pip:
+
+::
+
+    pip install patroni[dependencies]
+
+where dependencies can be either empty, or consist of one or more of the following:
+
+etcd
+    `python-etcd` module in order to use Etcd as DCS
+consul
+    `python-consul` module in order to use Consul as DCS
+zookeeper
+    `kazoo` module in order to use Zookeeper as DCS
+exhibitor
+    `kazoo` module in order to use Exhibitor as DCS (same dependencies as for Zookeeper)
+kubernetes
+    `kubernetes` module in order to use Kubernetes as DCS in Patroni
+aws
+    `boto` in order to use AWS callbacks
+
+For example, the command in order to install Patroni together with dependencies for Etcd as a DCS and AWS callbacks is:
+
+::
+
+    pip install patroni[etcd,aws]
+
+Note that external tools to call in the replica creation or custom bootstap scripts (i.e. WAL-E) should be installed independently of Patroni.
+
+=======================
+Running and Configuring
+=======================
 
 To get started, do the following from different terminals:
-
 ::
 
     > etcd --data-dir=data/etcd
     > ./patroni.py postgres0.yml
     > ./patroni.py postgres1.yml
 
-From there, you will see a high-availability cluster start up. Test
-different settings in the YAML files to see how its behavior changes. Kill
-some of the components to see how the system behaves.
+You will then see a high-availability cluster start up. Test different settings in the YAML files to see how the cluster's behavior changes. Kill some of the components to see how the system behaves.
 
 Add more ``postgres*.yml`` files to create an even larger cluster.
 
-We provide a haproxy configuration, which will give your application a
-single endpoint for connecting to the cluster's leader. To configure,
+Patroni provides an `HAProxy <http://www.haproxy.org/>`__ configuration, which will give your application a single endpoint for connecting to the cluster's leader. To configure,
 run:
 
 ::
@@ -39,161 +116,29 @@ run:
 
     > psql --host 127.0.0.1 --port 5000 postgres
 
-How Patroni Works
------------------
-
-For a diagram of the high availability decision loop, review this PDF:
-`postgres-ha.pdf <https://github.com/zalando/patroni/blob/master/postgres-ha.pdf>`__
-
+==================
 YAML Configuration
-------------------
+==================
 
-For an example file, see ``postgres0.yml``. Regarding settings:
+Go `here <https://github.com/zalando/patroni/blob/master/docs/SETTINGS.rst>`__ for comprehensive information about settings for etcd, consul, and ZooKeeper. And for an example, see `postgres0.yml <https://github.com/zalando/patroni/blob/master/postgres0.yml>`__.
 
--  *ttl*: the TTL to acquire the leader lock. Think of it as the length of time before initiation of the automatic failover process.
--  *loop\_wait*: the number of seconds the loop will sleep
+=========================
+Environment Configuration
+=========================
 
--  *restapi*:
-    -  *listen*: IP address + port that Patroni will listen to, to provide health-check information for haproxy.
-    -  *connect\_address*: IP address + port through which restapi is accessible.
-    -  *auth*: (optional) 'username:password' to protect dangerous REST API endpoints.
-    -  *certfile*: (optional) Specifies a 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 a file with the secret key in the PEM format.
-
--  *etcd*:
-    -  *scope*: the relative path used on etcd's HTTP API for this deployment; makes it possible to run multiple HA deployments from a single etcd.
-    -  *ttl*: the TTL to acquire the leader lock. Think of it as the length of time before initiation of the automatic failover process.
-    -  *host*: the host:port for the etcd endpoint.
-
--  *zookeeper*:
-    -  *scope*: the relative path used on etcd's HTTP API for this deployment; makes it possible to run multiple HA deployments from a single etcd.
-    -  *session\_timeout*: the TTL to acquire the leader lock. Think of it as the length of time before initiation of the automatic failover process.
-    -  *reconnect\_timeout*: how long we should try to reconnect to ZooKeeper after a connection loss. After this timeout, assume that you no longer have a lock and restart in read-only mode.
-    -  *hosts*: list of ZooKeeper cluster members in format: ['host1:port1', 'host2:port2', 'etc...']
-    -  *exhibitor*: if you are running a ZooKeeper cluster under the Exhibitor supervisory, the following section might interest you:
-        -  *poll\_interval*: how often the list of ZooKeeper and Exhibitor nodes should be updated from Exhibitor
-        -  *port*: Exhibitor port.
-        -  *hosts*: initial list of Exhibitor (ZooKeeper) nodes in format: ['host1', 'host2', 'etc...' ]. This list updates automatically whenever the Exhibitor (ZooKeeper) cluster topology changes.
-
--  *postgresql*:
-    -  *name*: the name of the Postgres host. Must be unique for the cluster.
-    -  *listen*: IP address + port that Postgres listens to; must be accessible from other nodes in the cluster, if you're using streaming replication.
-    -  *connect\_address*: IP address + port through which Postgres is accessible from other nodes and applications.
-    -  *data\_dir*: file path to initialize and store Postgres data files.
-    -  *maximum\_lag\_on\_failover*: the maximum bytes a follower may lag.
-    -  *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.
-
-    -  *initdb*:  List options to be passed on to initdb
-        -  *encoding*: default encoding for new databases
-        -  *locale*: default locale for new databases
-        -  *data-checksums*  # When pg_rewind is needed on 9.3, this needs to be enabled
-
-    -  *pg\_hba*: list of lines which should be added to pg\_hba.conf.
-        -  *- host all all 0.0.0.0/0 md5*.
-
-    -  *replication*:
-        -  *username*: replication username; user will be created during initialization.
-        -  *password*: replication password; user will be created during initialization.
-        -  *network*: network setting for replication in pg\_hba.conf.
-
-    -  *callbacks* callback scripts to run on certain actions. Patroni will pass the action, role and cluster name. See scripts/aws.py as an example on how to write them.
-        -  *on\_start*: a script to run when the cluster starts.
-        -  *on\_stop*: a script to run when the cluster stops.
-        -  *on\_restart*: a script to run when the cluster restarts.
-        -  *on\_reload*: a script to run when configuration reload is triggered.
-        -  *on\_role\_change*: a script to run when the cluster is being promoted or demoted.
-
-    -  *superuser*:
-        -  *password*: password for the Postgres user, set during initialization.
-
-    -  *admin*:
-        -  *username*: admin username; user is created during initialization. It will have CREATEDB and CREATEROLE privileges.
-        -  *password*: admin password; user is created during initialization.
-
-    -  *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.
-
-    -  *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.
-
-    -  *replica\_method* for each create_replica_method 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".
+Go `here <https://github.com/zalando/patroni/blob/master/docs/ENVIRONMENT.rst>`__ for comprehensive information about configuring(overriding) settings via environment variables.
 
+===================
 Replication Choices
--------------------
+===================
 
-Patroni uses Postgres' streaming replication. By default, this
-replication is asynchronous. For more information, see the `Postgres
-documentation on streaming
-replication <http://www.postgresql.org/docs/current/static/warm-standby.html#STREAMING-REPLICATION>`__.
-
-Patroni's asynchronous replication configuration allows for
-``maximum_lag_on_failover`` settings. This setting ensures failover will
-not occur if a follower is more than a certain number of bytes behind
-the follower. This setting should be increased or decreased based on
-business requirements.
-
-When asynchronous replication is not optimal for your use case, investigate
-how Postgres's `synchronous
-replication <http://www.postgresql.org/docs/current/static/warm-standby.html#SYNCHRONOUS-REPLICATION>`__
-works. Synchronous replication ensures consistency across a cluster by
-confirming that writes are written to a secondary before returning to
-the connecting client with a success. The cost of synchronous
-replication: reduced throughput on writes. This throughput will
-be entirely based on network performance. In hosted datacenter
-environments (like AWS, Rackspace, or any network you do not control),
-synchrous replication significantly increases the variability of write 
-performance. If followers become inaccessible from the leader, the
-leader effectively becomes readonly.
-
-To enable a simple synchronous replication test, add the follow lines to
-the ``parameters`` section of your YAML configuration files:
-
-.. code:: YAML
-
-        synchronous_commit: "on"
-        synchronous_standby_names: "*"
-
-When using synchronous replication, use at least three Postgres data nodes
-to ensure write availability if one host fails.
-
-Choosing your replication schema is dependent on your business
-considerations. Investigate both async and sync replication, as well as other
-HA solutions, to determine which solution is best for you.
+Patroni uses Postgres' streaming replication, which is asynchronous by default. Patroni's asynchronous replication configuration allows for ``maximum_lag_on_failover`` settings. This setting ensures failover will not occur if a follower is more than a certain number of bytes behind the leader. This setting should be increased or decreased based on business requirements. It's also possible to use synchronous replication for better durability guarantees. See `replication modes documentation <https://github.com/zalando/patroni/blob/master/docs/replication_modes.rst>`__ for details.
 
+======================================
 Applications Should Not Use Superusers
---------------------------------------
+======================================
 
-When connecting from an application, always use a non-superuser. Patroni
-requires access to the database to function properly. By using a
-superuser from an application, you can potentially use the entire
-connection pool, including the connections reserved for superusers with
-the ``superuser_reserved_connections`` setting. If Patroni cannot access
-the Primary because the connection pool is full, behavior will be
-undesireable.
-
-Requirements on a Mac
----------------------
-
-Run the following on a Mac to install requirements:
-
-::
-
-    brew install postgresql etcd haproxy libyaml python
-    pip install psycopg2 pyyaml
-
-Notice
-------
-
-There are many different ways to do HA with PostgreSQL: See `the
-PostgreSQL
-documentation <https://wiki.postgresql.org/wiki/Replication,_Clustering,_and_Connection_Pooling>`__
-for a complete list.
-
-We call Patroni a "template" because it is far from being a one-size-fits-all
-or plug-and-play replication system. It will have its own caveats. Use wisely.
+When connecting from an application, always use a non-superuser. Patroni requires access to the database to function properly. By using a superuser from an application, you can potentially use the entire connection pool, including the connections reserved for superusers, with the ``superuser_reserved_connections`` setting. If Patroni cannot access the Primary because the connection pool is full, behavior will be undesirable.
 
 .. |Build Status| image:: https://travis-ci.org/zalando/patroni.svg?branch=master
    :target: https://travis-ci.org/zalando/patroni

TODO.md

@@ -1,4 +1,12 @@
 Failover
 ========
-- When determining who should become master, include the minor version of PostgreSQL in the decision
-- Create a way to disable governance of a cluster, something like the existence of a "nogover" or "admin" file in PGDATA will stop governor from changing the cluster state
+- When determining who should become master, include the minor version of PostgreSQL in the decision.
+
+Configuration
+==============
+- Provide a way to change pg_hba.conf of a running cluster on the Patroni level, without changing individual nodes.
+- Provide hooks to store and retrieve cluster-wide passwords without exposing them in a plain-text form to unauthorized users.
+
+Documentation
+==============
+- Document how to run cascading replication and possibly initialize the cluster without an access to the master node.

docker-compose.yml

@@ -0,0 +1,58 @@
+# docker compose file for running a 3-node PostgreSQL cluster
+# with etcd as the SIS
+
+patroni_etcd:
+    container_name: patroni_etcd
+    image: patroni
+    command: --etcd
+
+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
+
+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
+
+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
+
+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

docker/dev_patroni_cluster.sh

@@ -67,24 +67,38 @@ while getopts "$optspec" optchar; do
     esac
 done
 
-function random_name()
-{
-    cat /dev/urandom | LC_ALL=C tr -dc 'a-zA-Z0-9' | head -c 8
-}
-
-if [ -z ${PATRONI_SCOPE} ]
-then
-    PATRONI_SCOPE=$(random_name)
+if [ -z ${PATRONI_SCOPE} ]; then
+    PATRONI_SCOPE=$(cat /dev/urandom | LC_ALL=C tr -dc 'a-z0-9' | head -c 8)
 fi
 
-etcd_container=$(docker run -P -d --name="${PATRONI_SCOPE}_etcd" "${DOCKER_IMAGE}" --etcd-only)
-etcd_container_ip=$(docker inspect --format '{{ .NetworkSettings.IPAddress }}' ${etcd_container})
-echo "The etcd container is ${etcd_container}, ip=${etcd_container_ip}"
+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}"
+}
 
-for i in $(seq 1 "${MEMBERS}")
-do
-    container_name=$(random_name)
-    patroni_container=$(docker run -P -d --name="${PATRONI_SCOPE}_${container_name}" "${DOCKER_IMAGE}" --etcd="${etcd_container_ip}:4001" --name="${PATRONI_SCOPE}")
-    patroni_container_ip=$(docker inspect --format '{{ .NetworkSettings.IPAddress }}' ${patroni_container})
-    echo "Started Patroni container ${patroni_container}, ip=${patroni_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

@@ -7,49 +7,52 @@ Usage: $0
 
 Options:
 
-    --etcd      ETCD    Provide an external etcd to connect to
-    --name      NAME    Give the cluster a specific name
-    --etcd-only         Do not run Patroni, run a standalone etcd
+    --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=127.17.0.84:4001
-    $0 --etcd-only
+    $0 --etcd
+    $0 --confd
+    $0 --zookeeper
     $0
-    $0 --name=true_scotsman
 __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 "${OPTARG}" in
-                etcd-only)
-                    exec etcd --data-dir /tmp/etcd.data \
-                        -advertise-client-urls=http://${DOCKER_IP}:4001 \
-                        -listen-client-urls=http://0.0.0.0:4001 \
-                        -listen-peer-urls=http://0.0.0.0:2380
-                    exit 0
+                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 $ETCD_ARGS
+                    ;;
+                zookeeper)
+                    exec /usr/share/zookeeper/bin/zkServer.sh start-foreground
                     ;;
                 cheat)
                     CHEAT=1
                     ;;
-                name)
-                    PATRONI_SCOPE="${!OPTIND}"; OPTIND=$(( $OPTIND + 1 ))
-                    ;;
-                name=*)
-                    PATRONI_SCOPE=${OPTARG#*=}
-                    ;;
-                etcd)
-                    ETCD_CLUSTER="${!OPTIND}"; OPTIND=$(( $OPTIND + 1 ))
-                    ;;
-                etcd=*)
-                    ETCD_CLUSTER=${OPTARG#*=}
-                    ;;
                 help)
                     usage
                     exit 0
@@ -70,70 +73,43 @@ while getopts "$optspec" optchar; do
     esac
 done
 
-if [ -z ${ETCD_CLUSTER} ]
-then
-    etcd --data-dir /tmp/etcd.data \
-        -advertise-client-urls=http://${DOCKER_IP}:4001 \
-        -listen-client-urls=http://0.0.0.0:4001 \
-        -listen-peer-urls=http://0.0.0.0:2380 > /var/log/etcd.log 2> /var/log/etcd.err &
-    ETCD_CLUSTER="127.0.0.1:4001"
+## 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 &
+    export PATRONI_ETCD_URL="http://127.0.0.1:2379"
 fi
 
-cat > /patroni/postgres.yml <<__EOF__
+export PATRONI_SCOPE
+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_OPTIONS="${PATRONI_admin_OPTIONS:-createdb, createrole}"
+export PATRONI_POSTGRESQL_CONNECT_ADDRESS="${DOCKER_IP}:5432"
+export PATRONI_POSTGRESQL_LISTEN="0.0.0.0:5432"
+export PATRONI_POSTGRESQL_DATA_DIR="data/${PATRONI_SCOPE}"
+export PATRONI_REPLICATION_USERNAME="${PATRONI_REPLICATION_USERNAME:-replicator}"
+export PATRONI_REPLICATION_PASSWORD="${PATRONI_REPLICATION_PASSWORD:-abcd}"
+export PATRONI_SUPERUSER_USERNAME="${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
 
-ttl: &ttl 30
-loop_wait: &loop_wait 10
-scope: &scope '${PATRONI_SCOPE}'
-namespace: 'patroni'
-restapi:
-  listen: 0.0.0.0:8008
-  connect_address: ${DOCKER_IP}:8008
-etcd:
-  scope: *scope
-  ttl: *ttl
-  host: ${ETCD_CLUSTER}
-postgresql:
-  name: ${HOSTNAME}
-  scope: *scope
-  listen: 0.0.0.0:5432
-  connect_address: ${DOCKER_IP}:5432
-  data_dir: data/postgresql0
-  maximum_lag_on_failover: 1048576 # 1 megabyte in bytes
   pg_hba:
   - host all all 0.0.0.0/0 md5
-  - hostssl all all 0.0.0.0/0 md5
   - host replication replicator ${DOCKER_IP}/16    md5
-  replication:
-    username: replicator
-    password: rep-pass
-    network:  127.0.0.1/32
-  superuser:
-    password: zalando
-  restore: patroni/scripts/restore.py
-  admin:
-    username: admin
-    password: admin
-  parameters:
-    archive_mode: "on"
-    wal_level: hot_standby
-    archive_command: 'true'
-    max_wal_senders: 20
-    listen_addresses: 0.0.0.0
-    checkpoint_segments: 64
-    wal_keep_segments: 64
-    archive_timeout: 1800s
-    max_replication_slots: 20
-    hot_standby: "on"
 __EOF__
 
-cat /patroni/postgres.yml
+mkdir -p "$HOME/.config/patroni"
+[ -h "$HOME/.config/patroni/patronictl.yaml" ] || ln -s /patroni.yml "$HOME/.config/patroni/patronictl.yaml"
 
-if [ ! -z $CHEAT ]
-then
-    while :
-    do
-        sleep 60
-    done
-else
-    exec python /patroni.py /patroni/postgres.yml
-fi
+[ -z $CHEAT ] && exec python3 /patroni.py /patroni.yml
+
+while true; do
+    sleep 60
+done

docker/patroni-secrets.env

@@ -0,0 +1,8 @@
+PATRONI_RESTAPI_USERNAME=admin
+PATRONI_RESTAPI_PASSWORD=admin
+PATRONI_SUPERUSER_USERNAME=postgres
+PATRONI_SUPERUSER_PASSWORD=postgres
+PATRONI_REPLICATION_USERNAME=replicator
+PATRONI_REPLICATION_PASSWORD=replicate
+PATRONI_admin_PASSWORD=admin
+PATRONI_admin_OPTIONS=createdb,createrole

docs/CONTRIBUTING.rst

@@ -0,0 +1,29 @@
+.. _contributing:
+
+Contributing guidelines
+=======================
+
+Wanna contribute to Patroni? Yay - here is how!
+
+Chatting
+--------
+
+Just want to chat with other Patroni users?  Looking for interactive troubleshooting help? Join us on channel #patroni in the `PostgreSQL Slack <https://postgres-slack.herokuapp.com/>`__.
+
+Reporting issues
+----------------
+
+If you have a question about patroni or have a problem using it, please read the :ref:`README <readme>` before filing an issue.
+Also double check with the current issues on our `Issues Tracker <https://github.com/zalando/patroni/issues>`__.
+
+Contributing a pull request
+---------------------------
+
+1) Submit a comment to the relevant issue or create a new issue describing your proposed change.
+2) Do a fork, develop and test your code changes.
+3) Include documentation
+4) Submit a pull request.
+
+You'll get feedback about your pull request as soon as possible.
+
+Happy Patroni hacking ;-)

docs/ENVIRONMENT.rst

@@ -0,0 +1,99 @@
+.. _environment:
+
+Environment Configuration Settings
+==================================
+
+It is possible to override some of the configuration parameters defined in the Patroni configuration file using the system environment variables. This document lists all environment variables handled by Patroni. The values set via those variables always take precedence over the ones set in the Patroni configuration file.
+
+Global/Universal
+----------------
+-  **PATRONI\_CONFIGURATION**: it is possible to set the entire configuration for the Patroni via ``PATRONI_CONFIGURATION`` environment variable. In this case any other environment variables will not be considered!
+-  **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\_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\_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\_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
+-----------------------
+It is possible to create new database users right after the successful initialization of a new cluster. This process is defined by the following variables:
+
+-  **PATRONI\_<username>\_PASSWORD='<password>'**
+-  **PATRONI\_<username>\_OPTIONS='list,of,options'**
+
+Example: defining ``PATRONI_admin_PASSWORD=strongpasswd`` and ``PATRONI_admin_OPTIONS='createrole,createdb'`` will cause creation of the user **admin** with the password **strongpasswd** that is allowed to create other users and databases.
+
+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\_PORT**: (optional) Consul port
+-  **PATRONI\_CONSUL\_SCHEME**: (optional) **http** or **https**, defaults to **http**
+-  **PATRONI\_CONSUL\_TOKEN**: (optional) ACL token
+-  **PATRONI\_CONSUL\_VERIFY**: (optional) whether to verify the SSL certificate for HTTPS requests
+-  **PATRONI\_CONSUL\_CACERT**: (optional) The ca certificate. If present it will enable validation.
+-  **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\_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\_SRV**: Domain to search the SRV record(s) for cluster autodiscovery.
+-  **PATRONI\_ETCD\_CACERT**: The ca certificate. If present it will enable validation.
+-  **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.
+
+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.
+-  **PATRONI\_EXHIBITOR\_PORT**: Exhibitor port.
+
+.. _kubernetes_environment:
+
+Kubernetes
+----------
+-  **PATRONI\_KUBERNETES\_NAMESPACE**: (optional) Kubernetes namespace where the Patroni pod is running. Default value is `default`.
+-  **PATRONI\_KUBERNETES\_LABELS**: Labels in format ``{label1: value1, label2: value2}``. These labels will be used to find existing objects (Pods and either Endpoints or ConfigMaps) associated with the current cluster. Also Patroni will set them on every object (Endpoint or ConfigMap) it creates.
+-  **PATRONI\_KUBERNETES\_SCOPE\_LABEL**: (optional) name of the label containing cluster name. Default value is `cluster-name`.
+-  **PATRONI\_KUBERNETES\_ROLE\_LABEL**: (optional) name of the label containing Postgres role (`master` or `replica`). Patroni will set this label on the pod it is running in. Default value is `role`.
+-  **PATRONI\_KUBERNETES\_USE\_ENDPOINTS**: (optional) if set to true, Patroni will use Endpoints instead of ConfigMaps to run leader elections and keep cluster state.
+-  **PATRONI\_KUBERNETES\_POD\_IP**: (optional) IP address of the pod Patroni is running in. This value is required when `PATRONI_KUBERNETES_USE_ENDPOINTS` is enabled and is used to populate the leader endpoint subsets when the pod's PostgreSQL is promoted.
+-  **PATRONI\_KUBERNETES\_PORTS**: (optional) if the Service object has the name for the port, the same name must appear in the Endpoint object, otherwise service won't work. For example, if your service is defined as ``{Kind: Service, spec: {ports: [{name: postgresql, port: 5432, targetPort: 5432}]}}``, then you have to set ``PATRONI_KUBERNETES_PORTS='{[{"name": "postgresql", "port": 5432}]}'`` and Patroni will use it for updating subsets of the leader Endpoint. This parameter is used only if `PATRONI_KUBERNETES_USE_ENDPOINTS` is set.
+
+PostgreSQL
+----------
+-  **PATRONI\_POSTGRESQL\_LISTEN**: IP address + port that Postgres listens to. 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.
+-  **PATRONI\_POSTGRESQL\_CONNECT\_ADDRESS**: IP address + port through which Postgres is accessible from other nodes and applications.
+-  **PATRONI\_POSTGRESQL\_DATA\_DIR**: The location of the Postgres data directory, either existing or to be initialized by Patroni.
+-  **PATRONI\_POSTGRESQL\_CONFIG\_DIR**: The location of the Postgres configuration directory, defaults to the data directory. Must be writable by Patroni.
+-  **PATRONI\_POSTGRESQL\_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.
+-  **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\_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).
+
+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.
+
+ZooKeeper
+---------
+-  **PATRONI\_ZOOKEEPER\_HOSTS**: comma separated list of ZooKeeper cluster members: "'host1:port1','host2:port2','etc...'". It is important to quote every single entity!

docs/Makefile

@@ -0,0 +1,20 @@
+# Minimal makefile for Sphinx documentation
+#
+
+# You can set these variables from the command line.
+SPHINXOPTS    =
+SPHINXBUILD   = sphinx-build
+SPHINXPROJ    = Patroni
+SOURCEDIR     = .
+BUILDDIR      = build
+
+# Put it first so that "make" without argument is like "make help".
+help:
+	@$(SPHINXBUILD) -M help "$(SOURCEDIR)" "$(BUILDDIR)" $(SPHINXOPTS) $(O)
+
+.PHONY: help Makefile
+
+# Catch-all target: route all unknown targets to Sphinx using the new
+# "make mode" option.  $(O) is meant as a shortcut for $(SPHINXOPTS).
+%: Makefile
+	@$(SPHINXBUILD) -M $@ "$(SOURCEDIR)" "$(BUILDDIR)" $(SPHINXOPTS) $(O)

docs/README.rst

@@ -0,0 +1,125 @@
+.. _readme:
+
+============
+Introduction
+============
+
+Patroni originated as a fork of `Governor <https://github.com/compose/governor>`__, the project from Compose. It includes plenty of new features.
+
+For an example of a Docker-based deployment with Patroni, see `Spilo <https://github.com/zalando/spilo>`__, currently in use at Zalando.
+
+For additional background info, see:
+
+* `PostgreSQL HA with Kubernetes and Patroni <https://www.youtube.com/watch?v=iruaCgeG7qs>`__, talk by Josh Berkus at KubeCon 2016 (video)
+* `Feb. 2016 Zalando Tech blog post <https://tech.zalando.de/blog/zalandos-patroni-a-template-for-high-availability-postgresql/>`__
+
+
+Development Status
+------------------
+
+Patroni is in active development and accepts contributions. See our :ref:`Contributing <contributing>` section below for more details.
+
+We report new releases information :ref:`here <releases>`.
+
+
+Technical Requirements/Installation
+-----------------------------------
+
+**Pre-requirements for Mac OS**
+
+To install requirements on a Mac, run the following:
+
+::
+
+    brew install postgresql etcd haproxy libyaml python
+
+**General installation for pip**
+
+Patroni can be installed with pip:
+
+::
+
+    pip install patroni[dependencies]
+
+where dependencies can be either empty, or consist of one or more of the following:
+
+etcd
+    `python-etcd` module in order to use Etcd as DCS
+consul
+    `python-consul` module in order to use Consul as DCS
+zookeeper
+    `kazoo` module in order to use Zookeeper as DCS
+exhibitor
+    `kazoo` module in order to use Exhibitor as DCS (same dependencies as for Zookeeper)
+kubernetes
+    `kubernetes` module in order to use Kubernetes as DCS in Patroni
+aws
+    `boto` in order to use AWS callbacks
+
+For example, the command in order to install Patroni together with dependencies for Etcd as a DCS and AWS callbacks is:
+
+::
+
+    pip install patroni[etcd,aws]
+
+Note that external tools to call in the replica creation or custom bootstrap scripts (i.e. WAL-E) should be installed
+independently of Patroni.
+
+
+Running and Configuring
+-----------------------
+
+The following section assumes Patroni repository as being cloned from https://github.com/zalando/patroni. Namely, you
+will need example configuration files `postgres0.yml` and `postgres1.yml`. If you installed Patroni with pip, you can
+obtain those files from the git repository and replace `./patroni.py` below with `patroni` command.
+
+To get started, do the following from different terminals:
+::
+
+    > etcd --data-dir=data/etcd
+    > ./patroni.py postgres0.yml
+    > ./patroni.py postgres1.yml
+
+You will then see a high-availability cluster start up. Test different settings in the YAML files to see how the cluster's behavior changes. Kill some of the components to see how the system behaves.
+
+Add more ``postgres*.yml`` files to create an even larger cluster.
+
+Patroni provides an `HAProxy <http://www.haproxy.org/>`__ configuration, which will give your application a single endpoint for connecting to the cluster's leader. To configure,
+run:
+
+::
+
+    > haproxy -f haproxy.cfg
+
+::
+
+    > psql --host 127.0.0.1 --port 5000 postgres
+
+
+YAML Configuration
+------------------
+
+Go :ref:`here <settings>` for comprehensive information about settings for etcd, consul, and ZooKeeper. And for an example, see `postgres0.yml <https://github.com/zalando/patroni/blob/master/postgres0.yml>`__.
+
+
+Environment Configuration
+-------------------------
+
+Go :ref:`here <environment>` for comprehensive information about configuring(overriding) settings via environment variables.
+
+
+Replication Choices
+-------------------
+
+Patroni uses Postgres' streaming replication, which is asynchronous by default. Patroni's asynchronous replication configuration allows for ``maximum_lag_on_failover`` settings. This setting ensures failover will not occur if a follower is more than a certain number of bytes behind the leader. This setting should be increased or decreased based on business requirements. It's also possible to use synchronous replication for better durability guarantees. See :ref:`replication modes documentation <replication_modes>` for details.
+
+
+Applications Should Not Use Superusers
+--------------------------------------
+
+When connecting from an application, always use a non-superuser. Patroni requires access to the database to function properly. By using a superuser from an application, you can potentially use the entire connection pool, including the connections reserved for superusers, with the ``superuser_reserved_connections`` setting. If Patroni cannot access the Primary because the connection pool is full, behavior will be undesirable.
+
+.. |Build Status| image:: https://travis-ci.org/zalando/patroni.svg?branch=master
+   :target: https://travis-ci.org/zalando/patroni
+.. |Coverage Status| image:: https://coveralls.io/repos/zalando/patroni/badge.svg?branch=master
+   :target: https://coveralls.io/r/zalando/patroni?branch=master

docs/SETTINGS.rst

@@ -0,0 +1,195 @@
+.. _settings:
+
+===========================
+YAML Configuration Settings
+===========================
+
+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>`_)
+-  **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>`_)
+-  **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 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
+    -  **slots**: define permanent replication slots. These slots will be preserved during switchover/failover. Patroni will try to create slots before opening connections to the cluster.
+        -  **my_slot_name**: the name of replication slot. It is the responsibility of the operator to make sure that there are no clashes in names between replication slots automatically created by Patroni for members and permanent replication slots.
+            -  **type**: slot type. Could be ``physical`` or ``logical``. If the slot is logical, you have to additionally define ``database`` and ``plugin``.
+               **database**: the database name where logical slots should be created.
+               **plugin**: the plugin name for the logical slot.
+-  **method**: custom script to use for bootstrapping this cluster.
+   See :ref:`custom bootstrap methods documentation <custom_bootstrap>` for details.
+   When ``initdb`` is specified revert to the default ``initdb`` command. ``initdb`` is also triggered when no ``method``
+   parameter is present in the configuration file.
+-  **initdb**: List options to be passed on to initdb.
+        -  **- data-checksums**: Must be enabled when pg_rewind is needed on 9.3.
+        -  **- encoding: UTF8**: default encoding for new databases.
+        -  **- locale: UTF8**: default locale for new databases.
+-  **pg\_hba**: list of lines that you should add to pg\_hba.conf.
+        -  **- host all all 0.0.0.0/0 md5**.
+        -  **- host replication replicator 127.0.0.1/32 md5**: A line like this is required for replication.
+-  **users**: Some additional users which need to be created after initializing new cluster
+    -  **admin**: the name of user
+        -  **password: zalando**:
+        -  **options**: list of options for CREATE USER statement
+            -  **- createrole**
+            -  **- createdb**
+-  **post\_bootstrap** or **post\_init**: An additional script that will be executed after initializing the cluster. The script receives a connection string URL (with the cluster superuser as a user name). The PGPASSFILE variable is set to the location of pgpass file.
+
+.. _consul_settings:
+
+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
+-  **cacert**: (optional) The ca certificate. If present it will enable validation.
+-  **cert**: (optional) file with the client certificate
+-  **key**: (optional) file with the client key. 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
+-  **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
+----
+Most of the parameters are optional, but you have to specify one of the **host**, **hosts**, **url**, **proxy** or **srv**
+
+-  **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**
+-  **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
+-  **password**: (optional) password for etcd authentication.
+-  **cacert**: (optional) The ca certificate. If present it will enable validation.
+-  **cert**: (optional) file with the client certificate
+-  **key**: (optional) file with the client key. Can be empty if the key is part of **cert**.
+
+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
+-  **port**: Exhibitor port.
+
+.. _kubernetes_settings:
+
+Kubernetes
+----------
+-  **namespace**: (optional) Kubernetes namespace where Patroni pod is running. Default value is `default`.
+-  **labels**: Labels in format ``{label1: value1, label2: value2}``. These labels will be used to find existing objects (Pods and either Endpoints or ConfigMaps) associated with the current cluster. Also Patroni will set them on every object (Endpoint or ConfigMap) it creates.
+-  **scope\_label**: (optional) name of the label containing cluster name. Default value is `cluster-name`.
+-  **role\_label**: (optional) name of the label containing role (master or replica). Patroni will set this label on the pod it runs in. Default value is ``role``.
+-  **use\_endpoints**: (optional) if set to true, Patroni will use Endpoints instead of ConfigMaps to run leader elections and keep cluster state.
+-  **pod\_ip**: (optional) IP address of the pod Patroni is running in. This value is required when `use_endpoints` is enabled and is used to populate the leader endpoint subsets when the pod's PostgreSQL is promoted.
+-  **ports**: (optional) if the Service object has the name for the port, the same name must appear in the Endpoint object, otherwise service won't work. For example, if your service is defined as ``{Kind: Service, spec: {ports: [{name: postgresql, port: 5432, targetPort: 5432}]}}``, then you have to set ``kubernetes.ports: {[{"name": "postgresql", "port": 5432}]}`` and Patroni will use it for updating subsets of the leader Endpoint. This parameter is used only if `kubernetes.use_endpoints` is set.
+
+.. _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".
+
+REST API
+-------- 
+-  **connect\_address**: IP address (or hostname) and port, to access the Patroni's REST API. All the members of the cluster must be able to connect to this address, so unless the Patroni setup is intended for a demo inside the localhost, this address must be a non "localhost" or loopback addres (ie: "localhost" or "127.0.0.1"). It can serve as a endpoint for HTTP health checks (read below about the "listen" REST API parameter), and also for user queries (either directly or via the REST API), as well as for the health checks done by the cluster members during leader elections (for example, to determine whether the master is still running, or if there is a node which has a WAL position that is ahead of the one doing the query; etc.) The connect_address is put in the member key in DCS, making it possible to translate the member name into the address to connect to its REST API.
+
+-  **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**:
+            -  **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.
+
+.. _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...'].
+
+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.

docs/_static/custom.css

@@ -0,0 +1,3 @@
+li {
+   margin-bottom: 0.5em
+}

docs/conf.py

@@ -0,0 +1,197 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+#
+# Patroni documentation build configuration file, created by
+# sphinx-quickstart on Mon Dec 19 16:54:09 2016.
+#
+# This file is execfile()d with the current directory set to its
+# containing dir.
+#
+# Note that not all possible configuration values are present in this
+# autogenerated file.
+#
+# All configuration values have a default; values that are commented out
+# serve to show the default.
+
+# If extensions (or modules to document with autodoc) are in another directory,
+# add these directories to sys.path here. If the directory is relative to the
+# documentation root, use os.path.abspath to make it absolute, like shown here.
+#
+import os
+
+import sys
+sys.path.insert(0, os.path.abspath('..'))
+
+from patroni.version import __version__
+
+# -- General configuration ------------------------------------------------
+
+# If your documentation needs a minimal Sphinx version, state it here.
+#
+# needs_sphinx = '1.0'
+
+# Add any Sphinx extension module names here, as strings. They can be
+# extensions coming with Sphinx (named 'sphinx.ext.*') or your custom
+# ones.
+extensions = ['sphinx.ext.intersphinx',
+    'sphinx.ext.todo',
+    'sphinx.ext.mathjax',
+    'sphinx.ext.ifconfig',
+    'sphinx.ext.viewcode']
+
+# Add any paths that contain templates here, relative to this directory.
+templates_path = ['_templates']
+
+# The suffix(es) of source filenames.
+# You can specify multiple suffix as a list of string:
+#
+# source_suffix = ['.rst', '.md']
+source_suffix = '.rst'
+
+# The master toctree document.
+master_doc = 'index'
+
+# General information about the project.
+project = 'Patroni'
+copyright = '2015 Compose, Zalando SE'
+author = 'Zalando SE'
+
+# The version info for the project you're documenting, acts as replacement for
+# |version| and |release|, also used in various other places throughout the
+# built documents.
+#
+# The short X.Y version.
+version = __version__[:__version__.rfind('.')]
+# The full version, including alpha/beta/rc tags.
+release = __version__
+
+# The language for content autogenerated by Sphinx. Refer to documentation
+# for a list of supported languages.
+#
+# This is also used if you do content translation via gettext catalogs.
+# Usually you set "language" from the command line for these cases.
+language = None
+
+# List of patterns, relative to source directory, that match files and
+# directories to ignore when looking for source files.
+# This patterns also effect to html_static_path and html_extra_path
+exclude_patterns = []
+
+# The name of the Pygments (syntax highlighting) style to use.
+pygments_style = 'sphinx'
+
+# If true, `todo` and `todoList` produce output, else they produce nothing.
+todo_include_todos = True
+
+
+# -- Options for HTML output ----------------------------------------------
+
+# The theme to use for HTML and HTML Help pages.  See the documentation for
+# a list of builtin themes.
+#
+
+on_rtd = os.environ.get('READTHEDOCS', None) == 'True'
+if not on_rtd:  # only import and set the theme if we're building docs locally
+    import sphinx_rtd_theme
+    html_theme = 'sphinx_rtd_theme'
+    html_theme_path = [sphinx_rtd_theme.get_html_theme_path()]
+
+# Theme options are theme-specific and customize the look and feel of a theme
+# further.  For a list of options available for each theme, see the
+# documentation.
+#
+# html_theme_options = {}
+
+# Add any paths that contain custom static files (such as style sheets) here,
+# relative to this directory. They are copied after the builtin static files,
+# so a file named "default.css" will overwrite the builtin "default.css".
+html_static_path = ['_static']
+
+
+# -- Options for HTMLHelp output ------------------------------------------
+
+# Output file base name for HTML help builder.
+htmlhelp_basename = 'Patronidoc'
+
+
+# -- Options for LaTeX output ---------------------------------------------
+
+latex_elements = {
+    # The paper size ('letterpaper' or 'a4paper').
+    #
+    # 'papersize': 'letterpaper',
+
+    # The font size ('10pt', '11pt' or '12pt').
+    #
+    # 'pointsize': '10pt',
+
+    # Additional stuff for the LaTeX preamble.
+    #
+    # 'preamble': '',
+
+    # Latex figure (float) alignment
+    #
+    # 'figure_align': 'htbp',
+}
+
+# Grouping the document tree into LaTeX files. List of tuples
+# (source start file, target name, title,
+#  author, documentclass [howto, manual, or own class]).
+latex_documents = [
+    (master_doc, 'Patroni.tex', 'Patroni Documentation',
+     'Zalando SE', 'manual'),
+]
+
+
+# -- Options for manual page output ---------------------------------------
+
+# One entry per manual page. List of tuples
+# (source start file, name, description, authors, manual section).
+man_pages = [
+    (master_doc, 'patroni', 'Patroni Documentation',
+     [author], 1)
+]
+
+
+# -- Options for Texinfo output -------------------------------------------
+
+# Grouping the document tree into Texinfo files. List of tuples
+# (source start file, target name, title, author,
+#  dir menu entry, description, category)
+texinfo_documents = [
+    (master_doc, 'Patroni', 'Patroni Documentation',
+     author, 'Patroni', 'One line description of project.',
+     'Miscellaneous'),
+]
+
+
+
+# -- Options for Epub output ----------------------------------------------
+
+# Bibliographic Dublin Core info.
+epub_title = project
+epub_author = author
+epub_publisher = author
+epub_copyright = copyright
+
+# The unique identifier of the text. This can be a ISBN number
+# or the project homepage.
+#
+# epub_identifier = ''
+
+# A unique identification for the text.
+#
+# epub_uid = ''
+
+# A list of files that should not be packed into the epub file.
+epub_exclude_files = ['search.html']
+
+
+
+# Example configuration for intersphinx: refer to the Python standard library.
+intersphinx_mapping = {'https://docs.python.org/': None}
+
+# A possibility to have an own stylesheet, to add new rules or override existing ones
+# For the latter case, the CSS specificity of the rules should be higher than the default ones
+def setup(app):
+   app.add_stylesheet("custom.css")

docs/dynamic_configuration.rst

@@ -0,0 +1,229 @@
+.. _dynamic_configuration:
+
+Patroni configuration
+=====================
+
+Patroni configuration is stored in the DCS (Distributed Configuration Store). There are 3 types of configuration:
+
+- Dynamic configuration.
+	These options can be set in DCS at any time. If the options changed are not part of the startup configuration,
+	they are applied asynchronously (upon the next wake up cycle) to every node, which gets subsequently reloaded.
+	If the node requires a restart to apply the configuration (for options with context postmaster, if their values
+	have changed), a special flag, ``pending_restart`` indicating this, is set in the members.data JSON.
+	Additionally, the node status also indicates this, by showing ``"restart_pending": true``.
+
+- 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.
+
+- Environment :ref:`configuration <environment>`.
+	It is possible to set/override some of the "Local" configuration parameters with environment variables.
+	Environment configuration is very useful when you are running in a dynamic environment and you don't know some of the parameters in advance (for example it's not possible to know your external IP address when you are running inside ``docker``).
+
+Some of the PostgreSQL parameters must hold the same values on the master and the replicas. For those, values set either in the local patroni configuration files or via the environment variables take no effect. To alter or set their values one must change the shared configuration in the DCS. Below is the actual list of such parameters together with the default values:
+
+- max_connections: 100
+- max_locks_per_transaction: 64
+- max_worker_processes: 8
+- max_prepared_transactions: 0
+- wal_level: hot_standby
+- wal_log_hints: on
+- track_commit_timestamp: off
+
+For the parameters below, PostgreSQL does not require equal values among the master and all the replicas. However, considering the possibility of a replica to become the master at any time, it doesn't really make sense to set them differently; therefore, Patroni restricts setting their values to the Dynamic configuration
+
+- max_wal_senders: 5
+- max_replication_slots: 5
+- wal_keep_segments: 8
+
+These parameters are validated to ensure they are sane, or meet a minimum value.
+
+There are some other Postgres parameters controlled by Patroni:
+
+- listen_addresses - is set either from ``postgresql.listen`` or from ``PATRONI_POSTGRESQL_LISTEN`` environment variable
+- port - is set either from ``postgresql.listen`` or from ``PATRONI_POSTGRESQL_LISTEN`` environment variable
+- cluster_name - is set either from ``scope`` or from ``PATRONI_SCOPE`` environment variable
+- hot_standby: on
+
+To be on the safe side parameters from the above lists are not written into ``postgresql.conf``, but passed as a list of arguments to the ``pg_ctl start`` which gives them the highest precedence, even above `ALTER SYSTEM <https://www.postgresql.org/docs/current/static/sql-altersystem.html>`__
+
+
+When applying the local or dynamic configuration options, the following actions are taken:
+
+- The node first checks if there is a postgresql.base.conf or if the ``custom_conf`` parameter is set.
+- If the `custom_conf` parameter is set, it will take the file specified on it as a base configuration, ignoring `postgresql.base.conf` and `postgresql.conf`.
+- If the `custom_conf` parameter is not set and `postgresql.base.conf` exists, it contains the renamed "original" configuration and it will be used as a base configuration.
+- If there is no `custom_conf` nor `postgresql.base.conf`, the original postgresql.conf is taken and renamed to postgresql.base.conf.
+- The dynamic options (with the exceptions above) are dumped into the postgresql.conf and an include is set in
+  postgresql.conf to the used base configuration (either postgresql.base.conf or what is on ``custom_conf``). Therefore, we would be able to apply new options without re-reading the configuration file to check if the include is present not.
+- Some parameters that are essential for Patroni to manage the cluster are overridden using the command line.
+- If some of the options that require restart are changed (we should look at the context in pg_settings and at the actual
+  values of those options), a pending_restart flag of a given node is set. This flag is reset on any restart.
+
+The parameters would be applied in the following order (run-time are given the highest priority):
+
+1. load parameters from file `postgresql.base.conf` (or from a `custom_conf` file, if set)
+2. load parameters from file `postgresql.conf`
+3. load parameters from file `postgresql.auto.conf`
+4. run-time parameter using `-o --name=value`
+
+This allows configuration for all the nodes (2), configuration for a specific node using `ALTER SYSTEM` (3) and ensures that parameters essential to the running of Patroni are enforced (4), as well as leaves room for configuration tools that manage `postgresql.conf` directly without involving Patroni (1).
+
+
+Also, the following Patroni configuration options can be changed only dynamically:
+
+- ttl: 30
+- loop_wait: 10
+- retry_timeouts: 10
+- maximum_lag_on_failover: 1048576
+- 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/ha_loop_diagram.dot

@@ -0,0 +1,148 @@
+// Graphviz source for ha_loop_diagram.png
+// recompile with:
+// dot -Tpng ha_loop_diagram.dot -o ha_loop_diagram.png
+
+digraph G {
+  rankdir=TB;
+  fontname="sans-serif";
+  penwidth="0.3";
+  layout="dot";
+  newrank=true;
+  edge [fontname="sans-serif", 
+        fontsize=12, 
+        color=black, 
+        fontcolor=black];
+  node [fontname=serif, 
+        fontsize=12, 
+        fillcolor=white, 
+        color=black, 
+        fontcolor=black,
+        style=filled];
+  "start" [label=Start, shape="rectangle", fillcolor="green"]
+  "start" -> "load_cluster_from_dcs";
+  "update_member" [label="Persist node state in DCS"]
+  "update_member" -> "start"
+  
+  subgraph cluster_run_cycle {
+    label="run_cycle"
+    "load_cluster_from_dcs" [label="Load cluster from DCS"];
+    "touch_member" [label="Persist node in DCS"];
+    "cluster.has_member" [shape="diamond", label="Is node registered on DCS?"]
+    "cluster.has_member" -> "touch_member"  [label="no" color="red"]
+    "long_action_in_progress?" [shape="diamond" label="Is the PostgreSQL currently being\nstopping/starting/restarting/reinitializing?"]
+    "load_cluster_from_dcs" -> "cluster.has_member";
+    "touch_member" -> "long_action_in_progress?";
+    "cluster.has_member" -> "long_action_in_progress?"  [label="yes" color="green"];
+    "long_action_in_progress?" -> "recovering?" [label="no" color="red"]
+    "recovering?" [label="Was cluster recovering and failed?", shape="diamond"];
+    "recovering?" -> "post_recover" [label="yes" color="green"];
+    "recovering?" -> "data_directory_empty" [label="no" color="red"];
+    "post_recover" [label="Remove leader key (if I was the leader)"];
+    "data_directory_empty" [label="Is data folder empty?", shape="diamond"];
+    "data_directory_empty" -> "cluster_initialize" [label="no" color="red"];
+    "data_belongs_to_cluster" [label="Does data dir belong to cluster?", shape="diamond"];
+    "data_belongs_to_cluster" -> "exit" [label="no" color="red"];
+    "data_belongs_to_cluster" -> "is_healthy" [label="yes" color="green"]
+    "exit" [label="Fail and exit", fillcolor=red];
+    "cluster_initialize" [label="Is cluster initialized on DCS?" shape="diamond"]
+    "cluster_initialize" -> "cluster.has_leader" [label="no" color="red"]
+    "cluster.has_leader" [label="Does the cluster has leader?", shape="diamond"]
+    "cluster.has_leader" -> "dcs.initialize" [label="no", color="red"]
+    "cluster.has_leader" -> "is_healthy" [label="yes", color="green"]
+    "cluster_initialize" -> "data_belongs_to_cluster" [label="yes" color="green"]
+    "dcs.initialize" [label="Initialize new cluster"];
+    "dcs.initialize" -> "is_healthy"
+    "is_healthy" [label="Is node healthy?\n(running Postgres)", shape="diamond"];
+    "recover" [label="Start as read-only\nand set Recover flag"]
+    "is_healthy" -> "recover" [label="no" color="red"]; 
+    "is_healthy" -> "cluster.is_unlocked" [label="yes" color="green"]; 
+    "cluster.is_unlocked" [label="Does the cluster has a leader?", shape="diamond"]
+  }
+
+  "post_recover" -> "update_member"
+  "recover" -> "update_member"
+  "long_action_in_progress?" -> "async_has_lock?" [label="yes" color="green"];
+  "cluster.is_unlocked" -> "unhealthy_is_healthiest" [label="no" color="red"]
+  "cluster.is_unlocked" -> "healthy_has_lock" [label="yes" color="green"]
+  "data_directory_empty" -> "bootstrap.is_unlocked" [label="yes" color="green"]
+
+  subgraph cluster_async {
+    label = "Long action in progress\n(Start/Stop/Restart/Reinitialize)"
+    "async_has_lock?" [label="Do I have the leader lock?", shape="diamond"]
+    "async_update_lock" [label="Renew leader lock"]
+    "async_has_lock?" -> "async_update_lock" [label="yes" color="green"]
+  }
+  "async_update_lock" -> "update_member"
+  "async_has_lock?" -> "update_member" [label="no" color="red"]
+
+  subgraph cluster_bootstrap {
+    label = "Node bootstrap";
+    "bootstrap.is_unlocked" [label="Does the cluster has a leader?", shape="diamond"]
+    "bootstrap.is_initialized" [label="Does the cluster has an initialize key?", shape="diamond"]
+    "bootstrap.is_unlocked" -> "bootstrap.is_initialized" [label="no" color="red"]
+    "bootstrap.is_unlocked" -> "bootstrap.select_node" [label="yes" color="green"]
+    "bootstrap.select_node" [label="Select a node to take a backup from"]
+    "bootstrap.do_bootstrap" [label="Run pg_basebackup\n(async)"]
+    "bootstrap.select_node" -> "bootstrap.do_bootstrap"
+    "bootstrap.is_initialized" -> "bootstrap.initialization_race" [label="no" color="red"]
+    "bootstrap.is_initialized" -> "bootstrap.wait_for_leader" [label="yes" color="green"]
+    "bootstrap.initialization_race" [label="Race for initialize key"]
+    "bootstrap.initialization_race" -> "bootstrap.won_initialize_race?"
+    "bootstrap.won_initialize_race?" [label="Do I won initialize race?", shape="diamond"]
+    "bootstrap.won_initialize_race?" -> "bootstrap.initdb_and_start" [label="yes" color="green"]
+    "bootstrap.won_initialize_race?" -> "bootstrap.wait_for_leader" [label="no" color="red"]
+    "bootstrap.wait_for_leader" [label="Need to wait for leader key"]
+    "bootstrap.initdb_and_start" [label="Run initdb, start postgres and create roles"]
+    "bootstrap.initdb_and_start" -> "bootstrap.success?"
+    "bootstrap.success?" [label="Success", shape="diamond"]
+    "bootstrap.success?" -> "bootstrap.take_leader_key" [label="yes" color="green"]
+    "bootstrap.success?" -> "bootstrap.clean" [label="no" color="red"]
+    "bootstrap.clean" [label="Remove initialize key from DCS\nand data directory from filesystem"]
+    "bootstrap.take_leader_key" [label="Take a leader key in DCS"]
+  }
+
+  "bootstrap.do_bootstrap" -> "update_member"
+  "bootstrap.wait_for_leader" -> "update_member"
+  "bootstrap.clean" -> "update_member"
+  "bootstrap.take_leader_key" -> "update_member"
+
+  subgraph cluster_process_healthy_cluster {
+    label = "process_healthy_cluster"
+    "healthy_has_lock" [label="Am I the owner of the leader lock?", shape=diamond]
+    "healthy_is_leader" [label="Is Postgres running as master?", shape=diamond]
+    "healthy_no_lock" [label="Follow the leader (async,\ncreate/update recovery.conf and restart if necessary)"]
+    "healthy_has_lock" -> "healthy_no_lock" [label="no" color="red"]
+    "healthy_has_lock" -> "healthy_update_leader_lock" [label="yes" color="green"]
+    "healthy_update_leader_lock" [label="Try to update leader lock"]
+    "healthy_update_leader_lock" -> "healthy_update_success"
+    "healthy_update_success" [label="Success?", shape=diamond]
+    "healthy_update_success" -> "healthy_is_leader" [label="yes" color="green"]
+    "healthy_update_success" -> "healthy_demote" [label="no" color="red"]
+    "healthy_demote" [label="Demote (async,\nrestart in read-only)"]
+    "healthy_failover" [label="Promote Postgres to master"]
+    "healthy_is_leader" -> "healthy_failover" [label="no" color="red"]
+  }
+  "healthy_demote" -> "update_member"
+  "healthy_is_leader" -> "update_member" [label="yes" color="green"]
+  "healthy_failover" -> "update_member"
+  "healthy_no_lock" -> "update_member"
+
+  subgraph cluster_process_unhealthy_cluster {
+    label = "process_unhealthy_cluster"
+    "unhealthy_is_healthiest" [label="Am I the healthiest node?", shape="diamond"]
+    "unhealthy_is_healthiest" -> "unhealthy_leader_race" [label="yes", color="green"]
+    "unhealthy_leader_race" [label="Try to create leader key"]
+    "unhealthy_leader_race" -> "unhealthy_acquire_lock"
+    "unhealthy_acquire_lock" [label="Was I able to get the lock?", shape="diamond"]
+    "unhealthy_is_leader" [label="Is Postgres running as master?", shape=diamond]
+    "unhealthy_acquire_lock" -> "unhealthy_is_leader" [label="yes" color="green"]
+    "unhealthy_is_leader" -> "unhealthy_promote" [label="no" color="red"]
+    "unhealthy_promote" [label="Promote to master"]
+    "unhealthy_is_healthiest" -> "unhealthy_follow" [label="no" color="red"]
+    "unhealthy_follow" [label="try to follow somebody else()"]
+    "unhealthy_acquire_lock" -> "unhealthy_follow" [label="no" color="red"]
+  }
+  "unhealthy_follow" -> "update_member"
+  "unhealthy_promote" -> "update_member"
+  "unhealthy_is_leader" -> "update_member" [label="yes" color="green"]
+}

docs/index.rst

@@ -0,0 +1,37 @@
+.. Patroni documentation master file, created by
+   sphinx-quickstart on Mon Dec 19 16:54:09 2016.
+   You can adapt this file completely to your liking, but it should at least
+   contain the root `toctree` directive.
+
+Introduction
+============
+
+Patroni is a template for you to create your own customized, high-availability solution using Python and - for maximum accessibility - a distributed configuration store like `ZooKeeper <https://zookeeper.apache.org/>`__, `etcd <https://github.com/coreos/etcd>`__, `Consul <https://github.com/hashicorp/consul>`__ or `Kubernetes <https://kubernetes.io>`__. Database engineers, DBAs, DevOps engineers, and SREs who are looking to quickly deploy HA PostgreSQL in the datacenter-or anywhere else-will hopefully find it useful.
+
+We call Patroni a "template" because it is far from being a one-size-fits-all or plug-and-play replication system. It will have its own caveats. Use wisely. There are many ways to run high availability with PostgreSQL; for a list, see the `PostgreSQL Documentation <https://wiki.postgresql.org/wiki/Replication,_Clustering,_and_Connection_Pooling>`__.
+
+**Note to Kubernetes users**: Patroni can run natively on top of Kubernetes. Take a look at the :ref:`Kubernetes <kubernetes>` chapter of the Patroni documentation.
+
+
+.. toctree::
+   :maxdepth: 2
+   :caption: Contents:
+
+   README
+   dynamic_configuration
+   ENVIRONMENT
+   SETTINGS
+   replica_bootstrap
+   replication_modes
+   pause
+   kubernetes
+   watchdog
+   releases
+   CONTRIBUTING
+
+Indices and tables
+==================
+
+* :ref:`genindex`
+* :ref:`modindex`
+* :ref:`search`

docs/kubernetes.rst

@@ -0,0 +1,53 @@
+.. _kubernetes:
+
+Using Patroni with Kubernetes
+=============================
+
+Patroni can use Kubernetes objects in order to store the state of the cluster and manage the leader key. That makes it
+capable of operating Postgres in Kubernetes environment without any consistency store, namely, one doesn't
+need to run an extra Etcd deployment. There are two different type of Kubernetes objects Patroni can use to store the
+leader and the configuration keys, they are configured with the `kubernetes.use_endpoints` or `PATRONI_KUBERNETES_USE_ENDPOINTS`
+environment variable.
+
+Use Endpoints
+-------------
+
+Despite the fact that this is the recommended mode, it is turned off by default for compatibility reasons. When it is on, Patroni stores
+the cluster configuration and the leader key in the `metadata: annotations` fields of the respective `Endpoints` it creates.
+Changing the leader is safer than when using `ConfigMaps`, since both the annotations, containing the leader information, and the actual addresses
+pointing to the running leader pod are updated simultaneously in one go.
+
+Use ConfigMaps
+--------------
+
+In this mode, Patroni will create ConfigMaps instead of Endpoints and store keys inside meta-data of those ConfigMaps.
+Changing the leader takes at least two updates, one to the leader ConfigMap and another to the respective Endpoint.
+
+There are two ways to direct the traffic to the Postgres master:
+
+- use the `callback script <https://github.com/zalando/patroni/blob/master/kubernetes/callback.py>`_ provided by Patroni
+- configure the Kubernetes Postgres service to use the label selector with the `role_label` (configured in patroni configuration).
+
+Note that in some cases, for instance, when running on OpenShift, there is no alternative to using ConfigMaps.
+
+Configuration
+-------------
+
+Patroni Kubernetes :ref:`settings <kubernetes_settings>` and :ref:`environment variables <kubernetes_environment>` are described in the general chapters of the documentation.
+
+Examples
+--------
+
+- The `kubernetes <https://github.com/zalando/patroni/tree/master/kubernetes>`__ folder of the Patroni repository contains
+  examples of the Docker image, the Kubernetes manifest and the callback script in order to test Patroni Kubernetes setup.
+  Note that in the current state it will not be able to use PersistentVolumes because of permission issues.
+
+- You can find the full-featured Docker image that can use Persistent Volumes in the
+  `Spilo Project <https://github.com/zalando/spilo>`_.
+
+- There is also a `Helm chart <https://github.com/kubernetes/charts/tree/master/incubator/patroni>`_
+  to deploy the Spilo image configured with Patroni running using Kubernetes.
+
+- In order to run your database clusters at scale using Patroni and Spilo, take a look at the
+  `postgres-operator <https://github.com/zalando-incubator/postgres-operator>`_ project. It implements the operator pattern
+  to manage Spilo clusters.

docs/pause.rst

@@ -0,0 +1,35 @@
+.. _pause:
+
+Pause/Resume mode for the cluster
+=================================
+
+The goal
+--------
+
+Under certain circumstances Patroni needs to temporary step down from managing the cluster, while still retaining the cluster state in DCS. Possible use cases are uncommon activities on the cluster, such as major version upgrades or corruption recovery. During those activities nodes are often started and stopped for the reason unknown to Patroni, some nodes can be even temporary promoted, violating the assumption of running only one master. Therefore, Patroni needs to be able to "detach" from the running cluster, implementing an equivalent of the maintenance mode in Pacemaker.
+
+
+
+The implementation
+------------------
+
+When Patroni runs in a paused mode, it does not change the state of PostgreSQL, except for the following cases:
+
+- For each node, the member key in DCS is updated with the current information about the cluster. This causes Patroni to run read-only queries on a member node if the member is running.
+
+- For the Postgres master with the leader lock Patroni updates the lock. If the node with the leader lock stops being the master (i.e. is demoted manually), Patroni will release the lock instead of promoting the node back.
+
+- Manual unscheduled restart, reinitialize and manual failover are allowed. Manual failover is only allowed if the node to failover to is specified. In the paused mode, manual failover does not require a running master node.
+
+- If 'parallel' masters are detected by Patroni, it emits a warning, but does not demote the masters without the leader lock.
+
+- If there is no leader lock in the cluster, the running master acquires the lock. If there is more than one master node, then the first master to acquire the lock wins. If there are no masters altogether, Patroni does not try to promote any replicas. There is an exception in this rule: if there is no leader lock because the old master has demoted itself due to the manual promotion, then only the candidate node mentioned in the promotion request may take the leader lock. When the new leader lock is granted (i.e. after promoting a replica manually), Patroni makes sure the replicas that were streaming from the previous leader will switch to the new one.
+
+- When Postgres is stopped, Patroni does not try to start it. When Patroni is stopped, it does not try to stop the Postgres instance it is managing.
+
+User guide
+----------
+
+``patronictl`` supports ``pause`` and ``resume`` commands.
+
+One can also issue a ``PATCH`` request to the ``{namespace}/{cluster}/config`` key with ``{"pause": true/false/null}``

docs/replica_bootstrap.rst

@@ -0,0 +1,199 @@
+Replica imaging and bootstrap
+=============================
+
+Patroni allows customizing creation of a new replica. It also supports defining what happens when the new empty cluster
+is being bootstrapped. The distinction between two is well defined: Patroni creates replicas only if the ``initialize``
+key is present in DCS for the cluster. If there is no ``initialize`` key - Patroni calls bootstrap exclusively on the
+first node that takes the initialize key lock.
+
+.. _custom_bootstrap:
+
+Bootstrap
+---------
+
+PostgreSQL provides ``initdb`` command to initialize a new cluster and Patroni calls it by default. In certain cases,
+particularly when creating a new cluster as a copy of an existing one, it is necessary to replace a built-in method with
+custom actions. Patroni supports executing user-defined scripts to bootstrap new clusters, supplying some required
+arguments to them, i.e. the name of the cluster and the path to the data directory. This is configured in the
+``bootstrap`` section of the Patroni configuration. For example:
+
+.. code:: YAML
+
+    bootstrap:
+        method: <custom_bootstrap_method_name>
+        <custom_bootstrap_method_name>:
+            command: <path_to_custom_bootstrap_script> [param1 [, ...]]
+            keep_existing_recovery_conf: False
+            recovery_conf:
+                recovery_target_action: promote
+                recovery_target_timeline: latest
+                restore_command: <method_specific_restore_command>
+
+
+Each bootstrap method must define at least a ``name`` and a ``command``. A special ``initdb`` method is available to trigger
+the default behavior, in which case ``method`` parameter can be omitted altogether. The ``command`` can be specified using either
+an absolute path, or the one relative to the ``patroni`` command location. In addition to the fixed parameters defined
+in the configuration files, Patroni supplies two cluster-specific ones:
+
+--scope
+    Name of the cluster to be bootstrapped
+--datadir
+    Path to the data directory of the cluster instance to be bootstrapped
+
+If the bootstrap script returns 0, Patroni tries to configure and start the PostgreSQL instance produced by it. If any
+of the intermediate steps fail, or the script returns a non-zero value, Patroni assumes that the bootstrap has failed,
+cleans up after itself and releases the initialize lock to give another node the opportunity to bootstrap.
+
+If a ``recovery_conf`` block is defined in the same section as the custom bootstrap method, Patroni will generate a
+``recovery.conf`` before starting the newly bootstrapped instance. Typically, such recovery.conf should contain at least
+one of the ``recovery_target_*`` parameters, together with the ``recovery_target_timeline`` set to ``promote``.
+
+If ``keep_existing_recovery_conf`` is defined and set to ``True``, Patroni will not remove the existing ``recovery.conf`` file if it exists.
+This is useful when bootstrapping from a backup with tools like pgBackRest that generate the appropriate ``recovery.conf`` for you.
+
+ .. note:: Bootstrap methods are neither chained, nor fallen-back to the default one in case the primary one fails
+
+
+.. _custom_replica_creation:
+
+Building replicas
+-----------------
+
+Patroni uses tried and proven ``pg_basebackup`` in order to create new replicas. One downside of it is that it requires
+a running master node. Another one is the lack of 'on-the-fly' compression for the backup data and no built-in cleanup
+for outdated backup files. Some people prefer other backup solutions, such as ``WAL-E``, ``pgBackRest``, ``Barman`` and
+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:
+        create_replica_methods:
+            - wal_e
+            - basebackup
+        wal_e:
+            command: patroni_wale_restore
+            no_master: 1
+            envdir: {{WALE_ENV_DIR}}
+            use_iam: 1
+        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
+to execute and any custom parameters that should be passed to that command. All parameters will be passed in a
+``--name=value`` format. Besides user-defined parameters, Patroni supplies a couple of cluster-specific ones:
+
+--scope
+    Which cluster this replica belongs to
+--datadir
+    Path to the data directory of the replica
+--role
+    Always 'replica'
+--connstring
+    Connection string to connect to the cluster member to clone from (master or other replica). The user in the
+    connection string can execute SQL and replication protocol commands.
+
+A special ``no_master`` parameter, if defined, allows Patroni to call the replica creation method even if there is no
+running master or replicas. In that case, an empty string will be passed in a connection string. This is useful for
+restoring the formerly running cluster from the binary backup.
+
+A special ``keep_data`` parameter, if defined, will instuct Patroni to  not clean PGDATA folder before calling restore.
+
+A special ``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
+``pg_basebackup``, the base backup is taken from the master unless there are replicas with ``clonefrom`` tag, in which case one
+of such replicas will be used as the origin for pg_basebackup. It works without any configuration; however, it is
+possible to specify a ``basebackup`` configuration section. Same rules as with the other method configuration apply,
+namely, only long (with --) options should be specified there. Not all parameters make sense, if you override a connection
+string or provide an option to created tar-ed or compressed base backups, patroni won't be able to make a replica out
+of it. There is no validation performed on the names or values of the parameters passed to the ``basebackup`` section.
+You can specify basebackup parameters as either a map (key-value pairs) or a list of elements, where each element
+could be either a key-value pair or a single key (for options that does not receive any values, for instance, ``--verbose``).
+Consider those 2 examples:
+
+.. code:: YAML
+
+    postgresql:
+        basebackup:
+            max-rate: '100M'
+            checkpoint: 'fast'
+
+and
+
+.. code:: YAML
+
+    postgresql:
+        basebackup:
+            - verbose
+            - max-rate: '100M'
+
+If all replica creation methods fail, Patroni will try again all methods in order during the next event loop cycle.
+
+.. _standby_cluster:
+
+Standby cluster
+---------------
+
+Another available option is to run a "standby cluster", that contains only of
+standby nodes replicating from some remote master. This type of clusters has:
+
+* "standby leader", that behaves pretty much like a regular cluster leader,
+  except it replicates from a remote master.
+
+* cascade replicas, that are replicating from standby leader.
+
+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.
+
+To configure such cluster you need to specify the section ``standby_cluster``
+in a patroni configuration:
+
+.. code:: YAML
+
+    bootstrap:
+        dcs:
+            standby_cluster:
+                host: 1.2.3.4
+                port: 5432
+                primary_slot_name: patroni
+
+Note, that these options will be applied only once during cluster bootstrap,
+and the only way to change them afterwards is through DCS.
+
+If you use replication slots on the standby cluster, you must also create the corresponding replication slot on the primary cluster.  It will not be done automatically by the standby cluster implementation.  You can use Patroni's permenant replication slots feature on the primary cluster to maintain a replication slot with the same name as ``primary_slot_name``, or its default value if ``primary_slot_name`` is not provided.

docs/replication_modes.rst

@@ -0,0 +1,79 @@
+.. _replication_modes:
+
+=================
+Replication modes
+=================
+
+Patroni uses PostgreSQL streaming replication. For more information about streaming replication, see the `Postgres documentation <http://www.postgresql.org/docs/current/static/warm-standby.html#STREAMING-REPLICATION>`__. By default Patroni configures PostgreSQL for asynchronous replication. Choosing your replication schema is dependent on your business considerations. Investigate both async and sync replication, as well as other HA solutions, to determine which solution is best for you.
+
+Asynchronous mode durability
+----------------------------
+
+In asynchronous mode the cluster is allowed to lose some committed transactions to ensure availability. When the primary server fails or becomes unavailable for any other reason Patroni will automatically promote a sufficiently healthy standby to primary. Any transactions that have not been replicated to that standby remain in a "forked timeline" on the primary, and are effectively unrecoverable [1]_.
+
+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
+----------------------------------
+
+You can use Postgres's `synchronous replication <http://www.postgresql.org/docs/current/static/warm-standby.html#SYNCHRONOUS-REPLICATION>`__ with Patroni. Synchronous replication ensures consistency across a cluster by confirming that writes are written to a secondary before returning to the connecting client with a success. The cost of synchronous replication: reduced throughput on writes. This throughput will be entirely based on network performance.
+
+In hosted datacenter environments (like AWS, Rackspace, or any network you do not control), synchronous replication significantly increases the variability of write performance. If followers become inaccessible from the leader, the leader effectively becomes read-only.
+
+To enable a simple synchronous replication test, add the following lines to the ``parameters`` section of your YAML configuration files:
+
+.. code:: YAML
+
+        synchronous_commit: "on"
+        synchronous_standby_names: "*"
+
+When using PostgreSQL synchronous replication, use at least three Postgres data nodes to ensure write availability if one host fails.
+
+Using PostgreSQL synchronous replication does not guarantee zero lost transactions under all circumstances. When the primary and the secondary that is currently acting as a synchronous replica fail simultaneously a third node that might not contain all transactions will be promoted.
+
+.. _synchronous_mode:
+
+Synchronous mode
+----------------
+
+For use cases where losing committed transactions is not permissible you can turn on Patroni's ``synchronous_mode``. When ``synchronous_mode`` is turned on Patroni will not promote a standby unless it is certain that the standby contains all transactions that may have returned a successful commit status to client [2]_. This means that the system may be unavailable for writes even though some servers are available. System administrators can still use manual failover commands to promote a standby even if it results in transaction loss.
+
+Turning on ``synchronous_mode`` does not guarantee multi node durability of commits under all circumstances. When no suitable standby is available, primary server will still accept writes, but does not guarantee their replication. When the primary fails in this mode no standby will be promoted. When the host that used to be the primary comes back it will get promoted automatically, unless system administrator performed a manual failover. This behavior makes synchronous mode usable with 2 node clusters.
+
+When ``synchronous_mode`` is on and a standby crashes, commits will block until next iteration of Patroni runs and switches the primary to standalone mode (worst case delay for writes ``ttl`` seconds, average case ``loop_wait``/2 seconds). Manually shutting down or restarting a standby will not cause a commit service interruption. Standby will signal the primary to release itself from synchronous standby duties before PostgreSQL shutdown is initiated.
+
+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 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``),
+blocking all client write requests until at least one synchronous replica comes
+up.
+
+You can ensure that a standby never becomes the synchronous standby by setting ``nosync`` tag to true. This is recommended to set for standbys that are behind slow network connections and would cause performance degradation when becoming a synchronous standby.
+
+Synchronous mode can be switched on and off via Patroni REST interface. See :ref:`dynamic configuration <dynamic_configuration>` for instructions.
+
+
+Synchronous mode implementation
+-------------------------------
+
+When in synchronous mode Patroni maintains synchronization state in the DCS, containing the latest primary and current synchronous standby. This state is updated with strict ordering constraints to ensure the following invariants:
+
+- A node must be marked as the latest leader whenever it can accept write transactions. Patroni crashing or PostgreSQL not shutting down can cause violations of this invariant.
+
+- A node must be set as the synchronous standby in PostgreSQL as long as it is published as the synchronous standby.
+
+- A node that is not the leader or current synchronous standby is not allowed to promote itself automatically.
+
+Patroni will only ever assign one standby to ``synchronous_standby_names`` because with multiple candidates it is not possible to know which node was acting as synchronous during the failure.
+
+On each HA loop iteration Patroni re-evaluates synchronous standby choice. If the current synchronous standby is connected and has not requested its synchronous status to be removed it remains picked. Otherwise the cluster member available for sync that is furthest ahead in replication is picked.
+
+
+.. [1] The data is still there, but recovering it requires a manual recovery effort by data recovery specialists. When Patroni is allowed to rewind with ``use_pg_rewind`` the forked timeline will be automatically erased to rejoin the failed primary with the cluster.
+
+.. [2] Clients can change the behavior per transaction using PostgreSQL's ``synchronous_commit`` setting. Transactions with ``synchronous_commit`` values of ``off`` and ``local`` may be lost on fail over, but will not be blocked by replication delays.

docs/watchdog.rst

@@ -0,0 +1,39 @@
+.. _watchdog:
+
+Watchdog support
+================
+
+Having multiple PostgreSQL servers running as master can result in transactions lost due to diverging timelines. This situation is also called a split-brain problem. To avoid split-brain Patroni needs to ensure PostgreSQL will not accept any transaction commits after leader key expires in the DCS. Under normal circumstances Patroni will try to achieve this by stopping PostgreSQL when leader lock update fails for any reason. However, this may fail to happen due to various reasons:
+
+- Patroni has crashed due to a bug, out-of-memory condition or by being accidentally killed by a system administrator.
+
+- Shutting down PostgreSQL is too slow.
+
+- Patroni does not get to run due to high load on the system, the VM being paused by the hypervisor, or other infrastructure issues.
+
+To guarantee correct behavior under these conditions Patroni supports watchdog devices. Watchdog devices are software or hardware mechanisms that will reset the whole system when they do not get a keepalive heartbeat within a specified timeframe. This adds an additional layer of fail safe in case usual Patroni split-brain protection mechanisms fail.
+
+Patroni will try to activate the watchdog before promoting PostgreSQL to master. If watchdog activation fails and watchdog mode is ``required`` then the node will refuse to become master. When deciding to participate in leader election Patroni will also check that watchdog configuration will allow it to become leader at all. After demoting PostgreSQL (for example due to a manual failover) Patroni will disable the watchdog again. Watchdog will also be disabled while Patroni is in paused state.
+
+By default Patroni will set up the watchdog to expire 5 seconds before TTL expires. With the default setup of ``loop_wait=10`` and ``ttl=30`` this gives HA loop at least 15 seconds (``ttl`` - ``safety_margin`` - ``loop_wait``) to complete before the system gets forcefully reset. By default accessing DCS is configured to time out after 10 seconds. This means that when DCS is unavailable, for example due to network issues, Patroni and PostgreSQL will have at least 5 seconds (``ttl`` - ``safety_margin`` - ``loop_wait`` - ``retry_timeout``) to come to a state where all client connections are terminated.
+
+Safety margin is the amount of time that Patroni reserves for time between leader key update and watchdog keepalive. Patroni will try to send a keepalive immediately after confirmation of leader key update. If Patroni process is suspended for extended amount of time at exactly the right moment the keepalive may be delayed for more than the safety margin without triggering the watchdog. This results in a window of time where watchdog will not trigger before leader key expiration, invalidating the guarantee. To be absolutely sure that watchdog will trigger under all circumstances set up the watchdog to expire after half of TTL by setting ``safety_margin`` to -1 to set watchdog timeout to ``ttl // 2``. If you need this guarantee you probably should increase ``ttl`` and/or reduce ``loop_wait`` and ``retry_timeout``.
+
+Currently watchdogs are only supported using Linux watchdog device interface.
+
+Setting up software watchdog on Linux
+-------------------------------------
+
+Default Patroni configuration will try to use ``/dev/watchdog`` on Linux if it is accessible to Patroni. For most use cases using software watchdog built into the Linux kernel is secure enough.
+
+To enable software watchdog issue the following commands as root before starting Patroni:
+
+.. code-block:: bash
+
+    modprobe softdog
+    # Replace postgres with the user you will be running patroni under
+    chown postgres /dev/watchdog
+
+For testing it may be helpful to disable rebooting by adding ``soft_noboot=1`` to the modprobe command line. In this case the watchdog will just log a line in kernel ring buffer, visible via `dmesg`.
+
+Patroni will log information about the watchdog when it is successfully enabled.

extras/README.md

@@ -0,0 +1,13 @@
+### confd
+
+`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 and pgbouncer.ini with the actual list of Patroni nodes from `etcd` and "reload" haproxy and pgbouncer.ini when it is necessary.
+
+
+### startup-scripts
+
+`startup-scripts` directory contains startup scripts for various OSes and management tools for Patroni.

extras/confd/conf.d/haproxy.toml

@@ -0,0 +1,13 @@
+[template]
+#prefix = "/service/batman"
+#owner = "haproxy"
+#mode = "0644"
+src = "haproxy.tmpl"
+dest = "/etc/haproxy/haproxy.cfg"
+
+check_cmd = "/usr/sbin/haproxy -c -f {{ .src }}"
+reload_cmd = "haproxy -f /etc/haproxy/haproxy.cfg -p /var/run/haproxy.pid -D -sf $(cat /var/run/haproxy.pid)"
+
+keys = [
+    "/members/",
+]

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/haproxy.tmpl

@@ -0,0 +1,32 @@
+global
+	maxconn 100
+
+defaults
+	log	global
+	mode	tcp
+	retries 2
+	timeout client 30m
+	timeout connect 4s
+	timeout server 30m
+	timeout check 5s
+
+listen stats
+	mode http
+	bind *:7000
+	stats enable
+	stats uri /
+
+listen master
+	bind *:5000
+	option httpchk OPTIONS /master
+	http-check expect status 200
+	default-server inter 3s fall 3 rise 2 on-marked-down shutdown-sessions
+{{range gets "/members/*"}}	server {{base .Key}} {{$data := json .Value}}{{base (replace (index (split $data.conn_url "/") 2) "@" "/" -1)}} maxconn 100 check port {{index (split (index (split $data.api_url "/") 2) ":") 1}}
+{{end}}
+listen replicas
+	bind *:5001
+	option httpchk OPTIONS /replica
+	http-check expect status 200
+	default-server inter 3s fall 3 rise 2 on-marked-down shutdown-sessions
+{{range gets "/members/*"}}	server {{base .Key}} {{$data := json .Value}}{{base (replace (index (split $data.conn_url "/") 2) "@" "/" -1)}} maxconn 100 check port {{index (split (index (split $data.api_url "/") 2) ":") 1}}
+{{end}}

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/README.md

@@ -8,3 +8,15 @@ Scripts supplied:
 ### patroni.upstart.conf
 
 Upstart job for Ubuntu 12.04 or 14.04.  Requires Upstart > 1.4. Intended for systems where Patroni has been installed on a base system, rather than in Docker.
+
+### patroni.service
+Systemd service file, to be copied to /etc/systemd/system/patroni.service, tested on Centos 7.1 with Patroni installed from pip. 
+
+### patroni
+Init.d service file for Debian-like distributions. Copy it to /etc/init.d/, make executable:
+```chmod 755 /etc/init.d/patroni``` and run with ```service patroni start```, or make it starting on boot with ```update-rc.d patroni defaults```. Also you might edit some configuration variables in it:
+PATRONI for patroni.py location
+CONF for configuration file
+LOGFILE for log (script creates it if does not exist)
+
+Note. If you have several versions of Postgres installed, please add to POSTGRES_VERSION the release number which you wish to run. Script uses this value to append PATH environment with correct path to Postgres bin.

extras/startup-scripts/patroni

@@ -0,0 +1,144 @@
+#!/bin/sh
+#
+### BEGIN INIT INFO
+# Provides:          patroni
+# Required-Start:    $remote_fs $syslog
+# Required-Stop:     $remote_fs $syslog
+# Default-Start:     2 3 4 5
+# Default-Stop:      0 1 6
+# Short-Description: Patroni init script
+# Description:       Runners to orchestrate a high-availability PostgreSQL
+### END INIT INFO
+
+### BEGIN USER CONFIGURATION
+
+CONF="/etc/patroni/postgres.yml"
+LOGFILE="/var/log/patroni.log"
+USER="postgres"
+GROUP="postgres"
+
+NAME=patroni
+PATRONI="/opt/patroni/$NAME.py"
+PIDFILE="/var/run/$NAME.pid"
+
+# Set this parameter, if you have several Postgres versions installed
+# POSTGRES_VERSION="9.4"
+POSTGRES_VERSION=""
+
+### END USER CONFIGURATION
+
+. /lib/lsb/init-functions
+
+# Loading this library for get_versions() function
+if test ! -e /usr/share/postgresql-common/init.d-functions; then
+  log_failure_msg "Probably postgresql-common does not installed."
+  exit 1
+else
+  . /usr/share/postgresql-common/init.d-functions
+fi
+
+# Is there Patroni executable?
+if test ! -e $PATRONI; then
+  log_failure_msg "Patroni executable $PATRONI does not exist."
+  exit 1
+fi
+
+# Is there Patroni configuration file?
+if test ! -e $CONF; then
+  log_failure_msg "Patroni configuration file $CONF does not exist."
+  exit 1
+fi
+
+# Create logfile if doesn't exist
+if test ! -e $LOGFILE; then
+  log_action_msg "Creating logfile for Patroni..."
+  touch $LOGFILE
+  chown $USER:$GROUP $LOGFILE
+fi
+
+prepare_pgpath() {
+  if [ "$POSTGRES_VERSION" != "" ]; then
+    if [ -x /usr/lib/postgresql/$POSTGRES_VERSION/bin/pg_ctl ]; then
+      PGPATH="/usr/lib/postgresql/$POSTGRES_VERSION/bin"
+    else
+      log_failure_msg "Postgres version incorrect, check POSTGRES_VERSION variable."
+      exit 0
+    fi
+  else
+    get_versions
+    if echo $versions | grep -q -e "\s"; then
+      log_warning_msg "You have several Postgres versions installed. Please, use POSTGRES_VERSION to define correct environment."
+    else
+      versions=`echo $versions | sed -e 's/^[ \t]*//'`
+      PGPATH="/usr/lib/postgresql/$versions/bin"
+    fi
+  fi
+}
+
+get_pid() {
+  if test -e $PIDFILE; then
+    PID=`cat $PIDFILE`
+    CHILDPID=`ps --ppid $PID -o %p --no-headers`
+  else
+    log_failure_msg "Could not find PID file. Patroni probably down."
+    exit 1
+  fi
+}
+
+
+case "$1" in
+  start)
+    prepare_pgpath
+    PGPATH=$PATH:$PGPATH
+    log_success_msg "Starting Patroni\n"
+    exec start-stop-daemon --start --quiet \
+        --background \
+        --pidfile $PIDFILE --make-pidfile \
+        --chuid $USER:$GROUP \
+        --chdir `eval echo ~$USER` \
+        --exec $PATRONI \
+        --startas /bin/sh -- \
+        -c "/usr/bin/env PATH=$PGPATH /usr/bin/python $PATRONI $CONF >> $LOGFILE 2>&1"
+    ;;
+
+  stop)
+    log_success_msg "Stopping Patroni"
+    get_pid
+    start-stop-daemon --stop --pid $CHILDPID
+    start-stop-daemon --stop --pidfile $PIDFILE --remove-pidfile --quiet
+    ;;
+
+  reload)
+    log_success_msg "Reloading Patroni configuration"
+    get_pid
+    kill -HUP $CHILDPID
+    ;;
+
+  status)
+    get_pid
+    if start-stop-daemon -T --pid $CHILDPID; then
+      log_success_msg "Patroni is running\n"
+      exit 0
+    else
+      log_warning_msg "Patroni in not running\n"
+    fi
+    ;;
+
+  restart)
+    $0 stop
+    $0 start
+    ;;
+
+  *)
+    echo "Usage: /etc/init.d/$NAME {start|stop|restart|reload|status}"
+    exit 1
+    ;;
+esac
+
+if [ $? -eq 0 ]; then
+        echo .
+        exit 0
+else
+        echo " failed"
+        exit 1
+fi

extras/startup-scripts/patroni.service

@@ -0,0 +1,45 @@
+# This is an example systemd config file for Patroni
+# You can copy it to "/etc/systemd/system/patroni.service",
+
+[Unit]
+Description=Runners to orchestrate a high-availability PostgreSQL
+After=syslog.target network.target
+
+[Service]
+Type=simple
+
+User=postgres
+Group=postgres
+
+# Read in configuration file if it exists, otherwise proceed
+EnvironmentFile=-/etc/patroni_env.conf
+
+WorkingDirectory=~
+
+# Where to send early-startup messages from the server
+# This is normally controlled by the global default set by systemd
+#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
+ExecReload=/bin/kill -s HUP $MAINPID
+
+# only kill the patroni process, not it's children, so it will gracefully stop postgres
+KillMode=process
+
+# Give a reasonable amount of time for the server to start up/shut down
+TimeoutSec=30
+
+# Do not restart the service if it crashes, we want to manually inspect database on failure
+Restart=no
+
+[Install]
+WantedBy=multi-user.target
+

features/backup_create.sh

@@ -0,0 +1,22 @@
+#!/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.sh

@@ -0,0 +1,21 @@
+#!/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

@@ -0,0 +1,53 @@
+Feature: basic replication
+  We should check that the basic bootstrapping, replication and failover works.
+
+  Scenario: check replication of a single table
+    Given I start postgres0
+    Then postgres0 is a leader after 10 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
+    And I configure and start postgres2 with a tag replicatefrom postgres0
+    And "sync" key in DCS has leader=postgres0 after 20 seconds
+    And I add the table foo to postgres0
+    Then table foo is present on postgres1 after 20 seconds
+    Then table foo is present on postgres2 after 20 seconds
+
+  Scenario: check restart of sync replica
+    Given I shut down postgres2
+    Then "sync" key in DCS has sync_standby=postgres1 after 5 seconds
+    When I start postgres2
+    And I shut down postgres1
+    Then "sync" key in DCS has sync_standby=postgres2 after 10 seconds
+    When I start postgres1
+    And "members/postgres1" key in DCS has state=running after 10 seconds
+    And I sleep for 2 seconds
+    When I issue a GET request to http://127.0.0.1:8010/sync
+    Then I receive a response code 200
+    When I issue a GET request to http://127.0.0.1:8009/async
+    Then I receive a response code 200
+
+  Scenario: check the basic failover in synchronous mode
+    Given I run patronictl.py pause batman
+    Then I receive a response returncode 0
+    When I sleep for 2 seconds
+    And I shut down postgres0
+    And I run patronictl.py resume batman
+    Then I receive a response returncode 0
+    And postgres2 role is the primary after 24 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
+
+  Scenario: check immediate failover when master_start_timeout=0
+    Given I kill postmaster on postgres2
+    Then postgres1 is a leader after 10 seconds
+    And postgres1 role is the primary after 10 seconds
+
+  Scenario: check rejoin of the former master with pg_rewind
+    Given I add the table splitbrain to postgres0
+    And I start postgres0
+    Then postgres0 role is the secondary after 20 seconds
+    When I add the table buz to postgres1
+    Then table buz is present on postgres0 after 20 seconds

features/callback.sh

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

features/cascading_replication.feature

@@ -0,0 +1,14 @@
+Feature: cascading replication
+	We should check that patroni can do base backup and streaming from the replica
+
+Scenario: check a base backup and streaming replication from a replica
+	Given I start postgres0
+	And postgres0 is a leader after 10 seconds
+	And I configure and start postgres1 with a tag clonefrom true
+	And replication works from postgres0 to postgres1 after 20 seconds
+	And I create label with "postgres0" in postgres0 data directory
+	And I create label with "postgres1" in postgres1 data directory
+	And "members/postgres1" key in DCS has state=running after 12 seconds
+	And I configure and start postgres2 with a tag replicatefrom postgres1
+	Then replication works from postgres0 to postgres2 after 30 seconds
+	And there is a label with "postgres1" in postgres2 data directory

features/custom_bootstrap.feature

@@ -0,0 +1,17 @@
+Feature: custom bootstrap
+    We should check that patroni can bootstrap a new cluster from a backup
+
+Scenario: clone existing cluster using pg_basebackup
+    Given I start postgres0
+    Then postgres0 is a leader after 10 seconds
+    When I add the table foo to postgres0
+    And I start postgres1 in a cluster batman1 as a clone of postgres0
+    Then postgres1 is a leader of batman1 after 10 seconds
+    Then table foo is present on postgres1 after 10 seconds
+
+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
+    And table bar is present on postgres2 after 10 seconds

features/environment.py

@@ -0,0 +1,825 @@
+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 tempfile
+import threading
+import time
+import yaml
+
+
+@six.add_metaclass(abc.ABCMeta)
+class AbstractController(object):
+
+    def __init__(self, context, name, work_directory, output_dir):
+        self._context = context
+        self._name = name
+        self._work_directory = work_directory
+        self._output_dir = output_dir
+        self._handle = None
+        self._log = None
+
+    def _has_started(self):
+        return self._handle and self._handle.pid and self._handle.poll() is None
+
+    def _is_running(self):
+        return self._has_started()
+
+    @abc.abstractmethod
+    def _is_accessible(self):
+        """process is accessible for queries"""
+
+    @abc.abstractmethod
+    def _start(self):
+        """start process"""
+
+    def start(self, max_wait_limit=5):
+        if self._is_running():
+            return True
+
+        self._log = open(os.path.join(self._output_dir, self._name + '.log'), 'a')
+        self._handle = self._start()
+
+        assert self._has_started(), "Process {0} is not running after being started".format(self._name)
+
+        max_wait_limit *= self._context.timeout_multiplier
+        for _ in range(max_wait_limit):
+            if self._is_accessible():
+                break
+            time.sleep(1)
+        else:
+            assert False,\
+                "{0} instance is not available for queries after {1} seconds".format(self._name, max_wait_limit)
+
+    def stop(self, kill=False, timeout=15, _=False):
+        term = False
+        start_time = time.time()
+
+        timeout *= self._context.timeout_multiplier
+        while self._handle and self._is_running():
+            if kill:
+                self._handle.kill()
+            elif not term:
+                self._handle.terminate()
+                term = True
+            time.sleep(1)
+            if not kill and time.time() - start_time > timeout:
+                kill = True
+
+        if self._log:
+            self._log.close()
+
+    def cancel_background(self):
+        pass
+
+
+class PatroniController(AbstractController):
+    __PORT = 5440
+    PATRONI_CONFIG = '{}.yml'
+    """ starts and stops individual patronis"""
+
+    def __init__(self, context, name, work_directory, output_dir, custom_config=None):
+        super(PatroniController, self).__init__(context, 'patroni_' + name, work_directory, output_dir)
+        PatroniController.__PORT += 1
+        self._data_dir = os.path.join(work_directory, 'data', name)
+        self._connstring = None
+        if custom_config and 'watchdog' in custom_config:
+            self.watchdog = WatchdogMonitor(name, work_directory, output_dir)
+            custom_config['watchdog'] = {'driver': 'testing', 'device': self.watchdog.fifo_path, 'mode': 'required'}
+        else:
+            self.watchdog = None
+
+        self._scope = (custom_config or {}).get('scope', 'batman')
+        self._config = self._make_patroni_test_config(name, custom_config)
+        self._closables = []
+
+        self._conn = None
+        self._curs = None
+
+    def write_label(self, content):
+        with open(os.path.join(self._data_dir, 'label'), 'w') as f:
+            f.write(content)
+
+    def read_label(self):
+        try:
+            with open(os.path.join(self._data_dir, 'label'), 'r') as f:
+                return f.read().strip()
+        except IOError:
+            return None
+
+    @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)
+            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:
+            self.watchdog.start()
+        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],
+                                stdout=self._log, stderr=subprocess.STDOUT, cwd=self._work_directory)
+
+    def stop(self, kill=False, timeout=15, postgres=False):
+        if postgres:
+            return subprocess.call(['pg_ctl', '-D', self._data_dir, 'stop', '-mi', '-w'])
+        super(PatroniController, self).stop(kill, timeout)
+        if isinstance(self._context.dcs_ctl, KubernetesController):
+            self._context.dcs_ctl.delete_pod(self._name[8:])
+        if self.watchdog:
+            self.watchdog.stop()
+
+    def _is_accessible(self):
+        cursor = self.query("SELECT 1", fail_ok=True)
+        if cursor is not None:
+            cursor.execute("SET synchronous_commit TO 'local'")
+            return True
+
+    def _make_patroni_test_config(self, name, custom_config):
+        patroni_config_name = self.PATRONI_CONFIG.format(name)
+        patroni_config_path = os.path.join(self._output_dir, patroni_config_name)
+
+        with open(patroni_config_name) as f:
+            config = yaml.safe_load(f)
+            config.pop('etcd', None)
+
+        host = config['postgresql']['listen'].split(':')[0]
+
+        config['postgresql']['listen'] = config['postgresql']['connect_address'] = '{0}:{1}'.format(host, self.__PORT)
+
+        config['name'] = name
+        config['postgresql']['data_dir'] = self._data_dir
+        config['postgresql']['use_unix_socket'] = True
+        config['postgresql']['parameters'].update({
+            'logging_collector': 'on', 'log_destination': 'csvlog', 'log_directory': self._output_dir,
+            'log_filename': name + '.log', 'log_statement': 'all', 'log_min_messages': 'debug1',
+            'unix_socket_directories': self._data_dir})
+
+        if 'bootstrap' in config:
+            config['bootstrap']['post_bootstrap'] = 'psql -w -c "SELECT 1"'
+            if 'initdb' in config['bootstrap']:
+                config['bootstrap']['initdb'].extend([{'auth': 'md5'}, {'auth-host': 'md5'}])
+
+        if custom_config is not None:
+            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)
+
+        user = config['postgresql'].get('authentication', config['postgresql']).get('superuser', {})
+        self._connkwargs = {k: user[n] for n, k in [('username', 'user'), ('password', 'password')] if n in user}
+        self._connkwargs.update({'host': host, 'port': self.__PORT, 'database': 'postgres'})
+
+        self._replication = config['postgresql'].get('authentication', config['postgresql']).get('replication', {})
+        self._replication.update({'host': host, 'port': self.__PORT, 'database': 'postgres'})
+
+        return patroni_config_path
+
+    def _connection(self):
+        if not self._conn or self._conn.closed != 0:
+            self._conn = psycopg2.connect(**self._connkwargs)
+            self._conn.autocommit = True
+        return self._conn
+
+    def _cursor(self):
+        if not self._curs or self._curs.closed or self._curs.connection.closed != 0:
+            self._curs = self._connection().cursor()
+        return self._curs
+
+    def query(self, query, fail_ok=False):
+        try:
+            cursor = self._cursor()
+            cursor.execute(query)
+            return cursor
+        except psycopg2.Error:
+            if not fail_ok:
+                raise
+
+    def check_role_has_changed_to(self, new_role, timeout=10):
+        bound_time = time.time() + timeout
+        recovery_status = new_role != 'primary'
+        while time.time() < bound_time:
+            cur = self.query("SELECT pg_is_in_recovery()", fail_ok=True)
+            if cur:
+                row = cur.fetchone()
+                if row and row[0] == recovery_status:
+                    return True
+            time.sleep(1)
+        return False
+
+    def get_watchdog(self):
+        return self.watchdog
+
+    def _get_pid(self):
+        try:
+            pidfile = os.path.join(self._data_dir, 'postmaster.pid')
+            if not os.path.exists(pidfile):
+                return None
+            return int(open(pidfile).readline().strip())
+        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])
+
+
+class ProcessHang(object):
+
+    """A background thread implementing a cancelable process hang via SIGSTOP."""
+
+    def __init__(self, pid, timeout):
+        self._cancelled = threading.Event()
+        self._thread = threading.Thread(target=self.run)
+        self.pid = pid
+        self.timeout = timeout
+
+    def start(self):
+        self._thread.start()
+
+    def run(self):
+        os.kill(self.pid, signal.SIGSTOP)
+        try:
+            self._cancelled.wait(self.timeout)
+        finally:
+            os.kill(self.pid, signal.SIGCONT)
+
+    def close(self):
+        self._cancelled.set()
+        self._thread.join()
+
+
+class AbstractDcsController(AbstractController):
+
+    _CLUSTER_NODE = '/service/{0}'
+
+    def __init__(self, context, mktemp=True):
+        work_directory = mktemp and tempfile.mkdtemp() or None
+        super(AbstractDcsController, self).__init__(context, self.name(), work_directory, context.pctl.output_dir)
+
+    def _is_accessible(self):
+        return self._is_running()
+
+    def stop(self, kill=False, timeout=15):
+        """ terminate process and wipe out the temp work directory, but only if we actually started it"""
+        super(AbstractDcsController, self).stop(kill=kill, timeout=timeout)
+        if self._work_directory:
+            shutil.rmtree(self._work_directory)
+
+    def path(self, key=None, scope='batman'):
+        return self._CLUSTER_NODE.format(scope) + (key and '/' + key or '')
+
+    @abc.abstractmethod
+    def query(self, key, scope='batman'):
+        """ query for a value of a given key """
+
+    @abc.abstractmethod
+    def cleanup_service_tree(self):
+        """ clean all contents stored in the tree used for the tests """
+
+    @classmethod
+    def get_subclasses(cls):
+        for subclass in cls.__subclasses__():
+            for subsubclass in subclass.get_subclasses():
+                yield subsubclass
+            yield subclass
+
+    @classmethod
+    def name(cls):
+        return cls.__name__[:-10].lower()
+
+
+class ConsulController(AbstractDcsController):
+
+    def __init__(self, context):
+        super(ConsulController, self).__init__(context)
+        os.environ['PATRONI_CONSUL_HOST'] = 'localhost:8500'
+        os.environ['PATRONI_CONSUL_REGISTER_SERVICE'] = 'on'
+        self._client = consul.Consul()
+        self._config_file = None
+
+    def _start(self):
+        self._config_file = self._work_directory + '.json'
+        with open(self._config_file, 'wb') as f:
+            f.write(b'{"session_ttl_min":"5s","server":true,"bootstrap":true,"advertise_addr":"127.0.0.1"}')
+        return subprocess.Popen(['consul', 'agent', '-config-file', self._config_file, '-data-dir',
+                                 self._work_directory], stdout=self._log, stderr=subprocess.STDOUT)
+
+    def stop(self, kill=False, timeout=15):
+        super(ConsulController, self).stop(kill=kill, timeout=timeout)
+        if self._config_file:
+            os.unlink(self._config_file)
+
+    def _is_running(self):
+        try:
+            return bool(self._client.status.leader())
+        except Exception:
+            return False
+
+    def path(self, key=None, scope='batman'):
+        return super(ConsulController, self).path(key, scope)[1:]
+
+    def query(self, key, scope='batman'):
+        _, value = self._client.kv.get(self.path(key, scope))
+        return value and value['Value'].decode('utf-8')
+
+    def cleanup_service_tree(self):
+        self._client.kv.delete(self.path(scope=''), recurse=True)
+
+    def start(self, max_wait_limit=15):
+        super(ConsulController, self).start(max_wait_limit)
+
+
+class EtcdController(AbstractDcsController):
+
+    """ handles all etcd related tasks, used for the tests setup and cleanup """
+
+    def __init__(self, context):
+        super(EtcdController, self).__init__(context)
+        os.environ['PATRONI_ETCD_HOST'] = 'localhost:2379'
+        self._client = etcd.Client(port=2379)
+
+    def _start(self):
+        return subprocess.Popen(["etcd", "--debug", "--data-dir", self._work_directory],
+                                stdout=self._log, stderr=subprocess.STDOUT)
+
+    def query(self, key, scope='batman'):
+        try:
+            return self._client.get(self.path(key, scope)).value
+        except etcd.EtcdKeyNotFound:
+            return None
+
+    def cleanup_service_tree(self):
+        try:
+            self._client.delete(self.path(scope=''), recursive=True)
+        except (etcd.EtcdKeyNotFound, etcd.EtcdConnectionFailed):
+            return
+        except Exception as e:
+            assert False, "exception when cleaning up etcd contents: {0}".format(e)
+
+    def _is_running(self):
+        # if etcd is running, but we didn't start it
+        try:
+            return bool(self._client.machines)
+        except Exception:
+            return False
+
+
+class KubernetesController(AbstractDcsController):
+
+    def __init__(self, context):
+        super(KubernetesController, self).__init__(context)
+        self._namespace = 'default'
+        self._labels = {"application": "patroni"}
+        self._label_selector = ','.join('{0}={1}'.format(k, v) for k, v in self._labels.items())
+        os.environ['PATRONI_KUBERNETES_LABELS'] = json.dumps(self._labels)
+        os.environ['PATRONI_KUBERNETES_USE_ENDPOINTS'] = 'true'
+
+        from kubernetes import client as k8s_client, config as k8s_config
+        k8s_config.load_kube_config(context='local')
+        self._client = k8s_client
+        self._api = self._client.CoreV1Api()
+
+    def _start(self):
+        pass
+
+    def create_pod(self, name, scope):
+        labels = self._labels.copy()
+        labels['cluster-name'] = scope
+        metadata = self._client.V1ObjectMeta(namespace=self._namespace, name=name, labels=labels)
+        spec = self._client.V1PodSpec(containers=[self._client.V1Container(name=name, image='empty')])
+        body = self._client.V1Pod(metadata=metadata, spec=spec)
+        self._api.create_namespaced_pod(self._namespace, body)
+
+    def delete_pod(self, name):
+        try:
+            self._api.delete_namespaced_pod(name, self._namespace, self._client.V1DeleteOptions())
+        except:
+            pass
+        while True:
+            try:
+                self._api.read_namespaced_pod(name, self._namespace)
+            except:
+                break
+
+    def query(self, key, scope='batman'):
+        if key.startswith('members/'):
+            pod = self._api.read_namespaced_pod(key[8:], self._namespace)
+            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':
+                    return e.metadata.annotations[key]
+                else:
+                    return json.dumps(e.metadata.annotations)
+            except:
+                return None
+
+    def cleanup_service_tree(self):
+        try:
+            self._api.delete_collection_namespaced_pod(self._namespace, label_selector=self._label_selector)
+        except:
+            pass
+        try:
+            self._api.delete_collection_namespaced_endpoints(self._namespace, label_selector=self._label_selector)
+        except:
+            pass
+
+        while True:
+            result = self._api.list_namespaced_pod(self._namespace, label_selector=self._label_selector)
+            if len(result.items) < 1:
+                break
+
+    def _is_running(self):
+        return True
+
+
+class ZooKeeperController(AbstractDcsController):
+
+    """ handles all zookeeper related tasks, used for the tests setup and cleanup """
+
+    def __init__(self, context, export_env=True):
+        super(ZooKeeperController, self).__init__(context, False)
+        if export_env:
+            os.environ['PATRONI_ZOOKEEPER_HOSTS'] = "'localhost:2181'"
+        self._client = kazoo.client.KazooClient()
+
+    def _start(self):
+        pass  # TODO: implement later
+
+    def query(self, key, scope='batman'):
+        try:
+            return self._client.get(self.path(key, scope))[0].decode('utf-8')
+        except kazoo.exceptions.NoNodeError:
+            return None
+
+    def cleanup_service_tree(self):
+        try:
+            self._client.delete(self.path(scope=''), recursive=True)
+        except (kazoo.exceptions.NoNodeError):
+            return
+        except Exception as e:
+            assert False, "exception when cleaning up zookeeper contents: {0}".format(e)
+
+    def _is_running(self):
+        # if zookeeper is running, but we didn't start it
+        if self._client.connected:
+            return True
+        try:
+            return self._client.start(1) or True
+        except Exception:
+            return False
+
+
+class ExhibitorController(ZooKeeperController):
+
+    def __init__(self, context):
+        super(ExhibitorController, self).__init__(context, False)
+        os.environ.update({'PATRONI_EXHIBITOR_HOSTS': 'localhost', 'PATRONI_EXHIBITOR_PORT': '8181'})
+
+
+class PatroniPoolController(object):
+
+    BACKUP_SCRIPT = 'features/backup_create.sh'
+
+    def __init__(self, context):
+        self._context = context
+        self._dcs = None
+        self._output_dir = None
+        self._patroni_path = None
+        self._processes = {}
+        self.create_and_set_output_directory('')
+        self.known_dcs = {subclass.name(): subclass for subclass in AbstractDcsController.get_subclasses()}
+
+    @property
+    def patroni_path(self):
+        if self._patroni_path is None:
+            cwd = os.path.realpath(__file__)
+            while True:
+                cwd, entry = os.path.split(cwd)
+                if entry == 'features' or cwd == '/':
+                    break
+            self._patroni_path = cwd
+        return self._patroni_path
+
+    @property
+    def output_dir(self):
+        return self._output_dir
+
+    def start(self, name, max_wait_limit=20, custom_config=None):
+        if name not in self._processes:
+            self._processes[name] = PatroniController(self._context, name, self.patroni_path,
+                                                      self._output_dir, custom_config)
+        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']:
+            raise AttributeError("PatroniPoolController instance has no attribute '{0}'".format(func))
+
+        def wrapper(name, *args, **kwargs):
+            return getattr(self._processes[name], func)(*args, **kwargs)
+        return wrapper
+
+    def stop_all(self):
+        for ctl in self._processes.values():
+            ctl.cancel_background()
+            ctl.stop()
+        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(' ', '_'))
+        if os.path.exists(feature_dir):
+            shutil.rmtree(feature_dir)
+        os.makedirs(feature_dir)
+        self._output_dir = feature_dir
+
+    def clone(self, from_name, cluster_name, to_name):
+        f = self._processes[from_name]
+        custom_config = {
+            'scope': cluster_name,
+            'bootstrap': {
+                'method': 'pg_basebackup',
+                'pg_basebackup': {
+                    'command': self.BACKUP_SCRIPT + ' --walmethod=stream --dbname=' + f.backup_source
+                },
+                'dcs': {
+                    'postgresql': {
+                        'parameters': {
+                            'max_connections': 101
+                        }
+                    }
+                }
+            },
+            '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'))
+                },
+                'authentication': {
+                    'superuser': {'password': 'zalando1'},
+                    'replication': {'password': 'rep-pass1'}
+                }
+            }
+        }
+        self.start(to_name, custom_config=custom_config)
+
+    def bootstrap_from_backup(self, name, cluster_name):
+        custom_config = {
+            'scope': cluster_name,
+            'bootstrap': {
+                'method': 'backup_restore',
+                'backup_restore': {
+                    'command': 'features/backup_restore.sh --sourcedir=' + os.path.join(self._output_dir, 'basebackup'),
+                    'recovery_conf': {
+                        'recovery_target_action': 'promote',
+                        'recovery_target_timeline': 'latest',
+                        'restore_command': 'cp {0}/wal_archive/%f %p'.format(self._output_dir)
+                    }
+                }
+            },
+            'postgresql': {
+                'authentication': {
+                    'superuser': {'password': 'zalando2'},
+                    'replication': {'password': 'rep-pass2'}
+                }
+            }
+        }
+        self.start(name, custom_config=custom_config)
+
+    @property
+    def dcs(self):
+        if self._dcs is None:
+            self._dcs = os.environ.pop('DCS', 'etcd')
+            assert self._dcs in self.known_dcs, 'Unsupported dcs: ' + self._dcs
+        return self._dcs
+
+
+class WatchdogMonitor(object):
+    """Testing harness for emulating a watchdog device as a named pipe. Because we can't easily emulate ioctl's we
+    require a custom driver on Patroni side. The device takes no action, only notes if it was pinged and/or triggered.
+    """
+    def __init__(self, name, work_directory, output_dir):
+        self.fifo_path = os.path.join(work_directory, 'data', 'watchdog.{0}.fifo'.format(name))
+        self.fifo_file = None
+        self._stop_requested = False  # Relying on bool setting being atomic
+        self._thread = None
+        self.last_ping = None
+        self.was_pinged = False
+        self.was_closed = False
+        self._was_triggered = False
+        self.timeout = 60
+        self._log_file = open(os.path.join(output_dir, 'watchdog.{0}.log'.format(name)), 'w')
+        self._log("watchdog {0} initialized".format(name))
+
+    def _log(self, msg):
+        tstamp = datetime.datetime.now().strftime("%Y-%m-%d %H:%M:%S,%f")
+        self._log_file.write("{0}: {1}\n".format(tstamp, msg))
+
+    def start(self):
+        assert self._thread is None
+        self._stop_requested = False
+        self._log("starting fifo {0}".format(self.fifo_path))
+        fifo_dir = os.path.dirname(self.fifo_path)
+        if os.path.exists(self.fifo_path):
+            os.unlink(self.fifo_path)
+        elif not os.path.exists(fifo_dir):
+            os.mkdir(fifo_dir)
+        os.mkfifo(self.fifo_path)
+        self.last_ping = time.time()
+
+        self._thread = threading.Thread(target=self.run)
+        self._thread.start()
+
+    def run(self):
+        try:
+            while not self._stop_requested:
+                self._log("opening")
+                self.fifo_file = os.open(self.fifo_path, os.O_RDONLY)
+                try:
+                    self._log("Fifo {0} connected".format(self.fifo_path))
+                    self.was_closed = False
+                    while not self._stop_requested:
+                        c = os.read(self.fifo_file, 1)
+
+                        if c == b'X':
+                            self._log("Stop requested")
+                            return
+                        elif c == b'':
+                            self._log("Pipe closed")
+                            break
+                        elif c == b'C':
+                            command = b''
+                            c = os.read(self.fifo_file, 1)
+                            while c != b'\n' and c != b'':
+                                command += c
+                                c = os.read(self.fifo_file, 1)
+                            command = command.decode('utf8')
+
+                            if command.startswith('timeout='):
+                                self.timeout = int(command.split('=')[1])
+                                self._log("timeout={0}".format(self.timeout))
+                        elif c in [b'V', b'1']:
+                            cur_time = time.time()
+                            if cur_time - self.last_ping > self.timeout:
+                                self._log("Triggered")
+                                self._was_triggered = True
+                            if c == b'V':
+                                self._log("magic close")
+                                self.was_closed = True
+                            elif c == b'1':
+                                self.was_pinged = True
+                                self._log("ping after {0} seconds".format(cur_time - (self.last_ping or cur_time)))
+                                self.last_ping = cur_time
+                        else:
+                            self._log('Unknown command {0} received from fifo'.format(c))
+                finally:
+                    self.was_closed = True
+                    self._log("closing")
+                    os.close(self.fifo_file)
+        except Exception as e:
+            self._log("Error {0}".format(e))
+        finally:
+            self._log("stopping")
+            self._log_file.flush()
+            if os.path.exists(self.fifo_path):
+                os.unlink(self.fifo_path)
+
+    def stop(self):
+        self._log("Monitor stop")
+        self._stop_requested = True
+        try:
+            if os.path.exists(self.fifo_path):
+                fd = os.open(self.fifo_path, os.O_WRONLY)
+                os.write(fd, b'X')
+                os.close(fd)
+        except Exception as e:
+            self._log("err while closing: {0}".format(str(e)))
+        if self._thread:
+            self._thread.join()
+            self._thread = None
+
+    def reset(self):
+        self._log("reset")
+        self.was_pinged = self.was_closed = self._was_triggered = False
+
+    @property
+    def was_triggered(self):
+        delta = time.time() - self.last_ping
+        triggered = self._was_triggered or not self.was_closed and delta > self.timeout
+        self._log("triggered={0}, {1}s left".format(triggered, self.timeout - delta))
+        return triggered
+
+
+# actions to execute on start/stop of the tests and before running invidual features
+def before_all(context):
+    os.environ.update({'PATRONI_RESTAPI_USERNAME': 'username', 'PATRONI_RESTAPI_PASSWORD': 'password'})
+    context.ci = 'TRAVIS_BUILD_NUMBER' in os.environ or 'BUILD_NUMBER' in os.environ
+    context.timeout_multiplier = 2 if context.ci else 1
+    context.pctl = PatroniPoolController(context)
+    context.dcs_ctl = context.pctl.known_dcs[context.pctl.dcs](context)
+    context.dcs_ctl.start()
+    try:
+        context.dcs_ctl.cleanup_service_tree()
+    except AssertionError:  # after_all handlers won't be executed in before_all
+        context.dcs_ctl.stop()
+        raise
+
+
+def after_all(context):
+    context.dcs_ctl.stop()
+    subprocess.call(['coverage', 'combine'])
+    subprocess.call(['coverage', 'report'])
+
+
+def before_feature(context, feature):
+    """ create per-feature output directory to collect Patroni and PostgreSQL logs """
+    context.pctl.create_and_set_output_directory(feature.name)
+
+
+def after_feature(context, feature):
+    """ stop all Patronis, remove their data directory and cleanup the keys in etcd """
+    context.pctl.stop_all()
+    shutil.rmtree(os.path.join(context.pctl.patroni_path, 'data'))
+    context.dcs_ctl.cleanup_service_tree()

features/patroni_api.feature

@@ -0,0 +1,115 @@
+Feature: patroni api
+	We should check that patroni correctly responds to valid and not-valid API requests.
+
+Scenario: check API requests on a stand-alone server
+	Given I start postgres0
+	And postgres0 is a leader after 10 seconds
+	When I issue a GET request to http://127.0.0.1:8008/
+	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/replica
+	Then I receive a response code 503
+	When I run patronictl.py reinit batman postgres0 --force
+	Then I receive a response returncode 0
+	And I receive a response output "Failed: reinitialize for member postgres0, status code=503, (I am the leader, can not reinitialize)"
+	When I run patronictl.py switchover batman --master postgres0 --force
+	Then I receive a response returncode 1
+	And I receive a response output "Error: No candidates found to switchover to"
+	When I issue a POST request to http://127.0.0.1:8008/switchover with {"leader": "postgres0"}
+	Then I receive a response code 412
+	And I receive a response text switchover is not possible: cluster does not have members except leader
+	When I issue an empty POST request to http://127.0.0.1:8008/failover
+	Then I receive a response code 400
+	When I issue a POST request to http://127.0.0.1:8008/failover with {"foo": "bar"}
+	Then I receive a response code 400
+	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
+	And I issue an empty POST request to http://127.0.0.1:8008/reload
+	Then I receive a response code 202
+
+Scenario: check dynamic configuration change via DCS
+	Given I run patronictl.py edit-config -s 'ttl=10' -s 'loop_wait=2' -p 'max_connections=101' --force batman
+	Then I receive a response returncode 0
+	And I receive a response output "+loop_wait: 2"
+	And Response on GET http://127.0.0.1:8008/patroni contains pending_restart after 11 seconds
+	When I issue a GET request to http://127.0.0.1:8008/config
+	Then I receive a response code 200
+	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 {'new_tag': 'new_value'}
+
+Scenario: check API requests for the primary-replica pair in the pause mode
+	Given I run patronictl.py pause batman
+	Then I receive a response returncode 0
+	When I start postgres1
+	Then replication works from postgres0 to postgres1 after 20 seconds
+	When I issue a GET request to http://127.0.0.1:8009/replica
+	Then I receive a response code 200
+	And I receive a response state running
+	And I receive a response role replica
+	When I run patronictl.py reinit batman postgres1 --force
+	Then I receive a response returncode 0
+	And I receive a response output "Success: reinitialize for member postgres1"
+	When I run patronictl.py restart batman postgres0 --force
+	Then I receive a response returncode 0
+	And I receive a response output "Success: restart on member postgres0"
+	And postgres0 role is the primary after 5 seconds
+	When I sleep for 10 seconds
+	Then postgres1 role is the secondary after 15 seconds
+
+Scenario: check the switchover via the API in the pause mode
+	Given I issue a POST request to http://127.0.0.1:8008/switchover with {"leader": "postgres0", "candidate": "postgres1"}
+	Then I receive a response code 200
+	And postgres1 is a leader after 5 seconds
+	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
+	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
+	Then I receive a response code 200
+	When I issue a GET request to http://127.0.0.1:8009/master
+	Then I receive a response code 200
+	When I issue a GET request to http://127.0.0.1:8009/replica
+	Then I receive a response code 503
+
+Scenario: check the scheduled switchover
+	Given I issue a scheduled switchover from postgres1 to postgres0 in 3 seconds
+	Then I receive a response returncode 1
+	And I receive a response output "Can't schedule switchover in the paused state"
+	When I run patronictl.py resume batman
+	Then I receive a response returncode 0
+	Given I issue a scheduled switchover from postgres1 to postgres0 in 3 seconds
+	Then I receive a response returncode 0
+	And postgres0 is a leader after 20 seconds
+	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
+	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
+	Then I receive a response code 503
+	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/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

@@ -0,0 +1,34 @@
+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 I sleep for 2 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 2 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
+    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 I sleep for 2 seconds
+    When I issue a GET request to http://127.0.0.1:8008/
+    Then I receive a response code 200
+    And there is a label with "test_logical" in postgres0 data directory
+
+  Scenario: check replication of a single table in a standby cluster
+    Given I start postgres1 in a standby cluster batman1 as a clone of postgres0
+    Then postgres1 is a leader of batman1 after 10 seconds
+    When I add the table foo to postgres0
+    Then table foo is present on postgres1 after 20 seconds
+    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
+    And I kill postmaster on postgres1
+    Then postgres2 is replicating from postgres0 after 20 seconds

features/steps/basic_replication.py

@@ -0,0 +1,61 @@
+import psycopg2 as pg
+
+from behave import step, then
+from time import sleep, time
+
+
+@step('I start {name:w}')
+def start_patroni(context, name):
+    return context.pctl.start(name)
+
+
+@step('I shut down {name:w}')
+def stop_patroni(context, name):
+    return context.pctl.stop(name, timeout=60)
+
+
+@step('I kill {name:w}')
+def kill_patroni(context, name):
+    return context.pctl.stop(name, kill=True)
+
+
+@step('I kill postmaster on {name:w}')
+def stop_postgres(context, name):
+    return context.pctl.stop(name, postgres=True)
+
+
+@step('I add the table {table_name:w} to {pg_name:w}')
+def add_table(context, table_name, pg_name):
+    # parse the configuration file and get the port
+    try:
+        context.pctl.query(pg_name, "CREATE TABLE {0}()".format(table_name))
+    except pg.Error as e:
+        assert False, "Error creating table {0} on {1}: {2}".format(table_name, pg_name, e)
+
+
+@then('Table {table_name:w} is present on {pg_name:w} after {max_replication_delay:d} seconds')
+def table_is_present_on(context, table_name, pg_name, max_replication_delay):
+    max_replication_delay *= context.timeout_multiplier
+    for _ in range(int(max_replication_delay)):
+        if context.pctl.query(pg_name, "SELECT 1 FROM {0}".format(table_name), fail_ok=True) is not None:
+            break
+        sleep(1)
+    else:
+        assert False,\
+            "Table {0} is not present on {1} after {2} seconds".format(table_name, pg_name, max_replication_delay)
+
+
+@then('{pg_name:w} role is the {pg_role:w} after {max_promotion_timeout:d} seconds')
+def check_role(context, pg_name, pg_role, max_promotion_timeout):
+    max_promotion_timeout *= context.timeout_multiplier
+    assert context.pctl.check_role_has_changed_to(pg_name, pg_role, timeout=int(max_promotion_timeout)),\
+        "{0} role didn't change to {1} after {2} seconds".format(pg_name, pg_role, max_promotion_timeout)
+
+
+@step('replication works from {master:w} to {replica:w} after {time_limit:d} seconds')
+@then('replication works from {master:w} to {replica:w} after {time_limit:d} seconds')
+def replication_works(context, master, replica, time_limit):
+    context.execute_steps(u"""
+        When I add the table test_{0} to {1}
+        Then table test_{0} is present on {2} after {3} seconds
+    """.format(int(time()), master, replica, time_limit))

features/steps/cascading_replication.py

@@ -0,0 +1,35 @@
+import json
+import time
+
+from behave import step, then
+
+
+@step('I configure and start {name:w} with a tag {tag_name:w} {tag_value:w}')
+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)
+
+
+@step('I create label with "{content:w}" in {name:w} data directory')
+def write_label(context, content, name):
+    context.pctl.write_label(name, content)
+
+
+@step('"{name}" key in DCS has {key:w}={value:w} after {time_limit:d} seconds')
+def check_member(context, name, key, value, time_limit):
+    time_limit *= context.timeout_multiplier
+    max_time = time.time() + int(time_limit)
+    while time.time() < max_time:
+        try:
+            response = json.loads(context.dcs_ctl.query(name))
+            if response.get(key) == value:
+                return
+        except Exception:
+            pass
+        time.sleep(1)
+    assert False, "{0} does not have {1}={2} in dcs after {3} seconds".format(name, key, value, time_limit)

features/steps/custom_bootstrap.py

@@ -0,0 +1,27 @@
+import time
+
+from behave import step, then
+
+
+@step('I start {name:w} in a cluster {cluster_name:w} as a clone of {name2:w}')
+def start_cluster_clone(context, name, cluster_name, name2):
+    context.pctl.clone(name2, cluster_name, name)
+
+
+@step('I start {name:w} in a cluster {cluster_name:w} from backup')
+def start_cluster_from_backup(context, name, cluster_name):
+    context.pctl.bootstrap_from_backup(name, cluster_name)
+
+
+@then('{name:w} is a leader of {cluster_name:w} after {time_limit:d} seconds')
+def is_a_leader(context, name, cluster_name, time_limit):
+    time_limit *= context.timeout_multiplier
+    max_time = time.time() + int(time_limit)
+    while (context.dcs_ctl.query("leader", scope=cluster_name) != name):
+        time.sleep(1)
+        assert time.time() < max_time, "{0} is not a leader in dcs after {1} seconds".format(name, time_limit)
+
+
+@step('I do a backup of {name:w}')
+def do_backup(context, name):
+    context.pctl.backup(name)

features/steps/patroni_api.py

@@ -0,0 +1,162 @@
+import base64
+import json
+import os
+import parse
+import requests
+import shlex
+import subprocess
+import time
+import yaml
+
+from behave import register_type, step, then
+from dateutil import tz
+from datetime import datetime, timedelta
+
+tzutc = tz.tzutc()
+
+
+@parse.with_pattern(r'https?://(?:\w|\.|:|/)+')
+def parse_url(text):
+    return text
+
+
+register_type(url=parse_url)
+
+
+# there is no way we can find out if the node has already
+# started as a leader without checking the DCS. We cannot
+# just rely on the database availability, since there is
+# a short gap between the time PostgreSQL becomes available
+# and Patroni assuming the leader role.
+@step('{name:w} is a leader after {time_limit:d} seconds')
+@then('{name:w} is a leader after {time_limit:d} seconds')
+def is_a_leader(context, name, time_limit):
+    time_limit *= context.timeout_multiplier
+    max_time = time.time() + int(time_limit)
+    while (context.dcs_ctl.query("leader") != name):
+        time.sleep(1)
+        assert time.time() < max_time, "{0} is not a leader in dcs after {1} seconds".format(name, time_limit)
+
+
+@step('I sleep for {value:d} seconds')
+def sleep_for_n_seconds(context, value):
+    time.sleep(int(value))
+
+
+def _set_response(context, response):
+    context.status_code = response.status_code
+    data = response.content.decode('utf-8')
+    ct = response.headers.get('content-type', '')
+    if ct.startswith('application/json') or\
+            ct.startswith('text/yaml') or\
+            ct.startswith('text/x-yaml') or\
+            ct.startswith('application/yaml') or\
+            ct.startswith('application/x-yaml'):
+        try:
+            context.response = yaml.safe_load(data)
+        except ValueError:
+            context.response = data
+    else:
+        context.response = data
+
+
+@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)
+
+
+@step('I issue an empty POST request to {url:url}')
+def do_post_empty(context, url):
+    do_request(context, 'POST', url, None)
+
+
+@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'}
+    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
+    else:
+        _set_response(context, r)
+
+
+@step('I run {cmd}')
+def do_run(context, cmd):
+    cmd = ['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()
+        env.update({'PATRONI_RESTAPI_USERNAME': 'username', 'PATRONI_RESTAPI_PASSWORD': 'password'})
+        response = subprocess.check_output(cmd, stderr=subprocess.STDOUT, env=env)
+        context.status_code = 0
+    except subprocess.CalledProcessError as e:
+        response = e.output
+        context.status_code = e.returncode
+    context.response = response.decode('utf-8').strip()
+
+
+@then('I receive a response {component:w} {data}')
+def check_response(context, component, data):
+    if component == 'code':
+        assert context.status_code == int(data),\
+            "status code {0} != {1}, response: {2}".format(context.status_code, data, context.response)
+    elif component == 'returncode':
+        assert context.status_code == int(data), "return code {0} != {1}, {2}".format(context.status_code,
+                                                                                      data, context.response)
+    elif component == 'text':
+        assert context.response == data.strip('"'), "response {0} does not contain {1}".format(context.response, data)
+    elif component == 'output':
+        assert data.strip('"') in context.response, "response {0} does not contain {1}".format(context.response, data)
+    else:
+        assert component in context.response, "{0} is not part of the response".format(component)
+        assert str(context.response[component]) == str(data), "{0} does not contain {1}".format(component, data)
+
+
+@step('I issue a scheduled switchover from {from_host:w} to {to_host:w} in {in_seconds:d} seconds')
+def scheduled_switchover(context, from_host, to_host, in_seconds):
+    context.execute_steps(u"""
+        Given I run patronictl.py switchover batman --master {0} --candidate {1} --scheduled "{2}" --force
+    """.format(from_host, to_host, datetime.now(tzutc) + timedelta(seconds=int(in_seconds))))
+
+
+@step('I issue a scheduled restart at {url:url} in {in_seconds:d} seconds with {data}')
+def scheduled_restart(context, url, in_seconds, data):
+    data = data and json.loads(data) or {}
+    data.update(schedule='{0}'.format((datetime.now(tzutc) + timedelta(seconds=int(in_seconds))).isoformat()))
+    context.execute_steps(u"""Given I issue a POST request to {0}/restart with {1}""".format(url, json.dumps(data)))
+
+
+@step('I add tag {tag:w} {value:w} to {pg_name:w} config')
+def add_tag_to_config(context, tag, value, pg_name):
+    context.pctl.add_tag_to_config(pg_name, tag, value)
+
+
+@then('Response on GET {url} contains {value} after {timeout:d} seconds')
+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:
+            break
+        time.sleep(1)
+    else:
+        assert False,\
+            "Value {0} is {1} present in response after {2} seconds".format(value, "not" if not negate else "", timeout)
+
+
+@then('Response on GET {url} does not contain {value} after {timeout:d} seconds')
+def check_not_in_http_response(context, url, value, timeout):
+    check_http_response(context, url, value, timeout, negate=True)

features/steps/standby_cluster.py

@@ -0,0 +1,67 @@
+import os
+import time
+
+from behave import step
+
+
+select_replication_query = """
+SELECT * FROM pg_catalog.pg_stat_replication
+WHERE application_name = '{0}'
+"""
+
+
+@step('I start {name:w} with callback configured')
+def start_patroni_with_callbacks(context, name):
+    return context.pctl.start(name, custom_config={
+        "postgresql": {
+            "callbacks": {
+                "on_role_change": "features/callback.sh"
+            }
+        }
+    })
+
+
+@step('I start {name:w} in a cluster {cluster_name:w}')
+def start_patroni(context, name, cluster_name):
+    return context.pctl.start(name, custom_config={
+        "scope": cluster_name
+    })
+
+
+@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):
+    # 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')
+    context.pctl._processes[name].update_config({
+        "scope": cluster_name,
+        "bootstrap": {
+            "dcs": {
+                "standby_cluster": {
+                    "host": "localhost",
+                    "port": port,
+                    "primary_slot_name": "pm_1",
+                }
+            }
+        }
+    })
+    return context.pctl.start(name)
+
+
+@step('{pg_name1:w} is replicating from {pg_name2:w} after {timeout:d} seconds')
+def check_replication_status(context, pg_name1, pg_name2, timeout):
+    bound_time = time.time() + timeout
+
+    while time.time() < bound_time:
+        cur = context.pctl.query(
+            pg_name2,
+            select_replication_query.format(pg_name1),
+            fail_ok=True
+        )
+
+        if cur and len(cur.fetchall()) != 0:
+            return True
+
+        time.sleep(1)
+
+    return False

features/steps/watchdog.py

@@ -0,0 +1,74 @@
+from behave import step, then
+import time
+
+
+def polling_loop(timeout, interval=1):
+    """Returns an iterator that returns values until timeout has passed. Timeout is measured from start of iteration."""
+    start_time = time.time()
+    iteration = 0
+    end_time = start_time + timeout
+    while time.time() < end_time:
+        yield iteration
+        iteration += 1
+        time.sleep(interval)
+
+
+@step('I start {name:w} with watchdog')
+def start_patroni_with_watchdog(context, name):
+    return context.pctl.start(name, custom_config={'watchdog': True})
+
+
+@step('{name:w} watchdog has been pinged after {timeout:d} seconds')
+def watchdog_was_pinged(context, name, timeout):
+    for _ in polling_loop(timeout):
+        if context.pctl.get_watchdog(name).was_pinged:
+            return True
+    return False
+
+
+@then('{name:w} watchdog has been closed')
+def watchdog_was_closed(context, name):
+    assert context.pctl.get_watchdog(name).was_closed
+
+
+@step('I reset {name:w} watchdog state')
+def watchdog_reset_pinged(context, name):
+    context.pctl.get_watchdog(name).reset()
+
+
+@then('{name:w} watchdog is triggered after {timeout:d} seconds')
+def watchdog_was_triggered(context, name, timeout):
+    for _ in polling_loop(timeout):
+        if context.pctl.get_watchdog(name).was_triggered:
+            return True
+    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)

features/watchdog.feature

@@ -0,0 +1,31 @@
+Feature: watchdog
+  Verify that watchdog gets pinged and triggered under appropriate circumstances.
+
+  Scenario: watchdog is opened and pinged
+    Given I start postgres0 with watchdog
+    Then postgres0 is a leader after 10 seconds
+    And postgres0 role is the primary after 10 seconds
+    And postgres0 watchdog has been pinged after 10 seconds
+
+  Scenario: watchdog is disabled during pause
+    Given I run patronictl.py pause batman
+    Then I receive a response returncode 0
+    When I sleep for 2 seconds
+    Then postgres0 watchdog has been closed
+
+  Scenario: watchdog is opened and pinged after resume
+    Given I reset postgres0 watchdog state
+    And I run patronictl.py resume batman
+    Then I receive a response returncode 0
+    And postgres0 watchdog has been pinged after 10 seconds
+
+  Scenario: watchdog is disabled when shutting down
+    Given I shut down postgres0
+    Then postgres0 watchdog has been closed
+
+  Scenario: watchdog is triggered if patroni stops responding
+    Given I reset postgres0 watchdog state
+    And I start postgres0 with watchdog
+    Then postgres0 role is the primary after 10 seconds
+    When postgres0 hangs for 30 seconds
+    Then postgres0 watchdog is triggered after 30 seconds

haproxy.cfg

@@ -1,21 +1,25 @@
 global
-	maxconn 100
+    maxconn 100
 
 defaults
-	log	global
-	mode	tcp
-	retries 2
-	timeout client 30m
-	timeout connect 4s
-	timeout server 30m
-	timeout check 5s
+    log global
+    mode tcp
+    retries 2
+    timeout client 30m
+    timeout connect 4s
+    timeout server 30m
+    timeout check 5s
 
-frontend ft_postgresql
-	bind *:5000
-	default_backend bk_db
-
-backend bk_db
-	option httpchk
+listen stats
+    mode http
+    bind *:7000
+    stats enable
+    stats uri /
 
+listen batman
+    bind *:5000
+    option httpchk
+    http-check expect status 200
+    default-server inter 3s fall 3 rise 2 on-marked-down shutdown-sessions
     server postgresql_127.0.0.1_5432 127.0.0.1:5432 maxconn 100 check port 8008
     server postgresql_127.0.0.1_5433 127.0.0.1:5433 maxconn 100 check port 8009

kubernetes/Dockerfile

@@ -0,0 +1,34 @@
+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-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 \
+    && 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 python3-pip python3-wheel \
+    && apt-get autoremove -y \
+    && apt-get clean -y \
+    && rm -rf /var/lib/apt/lists/* /root/.cache
+
+ADD entrypoint.sh /
+
+EXPOSE 5432 8008
+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/entrypoint.sh

@@ -0,0 +1,39 @@
+#!/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:
+    postgresql:
+      use_pg_rewind: true
+  initdb:
+  - auth-host: md5
+  - auth-local: trust
+  - encoding: UTF8
+  - locale: en_US.UTF-8
+  - data-checksums
+  pg_hba:
+  - host all all 0.0.0.0/0 md5
+  - host replication ${PATRONI_REPLICATION_USERNAME} ${PATRONI_KUBERNETES_POD_IP}/16 md5
+restapi:
+  connect_address: '${PATRONI_KUBERNETES_POD_IP}:8008'
+postgresql:
+  connect_address: '${PATRONI_KUBERNETES_POD_IP}:5432'
+  authentication:
+    superuser:
+      password: '${PATRONI_SUPERUSER_PASSWORD}'
+    replication:
+      password: '${PATRONI_REPLICATION_PASSWORD}'
+__EOF__
+
+unset PATRONI_SUPERUSER_PASSWORD PATRONI_REPLICATION_PASSWORD
+export KUBERNETES_NAMESPACE=$PATRONI_KUBERNETES_NAMESPACE
+export POD_NAME=$PATRONI_NAME
+
+exec /usr/bin/python3 /usr/local/bin/patroni /home/postgres/patroni.yml

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 +1,303 @@
+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:
+        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

@@ -0,0 +1,188 @@
+apiVersion: apps/v1beta1
+kind: StatefulSet
+metadata:
+  name: &cluster_name patronidemo
+  labels:
+    application: patroni
+    cluster-name: *cluster_name
+spec:
+  replicas: 3
+  serviceName: *cluster_name
+  template:
+    metadata:
+      labels:
+        application: patroni
+        cluster-name: *cluster_name
+    spec:
+      serviceAccountName: patronidemo
+      containers:
+      - name: *cluster_name
+        image: patroni  # docker build -t patroni .
+        imagePullPolicy: IfNotPresent
+        ports:
+        - containerPort: 8008
+          protocol: TCP
+        - containerPort: 5432
+          protocol: TCP
+        volumeMounts:
+        - mountPath: /home/postgres/pgdata
+          name: pgdata
+        env:
+        - name: PATRONI_KUBERNETES_POD_IP
+          valueFrom:
+            fieldRef:
+              fieldPath: status.podIP
+        - name: PATRONI_KUBERNETES_NAMESPACE
+          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
+          value: postgres
+        - name: PATRONI_SUPERUSER_PASSWORD
+          valueFrom:
+            secretKeyRef:
+              name: *cluster_name
+              key: superuser-password
+        - name: PATRONI_REPLICATION_USERNAME
+          value: standby
+        - name: PATRONI_REPLICATION_PASSWORD
+          valueFrom:
+            secretKeyRef:
+              name: *cluster_name
+              key: replication-password
+        - name: PATRONI_SCOPE
+          value: *cluster_name
+        - name: PATRONI_NAME
+          valueFrom:
+            fieldRef:
+              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'
+      terminationGracePeriodSeconds: 0
+      volumes:
+      - name: pgdata
+        emptyDir: {}
+#  volumeClaimTemplates:
+#  - metadata:
+#      labels:
+#        application: spilo
+#        spilo-cluster: *cluster_name
+#      annotations:
+#        volume.alpha.kubernetes.io/storage-class: anything
+#      name: pgdata
+#    spec:
+#      accessModes:
+#      - ReadWriteOnce
+#      resources:
+#        requests:
+#          storage: 5Gi
+
+---
+apiVersion: v1
+kind: Endpoints
+metadata:
+  name: &cluster_name patronidemo
+  labels:
+    application: patroni
+    cluster-name: *cluster_name
+subsets: []
+
+---
+apiVersion: v1
+kind: Service
+metadata:
+  name: &cluster_name patronidemo
+  labels:
+    application: patroni
+    cluster-name: *cluster_name
+spec:
+  type: ClusterIP
+  ports:
+  - port: 5432
+    targetPort: 5432
+
+---
+apiVersion: v1
+kind: Secret
+metadata:
+  name: &cluster_name patronidemo
+  labels:
+    application: patroni
+    cluster-name: *cluster_name
+type: Opaque
+data:
+  superuser-password: emFsYW5kbw==
+  replication-password: cmVwLXBhc3M=
+
+---
+apiVersion: v1
+kind: ServiceAccount
+metadata:
+  name: patronidemo
+
+---
+apiVersion: rbac.authorization.k8s.io/v1
+kind: Role
+metadata:
+  name: patronidemo
+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: patronidemo
+roleRef:
+  apiGroup: rbac.authorization.k8s.io
+  kind: Role
+  name: patronidemo
+subjects:
+- kind: ServiceAccount
+  name: patronidemo

mkbinary.sh

@@ -0,0 +1,5 @@
+#!/bin/sh
+set -e
+
+pip install --ignore-installed setuptools==19.2 pyinstaller
+pyinstaller --clean --onefile patroni.spec

patroni.spec

@@ -0,0 +1,39 @@
+# -*- mode: python -*-
+
+block_cipher = None
+
+
+def hiddenimports():
+    import sys
+    sys.path.insert(0, '.')
+    try:
+        import patroni.dcs
+        return patroni.dcs.dcs_modules()
+    finally:
+        sys.path.pop(0)
+
+
+a = Analysis(['patroni/__main__.py'],
+             pathex=[],
+             binaries=None,
+             datas=None,
+             hiddenimports=hiddenimports(),
+             hookspath=[],
+             runtime_hooks=[],
+             excludes=[],
+             win_no_prefer_redirects=False,
+             win_private_assemblies=False,
+             cipher=block_cipher)
+
+pyz = PYZ(a.pure, a.zipped_data, cipher=block_cipher)
+
+exe = EXE(pyz,
+    a.scripts,
+    a.binaries,
+    a.zipfiles,
+    a.datas,
+    name='patroni',
+    debug=False,
+    strip=False,
+    upx=True,
+    console=True)

patroni/__init__.py

@@ -1,79 +1,214 @@
 import logging
 import os
+import signal
 import sys
 import time
-import yaml
-
-from patroni.api import RestApiServer
-from patroni.etcd import Etcd
-from patroni.ha import Ha
-from patroni.postgresql import Postgresql
-from patroni.utils import setup_signal_handlers, reap_children
-from patroni.zookeeper import ZooKeeper
 
 logger = logging.getLogger(__name__)
 
 
-class Patroni:
+class Patroni(object):
 
-    def __init__(self, config):
-        self.nap_time = config['loop_wait']
-        self.tags = config.get('tags', dict())
-        self.postgresql = Postgresql(config['postgresql'])
-        self.dcs = self.get_dcs(self.postgresql.name, config)
-        self.api = RestApiServer(self, config['restapi'])
+    def __init__(self):
+        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.watchdog import Watchdog
+
+        self.setup_signal_handlers()
+
+        self.version = __version__
+        self.logger = PatroniLogger()
+        self.config = Config()
+        self.logger.reload_config(self.config.get('log', {}))
+        self.dcs = get_dcs(self.config)
+        self.watchdog = Watchdog(self.config)
+        self.load_dynamic_configuration()
+
+        self.postgresql = Postgresql(self.config['postgresql'])
+        self.api = RestApiServer(self, self.config['restapi'])
         self.ha = Ha(self)
+
+        self.tags = self.get_tags()
         self.next_run = time.time()
+        self.scheduled_restart = {}
+
+    def load_dynamic_configuration(self):
+        from patroni.exceptions import DCSError
+        while True:
+            try:
+                cluster = self.dcs.get_cluster()
+                if cluster and cluster.config and cluster.config.data:
+                    if self.config.set_dynamic_configuration(cluster.config):
+                        self.dcs.reload_config(self.config)
+                        self.watchdog.reload_config(self.config)
+                elif not self.config.dynamic_configuration and 'bootstrap' in self.config:
+                    if self.config.set_dynamic_configuration(self.config['bootstrap']['dcs']):
+                        self.dcs.reload_config(self.config)
+                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()
+                if tag not in ('clonefrom', 'nofailover', 'noloadbalance', 'nosync') or value}
 
     @property
     def nofailover(self):
-        return self.tags.get('nofailover', False)
+        return bool(self.tags.get('nofailover', False))
 
-    @staticmethod
-    def get_dcs(name, config):
-        if 'etcd' in config:
-            return Etcd(name, config['etcd'])
-        if 'zookeeper' in config:
-            return ZooKeeper(name, config['zookeeper'])
-        raise Exception('Can not find suitable configuration of distributed configuration store')
+    @property
+    def nosync(self):
+        return bool(self.tags.get('nosync', False))
+
+    def reload_config(self):
+        try:
+            self.tags = self.get_tags()
+            self.logger.reload_config(self.config.get('log', {}))
+            self.dcs.reload_config(self.config)
+            self.watchdog.reload_config(self.config)
+            self.api.reload_config(self.config['restapi'])
+            self.postgresql.reload_config(self.config['postgresql'])
+        except Exception:
+            logger.exception('Failed to reload config_file=%s', self.config.config_file)
+
+    @property
+    def replicatefrom(self):
+        return self.tags.get('replicatefrom')
+
+    def sighup_handler(self, *args):
+        self._received_sighup = True
+
+    def sigterm_handler(self, *args):
+        if not self._received_sigterm:
+            self._received_sigterm = True
+            sys.exit()
+
+    @property
+    def noloadbalance(self):
+        return bool(self.tags.get('noloadbalance', False))
 
     def schedule_next_run(self):
-        self.next_run += self.nap_time
+        self.next_run += self.dcs.loop_wait
         current_time = time.time()
         nap_time = self.next_run - current_time
         if nap_time <= 0:
             self.next_run = current_time
-        elif self.dcs.watch(nap_time):
+            # Release the GIL so we don't starve anyone waiting on async_executor lock
+            time.sleep(0.001)
+            # Warn user that Patroni is not keeping up
+            logger.warning("Loop time exceeded, rescheduling immediately.")
+        elif self.ha.watch(nap_time):
             self.next_run = time.time()
 
     def run(self):
         self.api.start()
         self.next_run = time.time()
 
-        while True:
+        while not self._received_sigterm:
+            if self._received_sighup:
+                self._received_sighup = False
+                if self.config.reload_local_configuration():
+                    self.reload_config()
+
             logger.info(self.ha.run_cycle())
-            reap_children()
+
+            if self.dcs.cluster and self.dcs.cluster.config and self.dcs.cluster.config.data \
+                    and self.config.set_dynamic_configuration(self.dcs.cluster.config):
+                self.reload_config()
+
+            if self.postgresql.role != 'uninitialized':
+                self.config.save_cache()
+
             self.schedule_next_run()
 
+    def setup_signal_handlers(self):
+        self._received_sighup = False
+        self._received_sigterm = False
+        if os.name != 'nt':
+            signal.signal(signal.SIGHUP, self.sighup_handler)
+        signal.signal(signal.SIGTERM, self.sigterm_handler)
 
-def main():
-    logging.basicConfig(format='%(asctime)s %(levelname)s: %(message)s', level=logging.INFO)
-    logging.getLogger('requests').setLevel(logging.WARNING)
-    setup_signal_handlers()
+    def shutdown(self):
+        try:
+            self.api.shutdown()
+        except Exception:
+            logger.exception('Exception during RestApi.shutdown')
+        self.ha.shutdown()
 
-    if len(sys.argv) < 2 or not os.path.isfile(sys.argv[1]):
-        print('Usage: {} config.yml'.format(sys.argv[0]))
-        return
 
-    with open(sys.argv[1], 'r') as f:
-        config = yaml.load(f)
-
-    patroni = Patroni(config)
+def patroni_main():
+    patroni = Patroni()
     try:
         patroni.run()
     except KeyboardInterrupt:
         pass
     finally:
-        patroni.api.shutdown()
-        patroni.postgresql.stop()
-        patroni.dcs.delete_leader()
+        patroni.shutdown()
+        logging.shutdown()
+
+
+def pg_ctl_start(args):
+    import subprocess
+    if os.name != 'nt':
+        os.setsid()
+    postmaster = subprocess.Popen(args)
+    print(postmaster.pid)
+
+
+def call_self(args, **kwargs):
+    """This function executes Patroni once again with provided arguments.
+
+    :args: list of arguments to call Patroni with.
+    :returns: `Popen` object"""
+
+    exe = [sys.executable]
+    if not getattr(sys, 'frozen', False):  # Binary distribution?
+        exe.append(sys.argv[0])
+
+    import subprocess
+    return subprocess.Popen(exe + args, **kwargs)
+
+
+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:])
+        return patroni_main()
+
+    pid = 0
+
+    # Looks like we are in a docker, so we will act like init
+    def sigchld_handler(signo, stack_frame):
+        try:
+            while True:
+                ret = os.waitpid(-1, os.WNOHANG)
+                if ret == (0, 0):
+                    break
+                elif ret[0] != pid:
+                    logger.info('Reaped pid=%s, exit status=%s', *ret)
+        except OSError:
+            pass
+
+    def passtochild(signo, stack_frame):
+        if pid:
+            os.kill(pid, signo)
+
+    signal.signal(signal.SIGCHLD, sigchld_handler)
+    if os.name != 'nt':
+        signal.signal(signal.SIGHUP, passtochild)
+        signal.signal(signal.SIGQUIT, 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:])
+    pid = patroni.pid
+    patroni.wait()

patroni/api.py

@@ -1,13 +1,16 @@
 import base64
-import fcntl
 import json
 import logging
 import psycopg2
-import socket
 import time
+import traceback
+import dateutil.parser
+import datetime
+import os
 
-from patroni.exceptions import PostgresConnectionException
-from patroni.utils import Retry, RetryFailedError
+from patroni.postgresql import PostgresConnectionException, PostgresException, Postgresql
+from patroni.utils import deep_compare, parse_bool, patch_config, Retry, \
+    RetryFailedError, parse_int, split_host_port, tzutc
 from six.moves.BaseHTTPServer import BaseHTTPRequestHandler, HTTPServer
 from six.moves.socketserver import ThreadingMixIn
 from threading import Thread
@@ -23,170 +26,360 @@ def check_auth(func):
     def do_PUT_foo():
         pass
     """
-    def wrapper(handler):
+    def wrapper(handler, *args, **kwargs):
         if handler.check_auth_header():
-            return func(handler)
+            return func(handler, *args, **kwargs)
     return wrapper
 
 
 class RestApiHandler(BaseHTTPRequestHandler):
 
-    def send_auth_request(self, body):
-        self.send_response(401)
-        self.send_header('WWW-Authenticate', 'Basic realm=\"Patroni\"')
-        self.send_header('Content-type', 'text/html')
+    def _write_response(self, status_code, body, content_type='text/html', headers=None):
+        self.send_response(status_code)
+        headers = headers or {}
+        if content_type:
+            headers['Content-Type'] = content_type
+        for name, value in headers.items():
+            self.send_header(name, value)
         self.end_headers()
         self.wfile.write(body.encode('utf-8'))
 
-    def finish(self, *args, **kwargs):
-        try:
-            if not self.wfile.closed:
-                self.wfile.flush()
-                self.wfile.close()
-        except socket.error:
-            pass
-        self.rfile.close()
+    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_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)
 
-    def do_OPTIONS(self):
-        self.do_GET(options=True)
+    def _write_status_response(self, status_code, response):
+        patroni = self.server.patroni
+        tags = patroni.ha.get_effective_tags()
+        if tags:
+            response['tags'] = tags
+        if patroni.postgresql.sysid:
+            response['database_system_identifier'] = patroni.postgresql.sysid
+        if patroni.postgresql.pending_restart:
+            response['pending_restart'] = True
+        response['patroni'] = {'version': patroni.version, 'scope': patroni.postgresql.scope}
+        if patroni.scheduled_restart and isinstance(patroni.scheduled_restart, dict):
+            response['scheduled_restart'] = patroni.scheduled_restart.copy()
+            del response['scheduled_restart']['postmaster_start_time']
+            response['scheduled_restart']['schedule'] = (response['scheduled_restart']['schedule']).isoformat()
+        if not patroni.ha.watchdog.is_healthy:
+            response['watchdog_failed'] = True
+        if patroni.ha.is_paused():
+            response['pause'] = True
+        self._write_json_response(status_code, response)
 
-    def do_GET(self, options=False):
+    def do_GET(self, write_status_code_only=False):
         """Default method for processing all GET requests which can not be routed to other methods"""
 
         path = '/master' if self.path == '/' else self.path
         response = self.get_postgresql_status()
-        response.update(self.get_tags())
 
         patroni = self.server.patroni
         cluster = patroni.dcs.cluster
-        if cluster:  # dcs available
-            if cluster.leader and cluster.leader.name == patroni.postgresql.name:  # is_leader
-                status_code = 200 if 'master' in path else 503
-            elif 'role' not in response:
-                status_code = 503
-            elif response['role'] == 'master':  # running as master but without leader lock!!!!
-                status_code = 503
-            elif response['role'] in path:
-                status_code = 200
-            else:
-                status_code = 503
-        elif 'role' in response and response['role'] in path:
-            status_code = 200
-        elif patroni.ha.restart_scheduled() and patroni.postgresql.role == 'master' and 'master' in path:
-            # exceptional case for master node when the postgres is being restarted via API
-            status_code = 200
-        else:
-            status_code = 503
 
-        self.send_response(status_code)
-        if not options:
-            self.send_header('Content-Type', 'application/json')
-            self.end_headers()
-            self.wfile.write(json.dumps(response).encode('utf-8'))
+        replica_status_code = 200 if not patroni.noloadbalance and response.get('role') == 'replica' else 503
+        status_code = 503
+
+        if patroni.ha.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:
+            # Round-robing across all masters in pause mode if DCS is not accessible
+            if not cluster and patroni.ha.is_paused():
+                status_code = 200 if response['role'] == 'master' else 503
+            else:
+                status_code = 200 if patroni.ha.is_leader() else 503
+        elif 'replica' in path:
+            status_code = replica_status_code
+        elif cluster:  # dcs is available
+            is_synchronous = cluster.is_synchronous_mode() and cluster.sync \
+                    and cluster.sync.sync_standby == patroni.postgresql.name
+            if path in ('/sync', '/synchronous') and is_synchronous:
+                status_code = replica_status_code
+            elif path in ('/async', '/asynchronous') and not is_synchronous:
+                status_code = replica_status_code
+
+        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'))
+        else:
+            self._write_status_response(status_code, response)
+
+    def do_OPTIONS(self):
+        self.do_GET(write_status_code_only=True)
 
     def do_GET_patroni(self):
         response = self.get_postgresql_status(True)
-        response.update(self.get_tags())
+        self._write_status_response(200, response)
 
-        self.send_response(200)
-        self.send_header('Content-Type', 'application/json')
-        self.end_headers()
-        self.wfile.write(json.dumps(response).encode('utf-8'))
+    def do_GET_config(self):
+        cluster = self.server.patroni.dcs.cluster or self.server.patroni.dcs.get_cluster()
+        if cluster.config:
+            self._write_json_response(200, cluster.config.data)
+        else:
+            self.send_error(502)
+
+    def _read_json_content(self, body_is_optional=False):
+        if 'content-length' not in self.headers:
+            return self.send_error(411) if not body_is_optional else {}
+        try:
+            content_length = int(self.headers.get('content-length'))
+            if content_length == 0 and body_is_optional:
+                return {}
+            request = json.loads(self.rfile.read(content_length).decode('utf-8'))
+            if isinstance(request, dict) and (request or body_is_optional):
+                return request
+        except Exception:
+            logger.exception('Bad request')
+        self.send_error(400)
+
+    @check_auth
+    def do_PATCH_config(self):
+        request = self._read_json_content()
+        if request:
+            cluster = self.server.patroni.dcs.get_cluster()
+            data = cluster.config.data.copy()
+            if patch_config(data, request):
+                value = json.dumps(data, separators=(',', ':'))
+                if not self.server.patroni.dcs.set_config_value(value, cluster.config.index):
+                    return self.send_error(409)
+            self.server.patroni.ha.wakeup()
+            self._write_json_response(200, data)
+
+    @check_auth
+    def do_PUT_config(self):
+        request = self._read_json_content()
+        if request:
+            cluster = self.server.patroni.dcs.get_cluster()
+            if not deep_compare(request, cluster.config.data):
+                value = json.dumps(request, separators=(',', ':'))
+                if not self.server.patroni.dcs.set_config_value(value):
+                    return self.send_error(502)
+            self._write_json_response(200, request)
+
+    @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)
+
+    @staticmethod
+    def parse_schedule(schedule, action):
+        """ parses the given schedule and validates at """
+        error = None
+        scheduled_at = None
+        try:
+            scheduled_at = dateutil.parser.parse(schedule)
+            if scheduled_at.tzinfo is None:
+                error = 'Timezone information is mandatory for the scheduled {0}'.format(action)
+                status_code = 400
+            elif scheduled_at < datetime.datetime.now(tzutc):
+                error = 'Cannot schedule {0} in the past'.format(action)
+                status_code = 422
+            else:
+                status_code = None
+        except (ValueError, TypeError):
+            logger.exception('Invalid scheduled %s time: %s', action, schedule)
+            error = 'Unable to parse scheduled timestamp. It should be in an unambiguous format, e.g. ISO 8601'
+            status_code = 422
+        return (status_code, error, scheduled_at)
 
     @check_auth
     def do_POST_restart(self):
-        status_code = 503
-        data = b'restart failed'
-        try:
-            status, msg = self.server.patroni.ha.restart()
-            status_code = 200 if status else 503
-            data = msg.encode('utf-8')
-        except:
-            logger.exception('Exception during restart')
+        status_code = 500
+        data = 'restart failed'
+        request = self._read_json_content(body_is_optional=True)
+        cluster = self.server.patroni.dcs.get_cluster()
+        if request is None:
+            # failed to parse the json
+            return
+        if request:
+            logger.debug("received restart request: {0}".format(request))
 
-        self.send_response(status_code)
-        self.send_header('Content-Type', 'text/html')
-        self.end_headers()
-        self.wfile.write(data)
+        if cluster.is_paused() and 'schedule' in request:
+            self._write_response(status_code, "Can't schedule restart in the paused state")
+            return
+
+        for k in request:
+            if k == 'schedule':
+                (_, data, request[k]) = self.parse_schedule(request[k], "restart")
+                if _:
+                    status_code = _
+                    break
+            elif k == 'role':
+                if request[k] not in ('master', 'replica'):
+                    status_code = 400
+                    data = "PostgreSQL role should be either master or replica"
+                    break
+            elif k == 'postgres_version':
+                try:
+                    Postgresql.postgres_version_to_int(request[k])
+                except PostgresException as e:
+                    status_code = 400
+                    data = e.value
+                    break
+            elif k == 'timeout':
+                request[k] = parse_int(request[k], 's')
+                if request[k] is None or request[k] <= 0:
+                    status_code = 400
+                    data = "Timeout should be a positive number of seconds"
+                    break
+            elif k != 'restart_pending':
+                status_code = 400
+                data = "Unknown filter for the scheduled restart: {0}".format(k)
+                break
+        else:
+            if 'schedule' not in request:
+                try:
+                    status, data = self.server.patroni.ha.restart(request)
+                    status_code = 200 if status else 503
+                except Exception:
+                    logger.exception('Exception during restart')
+                    status_code = 400
+            else:
+                if self.server.patroni.ha.schedule_future_restart(request):
+                    data = "Restart scheduled"
+                    status_code = 202
+                else:
+                    data = "Another restart is already scheduled"
+                    status_code = 409
+        self._write_response(status_code, data)
+
+    @check_auth
+    def do_DELETE_restart(self):
+        if self.server.patroni.ha.delete_future_restart():
+            data = "scheduled restart deleted"
+            code = 200
+        else:
+            data = "no restarts are scheduled"
+            code = 404
+        self._write_response(code, data)
 
     @check_auth
     def do_POST_reinitialize(self):
-        ha = self.server.patroni.ha
-        cluster = ha.dcs.get_cluster()
-        if cluster.is_unlocked():
-            status_code = 503
-            data = b'Cluster has no leader, can not reinitialize'
-        elif cluster.leader.name == ha.state_handler.name:
-            status_code = 503
-            data = b'I am the leader, can not reinitialize'
+        request = self._read_json_content(body_is_optional=True)
+
+        if request:
+            logger.debug('received reinitialize request: %s', request)
+
+        force = isinstance(request, dict) and parse_bool(request.get('force')) or False
+
+        data = self.server.patroni.ha.reinitialize(force)
+        if data is None:
+            status_code = 200
+            data = 'reinitialize started'
         else:
-            action = ha.schedule_reinitialize()
-            if action is not None:
-                status_code = 503
-                data = (action + ' already in progress').encode('utf-8')
-            else:
-                status_code = 200
-                data = b'reinitialize scheduled'
+            status_code = 503
+        self._write_response(status_code, data)
 
-        self.send_response(status_code)
-        self.send_header('Content-Type', 'text/html')
-        self.end_headers()
-        self.wfile.write(data)
-
-    def poll_failover_result(self, leader, member):
-        for a in range(0, 15):
+    def poll_failover_result(self, leader, candidate, action):
+        timeout = max(10, self.server.patroni.dcs.loop_wait)
+        for _ in range(0, timeout*2):
             time.sleep(1)
             try:
                 cluster = self.server.patroni.dcs.get_cluster()
-                if cluster.leader and cluster.leader.name != leader:
-                    return 200, ('Successfully failed over to ' + cluster.leader.name).encode('utf-8')
+                if not cluster.is_unlocked() and cluster.leader.name != leader:
+                    if not candidate or candidate == cluster.leader.name:
+                        return 200, 'Successfully {0}ed over to "{1}"'.format(action[:-4], cluster.leader.name)
+                    else:
+                        return 200, '{0}ed over to "{1}" instead of "{2}"'.format(action[:-4].title(),
+                                                                                  cluster.leader.name, candidate)
                 if not cluster.failover:
-                    return 503, b'Failover failed'
-            except:
-                pass
-        return 503, b'Failover status unknown'
+                    return 503, action.title() + ' failed'
+            except Exception as e:
+                logger.debug('Exception occured during polling %s result: %s', action, e)
+        return 503, action.title() + ' status unknown'
 
-    def is_failover_possible(self, cluster, leader, member):
-        if leader and not cluster.leader or cluster.leader.name != leader:
-            return b'leader name does not match'
-        if member:
-            members = [m for m in cluster.members if m.name == member]
+    def is_failover_possible(self, cluster, leader, candidate, action):
+        if leader and (not cluster.leader or cluster.leader.name != leader):
+            return 'leader name does not match'
+        if candidate:
+            if action == 'switchover' and cluster.is_synchronous_mode() and cluster.sync.sync_standby != candidate:
+                return 'candidate name does not match with sync_standby'
+            members = [m for m in cluster.members if m.name == candidate]
             if not members:
-                return b'member does not exists'
+                return 'candidate does not exists'
+        elif cluster.is_synchronous_mode():
+            members = [m for m in cluster.members if m.name == cluster.sync.sync_standby]
+            if not members:
+                return action + ' is not possible: can not find sync_standby'
         else:
             members = [m for m in cluster.members if m.name != cluster.leader.name and m.api_url]
             if not members:
-                return b'failover is not possible: cluster does not have members except leader'
-        for member, reachable, in_recovery, xlog_location, tags in self.server.patroni.ha.fetch_nodes_statuses(members):
-            if reachable and not tags.get('nofailover', False):
+                return action + ' is not possible: cluster does not have members except leader'
+        for st in self.server.patroni.ha.fetch_nodes_statuses(members):
+            if st.failover_limitation() is None:
                 return None
-        return b'failover is not possible: no good candidates have been found'
+        return action + ' is not possible: no good candidates have been found'
 
     @check_auth
-    def do_POST_failover(self):
-        content_length = int(self.headers.get('content-length', 0))
-        request = json.loads(self.rfile.read(content_length).decode('utf-8'))
-        leader = request.get('leader', None)
-        member = request.get('member', None)
-        cluster = self.server.patroni.ha.dcs.get_cluster()
-        status_code = 503
-        data = self.is_failover_possible(cluster, leader, member)
-        if not data:
-            if not self.server.patroni.dcs.manual_failover(leader, member):
-                data = b'failed to write failover key into DCS'
-            else:
-                self.server.patroni.dcs.event.set()
-                status_code, data = self.poll_failover_result(cluster.leader and cluster.leader.name, member)
+    def do_POST_failover(self, action='failover'):
+        request = self._read_json_content()
+        (status_code, data) = (400, '')
+        if not request:
+            return
 
-        self.send_response(status_code)
-        self.send_header('Content-Type', 'text/html')
-        self.end_headers()
-        self.wfile.write(data)
+        leader = request.get('leader')
+        candidate = request.get('candidate') or request.get('member')
+        scheduled_at = request.get('scheduled_at')
+        cluster = self.server.patroni.dcs.get_cluster()
+
+        logger.info("received %s request with leader=%s candidate=%s scheduled_at=%s",
+                    action, leader, candidate, scheduled_at)
+
+        if action == 'failover' and not candidate:
+            data = 'Failover could be performed only to a specific candidate'
+        elif action == 'switchover' and not leader:
+            data = 'Switchover could be performed only from a specific leader'
+
+        if not data and scheduled_at:
+            if not leader:
+                data = 'Scheduled {0} is possible only from a specific leader'.format(action)
+            if not data and cluster.is_paused():
+                data = "Can't schedule {0} in the paused state".format(action)
+            if not data:
+                (status_code, data, scheduled_at) = self.parse_schedule(scheduled_at, action)
+
+        if not data and cluster.is_paused() and not candidate:
+            data = action.title() + ' is possible only to a specific candidate in a paused state'
+
+        if not data and not scheduled_at:
+            data = self.is_failover_possible(cluster, leader, candidate, action)
+            if data:
+                status_code = 412
+
+        if not data:
+            if self.server.patroni.dcs.manual_failover(leader, candidate, scheduled_at=scheduled_at):
+                self.server.patroni.ha.wakeup()
+                if scheduled_at:
+                    data = action.title() + ' scheduled'
+                    status_code = 202
+                else:
+                    status_code, data = self.poll_failover_result(cluster.leader and cluster.leader.name,
+                                                                  candidate, action)
+            else:
+                data = 'failed to write {0} key into DCS'.format(action)
+                status_code = 503
+        self._write_response(status_code, data)
+
+    def do_POST_switchover(self):
+        self.do_POST_failover(action='switchover')
 
     def parse_request(self):
         """Override parse_request method to enrich basic functionality of `BaseHTTPRequestHandler` class
@@ -206,12 +399,6 @@ class RestApiHandler(BaseHTTPRequestHandler):
                 self.command = mname
         return ret
 
-    def handle_one_request(self):
-        try:
-            BaseHTTPRequestHandler.handle_one_request(self)
-        except socket.error:
-            pass
-
     def query(self, sql, *params, **kwargs):
         if not kwargs.get('retry', False):
             return self.server.query(sql, *params)
@@ -220,62 +407,73 @@ class RestApiHandler(BaseHTTPRequestHandler):
 
     def get_postgresql_status(self, retry=False):
         try:
-            row = self.query("""SELECT to_char(pg_postmaster_start_time(), 'YYYY-MM-DD HH24:MI:SS.MS TZ'),
-                                       pg_is_in_recovery(),
-                                       CASE WHEN pg_is_in_recovery()
-                                            THEN 0
-                                            ELSE pg_xlog_location_diff(pg_current_xlog_location(), '0/0')::bigint
-                                       END,
-                                       pg_xlog_location_diff(pg_last_xlog_receive_location(), '0/0')::bigint,
-                                       pg_xlog_location_diff(pg_last_xlog_replay_location(), '0/0')::bigint,
-                                       pg_is_in_recovery() AND pg_is_xlog_replay_paused()""", retry=retry)[0]
-            return {
+            cluster = self.server.patroni.dcs.cluster
+
+            if self.server.patroni.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_catalog.pg_stat_replication) SELECT"
+                    " pg_catalog.to_char(pg_catalog.pg_postmaster_start_time(), 'YYYY-MM-DD HH24:MI:SS.MS TZ'),"
+                    " CASE WHEN pg_catalog.pg_is_in_recovery() THEN 0"
+                    " ELSE ('x' || pg_catalog.substr(pg_catalog.pg_{0}file_name("
+                    "pg_catalog.pg_current_{0}_{1}()), 1, 8))::bit(32)::int END,"
+                    " CASE WHEN pg_catalog.pg_is_in_recovery() THEN 0"
+                    " ELSE pg_catalog.pg_{0}_{1}_diff(pg_catalog.pg_current_{0}_{1}(), '0/0')::bigint END,"
+                    " pg_catalog.pg_{0}_{1}_diff(COALESCE(pg_catalog.pg_last_{0}_receive_{1}(),"
+                    " pg_catalog.pg_last_{0}_replay_{1}()), '0/0')::bigint,"
+                    " pg_catalog.pg_{0}_{1}_diff(pg_catalog.pg_last_{0}_replay_{1}(), '0/0')::bigint,"
+                    " pg_catalog.to_char(pg_catalog.pg_last_xact_replay_timestamp(), 'YYYY-MM-DD HH24:MI:SS.MS TZ'),"
+                    " pg_catalog.pg_is_in_recovery() AND pg_catalog.pg_is_{0}_replay_paused(), "
+                    "(SELECT pg_catalog.array_to_json(pg_catalog.array_agg("
+                    "pg_catalog.row_to_json(ri))) FROM replication_info ri)")
+
+            row = self.query(stmt.format(self.server.patroni.postgresql.wal_name,
+                                         self.server.patroni.postgresql.lsn_name), retry=retry)[0]
+
+            result = {
                 'state': self.server.patroni.postgresql.state,
                 'postmaster_start_time': row[0],
-                'role': 'replica' if row[1] else 'master',
+                'role': 'replica' if row[1] == 0 else 'master',
+                'server_version': self.server.patroni.postgresql.server_version,
+                'cluster_unlocked': bool(not cluster or cluster.is_unlocked()),
                 'xlog': ({
                     'received_location': row[3],
                     'replayed_location': row[4],
-                    'paused': row[5]} if row[1] else {
+                    'replayed_timestamp': row[5],
+                    'paused': row[6]} if row[1] == 0 else {
                     'location': row[2]
                 })
             }
+
+            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)
+
+            if row[7]:
+                result['replication'] = row[7]
+
+            return result
         except (psycopg2.Error, RetryFailedError, PostgresConnectionException):
             state = self.server.patroni.postgresql.state
             if state == 'running':
                 logger.exception('get_postgresql_status')
                 state = 'unknown'
-            return {'state': state}
+            return {'state': state, 'role': self.server.patroni.postgresql.role}
 
-    def get_tags(self):
-        return {'tags': self.server.patroni.tags}
-
-    def log_message(self, format, *args):
-        logger.debug("API thread: " + format % args)
+    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):
 
     def __init__(self, patroni, config):
-        self._auth_key = base64.b64encode(config['auth'].encode('utf-8')).decode('utf-8') if 'auth' in config else None
-        host, port = config['listen'].split(':')
-        HTTPServer.__init__(self, (host, int(port)), RestApiHandler)
-        Thread.__init__(self, target=self.serve_forever)
-        self._set_fd_cloexec(self.socket)
-
-        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'.
-        options = {option: config[option] for option in ['certfile', 'keyfile'] if option in config}
-        if options.get('certfile', None):
-            import ssl
-            self.socket = ssl.wrap_socket(self.socket, server_side=True, **options)
-            protocol = 'https'
-
-        self.connection_string = '{}://{}/patroni'.format(protocol, config.get('connect_address', config['listen']))
-
         self.patroni = patroni
+        self.__listen = None
+        self.__initialize(config)
+        self.__set_config_parameters(config)
         self.daemon = True
 
     def query(self, sql, *params):
@@ -291,15 +489,68 @@ class RestApiServer(ThreadingMixIn, HTTPServer, Thread):
 
     @staticmethod
     def _set_fd_cloexec(fd):
-        flags = fcntl.fcntl(fd, fcntl.F_GETFD)
-        fcntl.fcntl(fd, fcntl.F_SETFD, flags | fcntl.FD_CLOEXEC)
+        if os.name != 'nt':
+            import fcntl
+            flags = fcntl.fcntl(fd, fcntl.F_GETFD)
+            fcntl.fcntl(fd, fcntl.F_SETFD, flags | fcntl.FD_CLOEXEC)
 
     def check_basic_auth_key(self, key):
-        return self._auth_key == key
+        return self.__auth_key == key
 
     def check_auth_header(self, auth_header):
-        if self._auth_key:
+        if self.__auth_key:
             if auth_header is None:
                 return 'no auth header received'
             if not auth_header.startswith('Basic ') or not self.check_basic_auth_key(auth_header[6:]):
                 return 'not authenticated'
+
+    @staticmethod
+    def __get_ssl_options(config):
+        return {option: config[option] for option in ['certfile', 'keyfile'] if option in config}
+
+    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 __initialize(self, config):
+        try:
+            host, port = split_host_port(config['listen'], None)
+        except Exception:
+            raise ValueError('Invalid "restapi" config: expected <HOST>:<PORT> for "listen", but got "{0}"'
+                             .format(config['listen']))
+
+        if self.__listen is not None:  # changing config in runtime
+            self.shutdown()
+
+        self.__listen = config['listen']
+        self.__ssl_options = self.__get_ssl_options(config)
+
+        HTTPServer.__init__(self, (host, port), RestApiHandler)
+        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'):
+            import ssl
+            self.socket = ssl.wrap_socket(self.socket, server_side=True, **self.__ssl_options)
+            self.__protocol = 'https'
+        return True
+
+    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)
+
+    @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

@@ -1,28 +1,78 @@
 import logging
-from threading import Lock, Thread
+from threading import Event, Lock, RLock, Thread
 
 logger = logging.getLogger(__name__)
 
 
-class AsyncExecutor:
+class CriticalTask(object):
+    """Represents a critical task in a background process that we either need to cancel or get the result of.
 
+    Fields of this object may be accessed only when holding a lock on it. To perform the critical task the background
+    thread must, while holding lock on this object, check `is_cancelled` flag, run the task and mark the task as
+    complete using `complete()`.
+
+    The main thread must hold async lock to prevent the task from completing, hold lock on critical task object,
+    call cancel. If the task has completed `cancel()` will return False and `result` field will contain the result of
+    the task. When cancel returns True it is guaranteed that the background task will notice the `is_cancelled` flag.
+    """
     def __init__(self):
-        Lock.__init__(self)
-        self._busy = False
-        self._thread_lock = Lock()
+        self._lock = Lock()
+        self.is_cancelled = False
+        self.result = None
+
+    def reset(self):
+        """Must be called every time the background task is finished.
+
+        Must be called from async thread. Caller must hold lock on async executor when calling."""
+        self.is_cancelled = False
+        self.result = None
+
+    def cancel(self):
+        """Tries to cancel the task, returns True if the task has already run.
+
+        Caller must hold lock on async executor and the task when calling."""
+        if self.result is not None:
+            return False
+        self.is_cancelled = True
+        return True
+
+    def complete(self, result):
+        """Mark task as completed along with a result.
+
+        Must be called from async thread. Caller must hold lock on task when calling."""
+        self.result = result
+
+    def __enter__(self):
+        self._lock.acquire()
+        return self
+
+    def __exit__(self, exc_type, exc_val, exc_tb):
+        self._lock.release()
+
+
+class AsyncExecutor(object):
+
+    def __init__(self, state_handler, ha_wakeup):
+        self.state_handler = state_handler
+        self._ha_wakeup = ha_wakeup
+        self._thread_lock = RLock()
         self._scheduled_action = None
-        self._scheduled_action_lock = Lock()
+        self._scheduled_action_lock = RLock()
+        self._is_cancelled = False
+        self._finish_event = Event()
+        self.critical_task = CriticalTask()
 
     @property
     def busy(self):
-        return self._busy
+        return self.scheduled_action is not None
 
-    def schedule(self, action, immediately=False):
+    def schedule(self, action):
         with self._scheduled_action_lock:
             if self._scheduled_action is not None:
                 return self._scheduled_action
             self._scheduled_action = action
-            self._busy = immediately
+            self._is_cancelled = False
+            self._finish_event.set()
         return None
 
     @property
@@ -35,21 +85,47 @@ class AsyncExecutor:
             self._scheduled_action = None
 
     def run(self, func, args=()):
+        wakeup = False
         try:
-            return func(*args) if args else func()
-        except:
+            with self:
+                if self._is_cancelled:
+                    return
+                self._finish_event.clear()
+
+            self.state_handler.reset_is_cancelled()
+            # if the func returned something (not None) - wake up main HA loop
+            wakeup = func(*args) if args else func()
+            return wakeup
+        except Exception:
             logger.exception('Exception during execution of long running task %s', self.scheduled_action)
         finally:
             with self:
-                self._busy = False
                 self.reset_scheduled_action()
+                self._finish_event.set()
+                with self.critical_task:
+                    self.critical_task.reset()
+            if wakeup is not None:
+                self._ha_wakeup()
 
     def run_async(self, func, args=()):
-        self._busy = True
         Thread(target=self.run, args=(func, args)).start()
 
+    def cancel(self):
+        with self:
+            with self._scheduled_action_lock:
+                if self._scheduled_action is None:
+                    return
+                logger.warning('Cancelling long running task %s', self._scheduled_action)
+            self._is_cancelled = True
+
+        self.state_handler.cancel()
+        self._finish_event.wait()
+
+        with self:
+            self.reset_scheduled_action()
+
     def __enter__(self):
         self._thread_lock.acquire()
 
-    def __exit__(self, type, value, traceback):
+    def __exit__(self, *args):
         self._thread_lock.release()

patroni/callback_executor.py

@@ -0,0 +1,37 @@
+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

@@ -0,0 +1,378 @@
+import json
+import logging
+import os
+import shutil
+import sys
+import tempfile
+import yaml
+
+from collections import defaultdict
+from copy import deepcopy
+from patroni.dcs import ClusterConfig
+from patroni.postgresql import Postgresql
+from patroni.utils import deep_compare, parse_bool, parse_int, patch_config
+from requests.structures import CaseInsensitiveDict
+
+logger = logging.getLogger(__name__)
+
+
+class Config(object):
+    """
+    This class is responsible for:
+
+      1) Building and giving access to `effective_configuration` from:
+         * `Config.__DEFAULT_CONFIG` -- some sane default values
+         * `dynamic_configuration` -- configuration stored in DCS
+         * `local_configuration` -- configuration from `config.yml` or environment
+
+      2) Saving and loading `dynamic_configuration` into 'patroni.dynamic.json' file
+         located in local_configuration['postgresql']['data_dir'] directory.
+         This is necessary to be able to restore `dynamic_configuration`
+         if DCS was accidentally wiped
+
+      3) Loading of configuration file in the old format and converting it into new format
+
+      4) Mimicking some of the `dict` interfaces to make it possible
+         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,
+        'synchronous_mode': False,
+        'synchronous_mode_strict': False,
+        'standby_cluster': {
+            'create_replica_methods': '',
+            'host': '',
+            'port': '',
+            'primary_slot_name': '',
+            'restore_command': '',
+            'archive_cleanup_command': '',
+            'recovery_min_apply_delay': ''
+        },
+        'postgresql': {
+            'bin_dir': '',
+            'use_slots': True,
+            'parameters': CaseInsensitiveDict({p: v[0] for p, v in Postgresql.CMDLINE_OPTIONS.items()})
+        },
+        'watchdog': {
+            'mode': 'automatic',
+        }
+    }
+
+    def __init__(self):
+        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]
+        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)
+
+        self.__effective_configuration = self._build_effective_configuration({}, self._local_configuration)
+        self._data_dir = self.__effective_configuration['postgresql']['data_dir']
+        self._cache_file = os.path.join(self._data_dir, self.__CACHE_FILENAME)
+        self._load_cache()
+        self._cache_needs_saving = False
+
+    @property
+    def config_file(self):
+        return self._config_file
+
+    @property
+    def dynamic_configuration(self):
+        return deepcopy(self._dynamic_configuration)
+
+    def check_mode(self, mode):
+        return bool(parse_bool(self._dynamic_configuration.get(mode)))
+
+    def _load_config_file(self):
+        """Loads config.yaml from filesystem and applies some values which were set via ENV"""
+        with open(self._config_file) as f:
+            config = yaml.safe_load(f)
+            patch_config(config, self.__environment_configuration)
+            return config
+
+    def _load_cache(self):
+        if os.path.isfile(self._cache_file):
+            try:
+                with open(self._cache_file) as f:
+                    self.set_dynamic_configuration(json.load(f))
+            except Exception:
+                logger.exception('Exception when loading file: %s', self._cache_file)
+
+    def save_cache(self):
+        if self._cache_needs_saving:
+            tmpfile = fd = None
+            try:
+                (fd, tmpfile) = tempfile.mkstemp(prefix=self.__CACHE_FILENAME, dir=self._data_dir)
+                with os.fdopen(fd, 'w') as f:
+                    fd = None
+                    json.dump(self.dynamic_configuration, f)
+                tmpfile = shutil.move(tmpfile, self._cache_file)
+                self._cache_needs_saving = False
+            except Exception:
+                logger.exception('Exception when saving file: %s', self._cache_file)
+                if fd:
+                    try:
+                        os.close(fd)
+                    except Exception:
+                        logger.error('Can not close temporary file %s', tmpfile)
+                if tmpfile and os.path.exists(tmpfile):
+                    try:
+                        os.remove(tmpfile)
+                    except Exception:
+                        logger.error('Can not remove temporary file %s', tmpfile)
+
+    # configuration could be either ClusterConfig or dict
+    def set_dynamic_configuration(self, configuration):
+        if isinstance(configuration, ClusterConfig):
+            if self._modify_index == configuration.modify_index:
+                return False  # If the index didn't changed there is nothing to do
+            self._modify_index = configuration.modify_index
+            configuration = configuration.data
+
+        if not deep_compare(self._dynamic_configuration, configuration):
+            try:
+                self.__effective_configuration = self._build_effective_configuration(configuration,
+                                                                                     self._local_configuration)
+                self._dynamic_configuration = configuration
+                self._cache_needs_saving = True
+                return True
+            except Exception:
+                logger.exception('Exception when setting dynamic_configuration')
+
+    def reload_local_configuration(self, dry_run=False):
+        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.')
+            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
+
+    def _safe_copy_dynamic_configuration(self, dynamic_configuration):
+        config = deepcopy(self.__DEFAULT_CONFIG)
+
+        for name, value in dynamic_configuration.items():
+            if name == 'postgresql':
+                for name, value in (value or {}).items():
+                    if name == 'parameters':
+                        config['postgresql'][name].update(self._process_postgresql_parameters(value))
+                    elif name not in ('connect_address', 'listen', 'data_dir', 'pgpass', 'authentication'):
+                        config['postgresql'][name] = deepcopy(value)
+            elif name == 'standby_cluster':
+                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
+                else:
+                    config[name] = int(value)
+        return config
+
+    @staticmethod
+    def _build_environment_configuration():
+        ret = defaultdict(dict)
+
+        def _popenv(name):
+            return os.environ.pop(Config.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 = Config.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('log', ['level', 'format', 'dateformat', 'dir', 'file_size', 'file_num', 'loggers'])
+
+        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):
+            ret = {}
+            for param in ('username', 'password'):
+                value = _popenv(name + '_' + param)
+                if value:
+                    ret[param] = value
+            return ret
+
+        restapi_auth = _get_auth('restapi')
+        if restapi_auth:
+            ret['restapi']['authentication'] = restapi_auth
+
+        authentication = {}
+        for user_type in ('replication', 'superuser'):
+            entry = _get_auth(user_type)
+            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)
+            try:
+                return yaml.safe_load(value)
+            except Exception:
+                logger.exception('Exception when parsing list %s', value)
+                return None
+
+        for param in list(os.environ.keys()):
+            if param.startswith(Config.PATRONI_ENV_PREFIX):
+                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', 'REGISTER_SERVICE', 'SERVICE_CHECK_INTERVAL', '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':
+                            value = _parse_dict(value)
+                        elif suffix == 'REGISTER_SERVICE':
+                            value = parse_bool(value)
+                        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 users:
+            ret['bootstrap']['users'] = users
+
+        return ret
+
+    def _build_effective_configuration(self, dynamic_configuration, local_configuration):
+        config = self._safe_copy_dynamic_configuration(dynamic_configuration)
+        for name, value in local_configuration.items():
+            if name == 'postgresql':
+                for name, value in (value or {}).items():
+                    if name == 'parameters':
+                        config['postgresql'][name].update(self._process_postgresql_parameters(value, True))
+                    elif name != 'use_slots':  # replication slots must be enabled/disabled globally
+                        config['postgresql'][name] = deepcopy(value)
+            elif name not in config or name in ['watchdog']:
+                config[name] = deepcopy(value) if value else {}
+
+        # restapi server expects to get restapi.auth = 'username:password'
+        if 'authentication' in config['restapi']:
+            config['restapi']['auth'] = '{username}:{password}'.format(**config['restapi']['authentication'])
+
+        # special treatment for old config
+
+        # 'exhibitor' inside 'zookeeper':
+        if 'zookeeper' in config and 'exhibitor' in config['zookeeper']:
+            config['exhibitor'] = config['zookeeper'].pop('exhibitor')
+            config.pop('zookeeper')
+
+        pg_config = config['postgresql']
+        # no 'authentication' in 'postgresql', but 'replication' and 'superuser'
+        if 'authentication' not in pg_config:
+            pg_config['use_pg_rewind'] = 'pg_rewind' in pg_config
+            pg_config['authentication'] = {u: pg_config[u] for u in ('replication', 'superuser') if u in pg_config}
+        # no 'superuser' in 'postgresql'.'authentication'
+        if 'superuser' not in pg_config['authentication'] and 'pg_rewind' in pg_config:
+            pg_config['authentication']['superuser'] = pg_config['pg_rewind']
+
+        # no 'name' in config
+        if 'name' not in config and 'name' in pg_config:
+            config['name'] = pg_config['name']
+
+        updated_fields = (
+            'name',
+            'scope',
+            'retry_timeout',
+            'synchronous_mode',
+            'synchronous_mode_strict',
+        )
+
+        pg_config.update({p: config[p] for p in updated_fields if p in config})
+
+        return config
+
+    def get(self, key, default=None):
+        return self.__effective_configuration.get(key, default)
+
+    def __contains__(self, key):
+        return key in self.__effective_configuration
+
+    def __getitem__(self, key):
+        return self.__effective_configuration[key]
+
+    def copy(self):
+        return deepcopy(self.__effective_configuration)

patroni/dcs.py

@@ -1,274 +0,0 @@
-import abc
-import json
-
-from collections import namedtuple
-from patroni.exceptions import DCSError
-from six.moves.urllib_parse import urlparse, urlunparse, parse_qsl
-from threading import Event, Lock
-
-
-def parse_connection_string(value):
-    """Original Governor stores connection strings for each cluster members if a following format:
-        postgres://{username}:{password}@{connect_address}/postgres
-    Since each of our patroni instances provides own REST API endpoint it's good to store this information
-    in DCS among with postgresql connection string. In order to not introduce new keys and be compatible with
-    original Governor we decided to extend original connection string in a following way:
-        postgres://{username}:{password}@{connect_address}/postgres?application_name={api_url}
-    This way original Governor could use such connection string as it is, because of feature of `libpq` library.
-
-    This method is able to split connection string stored in DCS into two parts, `conn_url` and `api_url`"""
-
-    scheme, netloc, path, params, query, fragment = urlparse(value)
-    conn_url = urlunparse((scheme, netloc, path, params, '', fragment))
-    api_url = ([v for n, v in parse_qsl(query) if n == 'application_name'] or [None])[0]
-    return conn_url, api_url
-
-
-class Member(namedtuple('Member', 'index,name,session,data')):
-
-    """Immutable object (namedtuple) which represents single member of PostgreSQL cluster.
-    Consists of the following fields:
-    :param index: modification index of a given member key in a Configuration Store
-    :param name: name of PostgreSQL cluster member
-    :param session: either session id or just ttl in seconds
-    :param data: arbitrary data i.e. conn_url, api_url, xlog location, state, role, tags, etc...
-
-    There are two mandatory keys in a data:
-    conn_url: connection string containing host, user and password which could be used to access this member.
-    api_url: REST API url of patroni instance"""
-
-    @staticmethod
-    def from_node(index, name, session, data):
-        """
-        >>> Member.from_node(-1, '', '', '{"conn_url": "postgres://foo@bar/postgres"}') is not None
-        True
-        >>> Member.from_node(-1, '', '', '{')
-        Member(index=-1, name='', session='', data={})
-        """
-        if data.startswith('postgres'):
-            conn_url, api_url = parse_connection_string(data)
-            data = {'conn_url': conn_url, 'api_url': api_url}
-        else:
-            try:
-                data = json.loads(data)
-            except:
-                data = {}
-        return Member(index, name, session, data)
-
-    @property
-    def conn_url(self):
-        return self.data.get('conn_url', None)
-
-    @property
-    def api_url(self):
-        return self.data.get('api_url', None)
-
-    @property
-    def nofailover(self):
-        return self.data.get('tags', {}).get('nofailover', False)
-
-
-class Leader(namedtuple('Leader', 'index,session,member')):
-
-    """Immutable object (namedtuple) which represents leader key.
-    Consists of the following fields:
-    :param index: modification index of a leader key in a Configuration Store
-    :param session: either session id or just ttl in seconds
-    :param member: reference to a `Member` object which represents current leader (see `Cluster.members`)"""
-
-    @property
-    def name(self):
-        return self.member.name
-
-    @property
-    def conn_url(self):
-        return self.member.conn_url
-
-
-class Failover(namedtuple('Failover', 'index,leader,member')):
-
-    @staticmethod
-    def from_node(index, value):
-        t = [a.strip() for a in value.split(':')] + ['']
-        return Failover(index, t[0], t[1]) if t[0] or t[1] else None
-
-
-class Cluster(namedtuple('Cluster', 'initialize,leader,last_leader_operation,members,failover')):
-
-    """Immutable object (namedtuple) which represents PostgreSQL cluster.
-    Consists of the following fields:
-    :param initialize: boolean, shows whether this cluster has initialization key stored in DC or not.
-    :param leader: `Leader` object which represents current leader of the cluster
-    :param last_leader_operation: int or long object containing position of last known leader operation.
-        This value is stored in `/optime/leader` key
-    :param members: list of Member object, all PostgreSQL cluster members including leader
-    :param failover: reference to `Failover` object"""
-
-    def is_unlocked(self):
-        return not (self.leader and self.leader.name)
-
-
-class AbstractDCS:
-
-    __metaclass__ = abc.ABCMeta
-
-    _INITIALIZE = 'initialize'
-    _LEADER = 'leader'
-    _FAILOVER = 'failover'
-    _MEMBERS = 'members/'
-    _OPTIME = 'optime'
-    _LEADER_OPTIME = _OPTIME + '/' + _LEADER
-
-    def __init__(self, name, config):
-        """
-        :param name: name of current instance (the same value as `~Postgresql.name`)
-        :param config: dict, reference to config section of selected DCS.
-            i.e.: `zookeeper` for zookeeper, `etcd` for etcd, etc...
-        """
-        self._name = name
-        self._namespace = '/{}'.format(config.get('namespace', '/service/').strip('/'))
-        self._base_path = '/'.join([self._namespace, config['scope']])
-
-        self._cluster = None
-        self._cluster_thread_lock = Lock()
-        self.event = Event()
-
-    def client_path(self, path):
-        return '/'.join([self._base_path, path.lstrip('/')])
-
-    @property
-    def initialize_path(self):
-        return self.client_path(self._INITIALIZE)
-
-    @property
-    def members_path(self):
-        return self.client_path(self._MEMBERS)
-
-    @property
-    def member_path(self):
-        return self.client_path(self._MEMBERS + self._name)
-
-    @property
-    def leader_path(self):
-        return self.client_path(self._LEADER)
-
-    @property
-    def failover_path(self):
-        return self.client_path(self._FAILOVER)
-
-    @property
-    def leader_optime_path(self):
-        return self.client_path(self._LEADER_OPTIME)
-
-    @abc.abstractmethod
-    def _load_cluster(self):
-        """Internally this method should build  `Cluster` object which
-           represents current state and topology of the cluster in DCS.
-           this method supposed to be called only by `get_cluster` method.
-
-           raise `~DCSError` in case of communication or other problems with DCS.
-           If the current node was running as a master and exception raised,
-           instance would be demoted."""
-
-    def get_cluster(self):
-        with self._cluster_thread_lock:
-            try:
-                self._load_cluster()
-            except:
-                self._cluster = None
-                raise
-            return self._cluster
-
-    @property
-    def cluster(self):
-        with self._cluster_thread_lock:
-            return self._cluster
-
-    def reset_cluster(self):
-        with self._cluster_thread_lock:
-            self._cluster = None
-
-    @abc.abstractmethod
-    def write_leader_optime(self, last_operation):
-        """write current xlog location into `/optime/leader` key in DCS
-        :param last_operation: absolute xlog location in bytes"""
-
-    @abc.abstractmethod
-    def update_leader(self):
-        """Update leader key (or session) ttl
-
-        :returns: `!True` if leader key (or session) has been updated successfully.
-            If not, `!False` must be returned and current instance would be demoted.
-
-        You have to use CAS (Compare And Swap) operation in order to update leader key,
-        for example for etcd `prevValue` parameter must be used."""
-
-    @abc.abstractmethod
-    def attempt_to_acquire_leader(self):
-        """Attempt to acquire leader lock
-        This method should create `/leader` key with value=`~self._name`
-        :returns: `!True` if key has been created successfully.
-
-        Key must be created atomically. In case if key already exists it should not be
-        overwritten and `!False` must be returned"""
-
-    @abc.abstractmethod
-    def set_failover_value(self, value, index=None):
-        """Create or update `/failover` key"""
-
-    def manual_failover(self, leader, member, index=None):
-        return self.set_failover_value(leader + (':' + member if member else ''), index)
-
-    def current_leader(self):
-        try:
-            cluster = self.get_cluster()
-            return None if cluster.is_unlocked() else cluster.leader
-        except DCSError:
-            return None
-
-    @abc.abstractmethod
-    def touch_member(self, connection_string, ttl=None):
-        """Update member key in DCS.
-        This method should create or update key with the name = '/members/' + `~self._name`
-        and value = connection_string in a given DCS.
-
-        :param connection_string: how this instance can be accessed by other instances
-        :param ttl: ttl for member key, optional parameter. If it is None `~self.member_ttl will be used`
-        :returns: `!True` on success otherwise `!False`
-        """
-
-    @abc.abstractmethod
-    def take_leader(self):
-        """This method should create leader key with value = `~self._name` and ttl=`~self.ttl`
-        Since it could be called only on initial cluster bootstrap it could create this key regardless,
-        overwriting the key if necessary."""
-
-    @abc.abstractmethod
-    def initialize(self, create_new=True, sysid=""):
-        """Race for cluster initialization.
-
-        :param create_new: False if the key should already exist (in the case we are setting the system_id)
-        :param sysid: PostgreSQL cluster system identifier, if specified, is written to the key
-        :returns: `!True` if key has been created successfully.
-
-        this method should create atomically initialize key and return `!True`
-        otherwise it should return `!False`"""
-
-    @abc.abstractmethod
-    def delete_leader(self):
-        """Voluntarily remove leader key from DCS
-        This method should remove leader key if current instance is the leader"""
-
-    @abc.abstractmethod
-    def cancel_initialization(self):
-        """ Removes the initialize key for a cluster """
-
-    def watch(self, timeout):
-        """If the current node is a master it should just sleep.
-        Any other node should watch for changes of leader key with a given timeout
-
-        :param timeout: timeout in seconds
-        :returns: `!True` if you would like to reschedule the next run of ha cycle"""
-
-        self.event.wait(timeout)
-        return self.event.isSet()

patroni/dcs/__init__.py

@@ -0,0 +1,759 @@
+import abc
+import dateutil
+import importlib
+import inspect
+import json
+import logging
+import os
+import pkgutil
+import re
+import six
+import sys
+
+from collections import defaultdict, namedtuple
+from copy import deepcopy
+from patroni.exceptions import PatroniException
+from patroni.utils import parse_bool
+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
+    Since each of our patroni instances provides own REST API endpoint it's good to store this information
+    in DCS among with postgresql connection string. In order to not introduce new keys and be compatible with
+    original Governor we decided to extend original connection string in a following way:
+        postgres://{username}:{password}@{connect_address}/postgres?application_name={api_url}
+    This way original Governor could use such connection string as it is, because of feature of `libpq` library.
+
+    This method is able to split connection string stored in DCS into two parts, `conn_url` and `api_url`"""
+
+    scheme, netloc, path, params, query, fragment = urlparse(value)
+    conn_url = urlunparse((scheme, netloc, path, params, '', fragment))
+    api_url = ([v for n, v in parse_qsl(query) if n == 'application_name'] or [None])[0]
+    return conn_url, api_url
+
+
+def dcs_modules():
+    """Get names of DCS modules, depending on execution environment. If being packaged with PyInstaller,
+    modules aren't discoverable dynamically by scanning source directory because `FrozenImporter` doesn't
+    implement `iter_modules` method. But it is still possible to find all potential DCS modules by
+    iterating through `toc`, which contains list of all "frozen" resources."""
+
+    dcs_dirname = os.path.dirname(__file__)
+    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]
+    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
+                        # 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:
+            if not config.get('patronictl'):
+                logger.info('Failed to import %s', module_name)
+    raise PatroniException("""Can not find suitable configuration of distributed configuration store
+Available implementations: """ + ', '.join(available_implementations))
+
+
+class Member(namedtuple('Member', 'index,name,session,data')):
+
+    """Immutable object (namedtuple) which represents single member of PostgreSQL cluster.
+    Consists of the following fields:
+    :param index: modification index of a given member key in a Configuration Store
+    :param name: name of PostgreSQL cluster member
+    :param session: either session id or just ttl in seconds
+    :param data: arbitrary data i.e. conn_url, api_url, xlog location, state, role, tags, etc...
+
+    There are two mandatory keys in a data:
+    conn_url: connection string containing host, user and password which could be used to access this member.
+    api_url: REST API url of patroni instance"""
+
+    @staticmethod
+    def from_node(index, name, session, data):
+        """
+        >>> Member.from_node(-1, '', '', '{"conn_url": "postgres://foo@bar/postgres"}') is not None
+        True
+        >>> Member.from_node(-1, '', '', '{')
+        Member(index=-1, name='', session='', data={})
+        """
+        if data.startswith('postgres'):
+            conn_url, api_url = parse_connection_string(data)
+            data = {'conn_url': conn_url, 'api_url': api_url}
+        else:
+            try:
+                data = json.loads(data)
+            except (TypeError, ValueError):
+                data = {}
+        return 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
+
+        if conn_kwargs:
+            conn_url = 'postgresql://{host}:{port}'.format(
+                host=conn_kwargs.get('host'),
+                port=conn_kwargs.get('port'),
+            )
+            self.data['conn_url'] = conn_url
+            return conn_url
+
+    def conn_kwargs(self, auth=None):
+        defaults = {
+            "host": "",
+            "port": "",
+            "database": ""
+        }
+        ret = self.data.get('conn_kwargs')
+        if ret:
+            defaults.update(ret)
+            ret = defaults
+        else:
+            r = urlparse(self.conn_url)
+            ret = {
+                'host': r.hostname,
+                'port': r.port or 5432,
+                'database': r.path[1:]
+            }
+            self.data['conn_kwargs'] = ret.copy()
+
+        if auth and isinstance(auth, dict):
+            if 'username' in auth:
+                ret['user'] = auth['username']
+            if 'password' in auth:
+                ret['password'] = auth['password']
+        return ret
+
+    @property
+    def api_url(self):
+        return self.data.get('api_url')
+
+    @property
+    def tags(self):
+        return self.data.get('tags', {})
+
+    @property
+    def nofailover(self):
+        return self.tags.get('nofailover', False)
+
+    @property
+    def replicatefrom(self):
+        return self.tags.get('replicatefrom')
+
+    @property
+    def clonefrom(self):
+        return self.tags.get('clonefrom', False) and bool(self.conn_url)
+
+    @property
+    def state(self):
+        return self.data.get('state', 'unknown')
+
+    @property
+    def is_running(self):
+        return self.state == 'running'
+
+
+class RemoteMember(Member):
+    """ Represents a remote master for a standby cluster
+    """
+    def __new__(cls, name, data):
+        return super(RemoteMember, cls).__new__(cls, None, name, None, data)
+
+    @staticmethod
+    def allowed_keys():
+        return ('primary_slot_name',
+                'create_replica_methods',
+                'restore_command',
+                'archive_cleanup_command',
+                'recovery_min_apply_delay',
+                'no_replication_slot')
+
+    def __getattr__(self, name):
+        if name in RemoteMember.allowed_keys():
+            return self.data.get(name)
+
+
+class Leader(namedtuple('Leader', 'index,session,member')):
+
+    """Immutable object (namedtuple) which represents leader key.
+    Consists of the following fields:
+    :param index: modification index of a leader key in a Configuration Store
+    :param session: either session id or just ttl in seconds
+    :param member: reference to a `Member` object which represents current leader (see `Cluster.members`)"""
+
+    @property
+    def name(self):
+        return self.member.name
+
+    def conn_kwargs(self, auth=None):
+        return self.member.conn_kwargs(auth)
+
+    @property
+    def conn_url(self):
+        return self.member.conn_url
+
+    @property
+    def timeline(self):
+        return self.member.data.get('timeline')
+
+
+class Failover(namedtuple('Failover', 'index,leader,candidate,scheduled_at')):
+
+    """
+    >>> 'Failover' in str(Failover.from_node(1, '{"leader": "cluster_leader"}'))
+    True
+    >>> 'Failover' in str(Failover.from_node(1, {"leader": "cluster_leader"}))
+    True
+    >>> 'Failover' in str(Failover.from_node(1, '{"leader": "cluster_leader", "member": "cluster_candidate"}'))
+    True
+    >>> Failover.from_node(1, 'null') is None
+    False
+    >>> n = '{"leader": "cluster_leader", "member": "cluster_candidate", "scheduled_at": "2016-01-14T10:09:57.1394Z"}'
+    >>> 'tzinfo=' in str(Failover.from_node(1, n))
+    True
+    >>> Failover.from_node(1, None) is None
+    False
+    >>> Failover.from_node(1, '{}') is None
+    False
+    >>> 'abc' in Failover.from_node(1, 'abc:def')
+    True
+    """
+    @staticmethod
+    def from_node(index, value):
+        if isinstance(value, dict):
+            data = value
+        elif value:
+            try:
+                data = json.loads(value)
+                if not isinstance(data, dict):
+                    data = {}
+            except ValueError:
+                t = [a.strip() for a in value.split(':')]
+                leader = t[0]
+                candidate = t[1] if len(t) > 1 else None
+                return Failover(index, leader, candidate, None) if leader or candidate else None
+        else:
+            data = {}
+
+        if data.get('scheduled_at'):
+            data['scheduled_at'] = dateutil.parser.parse(data['scheduled_at'])
+
+        return Failover(index, data.get('leader'), data.get('member'), data.get('scheduled_at'))
+
+    def __len__(self):
+        return int(bool(self.leader)) + int(bool(self.candidate))
+
+
+class ClusterConfig(namedtuple('ClusterConfig', 'index,data,modify_index')):
+
+    @staticmethod
+    def from_node(index, data, modify_index=None):
+        """
+        >>> ClusterConfig.from_node(1, '{') is None
+        False
+        """
+
+        try:
+            data = json.loads(data)
+        except (TypeError, ValueError):
+            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 {}
+
+
+class SyncState(namedtuple('SyncState', 'index,leader,sync_standby')):
+    """Immutable object (namedtuple) which represents last observed synhcronous replication state
+
+    :param index: modification index of a synchronization key in a Configuration Store
+    :param leader: reference to member that was leader
+    :param sync_standby: standby that was last synchronized to leader
+    """
+
+    @staticmethod
+    def from_node(index, value):
+        """
+        >>> SyncState.from_node(1, None).leader is None
+        True
+        >>> SyncState.from_node(1, '{}').leader is None
+        True
+        >>> SyncState.from_node(1, '{').leader is None
+        True
+        >>> SyncState.from_node(1, '[]').leader is None
+        True
+        >>> SyncState.from_node(1, '{"leader": "leader"}').leader == "leader"
+        True
+        >>> SyncState.from_node(1, {"leader": "leader"}).leader == "leader"
+        True
+        """
+        if isinstance(value, dict):
+            data = value
+        elif value:
+            try:
+                data = json.loads(value)
+                if not isinstance(data, dict):
+                    data = {}
+            except (TypeError, ValueError):
+                data = {}
+        else:
+            data = {}
+        return SyncState(index, data.get('leader'), data.get('sync_standby'))
+
+    def matches(self, name):
+        """
+        Returns if a node name matches one of the nodes in the sync state
+
+        >>> s = SyncState(1, 'foo', 'bar')
+        >>> s.matches('foo')
+        True
+        >>> s.matches('bar')
+        True
+        >>> s.matches('baz')
+        False
+        >>> s.matches(None)
+        False
+        >>> SyncState(1, None, None).matches('foo')
+        False
+        """
+        return name is not None and name in (self.leader, self.sync_standby)
+
+
+class TimelineHistory(namedtuple('TimelineHistory', 'index,value,lines')):
+    """Object representing timeline history file"""
+
+    @staticmethod
+    def from_node(index, value):
+        """
+        >>> h = TimelineHistory.from_node(1, 2)
+        >>> h.lines
+        []
+        """
+        try:
+            lines = json.loads(value)
+        except (TypeError, ValueError):
+            lines = None
+        if not isinstance(lines, list):
+            lines = []
+        return TimelineHistory(index, value, lines)
+
+
+class Cluster(namedtuple('Cluster', 'initialize,config,leader,last_leader_operation,members,failover,sync,history')):
+
+    """Immutable object (namedtuple) which represents PostgreSQL cluster.
+    Consists of the following fields:
+    :param initialize: shows whether this cluster has initialization key stored in DC or not.
+    :param config: global dynamic configuration, reference to `ClusterConfig` object
+    :param leader: `Leader` object which represents current leader of the cluster
+    :param last_leader_operation: int or long object containing position of last known leader operation.
+        This value is stored in `/optime/leader` key
+    :param members: list of Member object, all PostgreSQL cluster members including leader
+    :param failover: reference to `Failover` object
+    :param sync: reference to `SyncState` object, last observed synchronous replication state.
+    :param history: reference to `TimelineHistory` object
+    """
+
+    def is_unlocked(self):
+        return not (self.leader and self.leader.name)
+
+    def has_member(self, member_name):
+        return any(m for m in self.members if m.name == member_name)
+
+    def get_member(self, member_name, fallback_to_leader=True):
+        return ([m for m in self.members if m.name == member_name] or [self.leader if fallback_to_leader else None])[0]
+
+    def get_clone_member(self, exclude):
+        exclude = [exclude] + [self.leader.name] if self.leader else []
+        candidates = [m for m in self.members if m.clonefrom and m.is_running and m.name not in exclude]
+        return candidates[randint(0, len(candidates) - 1)] if candidates else self.leader
+
+    def check_mode(self, mode):
+        return bool(self.config and parse_bool(self.config.data.get(mode)))
+
+    def is_paused(self):
+        return self.check_mode('pause')
+
+    def is_synchronous_mode(self):
+        return self.check_mode('synchronous_mode')
+
+    def get_replication_slots(self, 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
+        if role in ('master', 'standby_leader'):
+            slot_members = [m.name for m in self.members if m.name != name and
+                            (m.replicatefrom is None or m.replicatefrom == 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 m.replicatefrom == 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
+
+            if name in slots:
+                logger.error("Permanent replication slot {'%s': %s} is conflicting with" +
+                             " physical replication slot for cluster member", name, value)
+                continue
+
+            value = deepcopy(value)
+            if not value:
+                value = {'type': 'physical'}
+
+            if isinstance(value, dict):
+                if 'type' not in value:
+                    value['type'] = 'logical' if value.get('database') and value.get('plugin') else 'physical'
+
+                if value['type'] == 'physical' or value['type'] == 'logical' \
+                        and value.get('database') and value.get('plugin'):
+                    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)
+class AbstractDCS(object):
+
+    _INITIALIZE = 'initialize'
+    _CONFIG = 'config'
+    _LEADER = 'leader'
+    _FAILOVER = 'failover'
+    _HISTORY = 'history'
+    _MEMBERS = 'members/'
+    _OPTIME = 'optime'
+    _LEADER_OPTIME = _OPTIME + '/' + _LEADER
+    _SYNC = 'sync'
+
+    def __init__(self, config):
+        """
+        :param config: dict, reference to config section of selected DCS.
+            i.e.: `zookeeper` for zookeeper, `etcd` for etcd, etc...
+        """
+        self._name = config['name']
+        self._base_path = re.sub('/+', '/', '/'.join(['', config.get('namespace', 'service'), config['scope']]))
+        self._set_loop_wait(config.get('loop_wait', 10))
+
+        self._ctl = bool(config.get('patronictl', False))
+        self._cluster = None
+        self._cluster_thread_lock = Lock()
+        self._last_leader_operation = ''
+        self.event = Event()
+
+    def client_path(self, path):
+        return '/'.join([self._base_path, path.lstrip('/')])
+
+    @property
+    def initialize_path(self):
+        return self.client_path(self._INITIALIZE)
+
+    @property
+    def config_path(self):
+        return self.client_path(self._CONFIG)
+
+    @property
+    def members_path(self):
+        return self.client_path(self._MEMBERS)
+
+    @property
+    def member_path(self):
+        return self.client_path(self._MEMBERS + self._name)
+
+    @property
+    def leader_path(self):
+        return self.client_path(self._LEADER)
+
+    @property
+    def failover_path(self):
+        return self.client_path(self._FAILOVER)
+
+    @property
+    def history_path(self):
+        return self.client_path(self._HISTORY)
+
+    @property
+    def leader_optime_path(self):
+        return self.client_path(self._LEADER_OPTIME)
+
+    @property
+    def sync_path(self):
+        return self.client_path(self._SYNC)
+
+    @abc.abstractmethod
+    def set_ttl(self, ttl):
+        """Set the new ttl value for leader key"""
+
+    @abc.abstractmethod
+    def set_retry_timeout(self, retry_timeout):
+        """Set the new value for retry_timeout"""
+
+    def _set_loop_wait(self, loop_wait):
+        self._loop_wait = loop_wait
+
+    def reload_config(self, config):
+        self._set_loop_wait(config['loop_wait'])
+        self.set_ttl(config['ttl'])
+        self.set_retry_timeout(config['retry_timeout'])
+
+    @property
+    def loop_wait(self):
+        return self._loop_wait
+
+    @abc.abstractmethod
+    def _load_cluster(self):
+        """Internally this method should build  `Cluster` object which
+           represents current state and topology of the cluster in DCS.
+           this method supposed to be called only by `get_cluster` method.
+
+           raise `~DCSError` in case of communication or other problems with DCS.
+           If the current node was running as a master and exception raised,
+           instance would be demoted."""
+
+    def get_cluster(self):
+        with self._cluster_thread_lock:
+            try:
+                self._load_cluster()
+            except Exception:
+                self._cluster = None
+                raise
+            return self._cluster
+
+    @property
+    def cluster(self):
+        with self._cluster_thread_lock:
+            return self._cluster
+
+    def reset_cluster(self):
+        with self._cluster_thread_lock:
+            self._cluster = None
+
+    @abc.abstractmethod
+    def _write_leader_optime(self, last_operation):
+        """write current xlog location into `/optime/leader` key in DCS
+        :param last_operation: absolute xlog location in bytes
+        :returns: `!True` on success."""
+
+    def write_leader_optime(self, last_operation):
+        if self._last_leader_operation != last_operation and self._write_leader_optime(last_operation):
+            self._last_leader_operation = last_operation
+
+    @abc.abstractmethod
+    def _update_leader(self):
+        """Update leader key (or session) ttl
+
+        :returns: `!True` if leader key (or session) has been updated successfully.
+            If not, `!False` must be returned and current instance would be demoted.
+
+        You have to use CAS (Compare And Swap) operation in order to update leader key,
+        for example for etcd `prevValue` parameter must be used."""
+
+    def update_leader(self, last_operation, access_is_restricted=False):
+        """Update leader key (or session) ttl and optime/leader
+
+        :param last_operation: absolute xlog location in bytes
+        :returns: `!True` if leader key (or session) has been updated successfully.
+            If not, `!False` must be returned and current instance would be demoted."""
+
+        ret = self._update_leader()
+        if ret and last_operation:
+            self.write_leader_optime(last_operation)
+        return ret
+
+    @abc.abstractmethod
+    def attempt_to_acquire_leader(self, permanent=False):
+        """Attempt to acquire leader lock
+        This method should create `/leader` key with value=`~self._name`
+        :param permanent: if set to `!True`, the leader key will never expire.
+         Used in patronictl for the external master
+        :returns: `!True` if key has been created successfully.
+
+        Key must be created atomically. In case if key already exists it should not be
+        overwritten and `!False` must be returned"""
+
+    @abc.abstractmethod
+    def set_failover_value(self, value, index=None):
+        """Create or update `/failover` key"""
+
+    def manual_failover(self, leader, candidate, scheduled_at=None, index=None):
+        failover_value = {}
+        if leader:
+            failover_value['leader'] = leader
+
+        if candidate:
+            failover_value['member'] = candidate
+
+        if scheduled_at:
+            failover_value['scheduled_at'] = scheduled_at.isoformat()
+        return self.set_failover_value(json.dumps(failover_value, separators=(',', ':')), index)
+
+    @abc.abstractmethod
+    def set_config_value(self, value, index=None):
+        """Create or update `/config` key"""
+
+    @abc.abstractmethod
+    def touch_member(self, data, ttl=None, 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.
+
+        :param data: information about instance (including connection strings)
+        :param ttl: ttl for member key, optional parameter. If it is None `~self.member_ttl will be used`
+        :param permanent: if set to `!True`, the member key will never expire.
+         Used in patronictl for the external master.
+        :returns: `!True` on success otherwise `!False`
+        """
+
+    @abc.abstractmethod
+    def take_leader(self):
+        """This method should create leader key with value = `~self._name` and ttl=`~self.ttl`
+        Since it could be called only on initial cluster bootstrap it could create this key regardless,
+        overwriting the key if necessary."""
+
+    @abc.abstractmethod
+    def initialize(self, create_new=True, sysid=""):
+        """Race for cluster initialization.
+
+        :param create_new: False if the key should already exist (in the case we are setting the system_id)
+        :param sysid: PostgreSQL cluster system identifier, if specified, is written to the key
+        :returns: `!True` if key has been created successfully.
+
+        this method should create atomically initialize key and return `!True`
+        otherwise it should return `!False`"""
+
+    @abc.abstractmethod
+    def delete_leader(self):
+        """Voluntarily remove leader key from DCS
+        This method should remove leader key if current instance is the leader"""
+
+    @abc.abstractmethod
+    def cancel_initialization(self):
+        """ Removes the initialize key for a cluster """
+
+    @abc.abstractmethod
+    def delete_cluster(self):
+        """Delete cluster from DCS"""
+
+    @staticmethod
+    def sync_state(leader, sync_standby):
+        """Build sync_state dict"""
+        return {'leader': leader, 'sync_standby': sync_standby}
+
+    def write_sync_state(self, leader, sync_standby, index=None):
+        sync_value = self.sync_state(leader, sync_standby)
+        return self.set_sync_state_value(json.dumps(sync_value, separators=(',', ':')), index)
+
+    @abc.abstractmethod
+    def set_history_value(self, value):
+        """"""
+
+    @abc.abstractmethod
+    def set_sync_state_value(self, value, index=None):
+        """"""
+
+    @abc.abstractmethod
+    def delete_sync_state(self, index=None):
+        """"""
+
+    def watch(self, leader_index, timeout):
+        """If the current node is a master it should just sleep.
+        Any other node should watch for changes of leader key with a given timeout
+
+        :param leader_index: index of a leader key
+        :param timeout: timeout in seconds
+        :returns: `!True` if you would like to reschedule the next run of ha cycle"""
+
+        self.event.wait(timeout)
+        return self.event.isSet()

patroni/dcs/consul.py

@@ -0,0 +1,520 @@
+from __future__ import absolute_import
+import json
+import logging
+import os
+import re
+import socket
+import ssl
+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
+
+logger = logging.getLogger(__name__)
+
+
+class ConsulError(DCSError):
+    pass
+
+
+class ConsulInternalError(ConsulException):
+    """An internal Consul server error occurred"""
+
+
+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)
+        kwargs = {}
+        if cert:
+            if isinstance(cert, tuple):
+                # Key and cert are separate
+                kwargs['cert_file'] = cert[0]
+                kwargs['key_file'] = cert[1]
+            else:
+                # combined certificate
+                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)
+        self._ttl = None
+
+    def set_read_timeout(self, timeout):
+        self._read_timeout = timeout/3.0
+
+    @property
+    def ttl(self):
+        return self._ttl
+
+    def set_ttl(self, ttl):
+        ret = self._ttl != ttl
+        self._ttl = ttl
+        return ret
+
+    @staticmethod
+    def response(response):
+        data = response.data.decode('utf-8')
+        if response.status == 500:
+            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)
+
+    def uri(self, path, params=None):
+        return '{0}{1}{2}'.format(self.base_uri, path, params and '?' + urlencode(params) or '')
+
+    def __getattr__(self, method):
+        if method not in ('get', 'post', 'put', 'delete'):
+            raise AttributeError("HTTPClient instance has no attribute '{0}'".format(method))
+
+        def wrapper(callback, path, params=None, data=''):
+            # python-consul doesn't allow to specify ttl smaller then 10 seconds
+            # because session_ttl_min defaults to 10s, so we have to do this ugly dirty hack...
+            if method == 'put' and path == '/v1/session/create':
+                ttl = '"ttl": "{0}s"'.format(self._ttl)
+                if not data or data == '{}':
+                    data = '{' + ttl + '}'
+                else:
+                    data = data[:-1] + ', ' + ttl + '}'
+            if isinstance(params, list):  # starting from v1.1.0 python-consul switched from `dict` to `list` for params
+                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
+            else:
+                kwargs['timeout'] = self._read_timeout
+            token = params.pop('token', self.token) if isinstance(params, dict) else self.token
+            if token:
+                kwargs['headers'] = {'X-Consul-Token': token}
+            return callback(self.response(self.http.request(method.upper(), self.uri(path, params), **kwargs)))
+        return wrapper
+
+
+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')
+        super(ConsulClient, self).__init__(*args, **kwargs)
+
+    def connect(self, *args, **kwargs):
+        kwargs.update(dict(zip(['host', 'port', 'scheme', 'verify'], args)))
+        if self._cert:
+            kwargs['cert'] = self._cert
+        if self._ca_cert:
+            kwargs['ca_cert'] = self._ca_cert
+        if self._token:
+            kwargs['token'] = self._token
+        return HTTPClient(**kwargs)
+
+
+def catch_consul_errors(func):
+    def wrapper(*args, **kwargs):
+        try:
+            return func(*args, **kwargs)
+        except (RetryFailedError, ConsulException, HTTPException, HTTPError, socket.error, socket.timeout):
+            return False
+    return wrapper
+
+
+def force_if_last_failed(func):
+    def wrapper(*args, **kwargs):
+        if wrapper.last_result is False:
+            kwargs['force'] = True
+        wrapper.last_result = func(*args, **kwargs)
+        return wrapper.last_result
+
+    wrapper.last_result = None
+    return wrapper
+
+
+def service_name_from_scope_name(scope_name):
+    """Translate scope name to service name which can be used in dns.
+
+    230 = 253 - len('replica.') - len('.service.consul')
+    """
+
+    def replace_char(match):
+        c = match.group(0)
+        return '-' if c in '. _' else "u{:04d}".format(ord(c))
+
+    service_name = re.sub(r'[^a-z0-9\-]', replace_char, scope_name.lower())
+    return service_name[0:230]
+
+
+class Consul(AbstractDCS):
+
+    def __init__(self, config):
+        super(Consul, self).__init__(config)
+        self._scope = config['scope']
+        self._session = None
+        self.__do_not_watch = False
+        self._retry = Retry(deadline=config['retry_timeout'], max_delay=1, max_tries=-1,
+                            retry_exceptions=(ConsulInternalError, HTTPException,
+                                              HTTPError, socket.error, socket.timeout))
+
+        kwargs = {}
+        if 'url' in config:
+            r = urlparse(config['url'])
+            config.update({'scheme': r.scheme, 'host': r.hostname, 'port': r.port or 8500})
+        elif 'host' in config:
+            host, port = split_host_port(config.get('host', '127.0.0.1:8500'), 8500)
+            config['host'] = host
+            if 'port' not in config:
+                config['port'] = int(port)
+
+        if config.get('cacert'):
+            config['ca_cert'] = config.pop('cacert')
+
+        if config.get('key') and config.get('cert'):
+            config['cert'] = (config['cert'], config['key'])
+
+        config_keys = ('host', 'port', 'token', 'scheme', 'cert', 'ca_cert', 'dc')
+        kwargs = {p: config.get(p) for p in config_keys if config.get(p)}
+
+        verify = config.get('verify')
+        if not isinstance(verify, bool):
+            verify = parse_bool(verify)
+        if isinstance(verify, bool):
+            kwargs['verify'] = verify
+
+        self._client = ConsulClient(**kwargs)
+        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._register_service = config.get('register_service', False)
+        if self._register_service:
+            self._service_name = service_name_from_scope_name(self._scope)
+            if self._scope != self._service_name:
+                logger.warning('Using %s as consul service name instead of scope name %s', self._service_name,
+                               self._scope)
+        self._service_check_interval = config.get('service_check_interval', '5s')
+        if not self._ctl:
+            self.create_session()
+
+    def retry(self, *args, **kwargs):
+        return self._retry.copy()(*args, **kwargs)
+
+    def create_session(self):
+        while not self._session:
+            try:
+                self.refresh_session()
+            except ConsulError:
+                logger.info('waiting on consul')
+                time.sleep(5)
+
+    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
+
+    def set_retry_timeout(self, retry_timeout):
+        self._retry.deadline = retry_timeout
+        self._client.http.set_read_timeout(retry_timeout)
+
+    def adjust_ttl(self):
+        try:
+            settings = self._client.agent.self()
+            min_ttl = (settings['Config']['SessionTTLMin'] or 10000000000)/1000000000.0
+            logger.warning('Changing Session TTL from %s to %s', self._client.http.ttl, min_ttl)
+            self._client.http.set_ttl(min_ttl)
+        except Exception:
+            logger.exception('adjust_ttl')
+
+    def _do_refresh_session(self):
+        """:returns: `!True` if it had to create new session"""
+        if self._session and self._last_session_refresh + self._loop_wait > time.time():
+            return False
+
+        if self._session:
+            try:
+                self._client.session.renew(self._session)
+            except NotFound:
+                self._session = None
+        ret = not self._session
+        if ret:
+            try:
+                self._session = self._client.session.create(name=self._scope + '-' + self._name,
+                                                            checks=self.__session_checks,
+                                                            lock_delay=0.001, behavior='delete')
+            except InvalidSessionTTL:
+                logger.exception('session.create')
+                self.adjust_ttl()
+                raise
+
+        self._last_session_refresh = time.time()
+        return ret
+
+    def refresh_session(self):
+        try:
+            return self.retry(self._do_refresh_session)
+        except (ConsulException, RetryFailedError):
+            logger.exception('refresh_session')
+        raise ConsulError('Failed to renew/create session')
+
+    def client_path(self, path):
+        return super(Consul, self).client_path(path)[1:]
+
+    @staticmethod
+    def member(node):
+        return Member.from_node(node['ModifyIndex'], os.path.basename(node['Key']), node.get('Session'), node['Value'])
+
+    def _load_cluster(self):
+        try:
+            path = self.client_path('/')
+            _, results = self.retry(self._client.kv.get, path, recurse=True)
+
+            if results is None:
+                raise NotFound
+
+            nodes = {}
+            for node in results:
+                node['Value'] = (node['Value'] or b'').decode('utf-8')
+                nodes[os.path.relpath(node['Key'], path).replace('\\', '/')] = node
+
+            # get initialize flag
+            initialize = nodes.get(self._INITIALIZE)
+            initialize = initialize and initialize['Value']
+
+            # get global dynamic configuration
+            config = nodes.get(self._CONFIG)
+            config = config and ClusterConfig.from_node(config['ModifyIndex'], config['Value'])
+
+            # get timeline history
+            history = nodes.get(self._HISTORY)
+            history = history and TimelineHistory.from_node(history['ModifyIndex'], history['Value'])
+
+            # get last leader operation
+            last_leader_operation = nodes.get(self._LEADER_OPTIME)
+            last_leader_operation = 0 if last_leader_operation is None else int(last_leader_operation['Value'])
+
+            # get list of members
+            members = [self.member(n) for k, n in nodes.items() if k.startswith(self._MEMBERS) and k.count('/') == 1]
+
+            # get leader
+            leader = nodes.get(self._LEADER)
+            if not self._ctl and leader and leader['Value'] == self._name \
+                    and self._session != leader.get('Session', 'x'):
+                logger.info('I am leader but not owner of the session. Removing leader node')
+                self._client.kv.delete(self.leader_path, cas=leader['ModifyIndex'])
+                leader = None
+
+            if leader:
+                member = Member(-1, leader['Value'], None, {})
+                member = ([m for m in members if m.name == leader['Value']] or [member])[0]
+                leader = Leader(leader['ModifyIndex'], leader.get('Session'), member)
+
+            # failover key
+            failover = nodes.get(self._FAILOVER)
+            if failover:
+                failover = Failover.from_node(failover['ModifyIndex'], failover['Value'])
+
+            # get synchronization state
+            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)
+        except NotFound:
+            self._cluster = Cluster(None, None, None, None, [], None, None, None)
+        except Exception:
+            logger.exception('get_cluster')
+            raise ConsulError('Consul is not responding properly')
+
+    @catch_consul_errors
+    def touch_member(self, data, ttl=None, 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()
+
+        if member and (create_member or member.session != self._session):
+            self._client.kv.delete(self.member_path)
+            create_member = True
+
+        if not create_member and member and deep_compare(data, member.data):
+            return True
+
+        try:
+            args = {} if permanent else {'acquire': self._session}
+            self._client.kv.put(self.member_path, json.dumps(data, separators=(',', ':')), **args)
+            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
+
+    @catch_consul_errors
+    def register_service(self, service_name, **kwargs):
+        logger.info('Register service %s, params %s', service_name, kwargs)
+        return self._client.agent.service.register(service_name, **kwargs)
+
+    @catch_consul_errors
+    def deregister_service(self, service_id):
+        logger.info('Deregister service %s', service_id)
+        # service_id can contain special characters, but is used as part of uri in deregister request
+        service_id = quote(service_id)
+        return self._client.agent.service.deregister(service_id)
+
+    def _update_service(self, data):
+        service_name = self._service_name
+        role = data['role'].replace('_', '-')
+        state = data['state']
+        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='{0}s'.format(self._client.http.ttl * 10))
+        params = {
+            'service_id': '{0}/{1}'.format(self._scope, self._name),
+            'address': conn_parts.hostname,
+            'port': conn_parts.port,
+            'check': check,
+            'tags': [role]
+        }
+
+        if state == 'stopped':
+            return self.deregister_service(params['service_id'])
+
+        if role in ['master', 'replica', 'standby-leader']:
+            if state != 'running':
+                return
+            return self.register_service(service_name, **params)
+
+        logger.warning('Could not register service: unknown role type %s', role)
+
+    @force_if_last_failed
+    def update_service(self, old_data, new_data, force=False):
+        update = False
+
+        for key in ['role', 'api_url', 'conn_url', 'state']:
+            if key not in new_data:
+                logger.warning('Could not register service: not enough params in member data')
+                return
+            if old_data.get(key) != new_data[key]:
+                update = True
+
+        if force or update:
+            return self._update_service(new_data)
+
+    @catch_consul_errors
+    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(permanent)
+        if not ret:
+            logger.info('Could not take out TTL lock')
+
+        return ret
+
+    def take_leader(self):
+        return self.attempt_to_acquire_leader()
+
+    @catch_consul_errors
+    def set_failover_value(self, value, index=None):
+        return self._client.kv.put(self.failover_path, value, cas=index)
+
+    @catch_consul_errors
+    def set_config_value(self, value, index=None):
+        return self._client.kv.put(self.config_path, value, cas=index)
+
+    @catch_consul_errors
+    def _write_leader_optime(self, last_operation):
+        return self._client.kv.put(self.leader_optime_path, last_operation)
+
+    @catch_consul_errors
+    def _update_leader(self):
+        if self._session:
+            self.retry(self._client.session.renew, self._session)
+            self._last_session_refresh = time.time()
+        return bool(self._session)
+
+    @catch_consul_errors
+    def initialize(self, create_new=True, sysid=''):
+        kwargs = {'cas': 0} if create_new else {}
+        return self.retry(self._client.kv.put, self.initialize_path, sysid, **kwargs)
+
+    @catch_consul_errors
+    def cancel_initialization(self):
+        return self.retry(self._client.kv.delete, self.initialize_path)
+
+    @catch_consul_errors
+    def delete_cluster(self):
+        return self.retry(self._client.kv.delete, self.client_path(''), recurse=True)
+
+    @catch_consul_errors
+    def set_history_value(self, value):
+        return self._client.kv.put(self.history_path, value)
+
+    @catch_consul_errors
+    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)
+
+    @catch_consul_errors
+    def set_sync_state_value(self, value, index=None):
+        return self.retry(self._client.kv.put, self.sync_path, value, cas=index)
+
+    @catch_consul_errors
+    def delete_sync_state(self, index=None):
+        return self.retry(self._client.kv.delete, self.sync_path, cas=index)
+
+    def watch(self, leader_index, timeout):
+        if self.__do_not_watch:
+            self.__do_not_watch = False
+            return True
+
+        if leader_index:
+            end_time = time.time() + timeout
+            while timeout >= 1:
+                try:
+                    idx, _ = self._client.kv.get(self.leader_path, index=leader_index, wait=str(timeout) + 's')
+                    return str(idx) != str(leader_index)
+                except (ConsulException, HTTPException, HTTPError, socket.error, socket.timeout):
+                    logger.exception('watch')
+
+                timeout = end_time - time.time()
+
+        try:
+            return super(Consul, self).watch(None, timeout)
+        finally:
+            self._last_session_refresh = 0
+            self.event.clear()

patroni/dcs/etcd.py

@@ -0,0 +1,592 @@
+from __future__ import absolute_import
+import etcd
+import json
+import logging
+import os
+import urllib3.util.connection
+import random
+import requests
+import six
+import socket
+import time
+
+from dns.exception import DNSException
+from dns import resolver
+from patroni.dcs import AbstractDCS, ClusterConfig, Cluster, Failover, Leader, Member, SyncState, TimelineHistory
+from 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 six.moves.queue import Queue
+from six.moves.http_client import HTTPException
+from six.moves.urllib_parse import urlparse
+from threading import Thread
+
+logger = logging.getLogger(__name__)
+
+
+def uri(protocol, host, port, endpoint=''):
+    return '{0}://{1}:{2}{3}'.format(protocol, host, port, endpoint)
+
+
+class EtcdError(DCSError):
+    pass
+
+
+class DnsCachingResolver(Thread):
+
+    def __init__(self, cache_time=600.0, cache_fail_time=30.0):
+        super(DnsCachingResolver, self).__init__()
+        self._cache = {}
+        self._cache_time = cache_time
+        self._cache_fail_time = cache_fail_time
+        self._resolve_queue = Queue()
+        self.daemon = True
+        self.start()
+
+    def run(self):
+        while True:
+            (host, port), attempt = self._resolve_queue.get()
+            response = self._do_resolve(host, port)
+            if response:
+                self._cache[(host, port)] = (time.time(), response)
+            else:
+                if attempt < 10:
+                    self.resolve_async(host, port, attempt + 1)
+                    time.sleep(1)
+
+    def resolve(self, host, port):
+        current_time = time.time()
+        cached_time, response = self._cache.get((host, port), (0, []))
+        time_passed = current_time - cached_time
+        if time_passed > self._cache_time or (not response and time_passed > self._cache_fail_time):
+            new_response = self._do_resolve(host, port)
+            if new_response:
+                self._cache[(host, port)] = (current_time, new_response)
+                response = new_response
+        return response
+
+    def resolve_async(self, host, port, attempt=0):
+        self._resolve_queue.put(((host, port), attempt))
+
+    @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)
+            return []
+
+
+class Client(etcd.Client):
+
+    def __init__(self, config, dns_resolver, cache_ttl=300):
+        self._dns_resolver = dns_resolver
+        self.set_machines_cache_ttl(cache_ttl)
+        self._machines_cache_updated = 0
+        args = {p: config.get(p) for p in ('host', 'port', 'protocol', 'use_proxies', 'username', 'password',
+                                           'cert', 'ca_cert') if config.get(p)}
+        super(Client, self).__init__(read_timeout=config['retry_timeout'], **args)
+        self._config = config
+        self._load_machines_cache()
+        self._allow_reconnect = not self._use_proxies
+
+    def _build_request_parameters(self):
+        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
+        return kwargs
+
+    def set_machines_cache_ttl(self, cache_ttl):
+        self._machines_cache_ttl = cache_ttl
+
+    @property
+    def machines(self):
+        """Original `machines` method(property) of `etcd.Client` class raise exception
+        when it failed to get list of etcd cluster members. This method is being called
+        only when request failed on one of the etcd members during `api_execute` call.
+        For us it's more important to execute original request rather then get new topology
+        of etcd cluster. So we will catch this exception and return empty list of machines.
+        Later, during next `api_execute` call we will forcefully update machines_cache.
+
+        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()
+
+        while True:
+            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(',')]
+                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
+            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 []
+
+    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 api_execute(self, path, method, params=None, timeout=None):
+        if not path.startswith('/'):
+            raise ValueError('Path does not start with /')
+
+        kwargs = {'fields': params, 'preload_content': False}
+
+        if method in [self._MGET, self._MDELETE]:
+            request_executor = self.http.request
+        elif method in [self._MPUT, self._MPOST]:
+            request_executor = self.http.request_encode_body
+            kwargs['encode_multipart'] = False
+        else:
+            raise etcd.EtcdException('HTTP method {0} not supported'.format(method))
+
+        # Update machines_cache if previous attempt of update has failed
+        if self._update_machines_cache:
+            self._load_machines_cache()
+        elif 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()
+
+        kwargs.update(self._build_request_parameters())
+
+        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:
+                raise
+        return self._handle_server_response(response)
+
+    @staticmethod
+    def get_srv_record(host):
+        try:
+            return [(r.target.to_text(True), r.port) for r in resolver.query(host, 'SRV')]
+        except DNSException:
+            return []
+
+    def _get_machines_cache_from_srv(self, srv):
+        """Fetch list of etcd-cluster member by resolving _etcd-server._tcp. SRV record.
+        This record should contain list of host and peer ports which could be used to run
+        'GET http://{host}:{port}/members' request (peer protocol)"""
+
+        ret = []
+        for r in ['-client-ssl', '-client', '-ssl', '', '-server-ssl', '-server']:
+            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)
+                if endpoint:
+                    try:
+                        response = requests.get(url, timeout=self.read_timeout, verify=False)
+                        if response.ok:
+                            for member in response.json():
+                                ret.extend(member['clientURLs'])
+                            break
+                    except RequestException:
+                        logger.exception('GET %s', url)
+                else:
+                    ret.append(url)
+            if ret:
+                self._protocol = protocol
+                break
+        else:
+            logger.warning('Can not resolve SRV for %s', srv)
+        return list(set(ret))
+
+    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))
+            if ret:
+                return list(set(ret))
+        return [uri(self.protocol, host, port)]
+
+    def _load_machines_cache(self):
+        """This method should fill up `_machines_cache` from scratch.
+        It could happen only in two cases:
+        1. During class initialization
+        2. When all etcd members failed"""
+
+        self._update_machines_cache = True
+
+        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'])
+
+        # Can not bootstrap list of etcd-cluster members, giving up
+        if not self._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
+
+        if self._base_uri in self._machines_cache:
+            self._machines_cache.remove(self._base_uri)
+
+        self._update_machines_cache = False
+        self._machines_cache_updated = time.time()
+
+
+class Etcd(AbstractDCS):
+
+    def __init__(self, config):
+        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))
+        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)
+
+    def _handle_exception(self, e, name='', do_sleep=False, raise_ex=None):
+        if not self._has_failed:
+            logger.exception(name)
+        else:
+            logger.error(e)
+            if do_sleep:
+                time.sleep(1)
+        self._has_failed = True
+        if isinstance(raise_ex, Exception):
+            raise raise_ex
+
+    def catch_etcd_errors(func):
+        def wrapper(self, *args, **kwargs):
+            try:
+                retval = func(self, *args, **kwargs) is not None
+                self._has_failed = False
+                return retval
+            except (RetryFailedError, etcd.EtcdException) as e:
+                self._handle_exception(e)
+                return False
+            except Exception as e:
+                self._handle_exception(e, raise_ex=EtcdError('unexpected error'))
+
+        return wrapper
+
+    @staticmethod
+    def get_etcd_client(config):
+        if 'proxy' in config:
+            config['use_proxies'] = True
+            config['url'] = config['proxy']
+
+        if 'url' in config:
+            r = urlparse(config['url'])
+            config.update({'protocol': r.scheme, 'host': r.hostname, 'port': r.port or 2379,
+                           'username': r.username, 'password': r.password})
+        elif 'hosts' in config:
+            hosts = config.pop('hosts')
+            default_port = config.pop('port', 2379)
+            protocol = config.get('protocol', 'http')
+
+            if isinstance(hosts, six.string_types):
+                hosts = hosts.split(',')
+
+            config['hosts'] = []
+            for value in hosts:
+                if isinstance(value, six.string_types):
+                    config['hosts'].append(uri(protocol, *split_host_port(value, default_port)))
+        elif 'host' in config:
+            host, port = split_host_port(config['host'], 2379)
+            config['host'] = host
+            if 'port' not in config:
+                config['port'] = int(port)
+
+        if config.get('cacert'):
+            config['ca_cert'] = config.pop('cacert')
+
+        if config.get('key') and config.get('cert'):
+            config['cert'] = (config['cert'], config['key'])
+
+        for p in ('discovery_srv', 'srv_domain'):
+            if p in config:
+                config['srv'] = config.pop(p)
+
+        dns_resolver = DnsCachingResolver()
+
+        def create_connection_patched(address, timeout=socket._GLOBAL_DEFAULT_TIMEOUT,
+                                      source_address=None, socket_options=None):
+            host, port = address
+            if host.startswith('['):
+                host = host.strip('[]')
+            err = None
+            for af, socktype, proto, _, sa in dns_resolver.resolve(host, port):
+                sock = None
+                try:
+                    sock = socket.socket(af, socktype, proto)
+                    if socket_options:
+                        for opt in socket_options:
+                            sock.setsockopt(*opt)
+                    if timeout is not socket._GLOBAL_DEFAULT_TIMEOUT:
+                        sock.settimeout(timeout)
+                    if source_address:
+                        sock.bind(source_address)
+                    sock.connect(sa)
+                    return sock
+
+                except socket.error as e:
+                    err = e
+                    if sock is not None:
+                        sock.close()
+                        sock = None
+
+            if err is not None:
+                raise err
+
+            raise socket.error("getaddrinfo returns an empty list")
+
+        urllib3.util.connection.create_connection = create_connection_patched
+
+        client = None
+        while not client:
+            try:
+                client = Client(config, dns_resolver)
+            except etcd.EtcdException:
+                logger.info('waiting on etcd')
+                time.sleep(5)
+        return client
+
+    def set_ttl(self, ttl):
+        ttl = int(ttl)
+        self.__do_not_watch = self._ttl != ttl
+        self._ttl = ttl
+        self._client.set_machines_cache_ttl(ttl*10)
+
+    def set_retry_timeout(self, retry_timeout):
+        self._retry.deadline = retry_timeout
+        self._client.set_read_timeout(retry_timeout)
+
+    @staticmethod
+    def member(node):
+        return Member.from_node(node.modifiedIndex, os.path.basename(node.key), node.ttl, node.value)
+
+    def _load_cluster(self):
+        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}
+
+            # get initialize flag
+            initialize = nodes.get(self._INITIALIZE)
+            initialize = initialize and initialize.value
+
+            # get global dynamic configuration
+            config = nodes.get(self._CONFIG)
+            config = config and ClusterConfig.from_node(config.modifiedIndex, config.value)
+
+            # get timeline history
+            history = nodes.get(self._HISTORY)
+            history = history and TimelineHistory.from_node(history.modifiedIndex, history.value)
+
+            # get last leader operation
+            last_leader_operation = nodes.get(self._LEADER_OPTIME)
+            last_leader_operation = 0 if last_leader_operation is None else int(last_leader_operation.value)
+
+            # get list of members
+            members = [self.member(n) for k, n in nodes.items() if k.startswith(self._MEMBERS) and k.count('/') == 1]
+
+            # get leader
+            leader = nodes.get(self._LEADER)
+            if leader:
+                member = Member(-1, leader.value, None, {})
+                member = ([m for m in members if m.name == leader.value] or [member])[0]
+                index = result.etcd_index if result.etcd_index > leader.modifiedIndex else leader.modifiedIndex + 1
+                leader = Leader(index, leader.ttl, member)
+
+            # failover key
+            failover = nodes.get(self._FAILOVER)
+            if failover:
+                failover = Failover.from_node(failover.modifiedIndex, failover.value)
+
+            # get synchronization state
+            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)
+        except etcd.EtcdKeyNotFound:
+            self._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
+
+    @catch_etcd_errors
+    def touch_member(self, data, ttl=None, permanent=False):
+        data = json.dumps(data, separators=(',', ':'))
+        return self._client.set(self.member_path, data, None if permanent else ttl or self._ttl)
+
+    @catch_etcd_errors
+    def take_leader(self):
+        return self.retry(self._client.set, self.leader_path, self._name, self._ttl)
+
+    def attempt_to_acquire_leader(self, permanent=False):
+        try:
+            return bool(self.retry(self._client.write,
+                                   self.leader_path,
+                                   self._name,
+                                   ttl=None if permanent else self._ttl,
+                                   prevExist=False))
+        except etcd.EtcdAlreadyExist:
+            logger.info('Could not take out TTL lock')
+        except (RetryFailedError, etcd.EtcdException):
+            pass
+        return False
+
+    @catch_etcd_errors
+    def set_failover_value(self, value, index=None):
+        return self._client.write(self.failover_path, value, prevIndex=index or 0)
+
+    @catch_etcd_errors
+    def set_config_value(self, value, index=None):
+        return self._client.write(self.config_path, value, prevIndex=index or 0)
+
+    @catch_etcd_errors
+    def _write_leader_optime(self, last_operation):
+        return self._client.set(self.leader_optime_path, last_operation)
+
+    @catch_etcd_errors
+    def _update_leader(self):
+        return self.retry(self._client.test_and_set, self.leader_path, self._name, self._name, 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):
+        return self._client.delete(self.leader_path, prevValue=self._name)
+
+    @catch_etcd_errors
+    def cancel_initialization(self):
+        return self.retry(self._client.delete, self.initialize_path)
+
+    @catch_etcd_errors
+    def delete_cluster(self):
+        return self.retry(self._client.delete, self.client_path(''), recursive=True)
+
+    @catch_etcd_errors
+    def set_history_value(self, value):
+        return self._client.write(self.history_path, value)
+
+    @catch_etcd_errors
+    def set_sync_state_value(self, value, index=None):
+        return self.retry(self._client.write, self.sync_path, value, prevIndex=index or 0)
+
+    @catch_etcd_errors
+    def delete_sync_state(self, index=None):
+        return self.retry(self._client.delete, self.sync_path, prevIndex=index or 0)
+
+    def watch(self, leader_index, timeout):
+        if self.__do_not_watch:
+            self.__do_not_watch = False
+            return True
+
+        if leader_index:
+            end_time = time.time() + timeout
+
+            while timeout >= 1:  # when timeout is too small urllib3 doesn't have enough time to connect
+                try:
+                    self._client.watch(self.leader_path, index=leader_index, timeout=timeout + 0.5)
+                    self._has_failed = False
+                    # Synchronous work of all cluster members with etcd is less expensive
+                    # than reestablishing http connection every time from every replica.
+                    return True
+                except etcd.EtcdWatchTimedOut:
+                    self._client.http.clear()
+                    self._has_failed = False
+                    return False
+                except (etcd.EtcdEventIndexCleared, etcd.EtcdWatcherCleared):  # Watch failed
+                    self._has_failed = False
+                    return True  # leave the loop, because watch with the same parameters will fail anyway
+                except etcd.EtcdException as e:
+                    self._handle_exception(e, 'watch', True)
+
+                timeout = end_time - time.time()
+
+        try:
+            return super(Etcd, self).watch(None, timeout)
+        finally:
+            self.event.clear()

patroni/dcs/exhibitor.py

@@ -0,0 +1,74 @@
+import logging
+import random
+import requests
+import time
+
+from patroni.dcs.zookeeper import ZooKeeper
+from requests.exceptions import RequestException
+
+logger = logging.getLogger(__name__)
+
+
+class ExhibitorEnsembleProvider(object):
+
+    TIMEOUT = 3.1
+
+    def __init__(self, hosts, port, uri_path='/exhibitor/v1/cluster/list', poll_interval=300):
+        self._exhibitor_port = port
+        self._uri_path = uri_path
+        self._poll_interval = poll_interval
+        self._exhibitors = hosts
+        self._master_exhibitors = hosts
+        self._zookeeper_hosts = ''
+        self._next_poll = None
+        while not self.poll():
+            logger.info('waiting on exhibitor')
+            time.sleep(5)
+
+    def poll(self):
+        if self._next_poll and self._next_poll > time.time():
+            return False
+
+        json = self._query_exhibitors(self._exhibitors)
+        if not json:
+            json = self._query_exhibitors(self._master_exhibitors)
+
+        if isinstance(json, dict) and 'servers' in json and 'port' in json:
+            self._next_poll = time.time() + self._poll_interval
+            zookeeper_hosts = ','.join([h + ':' + str(json['port']) for h in sorted(json['servers'])])
+            if self._zookeeper_hosts != zookeeper_hosts:
+                logger.info('ZooKeeper connection string has changed: %s => %s', self._zookeeper_hosts, zookeeper_hosts)
+                self._zookeeper_hosts = zookeeper_hosts
+                self._exhibitors = json['servers']
+                return True
+        return False
+
+    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
+        return None
+
+    @property
+    def zookeeper_hosts(self):
+        return self._zookeeper_hosts
+
+
+class Exhibitor(ZooKeeper):
+
+    def __init__(self, config):
+        interval = config.get('poll_interval', 300)
+        self._ensemble_provider = ExhibitorEnsembleProvider(config['hosts'], config['port'], poll_interval=interval)
+        config = config.copy()
+        config['hosts'] = self._ensemble_provider.zookeeper_hosts
+        super(Exhibitor, self).__init__(config)
+
+    def _load_cluster(self):
+        if self._ensemble_provider.poll():
+            self._client.set_hosts(self._ensemble_provider.zookeeper_hosts)
+        return super(Exhibitor, self)._load_cluster()

patroni/dcs/kubernetes.py

@@ -0,0 +1,415 @@
+from __future__ import absolute_import
+import datetime
+import functools
+import json
+import logging
+import socket
+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.exceptions import HTTPError
+from six.moves.http_client import HTTPException
+
+logger = logging.getLogger(__name__)
+
+
+class KubernetesError(DCSError):
+    pass
+
+
+class KubernetesRetriableException(k8s_client.rest.ApiException):
+
+    def __init__(self, orig):
+        super(KubernetesRetriableException, self).__init__(orig.status, orig.reason)
+        self.body = orig.body
+        self.headers = orig.headers
+
+
+class CoreV1ApiProxy(object):
+
+    def __init__(self, use_endpoints=False):
+        self._api = k8s_client.CoreV1Api()
+        self._request_timeout = None
+        self._use_endpoints = use_endpoints
+
+    def set_timeout(self, timeout):
+        self._request_timeout = (1, timeout / 3.0)
+
+    def __getattr__(self, func):
+        if func.endswith('_kind'):
+            func = func[:-4] + ('endpoints' if self._use_endpoints else 'config_map')
+
+        def wrapper(*args, **kwargs):
+            if '_request_timeout' not in kwargs:
+                kwargs['_request_timeout'] = self._request_timeout
+            try:
+                return getattr(self._api, func)(*args, **kwargs)
+            except k8s_client.rest.ApiException as e:
+                if e.status in (502, 503, 504):  # XXX
+                    raise KubernetesRetriableException(e)
+                raise
+        return wrapper
+
+
+def catch_kubernetes_errors(func):
+    def wrapper(*args, **kwargs):
+        try:
+            return func(*args, **kwargs)
+        except k8s_client.rest.ApiException as e:
+            if e.status == 403:
+                logger.exception('Permission denied')
+            elif e.status != 409:  # Object exists or conflict in resource_version
+                logger.exception('Unexpected error from Kubernetes API')
+            return False
+        except (RetryFailedError, HTTPException, HTTPError, socket.error, socket.timeout):
+            return False
+    return wrapper
+
+
+class Kubernetes(AbstractDCS):
+
+    def __init__(self, config):
+        self._labels = config['labels']
+        self._labels[config.get('scope_label', 'cluster-name')] = config['scope']
+        self._label_selector = ','.join('{0}={1}'.format(k, v) for k, v in self._labels.items())
+        self._namespace = config.get('namespace') or 'default'
+        self._role_label = config.get('role_label', 'role')
+        config['namespace'] = ''
+        super(Kubernetes, self).__init__(config)
+        self._retry = Retry(deadline=config['retry_timeout'], max_delay=1, max_tries=-1,
+                            retry_exceptions=(KubernetesRetriableException, HTTPException,
+                                              HTTPError, socket.error, socket.timeout))
+        self._ttl = None
+        try:
+            k8s_config.load_incluster_config()
+        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._leader_observed_record = {}
+        self._leader_observed_time = None
+        self._leader_resource_version = None
+        self._leader_observed_subsets = []
+        self.__do_not_watch = False
+
+    def retry(self, *args, **kwargs):
+        return self._retry.copy()(*args, **kwargs)
+
+    def client_path(self, path):
+        return super(Kubernetes, self).client_path(path)[1:].replace('/', '-')
+
+    @property
+    def leader_path(self):
+        return self._base_path[1:] if self.__subsets else super(Kubernetes, self).leader_path
+
+    def set_ttl(self, ttl):
+        ttl = int(ttl)
+        self.__do_not_watch = self._ttl != ttl
+        self._ttl = ttl
+
+    def set_retry_timeout(self, retry_timeout):
+        self._retry.deadline = retry_timeout
+        self._api.set_timeout(retry_timeout)
+
+    @staticmethod
+    def member(pod):
+        annotations = pod.metadata.annotations or {}
+        member = Member.from_node(pod.metadata.resource_version, pod.metadata.name, None, annotations.get('status', ''))
+        member.data['pod_labels'] = pod.metadata.labels
+        return member
+
+    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]
+
+            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}
+
+            config = nodes.get(self.config_path)
+            metadata = config and config.metadata
+            annotations = metadata and metadata.annotations or {}
+
+            # get initialize flag
+            initialize = annotations.get(self._INITIALIZE)
+
+            # get global dynamic configuration
+            config = ClusterConfig.from_node(metadata and metadata.resource_version,
+                                             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,
+                                                annotations.get(self._HISTORY) or '[]')
+
+            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
+            last_leader_operation = annotations.get(self._OPTIME)
+            last_leader_operation = 0 if last_leader_operation is None else int(last_leader_operation)
+
+            # get leader
+            leader_record = {n: annotations.get(n) for n in (self._LEADER, 'acquireTime',
+                             'ttl', 'renewTime', 'transitions') if n in annotations}
+            if (leader_record or self._leader_observed_record) and leader_record != self._leader_observed_record:
+                self._leader_observed_record = leader_record
+                self._leader_observed_time = time.time()
+
+            leader = leader_record.get(self._LEADER)
+            try:
+                ttl = int(leader_record.get('ttl')) or self._ttl
+            except (TypeError, ValueError):
+                ttl = self._ttl
+
+            if not metadata or not self._leader_observed_time or self._leader_observed_time + ttl < time.time():
+                leader = None
+
+            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)
+
+            # 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)
+
+            # 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)
+        except Exception:
+            logger.exception('get_cluster')
+            raise KubernetesError('Kubernetes API is not responding properly')
+
+    @staticmethod
+    def compare_ports(p1, p2):
+        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):
+        """
+        >>> Kubernetes.subsets_changed([], [])
+        False
+        >>> Kubernetes.subsets_changed([], [k8s_client.V1EndpointSubset()])
+        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)
+        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)
+        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)
+        True
+        >>> s2 = [k8s_client.V1EndpointSubset(addresses=a, ports=[p2, p1])]
+        >>> Kubernetes.subsets_changed(s1, s2)
+        False
+        """
+        if len(last_observed_subsets) != len(subsets):
+            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):
+            return True
+        if len(subsets[0].ports) == 1:
+            return not Kubernetes.compare_ports(last_observed_subsets[0].ports[0], subsets[0].ports[0])
+        observed_ports = {p.name: p for p in last_observed_subsets[0].ports}
+        for p in subsets[0].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):
+        metadata = {'namespace': self._namespace, 'name': name, 'labels': self._labels, 'annotations': annotations}
+        if patch or resource_version:
+            if resource_version is not None:
+                metadata['resource_version'] = resource_version
+            func = functools.partial(self._api.patch_namespaced_kind, name)
+        else:
+            func = functools.partial(self._api.create_namespaced_kind)
+            # skip annotations with null values
+            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:
+            endpoints = {'metadata': metadata}
+            if self.subsets_changed(self._leader_observed_subsets, subsets):
+                endpoints['subsets'] = subsets
+            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)
+
+    def _write_leader_optime(self, last_operation):
+        """Unused"""
+
+    def _update_leader(self):
+        """Unused"""
+
+    def update_leader(self, last_operation, access_is_restricted=False):
+        now = datetime.datetime.now(tzutc).isoformat()
+        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'}
+        if last_operation:
+            annotations[self._OPTIME] = last_operation
+
+        subsets = self.__subsets
+        if subsets is not None and access_is_restricted:
+            subsets = []
+
+        ret = self.patch_or_create(self.leader_path, annotations, self._leader_resource_version, subsets=subsets)
+        if ret:
+            self._leader_resource_version = ret.metadata.resource_version
+        return ret
+
+    def attempt_to_acquire_leader(self, permanent=False):
+        now = datetime.datetime.now(tzutc).isoformat()
+        annotations = {self._LEADER: self._name, 'ttl': str(sys.maxsize if permanent else self._ttl),
+                       'renewTime': now, 'acquireTime': now, 'transitions': '0'}
+        if self._leader_observed_record:
+            try:
+                transitions = int(self._leader_observed_record.get('transitions'))
+            except (TypeError, ValueError):
+                transitions = 0
+
+            if self._leader_observed_record.get(self._LEADER) != self._name:
+                transitions += 1
+            else:
+                annotations['acquireTime'] = self._leader_observed_record.get('acquireTime') or now
+            annotations['transitions'] = str(transitions)
+        subsets = [] if self.__subsets else None
+        ret = self.patch_or_create(self.leader_path, annotations, self._leader_resource_version, subsets=subsets)
+        if ret:
+            self._leader_resource_version = ret.metadata.resource_version
+        else:
+            logger.info('Could not take out TTL lock')
+        return ret
+
+    def take_leader(self):
+        return self.attempt_to_acquire_leader()
+
+    def set_failover_value(self, value, index=None):
+        """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}
+        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)
+
+    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)
+
+    @catch_kubernetes_errors
+    def touch_member(self, data, ttl=None, permanent=False):
+        cluster = self.cluster
+        if cluster and cluster.leader and cluster.leader.name == self._name:
+            role = 'master'
+        elif data['state'] == 'running' and data['role'] != 'master':
+            role = data['role']
+        else:
+            role = None
+
+        member = cluster and cluster.get_member(self._name, fallback_to_leader=False)
+        pod_labels = member and member.data.pop('pod_labels', None)
+        ret = pod_labels is not None and pod_labels.get(self._role_label) == role and deep_compare(data, member.data)
+
+        if not ret:
+            metadata = {'namespace': self._namespace, 'name': self._name, 'labels': {self._role_label: role},
+                        '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)
+        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)
+
+    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, [])
+            self.reset_cluster()
+
+    def cancel_initialization(self):
+        self.patch_or_create(self.config_path, {self._INITIALIZE: None}, self.cluster.config.index, 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)
+
+    def set_sync_state_value(self, value, index=None):
+        """Unused"""
+
+    def write_sync_state(self, leader, sync_standby, index=None):
+        return self.patch_or_create(self.sync_path, self.sync_state(leader, sync_standby), index, False)
+
+    def delete_sync_state(self, index=None):
+        return self.write_sync_state(None, None, index)
+
+    def watch(self, leader_index, timeout):
+        if self.__do_not_watch:
+            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)
+        finally:
+            self.event.clear()

patroni/dcs/zookeeper.py

@@ -0,0 +1,342 @@
+import json
+import logging
+import select
+import time
+
+from kazoo.client import KazooClient, KazooState, KazooRetry
+from kazoo.exceptions import NoNodeError, NodeExistsError
+from kazoo.handlers.threading import SequentialThreadingHandler
+from patroni.dcs import AbstractDCS, ClusterConfig, Cluster, Failover, Leader, Member, SyncState, TimelineHistory
+from patroni.exceptions import DCSError
+from patroni.utils import deep_compare
+
+logger = logging.getLogger(__name__)
+
+
+class ZooKeeperError(DCSError):
+    pass
+
+
+class PatroniSequentialThreadingHandler(SequentialThreadingHandler):
+
+    def __init__(self, connect_timeout):
+        super(PatroniSequentialThreadingHandler, self).__init__()
+        self.set_connect_timeout(connect_timeout)
+
+    def set_connect_timeout(self, connect_timeout):
+        self._connect_timeout = max(1.0, connect_timeout/2.0)  # try to connect to zookeeper node during loop_wait/2
+
+    def create_connection(self, *args, **kwargs):
+        """This method is trying to establish connection with one of the zookeeper nodes.
+           Somehow strategy "fail earlier and retry more often" works way better comparing to
+           the original strategy "try to connect with specified timeout".
+           Since we want to try connect to zookeeper more often (with the smaller connect_timeout),
+           he have to override `create_connection` method in the `SequentialThreadingHandler`
+           class (which is used by `kazoo.Client`).
+
+        :param args: always contains `tuple(host, port)` as the first element and could contain
+                     `connect_timeout` (negotiated session timeout) as the second element."""
+
+        args = list(args)
+        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):
+
+    def __init__(self, config):
+        super(ZooKeeper, self).__init__(config)
+
+        hosts = config.get('hosts', [])
+        if isinstance(hosts, list):
+            hosts = ','.join(hosts)
+
+        self._client = KazooClient(hosts, handler=PatroniSequentialThreadingHandler(config['retry_timeout']),
+                                   timeout=config['ttl'], connection_retry=KazooRetry(max_delay=1, max_tries=-1,
+                                   sleep_func=time.sleep), command_retry=KazooRetry(deadline=config['retry_timeout'],
+                                   max_delay=1, max_tries=-1, sleep_func=time.sleep))
+        self._client.add_listener(self.session_listener)
+
+        self._fetch_cluster = True
+
+        self._orig_kazoo_connect = self._client._connection._connect
+        self._client._connection._connect = self._kazoo_connect
+
+        self._client.start()
+
+    def _kazoo_connect(self, host, port):
+        """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
+        up to 2/3 from session timeout (ttl) to figure out that connection was dead and we had
+        only small time for reconnect and retry.
+
+        This method is needed to return different value of read_timeout, which is not calculated
+        from negotiated session timeout but from value of `loop_wait`. And it is 2 sec smaller
+        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)
+        return max(self.loop_wait - 2, 2)*1000, ret[1]
+
+    def session_listener(self, state):
+        if state in [KazooState.SUSPENDED, KazooState.LOST]:
+            self.cluster_watcher(None)
+
+    def cluster_watcher(self, event):
+        self._fetch_cluster = True
+        self.event.set()
+
+    def reload_config(self, config):
+        self.set_retry_timeout(config['retry_timeout'])
+
+        loop_wait = config['loop_wait']
+
+        loop_wait_changed = self._loop_wait != loop_wait
+        self._loop_wait = loop_wait
+        self._client.handler.set_connect_timeout(loop_wait)
+
+        # We need to reestablish connection to zookeeper if we want to change
+        # read_timeout (and Ping interval respectively), because read_timeout
+        # is calculated in `_kazoo_connect` method. If we are changing ttl at
+        # the same time, set_ttl method will reestablish connection and return
+        # `!True`, otherwise we will close existing connection and let kazoo
+        # open the new one.
+        if not self.set_ttl(int(config['ttl'] * 1000)) and loop_wait_changed:
+            self._client._connection._socket.close()
+
+    def set_ttl(self, ttl):
+        """It is not possible to change ttl (session_timeout) in zookeeper without
+        destroying old session and creating the new one. This method returns `!True`
+        if session_timeout has been changed (`restart()` has been called)."""
+        if self._client._session_timeout != ttl:
+            self._client._session_timeout = ttl
+            self._client.restart()
+            return True
+
+    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
+
+    def get_node(self, key, watch=None):
+        try:
+            ret = self._client.get(key, watch)
+            return (ret[0].decode('utf-8'), ret[1])
+        except NoNodeError:
+            return None
+
+    @staticmethod
+    def member(name, value, znode):
+        return Member.from_node(znode.version, name, znode.ephemeralOwner, value)
+
+    def get_children(self, key, watch=None):
+        try:
+            return self._client.get_children(key, watch)
+        except NoNodeError:
+            return []
+
+    def load_members(self, sync_standby):
+        members = []
+        for member in self.get_children(self.members_path, self.cluster_watcher):
+            watch = member == sync_standby and self.cluster_watcher or None
+            data = self.get_node(self.members_path + member, watch)
+            if data is not None:
+                members.append(self.member(member, *data))
+        return members
+
+    def _inner_load_cluster(self):
+        self._fetch_cluster = False
+        self.event.clear()
+        nodes = set(self.get_children(self.client_path(''), self.cluster_watcher))
+        if not nodes:
+            self._fetch_cluster = True
+
+        # get initialize flag
+        initialize = (self.get_node(self.initialize_path) or [None])[0] if self._INITIALIZE in nodes else None
+
+        # get global dynamic configuration
+        config = self.get_node(self.config_path, watch=self.cluster_watcher) if self._CONFIG in nodes else None
+        config = config and ClusterConfig.from_node(config[1].version, config[0], config[1].mzxid)
+
+        # get timeline history
+        history = self.get_node(self.history_path, watch=self.cluster_watcher) if self._HISTORY in nodes else None
+        history = history and TimelineHistory.from_node(history[1].mzxid, history[0])
+
+        # get last leader operation
+        last_leader_operation = self._OPTIME in nodes and self._fetch_cluster and self.get_node(self.leader_optime_path)
+        last_leader_operation = last_leader_operation and int(last_leader_operation[0]) or 0
+
+        # get synchronization state
+        sync = self.get_node(self.sync_path, watch=self.cluster_watcher) if self._SYNC in nodes else None
+        sync = SyncState.from_node(sync and sync[1].version, sync and sync[0])
+
+        # get list of members
+        sync_standby = sync.leader == self._name and sync.sync_standby or None
+        members = self.load_members(sync_standby) if self._MEMBERS[:-1] in nodes else []
+
+        # get leader
+        leader = self.get_node(self.leader_path) if self._LEADER in nodes else None
+        if leader:
+            client_id = self._client.client_id
+            if not self._ctl and leader[0] == self._name and client_id is not None \
+                    and client_id[0] != leader[1].ephemeralOwner:
+                logger.info('I am leader but not owner of the session. Removing leader node')
+                self._client.delete(self.leader_path)
+                leader = None
+
+            if leader:
+                member = Member(-1, leader[0], None, {})
+                member = ([m for m in members if m.name == leader[0]] or [member])[0]
+                leader = Leader(leader[1].version, leader[1].ephemeralOwner, member)
+                self._fetch_cluster = member.index == -1
+
+        # failover key
+        failover = self.get_node(self.failover_path, watch=self.cluster_watcher) if self._FAILOVER in nodes else None
+        failover = failover and Failover.from_node(failover[1].version, failover[0])
+
+        self._cluster = Cluster(initialize, config, leader, last_leader_operation, members, failover, sync, history)
+
+    def _load_cluster(self):
+        if self._fetch_cluster or self._cluster is None:
+            try:
+                self._client.retry(self._inner_load_cluster)
+            except Exception:
+                logger.exception('get_cluster')
+                self.cluster_watcher(None)
+                raise ZooKeeperError('ZooKeeper in not responding properly')
+
+    def _create(self, path, value, retry=False, ephemeral=False):
+        try:
+            if retry:
+                self._client.retry(self._client.create, path, value, makepath=True, ephemeral=ephemeral)
+            else:
+                self._client.create_async(path, value, makepath=True, ephemeral=ephemeral).get(timeout=1)
+            return True
+        except Exception:
+            logger.exception('Failed to create %s', path)
+        return False
+
+    def attempt_to_acquire_leader(self, permanent=False):
+        ret = self._create(self.leader_path, self._name.encode('utf-8'), retry=True, ephemeral=not permanent)
+        if not ret:
+            logger.info('Could not take out TTL lock')
+        return ret
+
+    def _set_or_create(self, key, value, index=None, retry=False, do_not_create_empty=False):
+        value = value.encode('utf-8')
+        try:
+            if retry:
+                self._client.retry(self._client.set, key, value, version=index or -1)
+            else:
+                self._client.set_async(key, value, version=index or -1).get(timeout=1)
+            return True
+        except NoNodeError:
+            if do_not_create_empty and not value:
+                return True
+            elif index is None:
+                return self._create(key, value, retry)
+            else:
+                return False
+        except Exception:
+            logger.exception('Failed to update %s', key)
+        return False
+
+    def set_failover_value(self, value, index=None):
+        return self._set_or_create(self.failover_path, value, index)
+
+    def set_config_value(self, value, index=None):
+        return self._set_or_create(self.config_path, value, index, retry=True)
+
+    def initialize(self, create_new=True, sysid=""):
+        sysid = sysid.encode('utf-8')
+        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):
+        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]:
+            try:
+                self._client.delete_async(self.member_path).get(timeout=1)
+            except NoNodeError:
+                pass
+            except Exception:
+                return False
+            member = None
+
+        if member:
+            if deep_compare(data, member.data):
+                return True
+        else:
+            try:
+                self._client.create_async(self.member_path, encoded_data, makepath=True,
+                                          ephemeral=not permanent).get(timeout=1)
+                return True
+            except Exception as e:
+                if not isinstance(e, NodeExistsError):
+                    logger.exception('touch_member')
+                    return False
+        try:
+            self._client.set_async(self.member_path, encoded_data).get(timeout=1)
+            return True
+        except Exception:
+            logger.exception('touch_member')
+
+        return False
+
+    def take_leader(self):
+        return self.attempt_to_acquire_leader()
+
+    def _write_leader_optime(self, last_operation):
+        return self._set_or_create(self.leader_optime_path, last_operation)
+
+    def _update_leader(self):
+        return True
+
+    def delete_leader(self):
+        self._client.restart()
+        return True
+
+    def _cancel_initialization(self):
+        node = self.get_node(self.initialize_path)
+        if node:
+            self._client.delete(self.initialize_path, version=node[1].version)
+
+    def cancel_initialization(self):
+        try:
+            self._client.retry(self._cancel_initialization)
+        except Exception:
+            logger.exception("Unable to delete initialize key")
+
+    def delete_cluster(self):
+        try:
+            return self._client.retry(self._client.delete, self.client_path(''), recursive=True)
+        except NoNodeError:
+            return True
+
+    def set_history_value(self, value):
+        return self._set_or_create(self.history_path, value)
+
+    def set_sync_state_value(self, value, index=None):
+        return self._set_or_create(self.sync_path, value, index, retry=True, do_not_create_empty=True)
+
+    def delete_sync_state(self, index=None):
+        return self.set_sync_state_value("{}", index)
+
+    def watch(self, leader_index, timeout):
+        if super(ZooKeeper, self).watch(leader_index, timeout):
+            self._fetch_cluster = True
+        return self._fetch_cluster

patroni/etcd.py

@@ -1,282 +0,0 @@
-from __future__ import absolute_import
-import etcd
-import logging
-import os
-import random
-import requests
-import socket
-import time
-import urllib3
-
-from dns.exception import DNSException
-from dns import resolver
-from patroni.dcs import AbstractDCS, Cluster, Failover, Leader, Member
-from patroni.exceptions import DCSError
-from patroni.utils import Retry, RetryFailedError, sleep
-from requests.exceptions import RequestException
-
-logger = logging.getLogger(__name__)
-
-
-class EtcdError(DCSError):
-    pass
-
-
-class Client(etcd.Client):
-
-    def __init__(self, config):
-        super(Client, self).__init__(read_timeout=5)
-        self._config = config
-        self._load_machines_cache()
-        self._allow_reconnect = True
-
-    @property
-    def machines(self):
-        """Original `machines` method(property) of `etcd.Client` class raise exception
-        when it failed to get list of etcd cluster members. This method is being called
-        only when request failed on one of the etcd members during `api_execute` call.
-        For us it's more important to execute original request rather then get new
-        topology of etcd cluster. So we will catch this exception and return valid list
-        of machines with setting flag `self._update_machines_cache` to `!True`.
-        Later, during next `api_execute` call we will forcefully update machines_cache"""
-        try:
-            ret = super(Client, self).machines
-            random.shuffle(ret)
-            return ret
-        except etcd.EtcdException:
-            if self._update_machines_cache:  # We are updating machines_cache
-                raise  # This exception is fatal, we should re-raise it.
-            self._update_machines_cache = True
-            return [self._base_uri]
-
-    def api_execute(self, path, method, **kwargs):
-        # Update machines_cache if previous attempt of update has failed
-        self._update_machines_cache and self._load_machines_cache()
-        try:
-            return super(Client, self).api_execute(path, method, **kwargs)
-        except etcd.EtcdConnectionFailed:
-            self._update_machines_cache = True
-            raise
-
-    @staticmethod
-    def get_srv_record(host):
-        try:
-            return [(str(r.target).rstrip('.'), r.port) for r in resolver.query('_etcd-server._tcp.' + host, 'SRV')]
-        except DNSException:
-            logger.exception('Can not resolve SRV for %s', host)
-        return []
-
-    # try to workarond bug in python-etcd: https://github.com/jplana/python-etcd/issues/81
-    def _result_from_response(self, response):
-        try:
-            response.data.decode('utf-8')
-        except urllib3.exceptions.TimeoutError:
-            raise
-        except Exception as e:
-            raise etcd.EtcdException('Unable to decode server response: %s' % e)
-        return super(Client, self)._result_from_response(response)
-
-    def _get_machines_cache_from_srv(self, discovery_srv):
-        """Fetch list of etcd-cluster member by resolving _etcd-server._tcp. SRV record.
-        This record should contain list of host and peer ports which could be used to run
-        'GET http://{host}:{port}/members' request (peer protocol)"""
-
-        ret = []
-        for host, port in self.get_srv_record(discovery_srv):
-            url = '{}://{}:{}/members'.format(self._protocol, host, port)
-            try:
-                response = requests.get(url, timeout=5)
-                if response.ok:
-                    for member in response.json():
-                        ret.extend(member['clientURLs'])
-                    break
-            except RequestException:
-                logger.exception('GET %s', url)
-        return list(set(ret))
-
-    def _get_machines_cache_from_dns(self, addr):
-        """One host might be resolved into multiple ip addresses. We will make list out of it"""
-
-        ret = []
-        host, port = addr.split(':')
-        try:
-            for r in set(socket.getaddrinfo(host, port, socket.AF_INET, socket.SOCK_STREAM, socket.IPPROTO_TCP)):
-                ret.append('{}://{}:{}'.format(self._protocol, r[4][0], r[4][1]))
-        except socket.error:
-            logger.exception('Can not resolve %s', host)
-        return list(set(ret)) if ret else ['{}://{}:{}'.format(self._protocol, host, port)]
-
-    def _load_machines_cache(self):
-        """This method should fill up `_machines_cache` from scratch.
-        It could happen only in two cases:
-        1. During class initialization
-        2. When all etcd members failed"""
-
-        self._update_machines_cache = True
-
-        if 'discovery_srv' not in self._config and 'host' not in self._config:
-            raise Exception('Neither discovery_srv nor host are defined in etcd section of config')
-
-        self._machines_cache = []
-
-        if 'discovery_srv' in self._config:
-            self._machines_cache = self._get_machines_cache_from_srv(self._config['discovery_srv'])
-
-        if not self._machines_cache and 'host' in self._config:
-            self._machines_cache = self._get_machines_cache_from_dns(self._config['host'])
-
-        # Can not bootstrap list of etcd-cluster members, giving up
-        if not self._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._machines_cache.pop(0)
-        self._machines_cache = self.machines
-        self._base_uri in self._machines_cache and self._machines_cache.remove(self._base_uri)
-
-        self._update_machines_cache = False
-
-
-def catch_etcd_errors(func):
-    def wrapper(*args, **kwargs):
-        try:
-            return not func(*args, **kwargs) is None
-        except (RetryFailedError, etcd.EtcdException):
-            return False
-        except:
-            logger.exception("")
-            raise EtcdError("unexpected error")
-
-    return wrapper
-
-
-class Etcd(AbstractDCS):
-
-    def __init__(self, name, config):
-        super(Etcd, self).__init__(name, config)
-        self.ttl = config.get('ttl', 30)
-        self._retry = Retry(deadline=10, max_delay=1, max_tries=-1,
-                            retry_exceptions=(etcd.EtcdConnectionFailed,
-                                              etcd.EtcdLeaderElectionInProgress,
-                                              etcd.EtcdWatcherCleared,
-                                              etcd.EtcdEventIndexCleared))
-        self.client = self.get_etcd_client(config)
-
-    def retry(self, *args, **kwargs):
-        return self._retry.copy()(*args, **kwargs)
-
-    def get_etcd_client(self, config):
-        client = None
-        while not client:
-            try:
-                client = Client(config)
-            except etcd.EtcdException:
-                logger.info('waiting on etcd')
-                sleep(5)
-        return client
-
-    @staticmethod
-    def member(node):
-        return Member.from_node(node.modifiedIndex, os.path.basename(node.key), node.ttl, node.value)
-
-    def _load_cluster(self):
-        try:
-            result = self.retry(self.client.read, self.client_path(''), recursive=True)
-            nodes = {os.path.relpath(node.key, result.key): node for node in result.leaves}
-
-            # get initialize flag
-            initialize = nodes.get(self._INITIALIZE, None)
-            initialize = initialize and initialize.value
-
-            # get last leader operation
-            last_leader_operation = nodes.get(self._LEADER_OPTIME, None)
-            last_leader_operation = 0 if last_leader_operation is None else int(last_leader_operation.value)
-
-            # get list of members
-            members = [self.member(n) for k, n in nodes.items() if k.startswith(self._MEMBERS) and k.count('/') == 1]
-
-            # get leader
-            leader = nodes.get(self._LEADER, None)
-            if leader:
-                member = Member(-1, leader.value, None, {})
-                member = ([m for m in members if m.name == leader.value] or [member])[0]
-                leader = Leader(leader.modifiedIndex, leader.ttl, member)
-
-            # failover key
-            failover = nodes.get(self._FAILOVER, None)
-            if failover:
-                failover = Failover.from_node(failover.modifiedIndex, failover.value)
-
-            self._cluster = Cluster(initialize, leader, last_leader_operation, members, failover)
-        except etcd.EtcdKeyNotFound:
-            self._cluster = Cluster(False, None, None, [], None)
-        except:
-            logger.exception('get_cluster')
-            raise EtcdError('Etcd is not responding properly')
-
-    @catch_etcd_errors
-    def touch_member(self, connection_string, ttl=None):
-        return self.retry(self.client.set, self.member_path, connection_string, ttl or self.ttl)
-
-    @catch_etcd_errors
-    def take_leader(self):
-        return self.retry(self.client.set, self.leader_path, self._name, self.ttl)
-
-    def attempt_to_acquire_leader(self):
-        try:
-            return bool(self.retry(self.client.write, self.leader_path, self._name, ttl=self.ttl, prevExist=False))
-        except etcd.EtcdAlreadyExist:
-            logger.info('Could not take out TTL lock')
-        except (RetryFailedError, etcd.EtcdException):
-            pass
-        return False
-
-    @catch_etcd_errors
-    def set_failover_value(self, value, index=None):
-        return self.client.write(self.failover_path, value, prevIndex=index or 0)
-
-    @catch_etcd_errors
-    def write_leader_optime(self, last_operation):
-        return self.client.set(self.leader_optime_path, last_operation)
-
-    @catch_etcd_errors
-    def update_leader(self):
-        return self.retry(self.client.test_and_set, self.leader_path, self._name, self._name, 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):
-        return self.client.delete(self.leader_path, prevValue=self._name)
-
-    @catch_etcd_errors
-    def cancel_initialization(self):
-        return self.retry(self.client.delete, self.initialize_path)
-
-    def watch(self, timeout):
-        cluster = self.cluster
-        # watch on leader key changes if it is defined and current node is not lock owner
-        if cluster and cluster.leader and cluster.leader.name != self._name:
-            end_time = time.time() + timeout
-            index = cluster.leader.index
-
-            while index and timeout >= 1:  # when timeout is too small urllib3 doesn't have enough time to connect
-                try:
-                    self.client.watch(self.leader_path, index=index + 1, timeout=timeout + 0.5)
-                    # Synchronous work of all cluster members with etcd is less expensive
-                    # than reestablishing http connection every time from every replica.
-                    return True
-                except urllib3.exceptions.TimeoutError:
-                    self.client.http.clear()
-                    return False
-                except etcd.EtcdException:
-                    logging.exception('watch')
-
-                timeout = end_time - time.time()
-
-        try:
-            return super(Etcd, self).watch(timeout)
-        finally:
-            self.event.clear()

patroni/exceptions.py

@@ -13,10 +13,6 @@ class PatroniException(Exception):
         return repr(self.value)
 
 
-class PatroniCtlException(Exception):
-    pass
-
-
 class PostgresException(PatroniException):
     pass
 
@@ -27,3 +23,7 @@ class DCSError(PatroniException):
 
 class PostgresConnectionException(PostgresException):
     pass
+
+
+class WatchdogError(PatroniException):
+    pass

patroni/log.py

@@ -0,0 +1,66 @@
+import logging
+import os
+
+from copy import deepcopy
+from logging.handlers import RotatingFileHandler
+from patroni.utils import deep_compare
+
+
+class PatroniLogger(object):
+
+    DEFAULT_LEVEL = 'INFO'
+    DEFAULT_FORMAT = '%(asctime)s %(levelname)s: %(message)s'
+
+    def __init__(self):
+        self.root_logger = logging.getLogger()
+        self.config = None
+        self.handler = None
+        self.reload_config({'level': 'DEBUG'})
+
+    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):
+            self.root_logger.setLevel(config.get('level', PatroniLogger.DEFAULT_LEVEL))
+
+            add_handler = None
+            if 'dir' in config:
+                if not isinstance(self.handler, RotatingFileHandler):
+                    add_handler = RotatingFileHandler(os.path.join(config['dir'], __name__))
+                handler = add_handler or self.handler
+                handler.maxBytes = int(config.get('file_size', 25000000))
+                handler.backupCount = int(config.get('file_num', 4))
+            else:
+                if self.handler is None or isinstance(self.handler, RotatingFileHandler):
+                    add_handler = logging.StreamHandler()
+                handler = add_handler or self.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 add_handler:
+                handler.setFormatter(logging.Formatter(logformat, dateformat))
+
+            if add_handler:
+                self.root_logger.addHandler(add_handler)
+
+                if self.handler is not None:
+                    self.root_logger.removeHandler(self.handler)
+                    self.handler.close()
+
+                self.handler = add_handler
+
+            self.config = config.copy()
+            self.update_loggers()

patroni/postmaster.py

@@ -0,0 +1,171 @@
+import logging
+import os
+import psutil
+import re
+import signal
+import subprocess
+
+from patroni import call_self
+
+logger = logging.getLogger(__name__)
+
+STOP_SIGNALS = {
+    'smart': signal.SIGTERM,
+    'fast': signal.SIGINT,
+    'immediate': signal.SIGQUIT if os.name != 'nt' else signal.SIGABRT,
+}
+
+
+class PostmasterProcess(psutil.Process):
+
+    def __init__(self, pid):
+        self.is_single_user = False
+        if pid < 0:
+            pid = -pid
+            self.is_single_user = True
+        super(PostmasterProcess, self).__init__(pid)
+
+    @staticmethod
+    def _read_postmaster_pidfile(data_dir):
+        """Reads and parses postmaster.pid from the data directory
+
+        :returns dictionary of values if successful, empty dictionary otherwise
+        """
+        pid_line_names = ['pid', 'data_dir', 'start_time', 'port', 'socket_dir', 'listen_addr', 'shmem_key']
+        try:
+            with open(os.path.join(data_dir, 'postmaster.pid')) as f:
+                return {name: line.rstrip('\n') for name, line in zip(pid_line_names, f)}
+        except IOError:
+            return {}
+
+    def _is_postmaster_process(self):
+        try:
+            start_time = int(self._postmaster_pid.get('start_time', 0))
+            if start_time and abs(self.create_time() - start_time) > 3:
+                logger.info('Process %s is not postmaster, too much difference between PID file start time %s and '
+                            'process start time %s', self.pid, self.create_time(), start_time)
+                return False
+        except ValueError:
+            logger.warning('Garbage start time value in pid file: %r', self._postmaster_pid.get('start_time'))
+
+        # Extra safety check. The process can't be ourselves, our parent or our direct child.
+        if self.pid == os.getpid() or self.pid == os.getppid() or self.ppid() == os.getpid():
+            logger.info('Patroni (pid=%s, ppid=%s), "fake postmaster" (pid=%s, ppid=%s)',
+                        os.getpid(), os.getppid(), self.pid, self.ppid())
+            return False
+
+        return True
+
+    @classmethod
+    def _from_pidfile(cls, data_dir):
+        postmaster_pid = PostmasterProcess._read_postmaster_pidfile(data_dir)
+        try:
+            pid = int(postmaster_pid.get('pid', 0))
+            if pid:
+                proc = cls(pid)
+                proc._postmaster_pid = postmaster_pid
+                return proc
+        except ValueError:
+            pass
+
+    @staticmethod
+    def from_pidfile(data_dir):
+        try:
+            proc = PostmasterProcess._from_pidfile(data_dir)
+            return proc if proc and proc._is_postmaster_process() else None
+        except psutil.NoSuchProcess:
+            return None
+
+    @classmethod
+    def from_pid(cls, pid):
+        try:
+            return cls(pid)
+        except psutil.NoSuchProcess:
+            return None
+
+    def signal_stop(self, mode):
+        """Signal postmaster process to stop
+
+        :returns None if signaled, True if process is already gone, False if error
+        """
+        if self.is_single_user:
+            logger.warning("Cannot stop server; single-user server is running (PID: {0})".format(self.pid))
+            return False
+        try:
+            self.send_signal(STOP_SIGNALS[mode])
+        except psutil.NoSuchProcess:
+            return True
+        except psutil.AccessDenied as e:
+            logger.warning("Could not send stop signal to PostgreSQL (error: {0})".format(e))
+            return False
+
+        return None
+
+    def wait_for_user_backends_to_close(self):
+        # These regexps are cross checked against versions PostgreSQL 9.1 .. 11
+        aux_proc_re = re.compile("(?:postgres:)( .*:)? (?:(?:archiver|startup|autovacuum launcher|autovacuum worker|"
+                                 "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:
+            children = self.children()
+        except psutil.Error:
+            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):
+        # Unfortunately `pg_ctl start` does not return postmaster pid to us. Without this information
+        # it is hard to know the current state of postgres startup, so we had to reimplement pg_ctl start
+        # in python. It will start postgres, wait for port to be open and wait until postgres will start
+        # accepting connections.
+        # Important!!! We can't just start postgres using subprocess.Popen, because in this case it
+        # will be our child for the rest of our live and we will have to take care of it (`waitpid`).
+        # So we will use the same approach as pg_ctl uses: start a new process, which will start postgres.
+        # This process will write postmaster pid to stdout and exit immediately. Now it's responsibility
+        # of init process to take care about postmaster.
+        # 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}
+        try:
+            proc = PostmasterProcess._from_pidfile(data_dir)
+            if proc and not proc._is_postmaster_process():
+                # Upon start postmaster process performs various safety checks if there is a postmaster.pid
+                # file in the data directory. Although Patroni already detected that the running process
+                # corresponding to the postmaster.pid is not a postmaster, the new postmaster might fail
+                # to start, because it thinks that postmaster.pid is already locked.
+                # Important!!! Unlink of postmaster.pid isn't an option, because it has a lot of nasty race conditions.
+                # Luckily there is a workaround to this problem, we can pass the pid from postmaster.pid
+                # in the `PG_GRANDPARENT_PID` environment variable and postmaster will ignore it.
+                logger.info("Telling pg_ctl that it is safe to ignore postmaster.pid for process %s", proc.pid)
+                env['PG_GRANDPARENT_PID'] = str(proc.pid)
+        except psutil.NoSuchProcess:
+            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, env=env)
+        pid = int(proc.stdout.readline().strip())
+        proc.wait()
+        logger.info('postmaster pid=%s', pid)
+
+        # TODO: In an extremely unlikely case, the process could have exited and the pid reassigned. The start
+        # initiation time is not accurate enough to compare to create time as start time would also likely
+        # be relatively close. We need the subprocess extract pid+start_time in a race free manner.
+        return PostmasterProcess.from_pid(pid)

patroni/scripts/aws.py

@@ -6,70 +6,73 @@ from requests.exceptions import RequestException
 import sys
 import boto.ec2
 
+from patroni.utils import Retry, RetryFailedError
+
 logger = logging.getLogger(__name__)
 
 
-class AWSConnection:
+class AWSConnection(object):
+
     def __init__(self, cluster_name):
         self.available = False
         self.cluster_name = cluster_name if cluster_name is not None else 'unknown'
+        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=0.1)
+            r = requests.get('http://169.254.169.254/latest/dynamic/instance-identity/document', timeout=2.1)
         except RequestException:
-            logger.info("cannot query AWS meta-data")
+            logger.error('cannot query AWS meta-data')
             return
+
         if r.ok:
             try:
                 content = r.json()
                 self.instance_id = content['instanceId']
                 self.region = content['region']
-            except Exception as e:
-                logger.info('unable to fetch instance id and region from AWS meta-data: {}'.format(e))
+            except Exception:
+                logger.exception('unable to fetch instance id and region from AWS meta-data')
                 return
             self.available = True
 
+    def retry(self, *args, **kwargs):
+        return self._retry.copy()(*args, **kwargs)
+
     def aws_available(self):
         return self.available
 
-    def _tag_ebs(self, role):
+    def _tag_ebs(self, conn, role):
         """ set tags, carrying the cluster name, instance role and instance id for the EBS storage """
-        if not self.available:
-            return False
-
         tags = {'Name': 'spilo_' + self.cluster_name, 'Role': role, 'Instance': self.instance_id}
-        try:
-            conn = boto.ec2.connect_to_region(self.region)
-            volumes = conn.get_all_volumes(filters={'attachment.instance-id': self.instance_id})
-            conn.create_tags([v.id for v in volumes], tags)
-        except Exception as e:
-            logger.info('could not set tags for EBS storage devices attached: {}'.format(e))
-            return False
-        return True
+        volumes = conn.get_all_volumes(filters={'attachment.instance-id': self.instance_id})
+        conn.create_tags([v.id for v in volumes], tags)
 
-    def _tag_ec2(self, role):
+    def _tag_ec2(self, conn, role):
         """ tag the current EC2 instance with a cluster role """
-        if not self.available:
-            return False
         tags = {'Role': role}
-        try:
-            conn = boto.ec2.connect_to_region(self.region)
-            conn.create_tags([self.instance_id], tags)
-        except Exception as e:
-            logger.info("could not set tags for EC2 instance {}: {}".format(self.instance_id, e))
-            return False
-        return True
+        conn.create_tags([self.instance_id], tags)
 
     def on_role_change(self, new_role):
-        ret = self._tag_ec2(new_role)
-        return self._tag_ebs(new_role) and ret
+        if not self.available:
+            return False
+        try:
+            conn = self.retry(boto.ec2.connect_to_region, self.region)
+            self.retry(self._tag_ec2, conn, new_role)
+            self.retry(self._tag_ebs, conn, new_role)
+        except RetryFailedError:
+            logger.warning("Unable to communicate to AWS "
+                           "when setting tags for the EC2 instance {0} "
+                           "and attached EBS volumes".format(self.instance_id))
+            return False
+        return True
 
 
 def main():
+    logging.basicConfig(format='%(asctime)s %(levelname)s: %(message)s', level=logging.INFO)
     if len(sys.argv) == 4 and sys.argv[1] in ('on_start', 'on_stop', 'on_role_change'):
         AWSConnection(cluster_name=sys.argv[3]).on_role_change(sys.argv[2])
     else:
         sys.exit("Usage: {0} action role name".format(sys.argv[0]))
 
+
 if __name__ == '__main__':
     main()

patroni/scripts/wale_restore.py

@@ -1,4 +1,4 @@
-#!/usr/bin/python
+#!/usr/bin/env python
 
 # sample script to clone new replicas using WAL-E restore
 # falls back to pg_basebackup if WAL-E restore fails, or if
@@ -23,77 +23,180 @@
 # currently also requires that you configure the restore_command to use wal_e, example:
 #       recovery_conf:
 #               restore_command: envdir /etc/wal-e.d/env wal-e wal-fetch "%f" "%p" -p 1
-
-from collections import namedtuple
+import argparse
+import csv
 import logging
 import os
 import psycopg2
 import subprocess
 import sys
-import argparse
+import time
 
+from collections import namedtuple
 
-if sys.hexversion >= 0x03000000:
-    long = int
-
-logging.basicConfig(format='%(asctime)s %(levelname)s: %(message)s', level=logging.INFO)
 logger = logging.getLogger(__name__)
 
+RETRY_SLEEP_INTERVAL = 1
+si_prefixes = ['K', 'M', 'G', 'T', 'P', 'E', 'Z', 'Y']
+
+
+# Meaningful names to the exit codes used by WALERestore
+ExitCode = type('Enum', (), {
+    'SUCCESS': 0,  #: Succeeded
+    'RETRY_LATER': 1,  #: External issue, retry later
+    'FAIL': 2  #: Don't try again unless configuration changes
+})
+
+
+# We need to know the current PG version in order to figure out the correct WAL directory name
+def get_major_version(data_dir):
+    version_file = os.path.join(data_dir, 'PG_VERSION')
+    if os.path.isfile(version_file):  # version file exists
+        try:
+            with open(version_file) as f:
+                return float(f.read())
+        except Exception:
+            logger.exception('Failed to read PG_VERSION from %s', data_dir)
+    return 0.0
+
+
+def repr_size(n_bytes):
+    """
+    >>> repr_size(1000)
+    '1000 Bytes'
+    >>> repr_size(8257332324597)
+    '7.5 TiB'
+    """
+    if n_bytes < 1024:
+        return '{0} Bytes'.format(n_bytes)
+    i = -1
+    while n_bytes > 1023:
+        n_bytes /= 1024.0
+        i += 1
+    return '{0} {1}iB'.format(round(n_bytes, 1), si_prefixes[i])
+
+
+def size_as_bytes(size_, prefix):
+    """
+    >>> size_as_bytes(7.5, 'T')
+    8246337208320
+    """
+    prefix = prefix.upper()
+
+    assert prefix in si_prefixes
+
+    exponent = si_prefixes.index(prefix) + 1
+
+    return int(size_ * (1024.0 ** exponent))
+
+
+WALEConfig = namedtuple(
+    'WALEConfig',
+    [
+        'env_dir',
+        'threshold_mb',
+        'threshold_pct',
+        'cmd',
+    ]
+)
+
 
 class WALERestore(object):
-
-    def __init__(self, scope, datadir, connstring, env_dir, threshold_mb, threshold_pct, use_iam):
+    def __init__(self, scope, datadir, connstring, env_dir, threshold_mb,
+                 threshold_pct, use_iam, no_master, retries):
         self.scope = scope
         self.master_connection = connstring
         self.data_dir = datadir
-        self.wal_e = namedtuple('wale', 'dir,threshold_mb,threshold_pct,iam_string,cmd')
-        self.wal_e.dir = env_dir
-        self.wal_e.threshold_mb = threshold_mb
-        self.wal_e.threshold_pct = threshold_pct
-        self.wal_e.iam_string = ' --aws-instance-profile ' if use_iam == 1 else ''
-        self.wal_e.cmd = 'envdir {0} wal-e {1} '.format(self.wal_e.dir, self.wal_e.iam_string)
-        self.init_error = (not os.path.exists(self.wal_e.dir))
+        self.no_master = no_master
+
+        wale_cmd = [
+            'envdir',
+            env_dir,
+            'wal-e',
+        ]
+
+        if use_iam == 1:
+            wale_cmd += ['--aws-instance-profile']
+
+        self.wal_e = WALEConfig(
+            env_dir=env_dir,
+            threshold_mb=threshold_mb,
+            threshold_pct=threshold_pct,
+            cmd=wale_cmd,
+        )
+
+        self.init_error = (not os.path.exists(self.wal_e.env_dir))
+        self.retries = retries
 
     def run(self):
-        """ creates a new replica using WAL-E """
-        if not self.init_error and self.should_use_s3_to_create_replica():
-            return self.create_replica_with_s3()
-        return 2
+        """
+        Creates a new replica using WAL-E
+
+        Returns
+        -------
+        ExitCode
+            0 = Success
+            1 = Error, try again
+            2 = Error, don't try again
+
+        """
+        if self.init_error:
+            logger.error('init error: %r did not exist at initialization time',
+                         self.wal_e.env_dir)
+            return ExitCode.FAIL
+
+        try:
+            should_use_s3 = self.should_use_s3_to_create_replica()
+            if should_use_s3 is None:  # Need to retry
+                return ExitCode.RETRY_LATER
+            elif should_use_s3:
+                return self.create_replica_with_s3()
+            elif not should_use_s3:
+                return ExitCode.FAIL
+        except Exception:
+            logger.exception("Unhandled exception when running WAL-E restore")
+        return ExitCode.FAIL
 
     def should_use_s3_to_create_replica(self):
         """ determine whether it makes sense to use S3 and not pg_basebackup """
 
         threshold_megabytes = self.wal_e.threshold_mb
-        threshold_backup_size_percentage = self.wal_e.threshold_pct
+        threshold_percent = self.wal_e.threshold_pct
 
         try:
-            latest_backup = subprocess.check_output(self.wal_e.cmd.split() + ['backup-list', '--detail', 'LATEST'])
-            # name    last_modified   expanded_size_bytes wal_segment_backup_start    wal_segment_offset_backup_start
-            #                                                   wal_segment_backup_stop wal_segment_offset_backup_stop
-            # base_00000001000000000000007F_00000040  2015-05-18T10:13:25.000Z
-            # 20310671    00000001000000000000007F    00000040
-            # 00000001000000000000007F    00000240
-            backup_strings = latest_backup.splitlines() if latest_backup else ()
-            if len(backup_strings) != 2:
+            cmd = self.wal_e.cmd + ['backup-list', '--detail', 'LATEST']
+
+            logger.debug('calling %r', cmd)
+            wale_output = subprocess.check_output(cmd)
+
+            reader = csv.DictReader(wale_output.decode('utf-8').splitlines(),
+                                    dialect='excel-tab')
+            rows = list(reader)
+            if not len(rows):
+                logger.warning('wal-e did not find any backups')
                 return False
 
-            names = backup_strings[0].split()
-            vals = backup_strings[1].split()
-            if (len(names) != len(vals)) or (len(names) != 7):
+            # This check might not add much, it was performed in the previous
+            # version of this code. since the old version rolled CSV parsing the
+            # check may have been part of the CSV parsing.
+            if len(rows) > 1:
+                logger.warning(
+                    'wal-e returned more than one row of backups: %r',
+                    rows)
                 return False
 
-            backup_info = dict(zip(names, vals))
-        except subprocess.CalledProcessError as e:
-            logger.error("could not query wal-e latest backup: {}".format(e))
-            return False
+            backup_info = rows[0]
+        except subprocess.CalledProcessError:
+            logger.exception("could not query wal-e latest backup")
+            return None
 
         try:
-            backup_size = backup_info['expanded_size_bytes']
+            backup_size = int(backup_info['expanded_size_bytes'])
             backup_start_segment = backup_info['wal_segment_backup_start']
             backup_start_offset = backup_info['wal_segment_offset_backup_start']
-        except Exception as e:
-            logger.error("unable to get some of WALE backup parameters: {}".format(e))
-            return False
+        except KeyError:
+            logger.exception("unable to get some of WALE backup parameters")
+            return None
 
         # WAL filename is XXXXXXXXYYYYYYYY000000ZZ, where X - timeline, Y - LSN logical log file,
         # ZZ - 2 high digits of LSN offset. The rest of the offset is the provided decimal offset,
@@ -101,45 +204,134 @@ class WALERestore(object):
 
         lsn_segment = backup_start_segment[8:16]
         # first 2 characters of the result are 0x and the last one is L
-        lsn_offset = hex((long(backup_start_segment[16:32], 16) << 24) + long(backup_start_offset))[2:-1]
+        lsn_offset = hex((int(backup_start_segment[16:32], 16) << 24) + int(backup_start_offset))[2:-1]
 
         # construct the LSN from the segment and offset
-        backup_start_lsn = '{}/{}'.format(lsn_segment, lsn_offset)
+        backup_start_lsn = '{0}/{1}'.format(lsn_segment, lsn_offset)
 
-        conn = None
-        cursor = None
-        diff_in_bytes = long(backup_size)
-        try:
-            # get the difference in bytes between the current WAL location and the backup start offset
-            conn = psycopg2.connect(self.master_connection)
-            conn.autocommit = True
-            cursor = conn.cursor()
-            cursor.execute("SELECT pg_xlog_location_diff(pg_current_xlog_location(), %s)", (backup_start_lsn,))
-            diff_in_bytes = long(cursor.fetchone()[0])
-        except psycopg2.Error as e:
-            logger.error('could not determine difference with the master location: {}'.format(e))
-            return False
-        finally:
-            cursor and cursor.close()
-            conn and conn.close()
+        diff_in_bytes = backup_size
+        attempts_no = 0
+        while True:
+            if self.master_connection:
+                try:
+                    # get the difference in bytes between the current WAL location and the backup start offset
+                    with psycopg2.connect(self.master_connection) as con:
+                        if con.server_version >= 100000:
+                            wal_name = 'wal'
+                            lsn_name = 'lsn'
+                        else:
+                            wal_name = 'xlog'
+                            lsn_name = 'location'
+                        con.autocommit = True
+                        with con.cursor() as cur:
+                            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])
+                except psycopg2.Error:
+                    logger.exception('could not determine difference with the master location')
+                    if attempts_no < self.retries:  # retry in case of a temporarily connection issue
+                        attempts_no = attempts_no + 1
+                        time.sleep(RETRY_SLEEP_INTERVAL)
+                        continue
+                    else:
+                        if not self.no_master:
+                            return False  # do no more retries on the outer level
+                        logger.info("continue with base backup from S3 since master is not available")
+                        diff_in_bytes = 0
+                        break
+            else:
+                # always try to use WAL-E if master connection string is not available
+                diff_in_bytes = 0
+            break
 
         # if the size of the accumulated WAL segments is more than a certan percentage of the backup size
         # or exceeds the pre-determined size - pg_basebackup is chosen instead.
-        return (diff_in_bytes < long(threshold_megabytes) * 1048576) and\
-            (diff_in_bytes < long(backup_size) * float(threshold_backup_size_percentage) / 100)
+        is_size_thresh_ok = diff_in_bytes < int(threshold_megabytes) * 1048576
+        threshold_pct_bytes = backup_size * threshold_percent / 100.0
+        is_percentage_thresh_ok = float(diff_in_bytes) < int(threshold_pct_bytes)
+        are_thresholds_ok = is_size_thresh_ok and is_percentage_thresh_ok
+
+        class Size(object):
+            def __init__(self, n_bytes, prefix=None):
+                self.n_bytes = n_bytes
+                self.prefix = prefix
+
+            def __repr__(self):
+                if self.prefix is not None:
+                    n_bytes = size_as_bytes(self.n_bytes, self.prefix)
+                else:
+                    n_bytes = self.n_bytes
+                return repr_size(n_bytes)
+
+        class HumanContext(object):
+            def __init__(self, items):
+                self.items = items
+
+            def __repr__(self):
+                return ', '.join('{}={!r}'.format(key, value)
+                                 for key, value in self.items)
+
+        human_context = repr(HumanContext([
+            ('threshold_size', Size(threshold_megabytes, 'M')),
+            ('threshold_percent', threshold_percent),
+            ('threshold_percent_size', Size(threshold_pct_bytes)),
+            ('backup_size', Size(backup_size)),
+            ('backup_diff', Size(diff_in_bytes)),
+            ('is_size_thresh_ok', is_size_thresh_ok),
+            ('is_percentage_thresh_ok', is_percentage_thresh_ok),
+        ]))
+
+        if not are_thresholds_ok:
+            logger.info('wal-e backup size diff is over threshold, falling back '
+                        'to other means of restore: %s', human_context)
+        else:
+            logger.info('Thresholds are OK, using wal-e basebackup: %s', human_context)
+        return are_thresholds_ok
+
+    def fix_subdirectory_path_if_broken(self, dirname):
+        # in case it is a symlink pointing to a non-existing location, remove it and create the actual directory
+        path = os.path.join(self.data_dir, dirname)
+        if not os.path.exists(path):
+            if os.path.islink(path):  # broken xlog symlink, to remove
+                try:
+                    os.remove(path)
+                except OSError:
+                    logger.exception("could not remove broken %s symlink pointing to %s",
+                                     dirname, os.readlink(path))
+                    return False
+            try:
+                os.mkdir(path)
+            except OSError:
+                logger.exception("coud not create missing %s directory path", dirname)
+                return False
+        return True
 
     def create_replica_with_s3(self):
         # if we're set up, restore the replica using fetch latest
         try:
-            ret = subprocess.call(self.wal_e.cmd.split() + ['backup-fetch', '{}'.format(self.data_dir), 'LATEST'])
+            cmd = self.wal_e.cmd + ['backup-fetch',
+                                    '{}'.format(self.data_dir),
+                                    'LATEST']
+            logger.debug('calling: %r', cmd)
+            exit_code = subprocess.call(cmd)
         except Exception as e:
             logger.error('Error when fetching backup with WAL-E: {0}'.format(e))
-            return 1
+            return ExitCode.RETRY_LATER
 
-        return ret
+        if (exit_code == 0 and not
+           self.fix_subdirectory_path_if_broken('pg_xlog' if get_major_version(self.data_dir) < 10 else 'pg_wal')):
+            return ExitCode.FAIL
+        return exit_code
 
 
 def main():
+    logging.basicConfig(format='%(asctime)s %(levelname)s: %(message)s', level=logging.INFO)
     parser = argparse.ArgumentParser(description='Script to image replicas using WAL-E')
     parser.add_argument('--scope', required=True)
     parser.add_argument('--role', required=False)
@@ -150,18 +342,30 @@ def main():
     parser.add_argument('--threshold_megabytes', type=int, default=10240)
     parser.add_argument('--threshold_backup_size_percentage', type=int, default=30)
     parser.add_argument('--use_iam', type=int, default=0)
+    parser.add_argument('--no_master', type=int, default=0)
     args = parser.parse_args()
 
-    # retry cloning in a loop
-    for retry in range(0, args.retries + 1):
+    exit_code = None
+    assert args.retries >= 0
+
+    # Retry cloning in a loop. We do separate retries for the master
+    # connection attempt inside should_use_s3_to_create_replica,
+    # because we need to differentiate between the last attempt and
+    # the rest and make a decision when the last attempt fails on
+    # whether to use WAL-E or not depending on the no_master flag.
+    for _ in range(0, args.retries + 1):
         restore = WALERestore(scope=args.scope, datadir=args.datadir, connstring=args.connstring,
                               env_dir=args.envdir, threshold_mb=args.threshold_megabytes,
-                              threshold_pct=args.threshold_backup_size_percentage, use_iam=args.use_iam)
-        ret = restore.run()
-        if ret == 0:
+                              threshold_pct=args.threshold_backup_size_percentage, use_iam=args.use_iam,
+                              no_master=args.no_master, retries=args.retries)
+        exit_code = restore.run()
+        if not exit_code == ExitCode.RETRY_LATER:  # only WAL-E failures lead to the retry
+            logger.debug('exit_code is %r, not retrying', exit_code)
             break
+        time.sleep(RETRY_SLEEP_INTERVAL)
+
+    return exit_code
 
-    sys.exit(ret)
 
 if __name__ == '__main__':
-    main()
+    sys.exit(main())

patroni/utils.py

@@ -1,95 +1,196 @@
-import datetime
-import os
 import random
-import re
-import signal
-import sys
 import time
 
+from dateutil import tz
 from patroni.exceptions import PatroniException
 
-ignore_sigterm = False
-interrupted_sleep = False
-reap_children = False
-
-_DATE_TIME_RE = re.compile(r'''^
-(?P<year>\d{4})\-(?P<month>\d{2})\-(?P<day>\d{2})  # date
-T
-(?P<hour>\d{2}):(?P<minute>\d{2}):(?P<second>\d{2})\.(?P<microsecond>\d{6})  # time
-\d*Z$''', re.X)
+tzutc = tz.tzutc()
 
 
-def parse_datetime(time_str):
+def deep_compare(obj1, obj2):
     """
-    >>> parse_datetime('2015-06-10T12:56:30.552539016Z')
-    datetime.datetime(2015, 6, 10, 12, 56, 30, 552539)
-    >>> parse_datetime('2015-06-10 12:56:30.552539016Z')
-    """
-    m = _DATE_TIME_RE.match(time_str)
-    if not m:
-        return None
-    p = dict((n, int(m.group(n))) for n in 'year month day hour minute second microsecond'.split(' '))
-    return datetime.datetime(**p)
-
-
-def calculate_ttl(expiration):
-    """
-    >>> calculate_ttl(None)
-    >>> calculate_ttl('2015-06-10 12:56:30.552539016Z')
-    >>> calculate_ttl('2015-06-10T12:56:30.552539016Z') < 0
+    >>> deep_compare({'1': None}, {})
+    False
+    >>> deep_compare({'1': {}}, {'1': None})
+    False
+    >>> deep_compare({'1': [1]}, {'1': [2]})
+    False
+    >>> deep_compare({'1': 2}, {'1': '2'})
+    True
+    >>> deep_compare({'1': {'2': [3, 4]}}, {'1': {'2': [3, 4]}})
     True
     """
-    if not expiration:
-        return None
-    expiration = parse_datetime(expiration)
-    if not expiration:
-        return None
-    now = datetime.datetime.utcnow()
-    return int((expiration - now).total_seconds())
+
+    if set(list(obj1.keys())) != set(list(obj2.keys())):  # Objects have different sets of keys
+        return False
+
+    for key, value in obj1.items():
+        if isinstance(value, dict):
+            if not (isinstance(obj2[key], dict) and deep_compare(value, obj2[key])):
+                return False
+        elif str(value) != str(obj2[key]):
+            return False
+    return True
 
 
-def sigterm_handler(signo, stack_frame):
-    global ignore_sigterm
-    if not ignore_sigterm:
-        ignore_sigterm = True
-        sys.exit()
+def patch_config(config, data):
+    """recursively 'patch' `config` with `data`
+    :returns: `!True` if the `config` was changed"""
+    is_changed = False
+    for name, value in data.items():
+        if value is None:
+            if config.pop(name, None) is not None:
+                is_changed = True
+        elif name in config:
+            if isinstance(value, dict):
+                if isinstance(config[name], dict):
+                    if patch_config(config[name], value):
+                        is_changed = True
+                else:
+                    config[name] = value
+                    is_changed = True
+            elif str(config[name]) != str(value):
+                config[name] = value
+                is_changed = True
+        else:
+            config[name] = value
+            is_changed = True
+    return is_changed
 
 
-def sigchld_handler(signo, stack_frame):
-    global interrupted_sleep, reap_children
-    reap_children = interrupted_sleep = True
+def parse_bool(value):
+    """
+    >>> parse_bool(1)
+    True
+    >>> parse_bool('off')
+    False
+    >>> parse_bool('foo')
+    """
+    value = str(value).lower()
+    if value in ('on', 'true', 'yes', '1'):
+        return True
+    if value in ('off', 'false', 'no', '0'):
+        return False
 
 
-def sleep(interval):
-    global interrupted_sleep
-    current_time = time.time()
-    end_time = current_time + interval
-    while current_time < end_time:
-        interrupted_sleep = False
-        time.sleep(end_time - current_time)
-        if not interrupted_sleep:  # we will ignore only sigchld
-            break
-        current_time = time.time()
-    interrupted_sleep = False
+def strtol(value, strict=True):
+    """As most as possible close equivalent of strtol(3) function (with base=0),
+       used by postgres to parse parameter values.
+    >>> strtol(0) == (0, '')
+    True
+    >>> strtol(1) == (1, '')
+    True
+    >>> strtol(9) == (9, '')
+    True
+    >>> strtol(' +0x400MB') == (1024, 'MB')
+    True
+    >>> strtol(' -070d') == (-56, 'd')
+    True
+    >>> strtol(' d ') == (None, '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
+
+    # 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"
+
+    return (None if strict else 1), value.strip()
 
 
-def setup_signal_handlers():
-    signal.signal(signal.SIGTERM, sigterm_handler)
-    signal.signal(signal.SIGCHLD, sigchld_handler)
+def parse_int(value, base_unit=None):
+    """
+    >>> parse_int('1') == 1
+    True
+    >>> parse_int(' 0x400 MB ', '16384kB') == 64
+    True
+    >>> parse_int('1MB', 'kB') == 1024
+    True
+    >>> parse_int('1000 ms', 's') == 1
+    True
+    >>> parse_int('1GB', 'MB') is None
+    True
+    >>> parse_int(0) == 0
+    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}
+    }
+
+    value, unit = strtol(value)
+    if value is not None:
+        if not unit:
+            return value
+
+        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)
 
 
-def reap_children():
-    global reap_children
-    if reap_children:
-        try:
-            while True:
-                ret = os.waitpid(-1, os.WNOHANG)
-                if ret == (0, 0):
-                    break
-        except OSError:
-            pass
-        finally:
-            reap_children = False
+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)
+    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)
+    return old_value is not None and new_value is not None and old_value == new_value
+
+
+def _sleep(interval):
+    time.sleep(interval)
 
 
 class RetryFailedError(PatroniException):
@@ -97,12 +198,12 @@ class RetryFailedError(PatroniException):
     """Raised when retrying an operation ultimately failed, after retrying the maximum number of attempts."""
 
 
-class Retry:
+class Retry(object):
 
     """Helper for retrying a method in the face of retry-able exceptions"""
 
     def __init__(self, max_tries=1, delay=0.1, backoff=2, max_jitter=0.8, max_delay=3600,
-                 sleep_func=sleep, deadline=None, retry_exceptions=PatroniException):
+                 sleep_func=_sleep, deadline=None, retry_exceptions=PatroniException):
         """Create a :class:`Retry` instance for retrying function calls
 
         :param max_tries: How many times to retry the command. -1 means infinite tries.
@@ -163,3 +264,20 @@ class Retry:
                 else:
                     self.sleep_func(sleeptime)
                 self._cur_delay = min(self._cur_delay * self.backoff, self.max_delay)
+
+
+def polling_loop(timeout, interval=1):
+    """Returns an iterator that returns values until timeout has passed. Timeout is measured from start of iteration."""
+    start_time = time.time()
+    iteration = 0
+    end_time = start_time + timeout
+    while time.time() < end_time:
+        yield iteration
+        iteration += 1
+        time.sleep(interval)
+
+
+def split_host_port(value, default_port):
+    t = value.rsplit(':', 1)
+    t.append(default_port)
+    return t[0], int(t[1])

patroni/version.py

@@ -1 +1 @@
-__version__ = '0.75'
+__version__ = '1.5.4'

patroni/watchdog/__init__.py

@@ -0,0 +1,2 @@
+from patroni.watchdog.base import WatchdogError, Watchdog
+__all__ = ['WatchdogError', 'Watchdog']

patroni/watchdog/base.py

@@ -0,0 +1,313 @@
+import abc
+import logging
+import platform
+import six
+import sys
+from threading import RLock
+
+from patroni.exceptions import WatchdogError
+
+__all__ = ['WatchdogError', 'Watchdog']
+
+logger = logging.getLogger(__name__)
+
+MODE_REQUIRED = 'required'    # Will not run if a watchdog is not available
+MODE_AUTOMATIC = 'automatic'  # Will use a watchdog if one is available
+MODE_OFF = 'off'              # Will not try to use a watchdog
+
+
+def parse_mode(mode):
+    if mode is False:
+        return MODE_OFF
+    mode = mode.lower()
+    if mode in ['require', 'required']:
+        return MODE_REQUIRED
+    elif mode in ['auto', 'automatic']:
+        return MODE_AUTOMATIC
+    else:
+        if mode not in ['off', 'disable', 'disabled']:
+            logger.warning("Watchdog mode {0} not recognized, disabling watchdog".format(mode))
+        return MODE_OFF
+
+
+def synchronized(func):
+    def wrapped(self, *args, **kwargs):
+        with self._lock:
+            return func(self, *args, **kwargs)
+    return wrapped
+
+
+class WatchdogConfig(object):
+    """Helper to contain a snapshot of configuration"""
+    def __init__(self, config):
+        self.mode = parse_mode(config['watchdog'].get('mode', 'automatic'))
+        self.ttl = config['ttl']
+        self.loop_wait = config['loop_wait']
+        self.safety_margin = config['watchdog'].get('safety_margin', 5)
+        self.driver = config['watchdog'].get('driver', 'default')
+        self.driver_config = dict((k, v) for k, v in config['watchdog'].items()
+                                  if k not in ['mode', 'safety_margin', 'driver'])
+
+    def __eq__(self, other):
+        return isinstance(other, WatchdogConfig) and \
+            all(getattr(self, attr) == getattr(other, attr) for attr in
+                ['mode', 'ttl', 'loop_wait', 'safety_margin', 'driver', 'driver_config'])
+
+    def __ne__(self, other):
+        return not self == other
+
+    def get_impl(self):
+        if self.driver == 'testing':
+            from patroni.watchdog.linux import TestingWatchdogDevice
+            return TestingWatchdogDevice.from_config(self.driver_config)
+        elif platform.system() == 'Linux' and self.driver == 'default':
+            from patroni.watchdog.linux import LinuxWatchdogDevice
+            return LinuxWatchdogDevice.from_config(self.driver_config)
+        else:
+            return NullWatchdog()
+
+    @property
+    def timeout(self):
+        if self.safety_margin == -1:
+            return int(self.ttl // 2)
+        else:
+            return self.ttl - self.safety_margin
+
+    @property
+    def timing_slack(self):
+        return self.timeout - self.loop_wait
+
+
+class Watchdog(object):
+    """Facade to dynamically manage watchdog implementations and handle config changes.
+
+    When activation fails underlying implementation will be switched to a Null implementation. To avoid log spam
+    activation will only be retried when watchdog configuration is changed."""
+    def __init__(self, config):
+        self.active_config = self.config = WatchdogConfig(config)
+        self._lock = RLock()
+        self.active = False
+
+        if self.config.mode == MODE_OFF:
+            self.impl = NullWatchdog()
+        else:
+            self.impl = self.config.get_impl()
+            if self.config.mode == MODE_REQUIRED and self.impl.is_null:
+                logger.error("Configuration requires a watchdog, but watchdog is not supported on this platform.")
+                sys.exit(1)
+
+    @synchronized
+    def reload_config(self, config):
+        self.config = WatchdogConfig(config)
+        # Turning a watchdog off can always be done immediately
+        if self.config.mode == MODE_OFF:
+            if self.active:
+                self._disable()
+            self.active_config = self.config
+            self.impl = NullWatchdog()
+        # If watchdog is not active we can apply config immediately to show any warnings early. Otherwise we need to
+        # delay until next time a keepalive is sent so timeout matches up with leader key update.
+        if not self.active:
+            if self.config.driver != self.active_config.driver or \
+               self.config.driver_config != self.active_config.driver_config:
+                self.impl = self.config.get_impl()
+            self.active_config = self.config
+
+    @synchronized
+    def activate(self):
+        """Activates the watchdog device with suitable timeouts. While watchdog is active keepalive needs
+        to be called every time loop_wait expires.
+
+        :returns False if a safe watchdog could not be configured, but is required.
+        """
+        self.active = True
+        return self._activate()
+
+    def _activate(self):
+        self.active_config = self.config
+
+        if self.config.timing_slack < 0:
+            logger.warning('Watchdog not supported because leader TTL {0} is less than 2x loop_wait {1}'
+                           .format(self.config.ttl, self.config.loop_wait))
+            self.impl = NullWatchdog()
+
+        try:
+            self.impl.open()
+            actual_timeout = self._set_timeout()
+        except WatchdogError as e:
+            logger.warning("Could not activate %s: %s", self.impl.describe(), e)
+            self.impl = NullWatchdog()
+
+        if self.impl.is_running and not self.impl.can_be_disabled:
+            logger.warning("Watchdog implementation can't be disabled."
+                           " Watchdog will trigger after Patroni loses leader key.")
+
+        if not self.impl.is_running or actual_timeout > self.config.timeout:
+            if self.config.mode == MODE_REQUIRED:
+                if self.impl.is_null:
+                    logger.error("Configuration requires watchdog, but watchdog could not be configured.")
+                else:
+                    logger.error("Configuration requires watchdog, but a safe watchdog timeout {0} could"
+                                 " not be configured. Watchdog timeout is {1}.".format(
+                                     self.config.timeout, actual_timeout))
+                return False
+            else:
+                if not self.impl.is_null:
+                    logger.warning("Watchdog timeout {0} seconds does not ensure safe termination within {1} seconds"
+                                   .format(actual_timeout, self.config.timeout))
+
+        if self.is_running:
+            logger.info("{0} activated with {1} second timeout, timing slack {2} seconds"
+                        .format(self.impl.describe(), actual_timeout, self.config.timing_slack))
+        else:
+            if self.config.mode == MODE_REQUIRED:
+                logger.error("Configuration requires watchdog, but watchdog could not be activated")
+                return False
+
+        return True
+
+    def _set_timeout(self):
+        if self.impl.has_set_timeout():
+            self.impl.set_timeout(self.config.timeout)
+
+        # Safety checks for watchdog implementations that don't support configurable timeouts
+        actual_timeout = self.impl.get_timeout()
+        if self.impl.is_running and actual_timeout < self.config.loop_wait:
+            logger.error('loop_wait of {0} seconds is too long for watchdog {1} second timeout'
+                         .format(self.config.loop_wait, actual_timeout))
+            if self.impl.can_be_disabled:
+                logger.info('Disabling watchdog due to unsafe timeout.')
+                self.impl.close()
+                self.impl = NullWatchdog()
+                return None
+        return actual_timeout
+
+    @synchronized
+    def disable(self):
+        self._disable()
+        self.active = False
+
+    def _disable(self):
+        try:
+            if self.impl.is_running and not self.impl.can_be_disabled:
+                # Give sysadmin some extra time to clean stuff up.
+                self.impl.keepalive()
+                logger.warning("Watchdog implementation can't be disabled. System will reboot after "
+                               "{0} seconds when watchdog times out.".format(self.impl.get_timeout()))
+            self.impl.close()
+        except WatchdogError as e:
+            logger.error("Error while disabling watchdog: %s", e)
+
+    @synchronized
+    def keepalive(self):
+        try:
+            if self.active:
+                self.impl.keepalive()
+            # In case there are any pending configuration changes apply them now.
+            if self.active and self.config != self.active_config:
+                if self.config.mode != MODE_OFF and self.active_config.mode == MODE_OFF:
+                    self.impl = self.config.get_impl()
+                    self._activate()
+                if self.config.driver != self.active_config.driver \
+                   or self.config.driver_config != self.active_config.driver_config:
+                    self._disable()
+                    self.impl = self.config.get_impl()
+                    self._activate()
+                if self.config.timeout != self.active_config.timeout:
+                    self.impl.set_timeout(self.config.timeout)
+        except WatchdogError as e:
+            logger.error("Error while sending keepalive: %s", e)
+
+    @property
+    @synchronized
+    def is_running(self):
+        return self.impl.is_running
+
+    @property
+    @synchronized
+    def is_healthy(self):
+        if self.config.mode != MODE_REQUIRED:
+            return True
+        return self.config.timing_slack >= 0 and self.impl.is_healthy
+
+
+@six.add_metaclass(abc.ABCMeta)
+class WatchdogBase(object):
+    """A watchdog object when opened requires periodic calls to keepalive.
+    When keepalive is not called within a timeout the system will be terminated."""
+    is_null = False
+
+    @property
+    def is_running(self):
+        """Returns True when watchdog is activated and capable of performing it's task."""
+        return False
+
+    @property
+    def is_healthy(self):
+        """Returns False when calling open() is known to fail."""
+        return False
+
+    @property
+    def can_be_disabled(self):
+        """Returns True when watchdog will be disabled by calling close(). Some watchdog devices
+        will keep running no matter what once activated. May raise WatchdogError if called without
+        calling open() first."""
+        return True
+
+    @abc.abstractmethod
+    def open(self):
+        """Open watchdog device.
+
+        When watchdog is opened keepalive must be called. Returns nothing on success
+        or raises WatchdogError if the device could not be opened."""
+
+    @abc.abstractmethod
+    def close(self):
+        """Gracefully close watchdog device."""
+
+    @abc.abstractmethod
+    def keepalive(self):
+        """Resets the watchdog timer.
+
+        Watchdog must be open when keepalive is called."""
+
+    @abc.abstractmethod
+    def get_timeout(self):
+        """Returns the current keepalive timeout in effect."""
+
+    @staticmethod
+    def has_set_timeout():
+        """Returns True if setting a timeout is supported."""
+        return False
+
+    def set_timeout(self, timeout):
+        """Set the watchdog timer timeout.
+
+        :param timeout: watchdog timeout in seconds"""
+        raise WatchdogError("Setting timeout is not supported on {0}".format(self.describe()))
+
+    def describe(self):
+        """Human readable name for this device"""
+        return self.__class__.__name__
+
+    @classmethod
+    def from_config(cls, config):
+        return cls()
+
+
+class NullWatchdog(WatchdogBase):
+    """Null implementation when watchdog is not supported."""
+    is_null = True
+
+    def open(self):
+        return
+
+    def close(self):
+        return
+
+    def keepalive(self):
+        return
+
+    def get_timeout(self):
+        # A big enough number to not matter
+        return 1000000000

patroni/watchdog/linux.py

@@ -0,0 +1,234 @@
+import collections
+import ctypes
+import fcntl
+import os
+import platform
+from patroni.watchdog.base import WatchdogBase, WatchdogError
+
+# Pythonification of linux/ioctl.h
+IOC_NONE = 0
+IOC_WRITE = 1
+IOC_READ = 2
+
+IOC_NRBITS = 8
+IOC_TYPEBITS = 8
+IOC_SIZEBITS = 14
+IOC_DIRBITS = 2
+
+# Non-generic platform special cases
+machine = platform.machine()
+if machine in ['mips', 'sparc', 'powerpc', 'ppc64']:
+    IOC_SIZEBITS = 13
+    IOC_DIRBITS = 3
+    IOC_NONE, IOC_WRITE, IOC_READ = 1, 2, 4
+elif machine == 'parisc':
+    IOC_WRITE, IOC_READ = 2, 1
+
+IOC_NRSHIFT = 0
+IOC_TYPESHIFT = IOC_NRSHIFT + IOC_NRBITS
+IOC_SIZESHIFT = IOC_TYPESHIFT + IOC_TYPEBITS
+IOC_DIRSHIFT = IOC_SIZESHIFT + IOC_SIZEBITS
+
+
+def IOW(type_, nr, size):
+    return IOC(IOC_WRITE, type_, nr, size)
+
+
+def IOR(type_, nr, size):
+    return IOC(IOC_READ, type_, nr, size)
+
+
+def IOWR(type_, nr, size):
+    return IOC(IOC_READ | IOC_WRITE, type_, nr, size)
+
+
+def IOC(dir_, type_, nr, size):
+    return (dir_ << IOC_DIRSHIFT) \
+        | (ord(type_) << IOC_TYPESHIFT) \
+        | (nr << IOC_NRSHIFT) \
+        | (size << IOC_SIZESHIFT)
+
+
+# Pythonification of linux/watchdog.h
+
+WATCHDOG_IOCTL_BASE = 'W'
+
+
+class watchdog_info(ctypes.Structure):
+    _fields_ = [
+        ('options', ctypes.c_uint32),           # Options the card/driver supports
+        ('firmware_version', ctypes.c_uint32),  # Firmware version of the card
+        ('identity', ctypes.c_uint8 * 32),      # Identity of the board
+    ]
+
+
+struct_watchdog_info_size = ctypes.sizeof(watchdog_info)
+int_size = ctypes.sizeof(ctypes.c_int)
+
+WDIOC_GETSUPPORT = IOR(WATCHDOG_IOCTL_BASE, 0, struct_watchdog_info_size)
+WDIOC_GETSTATUS = IOR(WATCHDOG_IOCTL_BASE, 1, int_size)
+WDIOC_GETBOOTSTATUS = IOR(WATCHDOG_IOCTL_BASE, 2, int_size)
+WDIOC_GETTEMP = IOR(WATCHDOG_IOCTL_BASE, 3, int_size)
+WDIOC_SETOPTIONS = IOR(WATCHDOG_IOCTL_BASE, 4, int_size)
+WDIOC_KEEPALIVE = IOR(WATCHDOG_IOCTL_BASE, 5, int_size)
+WDIOC_SETTIMEOUT = IOWR(WATCHDOG_IOCTL_BASE, 6, int_size)
+WDIOC_GETTIMEOUT = IOR(WATCHDOG_IOCTL_BASE, 7, int_size)
+WDIOC_SETPRETIMEOUT = IOWR(WATCHDOG_IOCTL_BASE, 8, int_size)
+WDIOC_GETPRETIMEOUT = IOR(WATCHDOG_IOCTL_BASE, 9, int_size)
+WDIOC_GETTIMELEFT = IOR(WATCHDOG_IOCTL_BASE, 10, int_size)
+
+
+WDIOF_UNKNOWN = -1  # Unknown flag error
+WDIOS_UNKNOWN = -1  # Unknown status error
+
+WDIOF = {
+    "OVERHEAT": 0x0001,       # Reset due to CPU overheat
+    "FANFAULT": 0x0002,       # Fan failed
+    "EXTERN1": 0x0004,        # External relay 1
+    "EXTERN2": 0x0008,        # External relay 2
+    "POWERUNDER": 0x0010,     # Power bad/power fault
+    "CARDRESET": 0x0020,      # Card previously reset the CPU
+    "POWEROVER": 0x0040,      # Power over voltage
+    "SETTIMEOUT": 0x0080,     # Set timeout (in seconds)
+    "MAGICCLOSE": 0x0100,     # Supports magic close char
+    "PRETIMEOUT": 0x0200,     # Pretimeout (in seconds), get/set
+    "ALARMONLY": 0x0400,      # Watchdog triggers a management or other external alarm not a reboot
+    "KEEPALIVEPING": 0x8000,  # Keep alive ping reply
+}
+
+WDIOS = {
+    "DISABLECARD": 0x0001,    # Turn off the watchdog timer
+    "ENABLECARD": 0x0002,     # Turn on the watchdog timer
+    "TEMPPANIC": 0x0004,      # Kernel panic on temperature trip
+}
+
+# Implementation
+
+
+class WatchdogInfo(collections.namedtuple('WatchdogInfo', 'options,version,identity')):
+    """Watchdog descriptor from the kernel"""
+    def __getattr__(self, name):
+        """Convenience has_XYZ attributes for checking WDIOF bits in options"""
+        if name.startswith('has_') and name[4:] in WDIOF:
+            return bool(self.options & WDIOF[name[4:]])
+
+        raise AttributeError("WatchdogInfo instance has no attribute '{0}'".format(name))
+
+
+class LinuxWatchdogDevice(WatchdogBase):
+    DEFAULT_DEVICE = '/dev/watchdog'
+
+    def __init__(self, device):
+        self.device = device
+        self._support_cache = None
+        self._fd = None
+
+    @classmethod
+    def from_config(cls, config):
+        device = config.get('device', cls.DEFAULT_DEVICE)
+        return cls(device)
+
+    @property
+    def is_running(self):
+        return self._fd is not None
+
+    @property
+    def is_healthy(self):
+        return os.path.exists(self.device) and os.access(self.device, os.W_OK)
+
+    def open(self):
+        try:
+            self._fd = os.open(self.device, os.O_WRONLY)
+        except OSError as e:
+            raise WatchdogError("Can't open watchdog device: {0}".format(e))
+
+    def close(self):
+        if self.is_running:
+            try:
+                os.write(self._fd, b'V')
+                os.close(self._fd)
+                self._fd = None
+            except OSError as e:
+                raise WatchdogError("Error while closing {0}: {1}".format(self.describe(), e))
+
+    @property
+    def can_be_disabled(self):
+        return self.get_support().has_MAGICCLOSE
+
+    def _ioctl(self, func, arg):
+        """Runs the specified ioctl on the underlying fd.
+
+        Raises WatchdogError if the device is closed.
+        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)
+
+    def get_support(self):
+        if self._support_cache is None:
+            info = watchdog_info()
+            try:
+                self._ioctl(WDIOC_GETSUPPORT, info)
+            except (WatchdogError, OSError, IOError) as e:
+                raise WatchdogError("Could not get information about watchdog device: {}".format(e))
+            self._support_cache = WatchdogInfo(info.options,
+                                               info.firmware_version,
+                                               bytearray(info.identity).decode(errors='ignore').rstrip('\x00'))
+        return self._support_cache
+
+    def describe(self):
+        dev_str = " at {0}".format(self.device) if self.device != self.DEFAULT_DEVICE else ""
+        ver_str = ""
+        identity = "Linux watchdog device"
+        if self._fd:
+            try:
+                _, version, identity = self.get_support()
+                ver_str = " (firmware {0})".format(version) if version else ""
+            except WatchdogError:
+                pass
+
+        return identity + ver_str + dev_str
+
+    def keepalive(self):
+        try:
+            os.write(self._fd, b'1')
+        except OSError as e:
+            raise WatchdogError("Could not send watchdog keepalive: {0}".format(e))
+
+    def has_set_timeout(self):
+        """Returns True if setting a timeout is supported."""
+        return self.get_support().has_SETTIMEOUT
+
+    def set_timeout(self, timeout):
+        timeout = int(timeout)
+        if not 0 < timeout < 0xFFFF:
+            raise WatchdogError("Invalid timeout {0}. Supported values are between 1 and 65535".format(timeout))
+        try:
+            self._ioctl(WDIOC_SETTIMEOUT, ctypes.c_int(timeout))
+        except (WatchdogError, OSError, IOError) as e:
+            raise WatchdogError("Could not set timeout on watchdog device: {}".format(e))
+
+    def get_timeout(self):
+        timeout = ctypes.c_int()
+        try:
+            self._ioctl(WDIOC_GETTIMEOUT, timeout)
+        except (WatchdogError, OSError, IOError) as e:
+            raise WatchdogError("Could not get timeout on watchdog device: {}".format(e))
+        return timeout.value
+
+
+class TestingWatchdogDevice(LinuxWatchdogDevice):
+    """Converts timeout ioctls to regular writes that can be intercepted from a named pipe."""
+    timeout = 60
+
+    def get_support(self):
+        return WatchdogInfo(WDIOF['MAGICCLOSE'] | WDIOF['SETTIMEOUT'], 0, "Watchdog test harness")
+
+    def set_timeout(self, timeout):
+        buf = "Ctimeout={0}\n".format(timeout).encode('utf8')
+        while len(buf):
+            buf = buf[os.write(self._fd, buf):]
+        self.timeout = timeout
+
+    def get_timeout(self):
+        return self.timeout

patroni/zookeeper.py

@@ -1,286 +0,0 @@
-import logging
-import random
-import requests
-import time
-
-from kazoo.client import KazooClient, KazooState
-from kazoo.exceptions import NoNodeError, NodeExistsError
-from patroni.dcs import AbstractDCS, Cluster, Failover, Leader, Member
-from patroni.exceptions import DCSError
-from patroni.utils import sleep
-from requests.exceptions import RequestException
-
-logger = logging.getLogger(__name__)
-
-
-class ZooKeeperError(DCSError):
-    pass
-
-
-class ExhibitorEnsembleProvider:
-
-    TIMEOUT = 3.1
-
-    def __init__(self, hosts, port, uri_path='/exhibitor/v1/cluster/list', poll_interval=300):
-        self._exhibitor_port = port
-        self._uri_path = uri_path
-        self._poll_interval = poll_interval
-        self._exhibitors = hosts
-        self._master_exhibitors = hosts
-        self._zookeeper_hosts = ''
-        self._next_poll = None
-        while not self.poll():
-            logger.info('waiting on exhibitor')
-            sleep(5)
-
-    def poll(self):
-        if self._next_poll and self._next_poll > time.time():
-            return False
-
-        json = self._query_exhibitors(self._exhibitors)
-        if not json:
-            json = self._query_exhibitors(self._master_exhibitors)
-
-        if isinstance(json, dict) and 'servers' in json and 'port' in json:
-            self._next_poll = time.time() + self._poll_interval
-            zookeeper_hosts = ','.join([h + ':' + str(json['port']) for h in sorted(json['servers'])])
-            if self._zookeeper_hosts != zookeeper_hosts:
-                logger.info('ZooKeeper connection string has changed: %s => %s', self._zookeeper_hosts, zookeeper_hosts)
-                self._zookeeper_hosts = zookeeper_hosts
-                self._exhibitors = json['servers']
-                return True
-        return False
-
-    def _query_exhibitors(self, exhibitors):
-        random.shuffle(exhibitors)
-        for host in exhibitors:
-            uri = 'http://{}:{}{}'.format(host, self._exhibitor_port, self._uri_path)
-            try:
-                response = requests.get(uri, timeout=self.TIMEOUT)
-                return response.json()
-            except RequestException:
-                pass
-        return None
-
-    @property
-    def zookeeper_hosts(self):
-        return self._zookeeper_hosts
-
-
-class ZooKeeper(AbstractDCS):
-
-    def __init__(self, name, config):
-        super(ZooKeeper, self).__init__(name, config)
-
-        hosts = config.get('hosts', [])
-        if isinstance(hosts, list):
-            hosts = ','.join(hosts)
-
-        self.exhibitor = None
-        if 'exhibitor' in config:
-            exhibitor = config['exhibitor']
-            interval = exhibitor.get('poll_interval', 300)
-            self.exhibitor = ExhibitorEnsembleProvider(exhibitor['hosts'], exhibitor['port'], poll_interval=interval)
-            hosts = self.exhibitor.zookeeper_hosts
-
-        self.client = KazooClient(hosts=hosts,
-                                  timeout=(config.get('session_timeout', None) or 30),
-                                  command_retry={
-                                      'deadline': (config.get('reconnect_timeout', None) or 10),
-                                      'max_delay': 1,
-                                      'max_tries': -1},
-                                  connection_retry={'max_delay': 1, 'max_tries': -1})
-        self.client.add_listener(self.session_listener)
-
-        self._my_member_data = None
-        self.fetch_cluster = True
-        self.last_leader_operation = 0
-
-        self.client.start(None)
-
-    def session_listener(self, state):
-        if state in [KazooState.SUSPENDED, KazooState.LOST]:
-            self.cluster_watcher(None)
-
-    def cluster_watcher(self, event):
-        self.fetch_cluster = True
-        self.event.set()
-
-    def get_node(self, key, watch=None):
-        try:
-            ret = self.client.get(key, watch)
-            return (ret[0].decode('utf-8'), ret[1])
-        except NoNodeError:
-            return None
-
-    @staticmethod
-    def member(name, value, znode):
-        return Member.from_node(znode.version, name, znode.ephemeralOwner, value)
-
-    def get_children(self, key, watch=None):
-        try:
-            return self.client.get_children(key, watch)
-        except NoNodeError:
-            return []
-
-    def load_members(self):
-        members = []
-        for member in self.get_children(self.members_path, self.cluster_watcher):
-            data = self.get_node(self.members_path + member)
-            if data is not None:
-                members.append(self.member(member, *data))
-        return members
-
-    def _inner_load_cluster(self):
-        self.fetch_cluster = False
-        self.event.clear()
-        nodes = set(self.get_children(self.client_path(''), self.cluster_watcher))
-        if not nodes:
-            self.fetch_cluster = True
-
-        # get initialize flag
-        initialize = (self.get_node(self.initialize_path) or [None])[0] if self._INITIALIZE in nodes else None
-
-        # get list of members
-        members = self.load_members() if self._MEMBERS[:-1] in nodes else []
-
-        # get leader
-        leader = self.get_node(self.leader_path) if self._LEADER in nodes else None
-        if leader:
-            client_id = self.client.client_id
-            if leader[0] == self._name and client_id is not None and client_id[0] != leader[1].ephemeralOwner:
-                logger.info('I am leader but not owner of the session. Removing leader node')
-                self.client.delete(self.leader_path)
-                leader = None
-
-            if leader:
-                member = Member(-1, leader[0], None, {})
-                member = ([m for m in members if m.name == leader[0]] or [member])[0]
-                leader = Leader(leader[1].version, leader[1].ephemeralOwner, member)
-                self.fetch_cluster = member.index == -1
-
-        # failover key
-        failover = self.get_node(self.failover_path, watch=self.cluster_watcher) if self._FAILOVER in nodes else None
-        if failover:
-            failover = Failover.from_node(failover[1].version, failover[0])
-
-        # get last leader operation
-        optime = self.get_node(self.leader_optime_path) if self._OPTIME in nodes and self.fetch_cluster else None
-        self.last_leader_operation = 0 if optime is None else int(optime[0])
-        self._cluster = Cluster(initialize, leader, self.last_leader_operation, members, failover)
-
-    def _load_cluster(self):
-        if self.exhibitor and self.exhibitor.poll():
-            self.client.set_hosts(self.exhibitor.zookeeper_hosts)
-
-        if self.fetch_cluster:
-            try:
-                self.client.retry(self._inner_load_cluster)
-            except:
-                logger.exception('get_cluster')
-                self.session_listener(KazooState.LOST)
-                raise ZooKeeperError('ZooKeeper in not responding properly')
-
-    def _create(self, path, value, **kwargs):
-        try:
-            self.client.retry(self.client.create, path, value.encode('utf-8'), **kwargs)
-            return True
-        except:
-            return False
-
-    def attempt_to_acquire_leader(self):
-        ret = self._create(self.leader_path, self._name, makepath=True, ephemeral=True)
-        ret or logger.info('Could not take out TTL lock')
-        return ret
-
-    def set_failover_value(self, value, index=None):
-        try:
-            self.client.retry(self.client.set, self.failover_path, value.encode('utf-8'), version=index or -1)
-            return True
-        except NoNodeError:
-            return value == '' or (not index and self._create(self.failover_path, value))
-        except:
-            logging.exception('set_failover_value')
-            return False
-
-    def initialize(self, create_new=True, sysid=""):
-        return self._create(self.initialize_path, sysid, makepath=True) if create_new \
-            else self.client.retry(self.client.set, self.initialize_path,  sysid.encode("utf-8"))
-
-    def touch_member(self, data, ttl=None):
-        cluster = self.cluster
-        me = cluster and ([m for m in cluster.members if m.name == self._name] or [None])[0]
-        path = self.member_path
-        data = data.encode('utf-8')
-        create = not me
-        if me and self.client.client_id is not None and me.session != self.client.client_id[0]:
-            try:
-                self.client.retry(self.client.delete, path)
-            except NoNodeError:
-                pass
-            except:
-                return False
-            create = True
-
-        if not create and data == self._my_member_data:
-            return True
-
-        try:
-            if create:
-                self.client.retry(self.client.create, path, data, makepath=True, ephemeral=True)
-            else:
-                self.client.retry(self.client.set, path, data)
-            self._my_member_data = data
-            return True
-        except NodeExistsError:
-            try:
-                self.client.retry(self.client.set, path, data)
-                self._my_member_data = data
-                return True
-            except:
-                logger.exception('touch_member')
-        except:
-            logger.exception('touch_member')
-        return False
-
-    def take_leader(self):
-        return self.attempt_to_acquire_leader()
-
-    def write_leader_optime(self, last_operation):
-        last_operation = last_operation.encode('utf-8')
-        if last_operation != self.last_leader_operation:
-            self.last_leader_operation = last_operation
-            path = self.leader_optime_path
-            try:
-                self.client.retry(self.client.set, path, last_operation)
-            except NoNodeError:
-                try:
-                    self.client.retry(self.client.create, path, last_operation, makepath=True)
-                except:
-                    logger.exception('Failed to create %s', path)
-            except:
-                logger.exception('Failed to update %s', path)
-
-    def update_leader(self):
-        return True
-
-    def delete_leader(self):
-        self.client.restart()
-        self._my_member_data = None
-        return True
-
-    def _cancel_initialization(self):
-        node = self.get_node(self.initialize_path)
-        if node:
-            self.client.delete(self.initialize_path, version=node[1].version)
-
-    def cancel_initialization(self):
-        try:
-            self.client.retry(self._cancel_initialization)
-        except:
-            logger.exception("Unable to delete initialize key")
-
-    def watch(self, timeout):
-        if super(ZooKeeper, self).watch(timeout):
-            self.fetch_cluster = True
-        return self.fetch_cluster

patronictl.py

@@ -2,4 +2,4 @@
 from patroni.ctl import ctl
 
 if __name__ == '__main__':
-    ctl()
+    ctl(None)

postgres0.yml

@@ -1,99 +1,99 @@
-ttl: &ttl 30
-loop_wait: &loop_wait 10
-scope: &scope batman
+scope: batman
+#namespace: /service/
+name: postgresql0
+
 restapi:
   listen: 127.0.0.1:8008
   connect_address: 127.0.0.1:8008
-  auth: 'username:password'
 #  certfile: /etc/ssl/certs/ssl-cert-snakeoil.pem
 #  keyfile: /etc/ssl/private/ssl-cert-snakeoil.key
+#  authentication:
+#    username: username
+#    password: password
+
+# ctl:
+#   insecure: false # Allow connections to SSL sites without certs
+#   certfile: /etc/ssl/certs/ssl-cert-snakeoil.pem
+#   cacert: /etc/ssl/certs/ssl-cacert-snakeoil.pem
+
 etcd:
-  scope: *scope
-  ttl: *ttl
-  host: 127.0.0.1:4001
-  #discovery_srv: my-etcd.domain
-#zookeeper:
-#  scope: *scope
-#  session_timeout: *ttl
-#  reconnect_timeout: *loop_wait
-#  hosts:
-#    - 127.0.0.1:2181
-#    - 127.0.0.2:2181
-#  exhibitor:
-#    poll_interval: 300
-#    port: 8181
-#    hosts:
-#      - host1
-#      - host2
-#      - host3
+  host: 127.0.0.1:2379
+
+bootstrap:
+  # this section will be written into Etcd:/<namespace>/<scope>/config after initializing new cluster
+  # and all other cluster members will use it as a `global configuration`
+  dcs:
+    ttl: 30
+    loop_wait: 10
+    retry_timeout: 10
+    maximum_lag_on_failover: 1048576
+#    master_start_timeout: 300
+#    synchronous_mode: false
+    #standby_cluster:
+      #host: 127.0.0.1
+      #port: 1111
+      #primary_slot_name: patroni
+    postgresql:
+      use_pg_rewind: true
+#      use_slots: true
+      parameters:
+#        wal_level: hot_standby
+#        hot_standby: "on"
+#        wal_keep_segments: 8
+#        max_wal_senders: 10
+#        max_replication_slots: 10
+#        wal_log_hints: "on"
+#        archive_mode: "on"
+#        archive_timeout: 1800s
+#        archive_command: mkdir -p ../wal_archive && test ! -f ../wal_archive/%f && cp %p ../wal_archive/%f
+#      recovery_conf:
+#        restore_command: cp ../wal_archive/%f %p
+
+  # some desired options for 'initdb'
+  initdb:  # Note: It needs to be a list (some options need values, others are switches)
+  - encoding: UTF8
+  - data-checksums
+
+  pg_hba:  # Add following lines to pg_hba.conf after running 'initdb'
+  - 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
+
+  # Additional script to be launched after initial cluster creation (will be passed the connection URL as parameter)
+# post_init: /usr/local/bin/setup_cluster.sh
+
+  # Some additional users users which needs to be created after initializing new cluster
+  users:
+    admin:
+      password: admin
+      options:
+        - createrole
+        - createdb
+
 postgresql:
-  name: postgresql0
-  scope: *scope
   listen: 127.0.0.1:5432
   connect_address: 127.0.0.1:5432
   data_dir: data/postgresql0
-  maximum_lag_on_failover: 1048576 # 1 megabyte in bytes
-  use_slots: True
+#  bin_dir:
+#  config_dir:
   pgpass: /tmp/pgpass0
-  initdb:  ## We allow the following options to be passed on to initdb
-  # - auth: authmethod
-  # - auth-host: authmethod
-  # - auth-local: authmethod
-  - encoding: UTF8
-  # - data-checksums  # When pg_rewind is needed on 9.3, this needs to be enabled
-  # - locale: locale
-  # - lc-collate: locale
-  # - lc-ctype: locale
-  # - lc-messages: locale
-  # - lc-monetary: locale
-  # - lc-numeric: locale
-  # - lc-time: locale
-  # - text-search-config: CFG
-  # - xlogdir: directory
-  # - debug
-  # - noclean
-  pg_rewind:
-    username: postgres
-    password: zalando
-  pg_hba:
-  - host all all 0.0.0.0/0 md5
-  - hostssl all all 0.0.0.0/0 md5
-  replication:
-    username: replicator
-    password: rep-pass
-    network:  127.0.0.1/32
-  superuser:
-    username: postgres
-    password: zalando
-  admin:
-    username: admin
-    password: admin
-  create_replica_method:
-    - basebackup
-#   - wal_e
-# commented-out example for wal-e provisioning
-  #create_replica_method: wal_e, basebackup
-  #wal_e:
-    #command: /patroni/scripts/wale_restore.py
-    #env_dir: /etc/wal-e.d/env
-    #threshold_megabytes: 10240
-    #threshold_backup_size_percentage: 30
-    #retries: 2
-    #use_iam: 1
-  #recovery_conf:
-    #restore_command: envdir /etc/wal-e.d/env wal-e wal-fetch "%f" "%p" -p 1
+  authentication:
+    replication:
+      username: replicator
+      password: rep-pass
+    superuser:
+      username: postgres
+      password: zalando
   parameters:
-    archive_mode: "on"
-    wal_level: hot_standby
-    archive_command: mkdir -p ../wal_archive && cp %p ../wal_archive/%f
-    max_wal_senders: 5
-    wal_keep_segments: 8
-    archive_timeout: 1800s
-    max_replication_slots: 5
-    hot_standby: "on"
-    wal_log_hints: "on"
+    unix_socket_directories: '.'
+
+#watchdog:
+#  mode: automatic # Allowed values: off, automatic, required
+#  device: /dev/watchdog
+#  safety_margin: 5
+
 tags:
-    nofailover: False
-    noloadbalance: False
-    clonefrom: False
-    replicatefrom: 127.0.0.1
+    nofailover: false
+    noloadbalance: false
+    clonefrom: false
+    nosync: false

postgres1.yml

@@ -1,99 +1,89 @@
-ttl: &ttl 30
-loop_wait: &loop_wait 10
-scope: &scope batman
+scope: batman
+#namespace: /service/
+name: postgresql1
+
 restapi:
   listen: 127.0.0.1:8009
   connect_address: 127.0.0.1:8009
-  auth: 'username:password'
 #  certfile: /etc/ssl/certs/ssl-cert-snakeoil.pem
 #  keyfile: /etc/ssl/private/ssl-cert-snakeoil.key
+#  authentication:
+#    username: username
+#    password: password
+
+# ctl:
+#   insecure: false # Allow connections to SSL sites without certs
+#   certfile: /etc/ssl/certs/ssl-cert-snakeoil.pem
+#   cacert: /etc/ssl/certs/ssl-cacert-snakeoil.pem
+
 etcd:
-  scope: *scope
-  ttl: *ttl
-  host: 127.0.0.1:4001
-  #discovery_srv: my-etcd.domain
-#zookeeper:
-#  scope: *scope
-#  session_timeout: *ttl
-#  reconnect_timeout: *loop_wait
-#  hosts:
-#    - 127.0.0.1:2181
-#    - 127.0.0.2:2181
-#  exhibitor:
-#    poll_interval: 300
-#    port: 8181
-#    hosts:
-#      - host1
-#      - host2
-#      - host3
+  host: 127.0.0.1:2379
+
+bootstrap:
+  # this section will be written into Etcd:/<namespace>/<scope>/config after initializing new cluster
+  # and all other cluster members will use it as a `global configuration`
+  dcs:
+    ttl: 30
+    loop_wait: 10
+    retry_timeout: 10
+    maximum_lag_on_failover: 1048576
+    postgresql:
+      use_pg_rewind: true
+#      use_slots: true
+      parameters:
+#        wal_level: hot_standby
+#        hot_standby: "on"
+#        wal_keep_segments: 8
+#        max_wal_senders: 10
+#        max_replication_slots: 10
+#        wal_log_hints: "on"
+#        archive_mode: "on"
+#        archive_timeout: 1800s
+#        archive_command: mkdir -p ../wal_archive && test ! -f ../wal_archive/%f && cp %p ../wal_archive/%f
+#      recovery_conf:
+#        restore_command: cp ../wal_archive/%f %p
+
+  # some desired options for 'initdb'
+  initdb:  # Note: It needs to be a list (some options need values, others are switches)
+  - encoding: UTF8
+  - data-checksums
+
+  pg_hba:  # Add following lines to pg_hba.conf after running 'initdb'
+  - 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
+
+  # Additional script to be launched after initial cluster creation (will be passed the connection URL as parameter)
+# post_init: /usr/local/bin/setup_cluster.sh
+
+  # Some additional users users which needs to be created after initializing new cluster
+  users:
+    admin:
+      password: admin
+      options:
+        - createrole
+        - createdb
+
 postgresql:
-  name: postgresql1
-  scope: *scope
   listen: 127.0.0.1:5433
   connect_address: 127.0.0.1:5433
   data_dir: data/postgresql1
-  maximum_lag_on_failover: 1048576 # 1 megabyte in bytes
-  use_slots: True
+#  bin_dir:
+#  config_dir:
   pgpass: /tmp/pgpass1
-  initdb:  ## We allow the following options to be passed on to initdb
-  # - auth: authmethod
-  # - auth-host: authmethod
-  # - auth-local: authmethod
-  - encoding: UTF8
-  # - data-checksums  # When pg_rewind is needed on 9.3, this needs to be enabled
-  # - locale: locale
-  # - lc-collate: locale
-  # - lc-ctype: locale
-  # - lc-messages: locale
-  # - lc-monetary: locale
-  # - lc-numeric: locale
-  # - lc-time: locale
-  # - text-search-config: CFG
-  # - xlogdir: directory
-  # - debug
-  # - noclean
-  pg_rewind:
-    username: postgres
-    password: zalando
-  pg_hba:
-  - host all all 0.0.0.0/0 md5
-  - hostssl all all 0.0.0.0/0 md5
-  replication:
-    username: replicator
-    password: rep-pass
-    network: 127.0.0.1/32
-  superuser:
-    user: postgres
-    password: zalando
-  admin:
-    username: admin
-    password: admin
-# commented-out example for wal-e provisioning
-  create_replica_method:
-    - basebackup
-#   - wal_e
-# commented-out example for wal-e provisioning
-  #wal_e:
-    #command: /patroni/scripts/wale_restore.py
-    #env_dir: /home/postgres/etc/wal-e.d/env
-    #threshold_megabytes: 10240
-    #threshold_backup_size_percentage: 30
-    #retries: 2
-    #use_iam: 1
-  #recovery_conf:
-    #restore_command: envdir /etc/wal-e.d/env wal-e wal-fetch "%f" "%p" -p 1
+  authentication:
+    replication:
+      username: replicator
+      password: rep-pass
+    superuser:
+      username: postgres
+      password: zalando
   parameters:
-    archive_mode: "on"
-    wal_level: hot_standby
-    archive_command: mkdir -p ../wal_archive && cp %p ../wal_archive/%f
-    max_wal_senders: 5
-    wal_keep_segments: 8
-    archive_timeout: 1800s
-    max_replication_slots: 5
-    hot_standby: "on"
-    wal_log_hints: "on"
+    unix_socket_directories: '.'
+  basebackup:
+      - verbose
+      - max-rate: 100M
 tags:
-    nofailover: False
-    noloadbalance: False
-    clonefrom: False
-    replicatefrom: 127.0.0.1
+    nofailover: false
+    noloadbalance: false
+    clonefrom: false

postgres2.yml

@@ -0,0 +1,84 @@
+scope: batman
+#namespace: /service/
+name: postgresql2
+
+restapi:
+  listen: 127.0.0.1:8010
+  connect_address: 127.0.0.1:8010
+#  certfile: /etc/ssl/certs/ssl-cert-snakeoil.pem
+#  keyfile: /etc/ssl/private/ssl-cert-snakeoil.key
+  authentication:
+    username: username
+    password: password
+
+# ctl:
+#   insecure: false # Allow connections to SSL sites without certs
+#   certfile: /etc/ssl/certs/ssl-cert-snakeoil.pem
+#   cacert: /etc/ssl/certs/ssl-cacert-snakeoil.pem
+
+etcd:
+  host: 127.0.0.1:2379
+
+bootstrap:
+  # this section will be written into Etcd:/<namespace>/<scope>/config after initializing new cluster
+  # and all other cluster members will use it as a `global configuration`
+  dcs:
+    ttl: 30
+    loop_wait: 10
+    retry_timeout: 10
+    maximum_lag_on_failover: 1048576
+    postgresql:
+      use_pg_rewind: true
+#      use_slots: true
+      parameters:
+#        wal_level: hot_standby
+#        hot_standby: "on"
+#        wal_keep_segments: 8
+#        max_wal_senders: 10
+#        max_replication_slots: 10
+#        wal_log_hints: "on"
+#        archive_mode: "on"
+#        archive_timeout: 1800s
+#        archive_command: mkdir -p ../wal_archive && test ! -f ../wal_archive/%f && cp %p ../wal_archive/%f
+#      recovery_conf:
+#        restore_command: cp ../wal_archive/%f %p
+
+  # some desired options for 'initdb'
+  initdb:  # Note: It needs to be a list (some options need values, others are switches)
+  - encoding: UTF8
+  - data-checksums
+
+  pg_hba:  # Add following lines to pg_hba.conf after running 'initdb'
+  - 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
+
+  # Some additional users users which needs to be created after initializing new cluster
+  users:
+    admin:
+      password: admin
+      options:
+        - createrole
+        - createdb
+
+postgresql:
+  listen: 127.0.0.1:5434
+  connect_address: 127.0.0.1:5434
+  data_dir: data/postgresql2
+#  bin_dir:
+#  config_dir:
+  pgpass: /tmp/pgpass2
+  authentication:
+    replication:
+      username: replicator
+      password: rep-pass
+    superuser:
+      username: postgres
+      password: zalando
+  parameters:
+    unix_socket_directories: '.'
+tags:
+    nofailover: false
+    noloadbalance: false
+    clonefrom: false
+    replicatefrom: postgres1

requirements-py2.txt

@@ -1,11 +0,0 @@
-boto
-dnspython
-mock
-psycopg2>=2.6.1
-PyYAML
-requests
-six >= 1.7
-kazoo>=2.2.1
-python-etcd>=0.4.1
-click>=4.1
-prettytable>=0.7

requirements-py3.txt

@@ -1,11 +0,0 @@
-boto
-mock
-dnspython3
-psycopg2>=2.6.1
-PyYAML
-requests
-six
-kazoo>=2.2.1
-python-etcd>=0.4.1
-click>=4.1
-prettytable>=0.7

requirements.txt

@@ -0,0 +1,16 @@
+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
+click>=4.1
+prettytable>=0.7
+tzlocal
+python-dateutil
+psutil
+cdiff
+kubernetes>=2.0.0,<=7.0.0,!=4.0.*,!=5.0.*

setup.py

@@ -4,13 +4,12 @@
     Setup file for patroni
 """
 
-import sys
-import os
 import inspect
+import os
+import sys
 
-import setuptools
 from setuptools.command.test import test as TestCommand
-from setuptools import setup
+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')
@@ -25,6 +24,7 @@ def read_version(package):
         exec(fd.read(), data)
     return data['__version__']
 
+
 NAME = 'patroni'
 MAIN_PACKAGE = NAME
 SCRIPTS = 'scripts'
@@ -32,9 +32,10 @@ VERSION = read_version(MAIN_PACKAGE)
 DESCRIPTION = 'PostgreSQL High-Available orchestrator and CLI'
 LICENSE = 'The MIT License'
 URL = 'https://github.com/zalando/patroni'
-AUTHOR = 'Alexander Kukushkin, Oleksii Kliukin, Feike Steenbergen'
-AUTHOR_EMAIL = '[email protected], [email protected], [email protected]'
-KEYWORDS = 'etcd governor patroni postgresql postgres ha zookeeper streaming replication'
+AUTHOR = 'Alexander Kukushkin, Dmitrii Dolgov, Oleksii Kliukin'
+AUTHOR_EMAIL = '[email protected], [email protected], [email protected]'
+KEYWORDS = 'etcd governor patroni postgresql postgres ha haproxy confd' +\
+    ' zookeeper exhibitor consul streaming replication kubernetes k8s'
 
 COVERAGE_XML = True
 COVERAGE_HTML = False
@@ -43,16 +44,22 @@ JUNIT_XML = True
 # Add here all kinds of additional classifiers as defined under
 # https://pypi.python.org/pypi?%3Aaction=list_classifiers
 CLASSIFIERS = [
-    'Development Status :: 4 - Beta',
+    'Development Status :: 5 - Production/Stable',
     'Environment :: Console',
     'Intended Audience :: Developers',
     'Intended Audience :: System Administrators',
     'License :: OSI Approved :: MIT License',
+    'Operating System :: MacOS',
     'Operating System :: POSIX :: Linux',
+    'Operating System :: POSIX :: BSD :: FreeBSD',
+    'Operating System :: Microsoft :: Windows',
     'Programming Language :: Python',
     'Programming Language :: Python :: 2.7',
-    'Programming Language :: Python :: 3.3',
+    'Programming Language :: Python :: 3',
     'Programming Language :: Python :: 3.4',
+    'Programming Language :: Python :: 3.5',
+    'Programming Language :: Python :: 3.6',
+    'Programming Language :: Python :: 3.7',
     'Programming Language :: Python :: Implementation :: CPython',
 ]
 
@@ -76,7 +83,7 @@ class PyTest(TestCommand):
     def finalize_options(self):
         TestCommand.finalize_options(self)
         if self.cov_xml or self.cov_html:
-            self.cov = ['--cov', MAIN_PACKAGE, '--cov', MAIN_PACKAGE, '--cov-report', 'term-missing']
+            self.cov = ['--cov', MAIN_PACKAGE, '--cov-report', 'term-missing']
             if self.cov_xml:
                 self.cov.extend(['--cov-report', 'xml'])
             if self.cov_html:
@@ -87,14 +94,21 @@ class PyTest(TestCommand):
     def run_tests(self):
         try:
             import pytest
-        except:
+        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, '-s', '-vv']
+        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)
         sys.exit(errno)
 
@@ -110,14 +124,23 @@ def read(fname):
 
 def setup_package():
     # Assemble additional setup commands
-    cmdclass = {}
-    cmdclass['test'] = PyTest
+    cmdclass = {'test': PyTest}
 
     # Some helper variables
     version = os.getenv('GO_PIPELINE_LABEL', VERSION)
 
-    requirements = 'requirements-py2.txt' if sys.version_info[0] == 2 else 'requirements-py3.txt'
-    install_reqs = get_install_requirements(requirements)
+    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'):
+        extra = False
+        for e, v in extras_require.items():
+            if r.startswith(v[0]):
+                extras_require[e] = [r]
+                extra = True
+        if not extra:
+            install_requires.append(r)
 
     command_options = {'test': {'test_suite': ('setup.py', 'tests')}}
     if JUNIT_XML:
@@ -139,12 +162,12 @@ def setup_package():
         long_description=read('README.rst'),
         classifiers=CLASSIFIERS,
         test_suite='tests',
-        packages=setuptools.find_packages(exclude=['tests', 'tests.*']),
+        packages=find_packages(exclude=['tests', 'tests.*']),
         package_data={MAIN_PACKAGE: ["*.json"]},
-        install_requires=install_reqs,
-        setup_requires=['six', 'flake8'],
+        install_requires=install_requires,
+        extras_require=extras_require,
         cmdclass=cmdclass,
-        tests_require=['pytest-cov', 'pytest'],
+        tests_require=['flake8', 'mock>=2.0.0', 'pytest-cov', 'pytest'],
         command_options=command_options,
         entry_points={'console_scripts': CONSOLE_SCRIPTS},
     )

tests/test_api.py

@@ -1,168 +1,406 @@
+import datetime
+import json
 import psycopg2
 import unittest
 
-from mock import Mock, patch
+from mock import Mock, PropertyMock, patch
 from patroni.api import RestApiHandler, RestApiServer
-from patroni.dcs import Member
+from patroni.dcs import ClusterConfig, Member
+from patroni.ha import _MemberStatus
+from patroni.utils import tzutc
 from six import BytesIO as IO
 from six.moves import BaseHTTPServer
-from six.moves.BaseHTTPServer import BaseHTTPRequestHandler
-import socket
 from test_postgresql import psycopg2_connect, MockCursor
 
 
-class MockPostgresql(Mock):
+future_restart_time = datetime.datetime.now(tzutc) + datetime.timedelta(days=5)
+postmaster_start_time = datetime.datetime.now(tzutc)
+
+
+class MockPostgresql(object):
 
     name = 'test'
     state = 'running'
     role = 'master'
+    server_version = '999999'
+    sysid = 'dummysysid'
+    scope = 'dummy'
+    pending_restart = True
+    wal_name = 'wal'
+    lsn_name = 'lsn'
 
-    def connection(self):
+    @staticmethod
+    def connection():
         return psycopg2_connect()
 
-    def is_running(self):
-        return True
+    @staticmethod
+    def postmaster_start_time():
+        return str(postmaster_start_time)
+
+    @staticmethod
+    def replica_cached_timeline(_):
+        return 2
 
 
-class MockHa(Mock):
+class MockWatchdog(object):
+    is_healthy = False
+
+
+class MockHa(object):
 
-    dcs = Mock()
     state_handler = MockPostgresql()
+    watchdog = MockWatchdog()
 
-    def schedule_restart(self):
-        return 'restart'
-
-    def schedule_reinitialize(self):
-        return 'reinitialize'
-
-    def restart(self):
-        return (True, '')
-
-    def restart_scheduled(self):
+    @staticmethod
+    def is_leader():
         return False
 
-    def fetch_nodes_statuses(self, members):
-        return [[None, True, None, None, {}]]
+    @staticmethod
+    def reinitialize(_):
+        return 'reinitialize'
+
+    @staticmethod
+    def restart():
+        return (True, '')
+
+    @staticmethod
+    def restart_scheduled():
+        return False
+
+    @staticmethod
+    def delete_future_restart():
+        return True
+
+    @staticmethod
+    def fetch_nodes_statuses(members):
+        return [_MemberStatus(None, True, None, 0, None, {}, False)]
+
+    @staticmethod
+    def schedule_future_restart(data):
+        return True
+
+    @staticmethod
+    def is_lagging(wal):
+        return False
+
+    @staticmethod
+    def get_effective_tags():
+        return {'nosync': True}
+
+    @staticmethod
+    def wakeup():
+        pass
+
+    @staticmethod
+    def is_paused():
+        return True
+
+    @staticmethod
+    def is_standby_cluster():
+        return False
 
 
-class MockPatroni:
+class MockPatroni(object):
 
-    postgresql = MockPostgresql()
     ha = MockHa()
+    config = Mock()
+    postgresql = ha.state_handler
     dcs = Mock()
     tags = {}
+    version = '0.00'
+    noloadbalance = PropertyMock(return_value=False)
+    scheduled_restart = {'schedule': future_restart_time,
+                         'postmaster_start_time': postgresql.postmaster_start_time()}
+
+    @staticmethod
+    def sighup_handler():
+        pass
 
 
-class MockRequest:
+class MockRequest(object):
 
-    def __init__(self, path):
-        self.path = path
+    def __init__(self, request):
+        self.request = request.encode('utf-8')
 
     def makefile(self, *args, **kwargs):
-        return IO(self.path)
+        return IO(self.request)
+
+    def sendall(self, *args, **kwargs):
+        pass
 
 
 class MockRestApiServer(RestApiServer):
 
-    def __init__(self, Handler, path):
+    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 = {'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'}
         super(MockRestApiServer, self).__init__(MockPatroni(), config)
-        Handler(MockRequest(path), ('0.0.0.0', 8080), self)
+        Handler(MockRequest(request), ('0.0.0.0', 8080), self)
 
 
 @patch('ssl.wrap_socket', Mock(return_value=0))
 class TestRestApiHandler(unittest.TestCase):
 
+    _authorization = '\nAuthorization: Basic dGVzdDp0ZXN0'
+
     def test_do_GET(self):
-        MockRestApiServer(RestApiHandler, b'GET /replica')
+        MockRestApiServer(RestApiHandler, 'GET /replica')
         with patch.object(RestApiHandler, 'get_postgresql_status', Mock(return_value={})):
-            MockRestApiServer(RestApiHandler, b'GET /replica')
+            MockRestApiServer(RestApiHandler, 'GET /replica')
         with patch.object(RestApiHandler, 'get_postgresql_status', Mock(return_value={'role': 'master'})):
-            MockRestApiServer(RestApiHandler, b'GET /replica')
-        MockRestApiServer(RestApiHandler, b'GET /master')
-        MockPatroni.dcs.cluster.leader.name = MockPostgresql.name
-        MockRestApiServer(RestApiHandler, b'GET /replica')
+            MockRestApiServer(RestApiHandler, 'GET /replica')
+        MockRestApiServer(RestApiHandler, 'GET /master')
+        MockPatroni.dcs.cluster.sync.sync_standby = MockPostgresql.name
+        MockPatroni.dcs.cluster.is_synchronous_mode = Mock(return_value=True)
+        with patch.object(RestApiHandler, 'get_postgresql_status', Mock(return_value={'role': 'replica'})):
+            MockRestApiServer(RestApiHandler, 'GET /synchronous')
+        with patch.object(RestApiHandler, 'get_postgresql_status', Mock(return_value={'role': 'replica'})):
+            MockPatroni.dcs.cluster.sync.sync_standby = ''
+            MockRestApiServer(RestApiHandler, 'GET /asynchronous')
+        MockPatroni.ha.is_leader = Mock(return_value=True)
+        MockRestApiServer(RestApiHandler, 'GET /replica')
         MockPatroni.dcs.cluster = None
         with patch.object(RestApiHandler, 'get_postgresql_status', Mock(return_value={'role': 'master'})):
-            MockRestApiServer(RestApiHandler, b'GET /master')
+            MockRestApiServer(RestApiHandler, 'GET /master')
         with patch.object(MockHa, 'restart_scheduled', Mock(return_value=True)):
-            MockRestApiServer(RestApiHandler, b'GET /master')
-        MockRestApiServer(RestApiHandler, b'GET /master')
+            MockRestApiServer(RestApiHandler, 'GET /master')
+        self.assertIsNotNone(MockRestApiServer(RestApiHandler, 'GET /master'))
+        with patch.object(RestApiServer, 'query', Mock(return_value=[('', 1, '', '', '', '', False, '')])):
+            self.assertIsNotNone(MockRestApiServer(RestApiHandler, 'GET /patroni'))
+        with patch.object(MockHa, 'is_standby_cluster', Mock(return_value=True)):
+            MockRestApiServer(RestApiHandler, 'GET /standby_leader')
 
     def test_do_OPTIONS(self):
-        MockRestApiServer(RestApiHandler, b'OPTIONS / HTTP/1.0')
-
-        with patch.object(BaseHTTPRequestHandler, 'handle_one_request') as mock_handle_request:
-            mock_handle_request.side_effect = socket.error("foo")
-            MockRestApiServer(RestApiHandler, b'OPTIONS / HTTP/1.0')
-
-        # make sure socket.error gets propagated via wfile object in finalize()
-        with patch.object(MockRequest, 'makefile') as makefile:
-            makefile.return_value.closed = False
-            makefile.return_value.readline.side_effect = lambda x: b"foo"
-            makefile.return_value.flush = Mock(side_effect=socket.error("foo"))
-            MockRestApiServer(RestApiHandler, b'OPTIONS / HTTP/1.0')
+        self.assertIsNotNone(MockRestApiServer(RestApiHandler, 'OPTIONS / HTTP/1.0'))
 
+    @patch.object(MockPostgresql, 'state', PropertyMock(return_value='stopped'))
     def test_do_GET_patroni(self):
-        MockRestApiServer(RestApiHandler, b'GET /patroni')
+        self.assertIsNotNone(MockRestApiServer(RestApiHandler, 'GET /patroni'))
 
     def test_basicauth(self):
-        MockRestApiServer(RestApiHandler, b'POST /restart HTTP/1.0')
-        MockRestApiServer(RestApiHandler, b'POST /restart HTTP/1.0\nAuthorization:')
+        self.assertIsNotNone(MockRestApiServer(RestApiHandler, 'POST /restart HTTP/1.0'))
+        MockRestApiServer(RestApiHandler, 'POST /restart HTTP/1.0\nAuthorization:')
 
-    def test_do_POST_restart(self):
-        request = b'POST /restart HTTP/1.0\nAuthorization: Basic dGVzdDp0ZXN0'
+    @patch.object(MockPatroni, 'dcs')
+    def test_do_GET_config(self, mock_dcs):
+        mock_dcs.cluster.config.data = {}
+        self.assertIsNotNone(MockRestApiServer(RestApiHandler, 'GET /config'))
+        mock_dcs.cluster.config = None
+        self.assertIsNotNone(MockRestApiServer(RestApiHandler, 'GET /config'))
+
+    @patch.object(MockPatroni, 'dcs')
+    def test_do_PATCH_config(self, mock_dcs):
+        config = {'postgresql': {'use_slots': False, 'use_pg_rewind': True, 'parameters': {'wal_level': 'logical'}}}
+        mock_dcs.get_cluster.return_value.config = ClusterConfig.from_node(1, json.dumps(config))
+        request = 'PATCH /config HTTP/1.0' + self._authorization
+        self.assertIsNotNone(MockRestApiServer(RestApiHandler, request))
+        request += '\nContent-Length: '
+        self.assertIsNotNone(MockRestApiServer(RestApiHandler, request + '34\n\n{"postgresql":{"use_slots":false}}'))
+        config['ttl'] = 5
+        config['postgresql'].update({'use_slots': {'foo': True}, "parameters": None})
+        config = json.dumps(config)
+        request += str(len(config)) + '\n\n' + config
         MockRestApiServer(RestApiHandler, request)
+        mock_dcs.set_config_value.return_value = False
+        MockRestApiServer(RestApiHandler, request)
+
+    @patch.object(MockPatroni, 'dcs')
+    def test_do_PUT_config(self, mock_dcs):
+        mock_dcs.get_cluster.return_value.config = ClusterConfig.from_node(1, '{}')
+        request = 'PUT /config HTTP/1.0' + self._authorization + '\nContent-Length: '
+        self.assertIsNotNone(MockRestApiServer(RestApiHandler, request + '2\n\n{}'))
+        config = '{"foo": "bar"}'
+        request += str(len(config)) + '\n\n' + config
+        MockRestApiServer(RestApiHandler, request)
+        mock_dcs.set_config_value.return_value = False
+        MockRestApiServer(RestApiHandler, request)
+        mock_dcs.get_cluster.return_value.config = ClusterConfig.from_node(1, config)
+        MockRestApiServer(RestApiHandler, request)
+
+    @patch.object(MockPatroni, 'sighup_handler', Mock(side_effect=Exception))
+    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')
+    def test_do_POST_restart(self, mock_dcs):
+        mock_dcs.get_cluster.return_value.is_paused.return_value = False
+        request = 'POST /restart HTTP/1.0' + self._authorization
+        self.assertIsNotNone(MockRestApiServer(RestApiHandler, request))
+
         with patch.object(MockHa, 'restart', Mock(side_effect=Exception)):
             MockRestApiServer(RestApiHandler, request)
 
-    @patch.object(MockHa, 'dcs')
-    def test_do_POST_reinitialize(self, dcs):
-        cluster = dcs.get_cluster.return_value
-        request = b'POST /reinitialize HTTP/1.0\nAuthorization: Basic dGVzdDp0ZXN0'
+        post = request + '\nContent-Length: '
+
+        def make_request(request=None, **kwargs):
+            request = json.dumps(kwargs) if request is None else request
+            return '{0}{1}\n\n{2}'.format(post, len(request), request)
+
+        # empty request
+        request = make_request('')
         MockRestApiServer(RestApiHandler, request)
-        cluster.is_unlocked.return_value = False
+        # invalid request
+        request = make_request('foobar=baz')
         MockRestApiServer(RestApiHandler, request)
-        with patch.object(MockHa, 'schedule_reinitialize', Mock(return_value=None)):
+        # wrong role
+        request = make_request(schedule=future_restart_time.isoformat(), role='unknown', postgres_version='9.5.3')
+        MockRestApiServer(RestApiHandler, request)
+        # wrong version
+        request = make_request(schedule=future_restart_time.isoformat(), role='master', postgres_version='9.5.3.1')
+        MockRestApiServer(RestApiHandler, request)
+        # unknown filter
+        request = make_request(schedule=future_restart_time.isoformat(), batman='lives')
+        MockRestApiServer(RestApiHandler, request)
+        # incorrect schedule
+        request = make_request(schedule='2016-08-42 12:45TZ+1', role='master')
+        MockRestApiServer(RestApiHandler, request)
+        # everything fine, but the schedule is missing
+        request = make_request(role='master', postgres_version='9.5.2')
+        MockRestApiServer(RestApiHandler, request)
+        for retval in (True, False):
+            with patch.object(MockHa, 'schedule_future_restart', Mock(return_value=retval)):
+                request = make_request(schedule=future_restart_time.isoformat())
+                MockRestApiServer(RestApiHandler, request)
+            with patch.object(MockHa, 'restart', Mock(return_value=(retval, "foo"))):
+                request = make_request(role='master', postgres_version='9.5.2')
+                MockRestApiServer(RestApiHandler, request)
+
+        mock_dcs.get_cluster.return_value.is_paused.return_value = True
+        MockRestApiServer(RestApiHandler, make_request(schedule='2016-08-42 12:45TZ+1', role='master'))
+        # Valid timeout
+        MockRestApiServer(RestApiHandler, make_request(timeout='60s'))
+        # Invalid timeout
+        MockRestApiServer(RestApiHandler, make_request(timeout='42towels'))
+
+    def test_do_DELETE_restart(self):
+        for retval in (True, False):
+            with patch.object(MockHa, 'delete_future_restart', Mock(return_value=retval)):
+                request = 'DELETE /restart HTTP/1.0' + self._authorization
+                self.assertIsNotNone(MockRestApiServer(RestApiHandler, request))
+
+    @patch.object(MockPatroni, 'dcs')
+    def test_do_POST_reinitialize(self, mock_dcs):
+        cluster = mock_dcs.get_cluster.return_value
+        cluster.is_paused.return_value = False
+        request = 'POST /reinitialize HTTP/1.0' + self._authorization + '\nContent-Length: 15\n\n{"force": true}'
+        MockRestApiServer(RestApiHandler, request)
+        with patch.object(MockHa, 'reinitialize', Mock(return_value=None)):
             MockRestApiServer(RestApiHandler, request)
-        cluster.leader.name = 'test'
-        MockRestApiServer(RestApiHandler, request)
 
     @patch('time.sleep', Mock())
     def test_RestApiServer_query(self):
         with patch.object(MockCursor, 'execute', Mock(side_effect=psycopg2.OperationalError)):
-            MockRestApiServer(RestApiHandler, b'GET /patroni')
+            self.assertIsNotNone(MockRestApiServer(RestApiHandler, 'GET /patroni'))
         with patch.object(MockPostgresql, 'connection', Mock(side_effect=psycopg2.OperationalError)):
-            MockRestApiServer(RestApiHandler, b'GET /patroni')
+            self.assertIsNotNone(MockRestApiServer(RestApiHandler, 'GET /patroni'))
 
     @patch('time.sleep', Mock())
-    @patch.object(MockHa, 'dcs')
-    def test_do_POST_failover(self, dcs):
+    @patch.object(MockPatroni, 'dcs')
+    def test_do_POST_switchover(self, dcs):
+        dcs.loop_wait = 10
         cluster = dcs.get_cluster.return_value
-        request = b'POST /failover HTTP/1.0\nAuthorization: Basic dGVzdDp0ZXN0\n' +\
-                  b'Content-Length: 25\n\n{"leader": "postgresql1"}'
+        cluster.is_synchronous_mode.return_value = False
+        cluster.is_paused.return_value = False
+
+        post = 'POST /switchover HTTP/1.0' + self._authorization + '\nContent-Length: '
+
+        MockRestApiServer(RestApiHandler, post + '7\n\n{"1":2}')
+
+        request = post + '0\n\n'
         MockRestApiServer(RestApiHandler, request)
+
         cluster.leader.name = 'postgresql1'
         MockRestApiServer(RestApiHandler, request)
-        cluster.members = [Member(0, 'postgresql0', 30, {'api_url': 'http'})]
+
+        request = post + '25\n\n{"leader": "postgresql1"}'
+
+        cluster.is_paused.return_value = True
         MockRestApiServer(RestApiHandler, request)
-        with patch.object(MockPatroni, 'dcs') as d:
-            cluster = d.get_cluster.return_value
-            cluster.leader.name = 'postgresql0'
+
+        cluster.is_paused.return_value = False
+        for cluster.is_synchronous_mode.return_value in (True, False):
             MockRestApiServer(RestApiHandler, request)
-            cluster.leader.name = 'postgresql1'
-            cluster.failover = None
-            MockRestApiServer(RestApiHandler, request)
-            d.get_cluster = Mock(side_effect=Exception())
-            MockRestApiServer(RestApiHandler, request)
-            d.manual_failover.return_value = False
+
+        cluster.leader.name = 'postgresql2'
+        request = post + '53\n\n{"leader": "postgresql1", "candidate": "postgresql2"}'
+        MockRestApiServer(RestApiHandler, request)
+
+        cluster.leader.name = 'postgresql1'
+        for cluster.is_synchronous_mode.return_value in (True, False):
             MockRestApiServer(RestApiHandler, request)
+
+        cluster.members = [Member(0, 'postgresql0', 30, {'api_url': 'http'}),
+                           Member(0, 'postgresql2', 30, {'api_url': 'http'})]
+        MockRestApiServer(RestApiHandler, request)
+
+        cluster.failover = None
+        MockRestApiServer(RestApiHandler, request)
+
+        dcs.get_cluster.side_effect = [cluster]
+        MockRestApiServer(RestApiHandler, request)
+
+        cluster2 = cluster.copy()
+        cluster2.leader.name = 'postgresql0'
+        cluster2.is_unlocked.return_value = False
+        dcs.get_cluster.side_effect = [cluster, cluster2]
+        MockRestApiServer(RestApiHandler, request)
+
+        cluster2.leader.name = 'postgresql2'
+        dcs.get_cluster.side_effect = [cluster, cluster2]
+        MockRestApiServer(RestApiHandler, request)
+
+        dcs.get_cluster.side_effect = None
+        dcs.manual_failover.return_value = False
+        MockRestApiServer(RestApiHandler, request)
+        dcs.manual_failover.return_value = True
+
         with patch.object(MockHa, 'fetch_nodes_statuses', Mock(return_value=[])):
             MockRestApiServer(RestApiHandler, request)
-        request = b'POST /failover HTTP/1.0\nAuthorization: Basic dGVzdDp0ZXN0\n' +\
-                  b'Content-Length: 50\n\n{"leader": "postgresql1", "member": "postgresql2"}'
+
+        # Valid future date
+        request = post + '103\n\n{"leader": "postgresql1", "member": "postgresql2",' +\
+                         ' "scheduled_at": "6016-02-15T18:13:30.568224+01:00"}'
         MockRestApiServer(RestApiHandler, request)
+        with patch.object(MockPatroni, 'dcs') as d:
+            d.manual_failover.return_value = False
+            MockRestApiServer(RestApiHandler, request)
+
+        # Exception: No timezone specified
+        request = post + '97\n\n{"leader": "postgresql1", "member": "postgresql2",' +\
+                         ' "scheduled_at": "6016-02-15T18:13:30.568224"}'
+        MockRestApiServer(RestApiHandler, request)
+
+        # Exception: Scheduled in the past
+        request = post + '103\n\n{"leader": "postgresql1", "member": "postgresql2", "scheduled_at": "'
+        MockRestApiServer(RestApiHandler, request + '1016-02-15T18:13:30.568224+01:00"}')
+
+        # Invalid date
+        self.assertIsNotNone(MockRestApiServer(RestApiHandler, request + '2010-02-29T18:13:30.568224+01:00"}'))
+
+    @patch.object(MockPatroni, 'dcs', Mock())
+    def test_do_POST_failover(self):
+        post = 'POST /failover HTTP/1.0' + self._authorization + '\nContent-Length: '
+        MockRestApiServer(RestApiHandler, post + '14\n\n{"leader":"1"}')
+        MockRestApiServer(RestApiHandler, post + '37\n\n{"candidate":"2","scheduled_at": "1"}')
+
+
+@patch('ssl.wrap_socket', Mock(return_value=0))
+class TestRestApiServer(unittest.TestCase):
+
+    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'})
+
+    def test_handle_error(self):
+        self.assertIsNone(MockRestApiServer.handle_error(None, ('127.0.0.1', 55555)))