Release 1.4.4 (#683)

bump version and update release notes

.travis.yml

@@ -1,17 +1,28 @@
-sudo: false
+sudo: true
 dist: trusty
 language: python
-python:
-- "3.4"  # 2.7 and 3.5 are preinstalled by default
 env:
   global:
-  - ETCDVERSION=3.0.17 ZKVERSION=3.4.9 CONSULVERSION=0.7.4
-  - PYVERSIONS="2.7 3.4 3.5"
-  matrix:
-  - TEST_SUITE="python setup.py"
-  - DCS="etcd" TEST_SUITE="behave"
-  - DCS="exhibitor" TEST_SUITE="behave"
-  - DCS="consul" TEST_SUITE="behave"
+  - 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
@@ -19,7 +30,7 @@ before_cache:
 - |
   rm -fr $HOME/mycache/python*
   for pv in $PYVERSIONS; do
-      if [[ $TEST_SUITE != "behave" || $pv != "3.4" ]]; then
+      if [[ $TEST_SUITE != "behave" || $pv != $EXCLUDE_BEHAVE ]]; then
           fpv=$(basename $(readlink $HOME/virtualenv/python${pv}))
           mv $HOME/virtualenv/${fpv} $HOME/mycache/${fpv}
       fi
@@ -51,6 +62,26 @@ install:
             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
@@ -76,7 +107,7 @@ install:
     fi
 
     for pv in $PYVERSIONS; do
-        if [[ $TEST_SUITE != "behave" || $pv != "3.4" ]]; then
+        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
@@ -92,13 +123,17 @@ install:
 script:
   - |
     for pv in $PYVERSIONS; do
+        if [[ $TEST_SUITE == "behave" && $pv == $EXCLUDE_BEHAVE ]]; then
+            continue
+        fi
+
         source ~/virtualenv/python${pv}/bin/activate
 
         if [[ $TEST_SUITE != "behave" ]]; then
             echo Running unit tests using python${pv}
             $TEST_SUITE test
             $TEST_SUITE flake8
-        elif [[ $pv != "3.4" ]]; then
+        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
@@ -110,5 +145,9 @@ script:
 
     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

@@ -3,14 +3,17 @@
 FROM postgres:9.6
 MAINTAINER Alexander Kukushkin <[email protected]>
 
-RUN echo 'APT::Install-Recommends "0";\nAPT::Install-Suggests "0";' > /etc/apt/apt.conf.d/01norecommend \
+RUN export DEBIAN_FRONTEND=noninteractive \
+    && echo 'APT::Install-Recommends "0";\nAPT::Install-Suggests "0";' > /etc/apt/apt.conf.d/01norecommend \
     && apt-get update -y \
     && apt-get upgrade -y \
-    && apt-get install -y curl jq haproxy python-psycopg2 python-yaml python-requests \
-        python-six python-dateutil python-urllib3 python-dnspython \
-        python-pip python-setuptools python-kazoo python-prettytable python-wheel python \
+    && apt-get install -y curl jq haproxy python-psycopg2 python-yaml python-requests python-six python-pysocks \
+        python-dateutil python-pip python-setuptools python-prettytable python-wheel python-psutil python locales \
 
-    && pip install python-etcd==0.4.3 python-consul==0.7.0 click tzlocal --upgrade \
+    ## Make sure we have a en_US.UTF-8 locale available
+    && localedef -i en_US -c -f UTF-8 -A /usr/share/locale/locale.alias en_US.UTF-8 \
+
+    && pip install 'python-etcd>=0.4.3,<0.5' click tzlocal cdiff \
 
     && mkdir -p /home/postgres \
     && chown postgres:postgres /home/postgres \
@@ -21,7 +24,7 @@ RUN echo 'APT::Install-Recommends "0";\nAPT::Install-Suggests "0";' > /etc/apt/a
     && apt-get clean -y \
     && rm -rf /var/lib/apt/lists/* /root/.cache
 
-ENV ETCDVERSION 3.1.2
+ENV ETCDVERSION 3.2.3
 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
 
@@ -40,5 +43,6 @@ RUN mkdir /data/ && touch /pgpass /patroni.yml \
 
 EXPOSE 2379 5432 8008
 
+ENV LC_ALL=en_US.UTF-8 LANG=en_US.UTF-8
 ENTRYPOINT ["/bin/bash", "/entrypoint.sh"]
 USER postgres

README.rst

@@ -1,27 +1,27 @@
 |Build Status| |Coverage Status|
 
 Patroni: A Template for PostgreSQL HA with ZooKeeper, etcd or Consul
-------------------------------------------------------------
+--------------------------------------------------------------------
 
 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; 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>`__ or `Consul <https://github.com/hashicorp/consul>`__. 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.
+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**: We're currently developing Patroni to be as useful as possible for teams running Kubernetes on top of Google Compute Engine; Patroni can be the HA solution for Postgres in such an environment. To this end, there is a `Helm chart <https://github.com/kubernetes/charts/tree/master/incubator/patroni>`__ that uses Patroni and  `Spilo  <https://github.com/zalando/spilo/>`__ to provision a five-node PostgreSQL HA cluster in a Kubernetes+GCE environment. (The Helm chart deploys Spilo Docker images, not just "bare" Patroni.)
+**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.
 
@@ -29,34 +29,70 @@ For an example of a Docker-based deployment with Patroni, see `Spilo <https://gi
 
 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 
+* `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>`__.
 
-===========================
-Technical Requirements/Installation
-===========================
+=========
+Community
+=========
 
-**For Mac**
+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
-    pip install psycopg2 pyyaml
 
-===================
+**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:
 ::
@@ -80,9 +116,9 @@ run:
 
     > psql --host 127.0.0.1 --port 5000 postgres
 
-===============
+==================
 YAML Configuration
-===============
+==================
 
 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>`__.
 
@@ -92,15 +128,15 @@ Environment Configuration
 
 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, 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 undesirable.
 

docker/entrypoint.sh

@@ -22,7 +22,7 @@ __EOF__
 
 DOCKER_IP=$(hostname --ip-address)
 PATRONI_SCOPE=${PATRONI_SCOPE:-batman}
-ETCD_ARGS="--data-dir /tmp/etcd.data -advertise-client-urls=http://${DOCKER_IP}:2379 -listen-client-urls=http://0.0.0.0:2379 -listen-peer-urls=http://0.0.0.0:2380"
+ETCD_ARGS="--data-dir /tmp/etcd.data -advertise-client-urls=http://${DOCKER_IP}:2379 -listen-client-urls=http://0.0.0.0:2379"
 
 optspec=":vh-:"
 while getopts "$optspec" optchar; do

docs/CONTRIBUTING.rst

@@ -5,6 +5,11 @@ 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
 ----------------
 

docs/ENVIRONMENT.rst

@@ -1,6 +1,5 @@
 .. _environment:
 
-==================================
 Environment Configuration Settings
 ==================================
 
@@ -12,6 +11,8 @@ Global/Universal
 -  **PATRONI\_NAME**: name of the node where the current instance of Patroni is running. Must be unique for the cluster.
 -  **PATRONI\_NAMESPACE**: path within the configuration store where Patroni will keep information about the cluster. Default value: "/service"
 -  **PATRONI\_SCOPE**: cluster name
+-  **PATRONI\_LOGLEVEL**: sets the general logging level (see `the docs for Python logging <https://docs.python.org/3.6/library/logging.html#levels>`_)
+-  **PATRONI\_REQUESTS_LOGLEVEL**: sets the logging level for all HTTP requests e.g. Kubernetes API calls (see `the docs for Python logging <https://docs.python.org/3.6/library/logging.html#levels>`_)
 
 Bootstrap configuration
 -----------------------
@@ -25,14 +26,25 @@ Example: defining ``PATRONI_admin_PASSWORD=strongpasswd`` and ``PATRONI_admin_OP
 Consul
 ------
 -  **PATRONI\_CONSUL\_HOST**: the host:port for the Consul endpoint.
+-  **PATRONI\_CONSUL\_URL**: url for the Consul, in format: http(s)://host:port
+-  **PATRONI\_CONSUL\_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.
 
 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 pressent it will enable validation.
+-  **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.
 
@@ -41,11 +53,24 @@ 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

docs/README.rst

@@ -1,8 +1,8 @@
 .. _readme:
 
-=================
-How Patroni Works
-=================
+============
+Introduction
+============
 
 Patroni originated as a fork of `Governor <https://github.com/compose/governor>`__, the project from Compose. It includes plenty of new features.
 
@@ -13,30 +13,65 @@ 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
-===================================
 
-**For Mac**
+Technical Requirements/Installation
+-----------------------------------
+
+**Pre-requirements for Mac OS**
 
 To install requirements on a Mac, run the following:
 
 ::
 
     brew install postgresql etcd haproxy libyaml python
-    pip install psycopg2 pyyaml
 
-=======================
+**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:
 ::
@@ -60,27 +95,27 @@ run:
 
     > 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.
 

docs/SETTINGS.rst

@@ -15,14 +15,19 @@ 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 succesfully committed transactions will not be lost at failover, at the cost of losing availability for writes when Patroni cannot ensure transaction durability. See `replication modes documentation <https://github.com/zalando/patroni/blob/master/docs/replication_modes.rst>`__ for details.
+    -  **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.
     -  **postgresql**:
-        -  **use\_pg\_rewind**:whether or not to use pg_rewind
+        -  **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.
+-  **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.
@@ -30,29 +35,45 @@ Bootstrap configuration
 -  **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 users which needs to be created after initializing new cluster
+-  **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_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.
+-  **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
 ------
--  **host**: the host:port for the Consul endpoint.
+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
 
 Etcd
 ----
-Most of the parameters are optional, but you have to specify one of the **host**, **url**, **proxy** or **srv**
+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 pressent it will enable validation.
+-  **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**.
 
@@ -62,6 +83,18 @@ Exhibitor
 -  **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
@@ -80,18 +113,25 @@ PostgreSQL
         -  **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.
+-  **create\_replica\_method**: 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.
--  **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.
+-  **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 overriden by Patroni's own configuration facilities - see `dynamic configuration <https://github.com/zalando/patroni/blob/master/docs/dynamic_configuration.rst>`__ for details.
+-  **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_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".
+-  **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".
 
 REST API
 -------- 
@@ -108,3 +148,9 @@ REST API
 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/conf.py

@@ -18,9 +18,11 @@
 # documentation root, use os.path.abspath to make it absolute, like shown here.
 #
 import os
-# import sys
-# sys.path.insert(0, os.path.abspath('.'))
 
+import sys
+sys.path.insert(0, os.path.abspath('..'))
+
+from patroni.version import __version__
 
 # -- General configuration ------------------------------------------------
 
@@ -51,7 +53,7 @@ master_doc = 'index'
 
 # General information about the project.
 project = 'Patroni'
-copyright = '2016, Zalando SE'
+copyright = '2015 Compose, Zalando SE'
 author = 'Zalando SE'
 
 # The version info for the project you're documenting, acts as replacement for
@@ -59,9 +61,9 @@ author = 'Zalando SE'
 # built documents.
 #
 # The short X.Y version.
-version = '1.2'
+version = __version__[:__version__.rfind('.')]
 # The full version, including alpha/beta/rc tags.
-release = '1.2.2'
+release = __version__
 
 # The language for content autogenerated by Sphinx. Refer to documentation
 # for a list of supported languages.

docs/dynamic_configuration.rst

@@ -16,9 +16,9 @@ Patroni configuration is stored in the DCS (Distributed Configuration Store). Th
 	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>` .
+- 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 you external IP address when you are running inside ``docker``).
+	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:
 
@@ -42,7 +42,7 @@ 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 ``PATRRONI_SCOPE`` 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>`__

docs/index.rst

@@ -6,11 +6,11 @@
 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>`__ or `Consul <https://github.com/hashicorp/consul>`__. 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.
+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**: We're currently developing Patroni to be as useful as possible for teams running Kubernetes on top of Google Compute Engine; Patroni can be the HA solution for Postgres in such an environment. To this end, we've created a `Helm Chart <https://github.com/kubernetes/charts/tree/master/incubator/patroni>`__ that enables you to deploy a five-node Patroni cluster using a Kubernetes PetSet.
+**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::
@@ -21,8 +21,11 @@ We call Patroni a "template" because it is far from being a one-size-fits-all or
    dynamic_configuration
    ENVIRONMENT
    SETTINGS
+   replica_bootstrap
    replication_modes
    pause
+   kubernetes
+   watchdog
    releases
    CONTRIBUTING
 
@@ -32,5 +35,3 @@ 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

@@ -25,7 +25,7 @@ When Patroni runs in a paused mode, it does not change the state of PostgreSQL,
 
 - 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 to stop Postgres instance it is managing.
+- 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
 ----------

docs/releases.rst

@@ -3,6 +3,497 @@
 Release notes
 =============
 
+Version 1.4.4
+-------------
+
+**Stability improvements**
+
+- Fix race condition in poll_failover_result (Alexander Kukushkin)
+
+  It didn't affect directly neither failover nor switchover, but in some rare cases it was reporting success too early, when the former leader released the lock, producing a 'Failed over to "None"' instead of 'Failed over to "desired-node"' message.
+
+- Treat Postgres parameter names as case insensitive (Alexander)
+
+  Most of the Postgres parameters have snake_case names, but there are three exceptions from this rule: DateStyle, IntervalStyle and TimeZone. Postgres accepts those parameters when written in a different case (e.g. timezone = 'some/tzn'); however, Patroni was unable to find case-insensitive matches of those parameter names in pg_settings and ignored such parameters as a result.
+
+- Abort start if attaching to running postgres and cluster not initialized (Alexander)
+
+  Patroni can attach itself to an already running Postgres instance. It is imperative to start running Patroni on the master node before getting to the replicas.
+
+- Fix behavior of patronictl scaffold (Alexander)
+
+  Pass dict object to touch_member instead of json encoded string, DCS implementation will take care of encoding it.
+
+- Don't demote master if failed to update leader key in pause (Alexander)
+
+  During maintenance a DCS may start failing write requests while continuing to responds to read ones. In that case, Patroni used to put the Postgres master node to a read-only mode after failing to update the leader lock in DCS.
+
+- Sync replication slots when Patroni notices a new postmaster process (Alexander)
+
+  If Postgres has been restarted, Patroni has to make sure that list of replication slots matches its expectations.
+
+- Verify sysid and sync replication slots after coming out of pause (Alexander)
+
+  During the `maintenance` mode it may happen that data directory was completely rewritten and therefore we have to make sure that `Database system identifier` still belongs to our cluster and replication slots are in sync with Patroni expectations.
+
+- Fix a possible failure to start not running Postgres on a data directory with postmaster lock file present (Alexander)
+
+  Detect reuse of PID from the postmaster lock file. More likely to hit such problem if you run Patroni and Postgres in the docker container.
+
+- Improve protection of DCS being accidentally wiped (Alexander)
+
+  Patroni has a lot of logic in place to prevent failover in such case; it can also restore all keys back; however, until this change an accidental removal of /config key was switching off pause mode for 1 cycle of HA loop.
+
+- Do not exit when encountering invalid system ID (Oleksii Kliukin)
+
+  Do not exit when the cluster system ID is empty or the one that doesn't pass the validation check. In that case, the cluster most likely needs a reinit; mention it in the result message. Avoid terminating Patroni, as otherwise reinit cannot happen.
+
+**Compatibility with Kubernetes 1.10+**
+
+- Added check for empty subsets (Cody Coons)
+
+  Kubernetes 1.10.0+ started returning `Endpoints.subsets` set to `None` instead of `[]`.
+
+**Bootstrap improvements**
+
+- Make deleting recovery.conf optional (Brad Nicholson)
+
+  If `bootstrap.<custom_bootstrap_method_name>.keep_existing_recovery_conf` is defined and set to ``True``, Patroni will not remove the existing ``recovery.conf`` file. This is useful when bootstrapping from a backup with tools like pgBackRest that generate the appropriate `recovery.conf` for you.
+
+- Allow options to the basebackup built-in method (Oleksii)
+
+  It is now possible to supply options to the built-in basebackup method by defining the `basebackup` section in the configuration, similar to how those are defined for custom replica creation methods. The difference is in the format accepted by the `basebackup` section: since pg_basebackup accepts both `--key=value` and `--key` options, the contents of the section could be either a dictionary of key-value pairs, or a list of either one-element dictionaries or just keys (for the options that don't accept values). See :ref:`replica creation method <custom_replica_creation>` section for additional examples.
+
+
+Version 1.4.3
+-------------
+
+**Improvements in logging**
+
+- Make log level configurable from environment variables (Andy Newton, Keyvan Hedayati)
+
+  `PATRONI_LOGLEVEL` - sets the general logging level
+  `PATRONI_REQUESTS_LOGLEVEL` - sets the logging level for all HTTP requests e.g. Kubernetes API calls
+  See `the docs for Python logging <https://docs.python.org/3.6/library/logging.html#levels>` to get the names of possible log levels
+
+**Stability improvements and bug fixes**
+
+- Don't rediscover etcd cluster topology when watch timed out (Alexander Kukushkin)
+
+  If we have only one host in etcd configuration and exactly this host is not accessible, Patroni was starting discovery of cluster topology and never succeeding. Instead it should just switch to the next available node.
+
+- Write content of bootstrap.pg_hba into a pg_hba.conf after custom bootstrap (Alexander)
+
+  Now it behaves similarly to the usual bootstrap with `initdb`
+
+- Single user mode was waiting for user input and never finish (Alexander)
+
+  Regression was introduced in https://github.com/zalando/patroni/pull/576
+
+
+Version 1.4.2
+-------------
+
+**Improvements in patronictl**
+
+- Rename scheduled failover to scheduled switchover (Alexander Kukushkin)
+
+  Failover and switchover functions were separated in version 1.4, but `patronictl list` was still reporting `Scheduled failover` instead of `Scheduled switchover`.
+
+- Show information about pending restarts (Alexander)
+
+  In order to apply some configuration changes sometimes it is necessary to restart postgres. Patroni was already giving a hint about that in the REST API and when writing node status into DCS, but there were no easy way to display it.
+
+- Make show-config to work with cluster_name from config file (Alexander)
+
+  It works similar to the `patronictl edit-config`
+
+**Stability improvements**
+
+- Avoid calling pg_controldata during bootstrap (Alexander)
+
+  During initdb or custom bootstrap there is a time window when pgdata is not empty but pg_controldata has not been written yet. In such case pg_controldata call was failing with error messages.
+
+- Handle exceptions raised from psutil (Alexander)
+
+  cmdline is read and parsed every time when `cmdline()` method is called. It could happen that the process being examined
+  has already disappeared, in that case `NoSuchProcess` is raised.
+
+**Kubernetes support improvements**
+
+- Don't swallow errors from k8s API (Alexander)
+
+  A call to Kubernetes API could fail for a different number of reasons. In some cases such call should be retried, in some other cases we should log the error message and the exception stack trace. The change here will help debug Kubernetes permission issues.
+
+- Update Kubernetes example Dockerfile to install Patroni from the master branch (Maciej Szulik)
+
+  Before that it was using `feature/k8s`, which became outdated.
+
+- Add proper RBAC to run patroni on k8s (Maciej)
+
+  Add the Service account that is assigned to the pods of the cluster, the role that holds only the necessary permissions, and the rolebinding that connects the Service account and the Role.
+
+
+Version 1.4.1
+-------------
+
+**Fixes in patronictl**
+
+- Don't show current leader in suggested list of members to failover to. (Alexander Kukushkin)
+
+  patronictl failover could still work when there is leader in the cluster and it should be excluded from the list of member where it is possible to failover to.
+
+- Make patronictl switchover compatible with the old Patroni api (Alexander)
+
+  In case if POST /switchover REST API call has failed with status code 501 it will do it once again, but to /failover endpoint.
+
+
+Version 1.4
+-----------
+
+This version adds support for using Kubernetes as a DCS, allowing to run Patroni as a cloud-native agent in Kubernetes without any additional deployments of Etcd, Zookeeper or Consul.
+
+**Upgrade notice**
+
+Installing Patroni via pip will no longer bring in dependencies for (such as libraries for Etcd, Zookeper, Consul or Kubernetes, or support for AWS). In order to enable them one need to list them in pip install command explicitely, for instance `pip install patroni[etcd,kubernetes]`.
+
+**Kubernetes support**
+
+Implement Kubernetes-based DCS. The endpoints meta-data is used in order to store the configuration and the leader key. The meta-data field inside the pods definition is used to store the member-related data.
+In addition to using Endpoints, Patroni supports ConfigMaps. You can find more information about this feature in the :ref:`Kubernetes chapter of the documentation <kubernetes>`
+
+**Stability improvements**
+
+- Factor out postmaster process into a separate object (Ants Aasma)
+
+  This object identifies a running postmaster process via pid and start time and simplifies detection (and resolution) of situations when the postmaster was restarted behind our back or when postgres directory disappeared from the file system.
+
+- Minimize the amount of SELECT's issued by Patroni on every loop of HA cylce (Alexander Kukushkin)
+
+  On every iteration of HA loop Patroni needs to know recovery status and absolute wal position. From now on Patroni will run only single SELECT to get this information instead of two on the replica and three on the master. 
+  
+- Remove leader key on shutdown only when we have the lock (Ants)
+
+  Unconditional removal was generating unnecessary and missleading exceptions.
+
+**Improvements in patronictl**
+
+- Add version command to patronictl (Ants)
+
+  It will show the version of installed Patroni and versions of running Patroni instances (if the cluster name is specified).
+  
+- Make optional specifying cluster_name argument for some of patronictl commands (Alexander, Ants)
+
+  It will work if patronictl is using usual Patroni configuration file with the ``scope`` defined.
+
+- Show information about scheduled switchover and maintenance mode (Alexander)
+
+  Before that it was possible to get this information only from Patroni logs or directly from DCS.
+  
+- Improve ``patronictl reinit`` (Alexander)
+
+  Sometimes ``patronictl reinit`` refused to proceed when Patroni was busy with other actions, namely trying to start postgres. `patronictl` didn't provide any commands to cancel such long running actions and the only (dangerous) workarond was removing a data directory manually. The new implementation of `reinit` forcefully cancells other long-running actions before proceeding with reinit.
+  
+- Implement ``--wait`` flag in ``patronictl pause`` and ``patronictl resume`` (Alexander)
+
+  It will make ``patronictl`` wait until the requested action is acknowledged by all nodes in the cluster.
+  Such behaviour is achieved by exposing the ``pause`` flag for every node in DCS and via the REST API.
+
+- Rename ``patronictl failover`` into ``patronictl switchover`` (Alexander)
+
+  The previous ``failover`` was actually only capable of doing a switchover; it refused to proceed in a cluster without the leader.
+
+- Alter the behavior of ``patronictl failover`` (Alexander)
+
+  It will work even if there is no leader, but in that case you will have to explicitely specify a node which should become the new leader.
+
+**Expose information about timeline and history**
+
+- Expose current timeline in DCS and via API (Alexander)
+
+  Store information about the current timeline for each member of the cluster. This information is accessible via the API and is stored in the DCS
+
+- Store promotion history in the /history key in DCS (Alexander)
+
+  In addition, store the timeline history enriched with the timestamp of the corresponding promotion in the /history key in DCS and update it with each promote.
+
+**Add endpoints for getting synchronous and asynchronous replicas**
+
+- Add new /sync and /async endpoints (Alexander, Oleksii Kliukin)
+
+ Those endpoints (also accessible as /synchronous and /asynchronous) return 200 only for synchronous and asynchornous replicas correspondingly (exclusing those marked as `noloadbalance`).
+
+**Allow multiple hosts for Etcd**
+
+- Add a new `hosts` parameter to Etcd configuration (Alexander)
+
+  This parameter should contain the initial list of hosts that will be used to discover and populate the list of the running etcd cluster members. If for some reason during work this list of discovered hosts is exhausted (no available hosts from that list), Patroni will return to the initial list from the `hosts` parameter.
+
+
+Version 1.3.6
+-------------
+
+**Stability improvements**
+
+- Verify process start time when checking if postgres is running. (Ants Aasma)
+
+  After a crash that doesn't clean up postmaster.pid there could be a new process with the same pid, resulting in a false positive for is_running(), which will lead to all kinds of bad behavior.
+
+- Shutdown postgresql before bootstrap when we lost data directory (ainlolcat)
+  
+  When data directory on the master is forcefully removed, postgres process can still stay alive for some time and prevent the replica created in place of that former master from starting or replicating.
+  The fix makes Patroni cache the postmaster pid and its start time and let it terminate the old postmaster in case it is still running after the corresponding data directory has been removed.
+
+- Perform crash recovery in a single user mode if postgres master dies (Alexander Kukushkin)
+
+  It is unsafe to start immediately as a standby and not possible to run ``pg_rewind`` if postgres hasn't been shut down cleanly.
+  The single user crash recovery only kicks in if ``pg_rewind`` is enabled or there is no master at the moment.
+
+**Consul improvements**
+
+- Make it possible to provide datacenter configuration for Consul (Vilius Okockis, Alexander)
+
+  Before that Patroni was always communicating with datacenter of the host it runs on.
+
+- Always send a token in X-Consul-Token http header (Alexander)
+
+  If ``consul.token`` is defined in Patroni configuration, we will always send it in the 'X-Consul-Token' http header.
+  python-consul module tries to be "consistent" with Consul REST API, which doesn't accept token as a query parameter for `session API <https://www.consul.io/api/session.html>`__, but it still works with 'X-Consul-Token' header.
+
+- Adjust session TTL if supplied value is smaller than the minimum possible (Stas Fomin, Alexander)
+
+  It could happen that the TTL provided in the Patroni configuration is smaller than the minimum one supported by Consul. In that case, Consul agent fails to create a new session.
+  Without a session Patroni cannot create member and leader keys in the Consul KV store, resulting in an unhealthy cluster.
+
+**Other improvements**
+
+- Define custom log format via environment variable ``PATRONI_LOGFORMAT`` (Stas)
+
+  Allow disabling timestamps and other similar fields in Patroni logs if they are already added by the system logger (usually when Patroni runs as a service).
+
+Version 1.3.5
+-------------
+
+**Bugfix**
+
+- Set role to 'uninitialized' if data directory was removed (Alexander Kukushkin)
+
+  If the node was running as a master it was preventing from failover.
+
+**Stability improvement**
+
+- Try to run postmaster in a single-user mode if we tried and failed to start postgres (Alexander)
+
+  Usually such problem happens when node running as a master was terminated and timelines were diverged.
+  If ``recovery.conf`` has ``restore_command`` defined, there are really high chances that postgres will abort startup and leave controldata unchanged.
+  It makes impossible to use ``pg_rewind``, which requires a clean shutdown.
+
+**Consul improvements**
+
+- Make it possible to specify health checks when creating session (Alexander)
+
+  If not specified, Consul will use "serfHealth". From one side it allows fast detection of isolated master, but from another side it makes it impossible for Patroni to tolerate short network lags.
+
+**Bugfix**
+
+- Fix watchdog on Python 3 (Ants Aasma)
+
+  A misunderstanding of the ioctl() call interface. If mutable=False then fcntl.ioctl() actually returns the arg buffer back.
+  This accidentally worked on Python2 because int and str comparison did not return an error.
+  Error reporting is actually done by raising IOError on Python2 and OSError on Python3.
+
+Version 1.3.4
+-------------
+
+**Different Consul improvements**
+
+- Pass the consul token as a header (Andrew Colin Kissa)
+
+  Headers are now the prefered way to pass the token to the consul `API <https://www.consul.io/api/index.html#authentication>`__.
+  
+
+- Advanced configuration for Consul (Alexander Kukushkin)
+
+  possibility to specify ``scheme``, ``token``, client and ca certificates :ref:`details <consul_settings>`.
+
+- compatibility with python-consul-0.7.1 and above (Alexander)
+
+  new python-consul module has changed signature of some methods
+
+- "Could not take out TTL lock" message was never logged (Alexander)
+
+  Not a critical bug, but lack of proper logging complicates investigation in case of problems.
+
+
+**Quote synchronous_standby_names using quote_ident**
+
+- When writing ``synchronous_standby_names`` into the ``postgresql.conf`` its value must be quoted (Alexander)
+
+  If it is not quoted properly, PostgreSQL will effectively disable synchronous replication and continue to work.
+
+
+**Different bugfixes around pause state, mostly related to watchdog** (Alexander)
+
+- Do not send keepalives if watchdog is not active
+- Avoid activating watchdog in a pause mode
+- Set correct postgres state in pause mode
+- Do not try to run queries from API if postgres is stopped
+
+
+Version 1.3.3
+-------------
+
+**Bugfixes**
+
+- synchronous replication was disabled shortly after promotion even when synchronous_mode_strict was turned on (Alexander Kukushkin)
+- create empty ``pg_ident.conf`` file if it is missing after restoring from the backup (Alexander)
+- open access in ``pg_hba.conf`` to all databases, not only postgres (Franco Bellagamba)
+
+
+Version 1.3.2
+-------------
+
+**Bugfix**
+
+- patronictl edit-config didn't work with ZooKeeper (Alexander Kukushkin)
+
+
+Version 1.3.1
+-------------
+
+**Bugfix**
+
+- failover via API was broken due to change in ``_MemberStatus`` (Alexander Kukushkin)
+
+
+Version 1.3
+-----------
+
+Version 1.3 adds custom bootstrap possibility, significantly improves support for pg_rewind, enhances the
+synchronous mode support, adds configuration editing to patronictl and implements watchdog support on Linux.
+In addition, this is the first version to work correctly with PostgreSQL 10.
+
+**Upgrade notice**
+
+There are no known compatibility issues with the new version of Patroni. Configuration from version 1.2 should work
+without any changes. It is possible to upgrade by installing new packages and either  restarting Patroni (will cause
+PostgreSQL restart), or by putting Patroni into a :ref:`pause mode <pause>` first and then restarting Patroni on all
+nodes in the cluster (Patroni in a pause mode will not attempt to stop/start PostgreSQL), resuming from the pause mode
+at the end.
+
+**Custom bootstrap**
+
+- Make the process of bootstrapping the cluster configurable (Alexander Kukushkin)
+
+  Allow custom bootstrap scripts instead of ``initdb`` when initializing the very first node in the cluster.
+  The bootstrap command receives the name of the cluster and the path to the data directory. The resulting cluster can
+  be configured to perform recovery, making it possible to bootstrap from a backup and do point in time recovery. Refer
+  to the :ref:`documentaton page <custom_bootstrap>` for more detailed description of this feature.
+
+**Smarter pg_rewind support**
+
+-  Decide on whether to run pg_rewind by looking at the timeline differences from the current master (Alexander)
+
+   Previously, Patroni had a fixed set of conditions to trigger pg_rewind, namely when starting a former master, when
+   doing a switchover to the designated node for every other node in the cluster or when there is a replica with the
+   nofailover tag. All those cases have in common a chance that some replica may be ahead of the new master. In some cases,
+   pg_rewind did nothing, in some other ones it was not running when necessary. Instead of relying on this limited list
+   of rules make Patroni compare the master and the replica WAL positions (using the streaming replication protocol)
+   in order to reliably decide if rewind is necessary for the replica.
+
+**Synchronous replication mode strict**
+
+-  Enhance synchronous replication support by adding the strict mode (James Sewell, Alexander)
+
+   Normally, when ``synchronous_mode`` is enabled and there are no replicas attached to the master, Patroni will disable
+   synchronous replication in order to keep the master available for writes. The ``synchronous_mode_strict`` option
+   changes that, when it is set Patroni will not disable the synchronous replication in a lack of replicas, effectively
+   blocking all clients writing data to the master. In addition to the synchronous mode guarantee of preventing any data
+   loss due to automatic failover, the strict mode ensures that each write is either durably stored on two nodes or not
+   happening altogether if there is only one node in the cluster.
+
+**Configuration editing with patronictl**
+
+- Add configuration editing to patronictl (Ants Aasma, Alexander)
+
+  Add the ability to patronictl of editing dynamic cluster configuration stored in DCS. Support either specifying the
+  parameter/values from the command-line, invoking the $EDITOR, or applying configuration from the yaml file.
+
+**Linux watchdog support**
+
+- Implement watchdog support for Linux (Ants)
+
+  Support Linux software watchdog in order to reboot the node where Patroni is not running or not responding (e.g because
+  of the high load) The Linux software watchdog reboots the non-responsive node. It is possible to configure the watchdog
+  device to use (`/dev/watchdog` by default) and the mode (on, automatic, off) from the watchdog section of the Patroni
+  configuration. You can get more information from the :ref:`watchdog documentation <watchdog>`.
+
+**Add support for PostgreSQL 10**
+
+- Patroni is compatible with all beta versions of PostgreSQL 10 released so far and we expect it to be compatible with
+  the PostgreSQL 10 when it will be released.
+
+**PostgreSQL-related minor improvements**
+
+- Define pg_hba.conf via the Patroni configuration file or the dynamic configuration in DCS (Alexander)
+
+  Allow to define the contents of ``pg_hba.conf`` in the ``pg_hba`` sub-section of the ``postgresql`` section of the
+  configuration. This simplifies managing ``pg_hba.conf`` on multiple nodes, as one needs to define it only ones in DCS
+  instead of logging to every node, changing it manually and reload the configuration.
+
+  When defined, the contents of this section will replace the current ``pg_hba.conf`` completely. Patroni ignores it
+  if ``hba_file`` PostgreSQL parameter is set.
+
+- Support connecting via a UNIX socket to the local PostgreSQL cluster (Alexander)
+
+  Add the ``use_unix_socket`` option to the ``postgresql`` section of Patroni configuration. When set to true and the
+  PostgreSQL ``unix_socket_directories`` option is not empty, enables Patroni to use the first value from it to connect
+  to the local PostgreSQL cluster. If ``unix_socket_directories`` is not defined, Patroni will assume its default value
+  and omit the ``host`` parameter in the PostgreSQL connection string altogether.
+
+- Support change of superuser and replication credentials on reload (Alexander)
+
+- Support storing of configuration files outside of PostgreSQL data directory (@jouir)
+
+  Add the new configuration ``postgresql`` configuration directive ``config_dir``.
+  It defaults to the data directory and must be writable by Patroni.
+
+**Bug fixes and stability improvements**
+
+- Handle EtcdEventIndexCleared and EtcdWatcherCleared exceptions (Alexander)
+
+  Faster recovery when the watch operation is ended by Etcd by avoiding useless retries.
+
+- Remove error spinning on Etcd failure and reduce log spam (Ants)
+
+  Avoid immediate retrying and emitting stack traces in the log on the second and subsequent Etcd connection failures.
+
+- Export locale variables when forking PostgreSQL processes (Oleksii Kliukin)
+
+  Avoid the `postmaster became multithreaded during startup` fatal error on non-English locales for PostgreSQL built with NLS.
+
+- Extra checks when dropping the replication slot (Alexander)
+
+  In some cases Patroni is prevented from dropping the replication slot by the WAL sender.
+
+- Truncate the replication slot name to 63  (NAMEDATALEN - 1) characters to comply with PostgreSQL naming rules (Nick Scott)
+
+- Fix a race condition resulting in extra connections being opened to the PostgreSQL cluster from Patroni (Alexander)
+
+- Release the leader key when the node restarts with an empty data directory (Alex Kerney)
+
+- Set asynchronous executor busy when running bootstrap without a leader (Alexander)
+
+  Failure to do so could have resulted in errors stating the node belonged to a different cluster, as Patroni proceeded with
+  the normal business while being bootstrapped by a bootstrap method that doesn't require a leader to be present in the
+  cluster.
+
+- Improve WAL-E replica creation method (Joar Wandborg, Alexander).
+
+  - Use csv.DictReader when parsing WAL-E base backup, accepting ISO dates with space-delimited date and time.
+  - Support fetching current WAL position from the replica to estimate the amount of WAL to restore. Previously, the code used to call system information functions that were available only on the master node.
+
+
 Version 1.2
 -----------
 
@@ -391,4 +882,4 @@ This release adds support for *cascading replication* and simplifies Patroni man
 
   The tests can be launched manually using the *behave* command. They are also launched automatically for pull requests and after commits.
 
-  Releases notes for some older versions can be found on `project's github page <https://github.com/zalando/patroni/releases>`__.
+  Release notes for some older versions can be found on `project's github page <https://github.com/zalando/patroni/releases>`__.

docs/replica_bootstrap.rst

@@ -0,0 +1,130 @@
+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_method:
+            - wal_e
+            - basebackup
+        wal_e:
+            command: patroni_wale_restore
+            no_master: 1
+            envdir: {{WALE_ENV_DIR}}
+            use_iam: 1
+        basebackup:
+            max-rate: '100M'
+
+
+The ``create_replica_method`` 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 ``basebackup`` method is a special case: it will be used if ``create_replica_method`` is empty, although it is possible
+to list it explicitly among the ``create_replica_method`` 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.

docs/replication_modes.rst

@@ -9,9 +9,9 @@ Patroni uses PostgreSQL streaming replication. For more information about stream
 Asynchronous mode durability
 ----------------------------
 
-In asynchronous mode the cluster is allowed to lose some committed transactions to ensure availability. When master server fails or becomes unavailable for any other reason Patroni will automatically promote a sufficiently healthy standby to master. Any transactions that have not been replicated to that standby remain in a "forked timeline" on the master, and are effectively unrecoverable [1]_.
+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 master 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.
+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.
 
 PostgreSQL synchronous replication
 ----------------------------------
@@ -20,7 +20,7 @@ You can use Postgres's `synchronous replication <http://www.postgresql.org/docs/
 
 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 follow lines to the ``parameters`` section of your YAML configuration files:
+To enable a simple synchronous replication test, add the following lines to the ``parameters`` section of your YAML configuration files:
 
 .. code:: YAML
 
@@ -29,28 +29,28 @@ To enable a simple synchronous replication test, add the follow lines to the ``p
 
 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 master and standby that is currently acting as synchronous fail simultaneously a third node that might not contain all transactions will be promoted.
+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 Patronis ``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 commmands to promote a standby even if it results in transaction loss.
+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, master server will still accept writes, but does not guarantee their replication. When the master fails in this mode no standby will be promote.  When the host that used to be master comes back it will get promoted automatically, unless system administrator performed a manual failover. This behavior makes synchronous mode usable with 2 node clusters.
+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 master 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 master to release itself from synchronous standby duties before PostgreSQL shutdown is initiated.
+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.
 
 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 `dynamic configuration <https://github.com/zalando/patroni/blob/master/docs/dynamic_configuration.rst>`__ for instructions.
+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 master and current synchronous standby. This state is updated with strict ordering constraints to ensure the following invariants:
+When in synchronous mode Patroni maintains synchronization state in the DCS, containing the latest primary and current synchronous standby. 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.
 
@@ -60,9 +60,9 @@ When in synchronous mode Patroni maintains synchronization state in the DCS, con
 
 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 avaiable for sync that is furthest ahead in replication is picked.
+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 master with the cluster.
+.. [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.
+.. [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/confd/templates/haproxy.tmpl

@@ -10,19 +10,23 @@ defaults
 	timeout server 30m
 	timeout check 5s
 
-frontend master_postgresql
+listen stats
+	mode http
+	bind *:7000
+	stats enable
+	stats uri /
+
+listen master
 	bind *:5000
-	default_backend backend_master
-
-frontend replicas_postgresql
-	bind *:5001
-	default_backend backend_replicas
-
-backend backend_master
 	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 $data.api_url "/") 2) ":") 1}}
 {{end}}
-backend backend_replicas
+listen replicas
+	bind *:5001
 	option httpchk OPTIONS /replica
+	http-check expect status 200
+	default-server inter 3s fall 3 rise 2 on-marked-down shutdown-sessions
 {{range gets "/members/*"}}	server {{base .Key}} {{$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/startup-scripts/README.md

@@ -11,3 +11,12 @@ Upstart job for Ubuntu 12.04 or 14.04.  Requires Upstart > 1.4. Intended for sys
 
 ### 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

@@ -21,7 +21,7 @@ ExecStart=/bin/patroni /etc/patroni.yml
 KillMode=process
 
 # Give a reasonable amount of time for the server to start up/shut down
-TimeoutSec=10
+TimeoutSec=30
 
 # Do not restart the service if it crashes, we want to manually inspect database on failure
 Restart=no

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

@@ -19,11 +19,21 @@ Feature: basic replication
     And I run patronictl.py restart batman postgres1 --force
     Then I receive a response returncode 0
     And "sync" key in DCS has sync_standby=postgres2 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
-    When I kill postgres0
-    Then postgres2 role is the primary after 22 seconds
-    When I issue a PATCH request to http://127.0.0.1:8009/config with {"synchronous_mode": null, "master_start_timeout": 0}
+    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
@@ -34,7 +44,8 @@ Feature: basic replication
     And postgres1 role is the primary after 10 seconds
 
   Scenario: check rejoin of the former master with pg_rewind
-    Given I start postgres0
+    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/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

@@ -1,14 +1,19 @@
 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
 
@@ -74,18 +79,29 @@ class AbstractController(object):
         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, tags=None):
+    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
-        self._config = self._make_patroni_test_config(name, tags)
+        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
@@ -109,6 +125,11 @@ class PatroniController(AbstractController):
                 yaml.safe_dump(config, w, default_flow_style=False)
 
     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)
 
@@ -116,40 +137,60 @@ class PatroniController(AbstractController):
         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):
-        return self.query("SELECT 1", fail_ok=True) is not None
+        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, tags):
+    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')
+            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:
+            def recursive_update(dst, src):
+                for k, v in src.items():
+                    if k in dst and isinstance(dst[k], dict):
+                        recursive_update(dst[k], v)
+                    else:
+                        dst[k] = v
+            recursive_update(config, custom_config)
+
+        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'})
 
-        config['name'] = name
-        config['postgresql']['data_dir'] = self._data_dir
-        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'})
-
-        if 'bootstrap' in config and 'initdb' in config['bootstrap']:
-            config['bootstrap']['initdb'].extend([{'auth': 'md5'}, {'auth-host': 'md5'}])
-
-        if tags:
-            config['tags'] = tags
-
-        with open(patroni_config_path, 'w') as f:
-            yaml.safe_dump(config, f, default_flow_style=False)
+        self._replication = config['postgresql'].get('authentication', config['postgresql']).get('replication', {})
+        self._replication.update({'host': host, 'port': self.__PORT, 'database': 'postgres'})
 
         return patroni_config_path
 
@@ -185,10 +226,101 @@ class PatroniController(AbstractController):
             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/batman'
+    _CLUSTER_NODE = '/service/{0}'
 
     def __init__(self, context, mktemp=True):
         work_directory = mktemp and tempfile.mkdtemp() or None
@@ -203,17 +335,13 @@ class AbstractDcsController(AbstractController):
         if self._work_directory:
             shutil.rmtree(self._work_directory)
 
-    def path(self, key=None):
-        return self._CLUSTER_NODE + (key and '/' + key or '')
+    def path(self, key=None, scope='batman'):
+        return self._CLUSTER_NODE.format(scope) + (key and '/' + key or '')
 
     @abc.abstractmethod
-    def query(self, key):
+    def query(self, key, scope='batman'):
         """ query for a value of a given key """
 
-    @abc.abstractmethod
-    def set(self, key, value):
-        """ set a value to a given key """
-
     @abc.abstractmethod
     def cleanup_service_tree(self):
         """ clean all contents stored in the tree used for the tests """
@@ -236,18 +364,19 @@ class ConsulController(AbstractDcsController):
         super(ConsulController, self).__init__(context)
         os.environ['PATRONI_CONSUL_HOST'] = 'localhost:8500'
         self._client = consul.Consul()
+        self._config_file = None
 
     def _start(self):
-        config_file = self._work_directory + '.json'
-        with open(config_file, 'wb') as f:
+        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', config_file, '-data-dir', self._work_directory],
-                                stdout=self._log, stderr=subprocess.STDOUT)
+        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._work_directory:
-            os.unlink(self._work_directory + '.json')
+        if self._config_file:
+            os.unlink(self._config_file)
 
     def _is_running(self):
         try:
@@ -255,18 +384,15 @@ class ConsulController(AbstractDcsController):
         except Exception:
             return False
 
-    def path(self, key=None):
-        return super(ConsulController, self).path(key)[1:]
+    def path(self, key=None, scope='batman'):
+        return super(ConsulController, self).path(key, scope)[1:]
 
-    def query(self, key):
-        _, value = self._client.kv.get(self.path(key))
+    def query(self, key, scope='batman'):
+        _, value = self._client.kv.get(self.path(key, scope))
         return value and value['Value'].decode('utf-8')
 
-    def set(self, key, value):
-        self._client.kv.put(self.path(key), value)
-
     def cleanup_service_tree(self):
-        self._client.kv.delete(self.path(), recurse=True)
+        self._client.kv.delete(self.path(scope=''), recurse=True)
 
     def start(self, max_wait_limit=15):
         super(ConsulController, self).start(max_wait_limit)
@@ -285,18 +411,15 @@ class EtcdController(AbstractDcsController):
         return subprocess.Popen(["etcd", "--debug", "--data-dir", self._work_directory],
                                 stdout=self._log, stderr=subprocess.STDOUT)
 
-    def query(self, key):
+    def query(self, key, scope='batman'):
         try:
-            return self._client.get(self.path(key)).value
+            return self._client.get(self.path(key, scope)).value
         except etcd.EtcdKeyNotFound:
             return None
 
-    def set(self, key, value):
-        self._client.set(self.path(key), value)
-
     def cleanup_service_tree(self):
         try:
-            self._client.delete(self.path(), recursive=True)
+            self._client.delete(self.path(scope=''), recursive=True)
         except (etcd.EtcdKeyNotFound, etcd.EtcdConnectionFailed):
             return
         except Exception as e:
@@ -310,6 +433,76 @@ class EtcdController(AbstractDcsController):
             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 """
@@ -323,18 +516,15 @@ class ZooKeeperController(AbstractDcsController):
     def _start(self):
         pass  # TODO: implement later
 
-    def query(self, key):
+    def query(self, key, scope='batman'):
         try:
-            return self._client.get(self.path(key))[0].decode('utf-8')
+            return self._client.get(self.path(key, scope))[0].decode('utf-8')
         except kazoo.exceptions.NoNodeError:
             return None
 
-    def set(self, key, value):
-        self._client.set(self.path(key), value.encode('utf-8'))
-
     def cleanup_service_tree(self):
         try:
-            self._client.delete(self.path(), recursive=True)
+            self._client.delete(self.path(scope=''), recursive=True)
         except (kazoo.exceptions.NoNodeError):
             return
         except Exception as e:
@@ -359,6 +549,8 @@ class ExhibitorController(ZooKeeperController):
 
 class PatroniPoolController(object):
 
+    BACKUP_SCRIPT = 'features/backup_create.sh'
+
     def __init__(self, context):
         self._context = context
         self._dcs = None
@@ -383,13 +575,16 @@ class PatroniPoolController(object):
     def output_dir(self):
         return self._output_dir
 
-    def start(self, name, max_wait_limit=20, tags=None):
+    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, tags)
+            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']:
+        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):
@@ -398,6 +593,7 @@ class PatroniPoolController(object):
 
     def stop_all(self):
         for ctl in self._processes.values():
+            ctl.cancel_background()
             ctl.stop()
         self._processes.clear()
 
@@ -408,6 +604,53 @@ class PatroniPoolController(object):
         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
+                }
+            },
+            '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:
@@ -416,8 +659,125 @@ class PatroniPoolController(object):
         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)

features/patroni_api.feature

@@ -13,17 +13,17 @@ Scenario: check API requests on a stand-alone server
 	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 failover batman --master postgres0 --force
+	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 failover to"
-	When I issue a POST request to http://127.0.0.1:8008/failover with {"leader": "postgres0"}
-	Then I receive a response code 500
-	And I receive a response text failover is not possible: cluster does not have members except leader
+	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 "No values given for required parameters leader and candidate"
+	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
@@ -34,9 +34,9 @@ Scenario: check local configuration reload
 	Then I receive a response code 202
 
 Scenario: check dynamic configuration change via DCS
-	Given I issue a PATCH request to http://127.0.0.1:8008/config with {"ttl": 10, "loop_wait": 2, "postgresql": {"parameters": {"max_connections": 101}}}
-	Then I receive a response code 200
-	And I receive a response loop_wait 2
+	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
@@ -64,21 +64,21 @@ Scenario: check API requests for the primary-replica pair in the pause mode
 	When I sleep for 10 seconds
 	Then postgres1 role is the secondary after 15 seconds
 
-Scenario: check the failover via the API in the pause mode
-	Given I run patronictl.py failover batman --master postgres0 --candidate postgres1 --force
-	Then I receive a response returncode 0
+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
 
-Scenario: check the scheduled failover
-	Given I issue a scheduled failover from postgres1 to postgres0 in 3 seconds
+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 failover in the paused state"
+	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 failover from postgres1 to postgres0 in 3 seconds
+	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

features/steps/basic_replication.py

@@ -11,7 +11,7 @@ def start_patroni(context, name):
 
 @step('I shut down {name:w}')
 def stop_patroni(context, name):
-    return context.pctl.stop(name)
+    return context.pctl.stop(name, timeout=60)
 
 
 @step('I kill {name:w}')

features/steps/cascading_replication.py

@@ -6,7 +6,7 @@ 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, tags={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')

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

@@ -1,3 +1,4 @@
+import base64
 import json
 import os
 import parse
@@ -78,11 +79,13 @@ def do_post_empty(context, url):
 @step('I issue a {request_method:w} request to {url:url} with {data}')
 def do_request(context, request_method, url, data):
     data = data and json.loads(data) or {}
+    headers = {'Authorization': 'Basic ' + base64.b64encode('username:password'.encode('utf-8')).decode('utf-8'),
+               'Content-Type': 'application/json'}
     try:
         if request_method == 'PATCH':
-            r = requests.patch(url, json=data)
+            r = requests.patch(url, headers=headers, json=data)
         else:
-            r = requests.post(url, json=data)
+            r = requests.post(url, headers=headers, json=data)
     except requests.exceptions.RequestException:
         context.status_code = None
         context.response = None
@@ -111,7 +114,8 @@ def check_response(context, component, data):
         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}".format(context.status_code, data)
+        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':
@@ -121,10 +125,10 @@ def check_response(context, component, data):
         assert str(context.response[component]) == str(data), "{0} does not contain {1}".format(component, data)
 
 
-@step('I issue a scheduled failover from {from_host:w} to {to_host:w} in {in_seconds:d} seconds')
-def scheduled_failover(context, from_host, to_host, in_seconds):
+@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 failover batman --master {0} --candidate {1} --scheduled "{2}" --force
+        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))))
 
 

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 +1,32 @@
+FROM postgres:9.6
+MAINTAINER Alexander Kukushkin <[email protected]>
+
+RUN export DEBIAN_FRONTEND=noninteractive \
+    && echo 'APT::Install-Recommends "0";\nAPT::Install-Suggests "0";' > /etc/apt/apt.conf.d/01norecommend \
+    && apt-get update -y \
+    && apt-get upgrade -y \
+    && apt-get install -y git curl jq python-psycopg2 python-yaml python-requests python-six python-pysocks \
+        python-dateutil python-pip python-prettytable python-wheel python-psutil python locales \
+
+    ## Make sure we have a en_US.UTF-8 locale available
+    && localedef -i en_US -c -f UTF-8 -A /usr/share/locale/locale.alias en_US.UTF-8 \
+
+    && pip install setuptools pip --upgrade \
+    && pip install 'git+https://github.com/zalando/patroni.git#egg=patroni[kubernetes]' \
+
+    && mkdir -p /home/postgres \
+    && chown postgres:postgres /home/postgres \
+
+    # Clean up
+    && apt-get remove -y git python-pip python-setuptools \
+    && apt-get autoremove -y \
+    && apt-get clean -y \
+    && rm -rf /var/lib/apt/lists/* /root/.cache
+
+ADD entrypoint.sh callback.py /
+
+EXPOSE 5432 8008
+ENV LC_ALL=en_US.UTF-8 LANG=en_US.UTF-8
+USER postgres
+WORKDIR /home/postgres
+CMD ["/bin/bash", "/entrypoint.sh"]

kubernetes/callback.py

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

kubernetes/entrypoint.sh

@@ -0,0 +1,36 @@
+#!/bin/bash
+
+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} ${POD_IP}/16    md5
+restapi:
+  connect_address: '${POD_IP}:8008'
+postgresql:
+  connect_address: '${POD_IP}:5432'
+  authentication:
+    superuser:
+      password: '${PATRONI_SUPERUSER_PASSWORD}'
+    replication:
+      password: '${PATRONI_REPLICATION_PASSWORD}'
+  callbacks:
+    on_start: /callback.py
+    on_stop: /callback.py
+    on_role_change: /callback.py
+__EOF__
+
+unset PATRONI_SUPERUSER_PASSWORD PATRONI_REPLICATION_PASSWORD
+export KUBERNETES_NAMESPACE=$PATRONI_KUBERNETES_NAMESPACE
+export POD_NAME=$PATRONI_NAME
+
+exec /usr/bin/python /usr/local/bin/patroni /home/postgres/patroni.yml

kubernetes/patroni_k8s.yaml

@@ -0,0 +1,182 @@
+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: POD_IP
+          valueFrom:
+            fieldRef:
+              fieldPath: status.podIP
+        - name: PATRONI_KUBERNETES_NAMESPACE
+          valueFrom:
+            fieldRef:
+              fieldPath: metadata.namespace
+        - 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
+- apiGroups:
+  - ""
+  resources:
+  - endpoints
+  verbs:
+  - get
+  - patch
+  - update
+  # the following three privileges are necessary only when using endpoints
+  - create
+  - list
+  - watch
+- 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

patroni/__init__.py

@@ -16,12 +16,14 @@ class Patroni(object):
         from patroni.ha import Ha
         from patroni.postgresql import Postgresql
         from patroni.version import __version__
+        from patroni.watchdog import Watchdog
 
         self.setup_signal_handlers()
 
         self.version = __version__
         self.config = Config()
         self.dcs = get_dcs(self.config)
+        self.watchdog = Watchdog(self.config)
         self.load_dynamic_configuration()
 
         self.postgresql = Postgresql(self.config['postgresql'])
@@ -37,9 +39,10 @@ class Patroni(object):
         while True:
             try:
                 cluster = self.dcs.get_cluster()
-                if cluster and cluster.config:
+                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)
@@ -63,6 +66,7 @@ class Patroni(object):
         try:
             self.tags = self.get_tags()
             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:
@@ -109,11 +113,11 @@ class Patroni(object):
 
             logger.info(self.ha.run_cycle())
 
-            cluster = self.dcs.cluster
-            if cluster and cluster.config and self.config.set_dynamic_configuration(cluster.config):
+            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 not self.postgresql.data_directory_empty():
+            if self.postgresql.role != 'uninitialized':
                 self.config.save_cache()
 
             self.schedule_next_run()
@@ -124,10 +128,17 @@ class Patroni(object):
         signal.signal(signal.SIGHUP, self.sighup_handler)
         signal.signal(signal.SIGTERM, self.sigterm_handler)
 
+    def shutdown(self):
+        self.api.shutdown()
+        self.ha.shutdown()
+
 
 def patroni_main():
-    logging.basicConfig(format='%(asctime)s %(levelname)s: %(message)s', level=logging.INFO)
-    logging.getLogger('requests').setLevel(logging.WARNING)
+    logformat = os.environ.get('PATRONI_LOGFORMAT', '%(asctime)s %(levelname)s: %(message)s')
+    loglevel = os.environ.get('PATRONI_LOGLEVEL', 'INFO')
+    requests_loglevel = os.environ.get('PATRONI_REQUESTS_LOGLEVEL', 'WARNING')
+    logging.basicConfig(format=logformat, level=loglevel)
+    logging.getLogger('requests').setLevel(requests_loglevel)
 
     patroni = Patroni()
     try:
@@ -135,12 +146,7 @@ def patroni_main():
     except KeyboardInterrupt:
         pass
     finally:
-        patroni.api.shutdown()
-        if patroni.ha.is_paused():
-            logger.info('Leader key is not deleted and Postgresql is not stopped due paused state')
-        else:
-            patroni.ha.while_not_sync_standby(lambda: patroni.postgresql.stop(checkpoint=False))
-            patroni.dcs.delete_leader()
+        patroni.shutdown()
 
 
 def pg_ctl_start(args):

patroni/api.py

@@ -7,8 +7,8 @@ import time
 import dateutil.parser
 import datetime
 
-from patroni.exceptions import PostgresConnectionException
-from patroni.utils import deep_compare, patch_config, Retry, RetryFailedError, is_valid_pg_version, parse_int, tzutc
+from patroni.postgresql import PostgresConnectionException, PostgresException, Postgresql
+from patroni.utils import deep_compare, parse_bool, patch_config, Retry, RetryFailedError, parse_int, tzutc
 from six.moves.BaseHTTPServer import BaseHTTPRequestHandler, HTTPServer
 from six.moves.socketserver import ThreadingMixIn
 from threading import Thread
@@ -24,9 +24,9 @@ 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
 
 
@@ -68,6 +68,10 @@ class RestApiHandler(BaseHTTPRequestHandler):
             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, write_status_code_only=False):
@@ -78,6 +82,14 @@ class RestApiHandler(BaseHTTPRequestHandler):
 
         patroni = self.server.patroni
         cluster = patroni.dcs.cluster
+
+        def is_synchronous():
+            return (cluster.is_synchronous_mode() and cluster.sync
+                    and cluster.sync.sync_standby == patroni.postgresql.name)
+
+        def is_balanceable_replica():
+            return response.get('role') == 'replica' and not patroni.noloadbalance
+
         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
@@ -85,6 +97,10 @@ class RestApiHandler(BaseHTTPRequestHandler):
                 status_code = 503
             elif response['role'] == 'master':  # running as master but without leader lock!!!!
                 status_code = 503
+            elif path in ('/sync', '/synchronous'):
+                status_code = 200 if is_balanceable_replica() and is_synchronous() else 503
+            elif path in ('/async', '/asynchronous'):
+                status_code = 200 if is_balanceable_replica() and not is_synchronous() else 503
             elif response['role'] in path:  # response['role'] != 'master'
                 status_code = 503 if patroni.noloadbalance else 200
             else:
@@ -219,9 +235,11 @@ class RestApiHandler(BaseHTTPRequestHandler):
                     data = "PostgreSQL role should be either master or replica"
                     break
             elif k == 'postgres_version':
-                if not is_valid_pg_version(request[k]):
+                try:
+                    Postgresql.postgres_version_to_int(request[k])
+                except PostgresException as e:
                     status_code = 400
-                    data = "PostgreSQL version should be in the first.major.minor format"
+                    data = e.value
                     break
             elif k == 'timeout':
                 request[k] = parse_int(request[k], 's')
@@ -262,7 +280,14 @@ class RestApiHandler(BaseHTTPRequestHandler):
 
     @check_auth
     def do_POST_reinitialize(self):
-        data = self.server.patroni.ha.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'
@@ -270,43 +295,50 @@ class RestApiHandler(BaseHTTPRequestHandler):
             status_code = 503
         self._write_response(status_code, data)
 
-    def poll_failover_result(self, leader, candidate):
+    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:
+                if not cluster.is_unlocked() and cluster.leader.name != leader:
                     if not candidate or candidate == cluster.leader.name:
-                        return 200, 'Successfully failed over to "{0}"'.format(cluster.leader.name)
+                        return 200, 'Successfully {0}ed over to "{1}"'.format(action[:-4], cluster.leader.name)
                     else:
-                        return 200, 'Failed over to "{0}" instead of "{1}"'.format(cluster.leader.name, candidate)
+                        return 200, '{0}ed over to "{1}" instead of "{2}"'.format(action[:-4].title(),
+                                                                                  cluster.leader.name, candidate)
                 if not cluster.failover:
-                    return 503, 'Failover failed'
+                    return 503, action.title() + ' failed'
             except Exception as e:
-                logger.debug('Exception occured during polling failover result: %s', e)
-        return 503, 'Failover status unknown'
+                logger.debug('Exception occured during polling %s result: %s', action, e)
+        return 503, action.title() + ' status unknown'
 
-    def is_failover_possible(self, cluster, leader, candidate):
+    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 '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 'failover is not possible: cluster does not have members except leader'
-        for _, reachable, _, _, 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 '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):
+    def do_POST_failover(self, action='failover'):
         request = self._read_json_content()
-        status_code = 500
+        (status_code, data) = (400, '')
         if not request:
             return
 
@@ -315,39 +347,47 @@ class RestApiHandler(BaseHTTPRequestHandler):
         scheduled_at = request.get('scheduled_at')
         cluster = self.server.patroni.dcs.get_cluster()
 
-        if scheduled_at and cluster.is_paused():
-            self._write_response(status_code, "Can't schedule failover in the paused state")
+        logger.info("received %s request with leader=%s candidate=%s scheduled_at=%s",
+                    action, leader, candidate, scheduled_at)
 
-        logger.info("received failover request with leader=%s candidate=%s scheduled_at=%s",
-                    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'
 
-        data = ''
-        if leader or candidate:
-            if scheduled_at:
-                (_, data, scheduled_at) = self.parse_schedule(scheduled_at, "failover")
-                if _:
-                    status_code = _
-                elif self.server.patroni.dcs.manual_failover(leader, candidate, scheduled_at=scheduled_at):
-                    self.server.patroni.ha.wakeup()
-                    data = 'Failover scheduled'
+        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:
-                    data = 'failed to write failover key into DCS'
-                    status_code = 503
+                    status_code, data = self.poll_failover_result(cluster.leader and cluster.leader.name,
+                                                                  candidate, action)
             else:
-                data = self.is_failover_possible(cluster, leader, candidate)
-                if not data:
-                    if self.server.patroni.dcs.manual_failover(leader, candidate):
-                        self.server.patroni.ha.wakeup()
-                        status_code, data = self.poll_failover_result(cluster.leader and cluster.leader.name, candidate)
-                    else:
-                        data = 'failed to write failover key into DCS'
-                        status_code = 503
-        else:
-            status_code = 400
-            data = 'No values given for required parameters leader and candidate'
+                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
 
@@ -374,38 +414,46 @@ class RestApiHandler(BaseHTTPRequestHandler):
 
     def get_postgresql_status(self, retry=False):
         try:
-            row = self.query("""WITH replication_info AS (
-                                    SELECT usename, application_name, client_addr, state, sync_state, sync_priority
-                                      FROM pg_stat_replication
-                                )
-                                SELECT to_char(pg_postmaster_start_time(), 'YYYY-MM-DD HH24:MI:SS.MS TZ'),
-                                       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(COALESCE(pg_last_xlog_receive_location(),
-                                                                      pg_last_xlog_replay_location()), '0/0')::bigint,
-                                       pg_xlog_location_diff(pg_last_xlog_replay_location(), '0/0')::bigint,
-                                       to_char(pg_last_xact_replay_timestamp(), 'YYYY-MM-DD HH24:MI:SS.MS TZ'),
-                                       pg_is_in_recovery() AND pg_is_xlog_replay_paused(),
-                                       (SELECT array_to_json(array_agg(row_to_json(ri))) FROM replication_info ri)""",
-                             retry=retry)[0]
+            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_stat_replication) SELECT"
+                    " to_char(pg_postmaster_start_time(), 'YYYY-MM-DD HH24:MI:SS.MS TZ'),"
+                    " CASE WHEN pg_is_in_recovery() THEN 0"
+                    " ELSE ('x' || SUBSTR(pg_{0}file_name(pg_current_{0}_{1}()), 1, 8))::bit(32)::int END,"
+                    " CASE WHEN pg_is_in_recovery() THEN 0"
+                    " ELSE pg_{0}_{1}_diff(pg_current_{0}_{1}(), '0/0')::bigint END,"
+                    " pg_{0}_{1}_diff(COALESCE(pg_last_{0}_receive_{1}(), pg_last_{0}_replay_{1}()), '0/0')::bigint,"
+                    " pg_{0}_{1}_diff(pg_last_{0}_replay_{1}(), '0/0')::bigint,"
+                    " to_char(pg_last_xact_replay_timestamp(), 'YYYY-MM-DD HH24:MI:SS.MS TZ'),"
+                    " pg_is_in_recovery() AND pg_is_{0}_replay_paused(),"
+                    " (SELECT array_to_json(array_agg(row_to_json(ri))) FROM replication_info ri)")
+
+            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,
                 'xlog': ({
                     'received_location': row[3],
                     'replayed_location': row[4],
                     'replayed_timestamp': row[5],
-                    'paused': row[6]} if row[1] else {
+                    'paused': row[6]} if row[1] == 0 else {
                     'location': row[2]
                 })
             }
 
+            if row[1] > 0:
+                result['timeline'] = row[1]
+            else:
+                cluster = self.server.patroni.dcs.cluster
+                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]
 
@@ -469,7 +517,7 @@ class RestApiServer(ThreadingMixIn, HTTPServer, Thread):
     def __initialize(self, config):
         self.__ssl_options = self.__get_ssl_options(config)
         self.__listen = config['listen']
-        host, port = config['listen'].split(':')
+        host, port = config['listen'].rsplit(':', 1)
         HTTPServer.__init__(self, (host, int(port)), RestApiHandler)
         Thread.__init__(self, target=self.serve_forever)
         self._set_fd_cloexec(self.socket)

patroni/async_executor.py

@@ -1,26 +1,78 @@
 import logging
-from threading import RLock, Thread
+from threading import Event, Lock, RLock, Thread
 
 logger = logging.getLogger(__name__)
 
 
+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):
+        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, ha_wakeup):
+    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 = RLock()
+        self._is_cancelled = False
+        self._finish_event = Event()
+        self.critical_task = CriticalTask()
 
     @property
     def busy(self):
         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._is_cancelled = False
+            self._finish_event.set()
         return None
 
     @property
@@ -35,20 +87,43 @@ class AsyncExecutor(object):
     def run(self, func, args=()):
         wakeup = False
         try:
+            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:
+        except Exception:
             logger.exception('Exception during execution of long running task %s', self.scheduled_action)
         finally:
             with self:
                 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=()):
         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()
 

patroni/config.py

@@ -9,7 +9,8 @@ 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_int, patch_config
+from patroni.utils import deep_compare, parse_bool, parse_int, patch_config
+from requests.structures import CaseInsensitiveDict
 
 logger = logging.getLogger(__name__)
 
@@ -43,10 +44,14 @@ class Config(object):
         'maximum_lag_on_failover': 1048576,
         'master_start_timeout': 300,
         'synchronous_mode': False,
+        'synchronous_mode_strict': False,
         'postgresql': {
             'bin_dir': '',
             'use_slots': True,
-            'parameters': {p: v[0] for p, v in Postgresql.CMDLINE_OPTIONS.items()}
+            'parameters': CaseInsensitiveDict({p: v[0] for p, v in Postgresql.CMDLINE_OPTIONS.items()})
+        },
+        'watchdog': {
+            'mode': 'automatic',
         }
     }
 
@@ -83,6 +88,9 @@ class Config(object):
     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:
@@ -174,7 +182,7 @@ class Config(object):
                     elif name not in ('connect_address', 'listen', 'data_dir', 'pgpass', 'authentication'):
                         config['postgresql'][name] = deepcopy(value)
             elif name in config:  # only variables present in __DEFAULT_CONFIG allowed to be overriden from DCS
-                if name == 'synchronous_mode':
+                if name in ('synchronous_mode', 'synchronous_mode_strict'):
                     config[name] = value
                 else:
                     config[name] = int(value)
@@ -235,16 +243,25 @@ class Config(object):
 
         for param in list(os.environ.keys()):
             if param.startswith(Config.PATRONI_ENV_PREFIX):
-                name, suffix = (param[8:].rsplit('_', 1) + [''])[:2]
+                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') \
-                            and '_' not in name:
+                    if suffix in ('HOST', 'HOSTS', 'PORT', 'SRV', 'URL', 'PROXY', 'CACERT', 'CERT',
+                                  'KEY', 'VERIFY', 'TOKEN', 'CHECKS', 'DC', 'NAMESPACE', 'CONTEXT',
+                                  'USE_ENDPOINTS', 'SCOPE_LABEL', 'ROLE_LABEL', 'POD_IP', 'PORTS', 'LABELS'):
                         value = os.environ.pop(param)
                         if suffix == 'PORT':
                             value = value and parse_int(value)
-                        elif suffix == 'HOSTS':
+                        elif suffix in ('HOSTS', 'PORTS', 'CHECKS'):
                             value = value and _parse_list(value)
+                        elif suffix == 'LABELS':
+                            if not value.strip().startswith('{'):
+                                value = '{{{0}}}'.format(value)
+                            try:
+                                value = yaml.safe_load(value)
+                            except Exception:
+                                logger.exception('Exception when parsing dict %s', value)
+                                value = None
                         if value:
                             ret[name.lower()][suffix.lower()] = value
                     # PATRONI_<username>_PASSWORD=<password>, PATRONI_<username>_OPTIONS=<option1,option2,...>
@@ -271,7 +288,7 @@ class Config(object):
                         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:
+            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'

patroni/ctl.py

@@ -4,27 +4,37 @@ Patroni Control
 
 import base64
 import click
+import codecs
 import datetime
 import dateutil.parser
+import cdiff
+import copy
+import difflib
+import io
 import json
 import logging
 import os
 import psycopg2
 import random
 import requests
+import subprocess
 import sys
+import tempfile
 import time
 import tzlocal
 import yaml
 
 from click import ClickException
+from contextlib import contextmanager
 from patroni.config import Config
 from patroni.dcs import get_dcs as _get_dcs
 from patroni.exceptions import PatroniException
 from patroni.postgresql import Postgresql
-from patroni.utils import is_valid_pg_version
+from patroni.utils import patch_config, polling_loop
+from patroni.version import __version__
 from prettytable import PrettyTable
 from six.moves.urllib_parse import urlparse
+from six import text_type
 
 CONFIG_DIR_PATH = click.get_app_dir('patroni')
 CONFIG_FILE_PATH = os.path.join(CONFIG_DIR_PATH, 'patronictl.yaml')
@@ -87,10 +97,12 @@ def store_config(config, path):
         yaml.dump(config, fd)
 
 
-option_format = click.option('--format', '-f', 'fmt', help='Output format (pretty, json)', default='pretty')
+option_format = click.option('--format', '-f', 'fmt', help='Output format (pretty, json, yaml)', default='pretty')
 option_watchrefresh = click.option('-w', '--watch', type=float, help='Auto update the screen every X seconds')
 option_watch = click.option('-W', is_flag=True, help='Auto update the screen every 2 seconds')
 option_force = click.option('--force', is_flag=True, help='Do not ask for confirmation at any point')
+arg_cluster_name = click.argument('cluster_name', required=False,
+                                  default=lambda: click.get_current_context().obj.get('scope'))
 
 
 @click.group()
@@ -140,9 +152,12 @@ def print_output(columns, rows=None, alignment=None, fmt='pretty', header=True,
         click.echo(t)
         return
 
-    if fmt == 'json':
+    if fmt in ['json', 'yaml', 'yml']:
         elements = [dict(zip(columns, r)) for r in rows]
-        click.echo(json.dumps(elements))
+        if fmt == 'json':
+            click.echo(json.dumps(elements))
+        elif fmt in ('yaml', 'yml'):
+            click.echo(yaml.safe_dump(elements, encoding=None, default_flow_style=False, allow_unicode=True, width=200))
 
     if fmt == 'tsv':
         if columns is not None and header:
@@ -248,9 +263,9 @@ def get_members(cluster, cluster_name, member_names, role, force, action):
         member_names = [click.prompt('Which member do you want to {0} [{1}]?'.format(action,
                         ', '.join(candidates.keys())), type=str, default='')]
 
-    for mn in member_names:
-        if mn not in candidates:
-            raise PatroniCtlException('{0} is not a member of cluster'.format(mn))
+    for member_name in member_names:
+        if member_name not in candidates:
+            raise PatroniCtlException('{0} is not a member of cluster'.format(member_name))
 
     if not force:
         confirm = click.confirm('Are you sure you want to {0} members {1}?'.format(action, ', '.join(member_names)))
@@ -264,7 +279,7 @@ def get_members(cluster, cluster_name, member_names, role, force, action):
 @click.option('--role', '-r', help='Give a dsn of any member with this role', type=click.Choice(['master', 'replica',
               'any']), default=None)
 @click.option('--member', '-m', help='Generate a dsn for this member', type=str)
-@click.argument('cluster_name')
+@arg_cluster_name
 @click.pass_obj
 def dsn(obj, cluster_name, role, member):
     if role is not None and member is not None:
@@ -282,7 +297,7 @@ def dsn(obj, cluster_name, role, member):
 
 
 @ctl.command('query', help='Query a Patroni PostgreSQL member')
-@click.argument('cluster_name')
+@arg_cluster_name
 @option_format
 @click.option('--format', 'fmt', help='Output format (pretty, json)', default='tsv')
 @click.option('--file', '-f', 'p_file', help='Execute the SQL commands from this file', type=click.File('rb'))
@@ -397,7 +412,7 @@ def remove(obj, cluster_name, fmt):
     if message != confirm:
         raise PatroniCtlException('You did not exactly type "{0}"'.format(message))
 
-    if cluster.leader:
+    if cluster.leader and cluster.leader.name:
         confirm = click.prompt('This cluster currently is healthy. Please specify the master name to continue')
         if confirm != cluster.leader.name:
             raise PatroniCtlException('You did not specify the current master of the cluster')
@@ -410,8 +425,10 @@ def check_response(response, member_name, action_name, silent_success=False):
         click.echo('Failed: {0} for member {1}, status code={2}, ({3})'.format(
             action_name, member_name, response.status_code, response.text
         ))
+        return False
     elif not silent_success:
         click.echo('Success: {0} for member {1}'.format(action_name, member_name))
+    return True
 
 
 def parse_scheduled(scheduled):
@@ -460,11 +477,12 @@ def restart(obj, cluster_name, member_names, force, role, p_any, scheduled, vers
         content['restart_pending'] = True
 
     if version:
-        if not is_valid_pg_version(version):
-            message = 'PostgreSQL version should be in the first.major.minor format'
-            raise PatroniCtlException(message)
-        else:
-            content['postgres_version'] = version
+        try:
+            Postgresql.postgres_version_to_int(version)
+        except PatroniException as e:
+            raise PatroniCtlException(e.value)
+
+        content['postgres_version'] = version
 
     if scheduled is None and not force:
         scheduled = click.prompt('When should the restart take place (e.g. 2015-10-01T14:30) ', type=str, default='now')
@@ -507,35 +525,33 @@ def reinit(obj, cluster_name, member_names, force):
     members = get_members(cluster, cluster_name, member_names, None, force, 'reinitialize')
 
     for member in members:
-        r = request_patroni(member, 'post', 'reinitialize', headers=auth_header(obj))
-        check_response(r, member.name, 'reinitialize')
+        body = {'force': force}
+        while True:
+            r = request_patroni(member, 'post', 'reinitialize', body, auth_header(obj))
+            if not check_response(r, member.name, 'reinitialize') and r.text.endswith(' already in progress') \
+                    and not force and click.confirm('Do you want to cancel it and reinitialize anyway?'):
+                body['force'] = True
+                continue
+            break
 
 
-@ctl.command('failover', help='Failover to a replica')
-@click.argument('cluster_name')
-@click.option('--master', help='The name of the current master', default=None)
-@click.option('--candidate', help='The name of the candidate', default=None)
-@click.option('--scheduled', help='Timestamp of a scheduled failover in unambiguous format (e.g. ISO 8601)',
-              default=None)
-@option_force
-@click.pass_obj
-def failover(obj, cluster_name, master, candidate, force, scheduled):
+def _do_failover_or_switchover(obj, action, cluster_name, master, candidate, force, scheduled=None):
     """
-        We want to trigger a failover for the specified cluster name.
+        We want to trigger a failover or switchover for the specified cluster name.
 
         We verify that the cluster name, master name and candidate name are correct.
-        If so, we trigger a failover and keep the client up to date.
+        If so, we trigger an action and keep the client up to date.
     """
 
     dcs = get_dcs(obj, cluster_name)
     cluster = dcs.get_cluster()
 
-    if cluster.leader is None and not cluster.is_paused():
+    if action == 'switchover' and cluster.leader is None:
         raise PatroniCtlException('This cluster has no master')
 
-    if master is None and (not cluster.is_paused() or cluster.leader):
-        if force:
-            master = cluster.leader.member.name
+    if master is None:
+        if force or action == 'failover':
+            master = cluster.leader and cluster.leader.name
         else:
             master = click.prompt('Master', type=str, default=cluster.leader.member.name)
 
@@ -547,28 +563,32 @@ def failover(obj, cluster_name, master, candidate, force, scheduled):
     candidate_names.sort()
 
     if not candidate_names:
-        raise PatroniCtlException('No candidates found to failover to')
+        raise PatroniCtlException('No candidates found to {0} to'.format(action))
 
     if candidate is None and not force:
         candidate = click.prompt('Candidate ' + str(candidate_names), type=str, default='')
 
+    if action == 'failover' and not candidate:
+        raise PatroniCtlException('Failover could be performed only to a specific candidate')
+
     if candidate == master:
-        raise PatroniCtlException('Failover target and source are the same.')
+        raise PatroniCtlException(action.title() + ' target and source are the same.')
 
     if candidate and candidate not in candidate_names:
         raise PatroniCtlException('Member {0} does not exist in cluster {1}'.format(candidate, cluster_name))
 
-    if scheduled is None and not force:
-        scheduled = click.prompt('When should the failover take place (e.g. 2015-10-01T14:30) ', type=str,
-                                 default='now')
-
-    scheduled_at = parse_scheduled(scheduled)
-
     scheduled_at_str = None
-    if scheduled_at:
-        if cluster.is_paused():
-            raise PatroniCtlException("Can't schedule failover in the paused state")
-        scheduled_at_str = scheduled_at.isoformat()
+
+    if action == 'switchover':
+        if scheduled is None and not force:
+            scheduled = click.prompt('When should the switchover take place (e.g. 2015-10-01T14:30) ',
+                                     type=str, default='now')
+
+        scheduled_at = parse_scheduled(scheduled)
+        if scheduled_at:
+            if cluster.is_paused():
+                raise PatroniCtlException("Can't schedule switchover in the paused state")
+            scheduled_at_str = scheduled_at.isoformat()
 
     failover_value = {'leader': master, 'candidate': candidate, 'scheduled_at': scheduled_at_str}
 
@@ -579,46 +599,76 @@ def failover(obj, cluster_name, master, candidate, force, scheduled):
     output_members(dcs.get_cluster(), cluster_name)
 
     if not force:
-        a = \
-            click.confirm('Are you sure you want to failover cluster {0}, demoting current master {1}?'.format(
-                cluster_name, master))
-        if not a:
-            raise PatroniCtlException('Aborting failover')
+        demote_msg = ', demoting current master ' + master if master else ''
+
+        if not click.confirm('Are you sure you want to {0} cluster {1}{2}?'.format(action, cluster_name, demote_msg)):
+            raise PatroniCtlException('Aborting ' + action)
 
     r = None
     try:
-        member = cluster.leader.member if cluster.leader else [m for m in cluster.members if m.name == candidate][0]
+        member = cluster.leader.member if cluster.leader else cluster.get_member(candidate, False)
+
+        r = request_patroni(member, 'post', action, failover_value, auth_header(obj))
+
+        # probably old patroni, which doesn't support switchover yet
+        if r.status_code == 501 and action == 'switchover' and 'Server does not support this operation' in r.text:
+            r = request_patroni(member, 'post', 'failover', failover_value, auth_header(obj))
 
-        r = request_patroni(member, 'post', 'failover', failover_value, auth_header(obj))
         if r.status_code in (200, 202):
             logging.debug(r)
             cluster = dcs.get_cluster()
             logging.debug(cluster)
             click.echo('{0} {1}'.format(timestamp(), r.text))
         else:
-            click.echo('Failover failed, details: {0}, {1}'.format(r.status_code, r.text))
+            click.echo('{0} failed, details: {1}, {2}'.format(action.title(), r.status_code, r.text))
             return
     except Exception:
         logging.exception(r)
         logging.warning('Failing over to DCS')
-        click.echo(timestamp() + ' Could not failover using Patroni api, falling back to DCS')
-        click.echo(timestamp() + ' Initializing failover from master {0}'.format(master))
+        click.echo('{0} Could not {1} using Patroni api, falling back to DCS'.format(timestamp(), action))
         dcs.manual_failover(master, candidate, scheduled_at=scheduled_at)
 
     output_members(cluster, cluster_name)
 
 
+@ctl.command('failover', help='Failover to a replica')
+@arg_cluster_name
+@click.option('--master', help='The name of the current master', default=None)
+@click.option('--candidate', help='The name of the candidate', default=None)
+@option_force
+@click.pass_obj
+def failover(obj, cluster_name, master, candidate, force):
+    action = 'switchover' if master else 'failover'
+    _do_failover_or_switchover(obj, action, cluster_name, master, candidate, force)
+
+
+@ctl.command('switchover', help='Switchover to a replica')
+@arg_cluster_name
+@click.option('--master', help='The name of the current master', default=None)
+@click.option('--candidate', help='The name of the candidate', default=None)
+@click.option('--scheduled', help='Timestamp of a scheduled switchover in unambiguous format (e.g. ISO 8601)',
+              default=None)
+@option_force
+@click.pass_obj
+def switchover(obj, cluster_name, master, candidate, force, scheduled):
+    _do_failover_or_switchover(obj, 'switchover', cluster_name, master, candidate, force, scheduled)
+
+
 def output_members(cluster, name, extended=False, fmt='pretty'):
     rows = []
     logging.debug(cluster)
     leader_name = None
     if cluster.leader:
-        leader_name = cluster.leader.member.name
+        leader_name = cluster.leader.name
 
     xlog_location_cluster = cluster.last_leader_operation or 0
 
     # Mainly for consistent pretty printing and watching we sort the output
     cluster.members.sort(key=lambda x: x.name)
+
+    has_scheduled_restarts = any(m.data.get('scheduled_restart') for m in cluster.members)
+    has_pending_restarts = any(m.data.get('pending_restart') for m in cluster.members)
+
     for m in cluster.members:
         logging.debug(m)
 
@@ -628,22 +678,19 @@ def output_members(cluster, name, extended=False, fmt='pretty'):
         elif m.name == cluster.sync.sync_standby:
             role = 'Sync standby'
 
-        host = m.conn_kwargs()['host']
-
-        xlog_location = m.data.get('xlog_location') or 0
+        xlog_location = m.data.get('xlog_location')
         lag = ''
-        if xlog_location_cluster >= xlog_location:
+        if xlog_location is None:
+            lag = 'unknown'
+        elif xlog_location_cluster >= xlog_location:
             lag = round((xlog_location_cluster - xlog_location)/1024/1024)
 
-        row = [
-            name,
-            m.name,
-            host,
-            role,
-            m.data.get('state', ''),
-            lag,
-        ]
-        if extended:
+        row = [name, m.name, m.conn_kwargs()['host'], role, m.data.get('state', ''), lag]
+
+        if extended or has_pending_restarts:
+            row.append('*' if m.data.get('pending_restart') else '')
+
+        if extended or has_scheduled_restarts:
             value = ''
             scheduled_restart = m.data.get('scheduled_restart')
             if scheduled_restart:
@@ -655,39 +702,55 @@ def output_members(cluster, name, extended=False, fmt='pretty'):
 
         rows.append(row)
 
-    columns = [
-        'Cluster',
-        'Member',
-        'Host',
-        'Role',
-        'State',
-        'Lag in MB',
-    ]
-    alignment = {'Cluster': 'l', 'Member': 'l', 'Host': 'l', 'Lag in MB': 'r'}
+    columns = ['Cluster', 'Member', 'Host', 'Role', 'State', 'Lag in MB']
+    alignment = {'Lag in MB': 'r'}
 
-    if extended:
+    if extended or has_pending_restarts:
+        columns.append('Pending restart')
+
+    if extended or has_scheduled_restarts:
         columns.append('Scheduled restart')
-        alignment['Scheduled restart'] = 'l'
 
     print_output(columns, rows, alignment, fmt)
 
+    service_info = []
+    if cluster.is_paused():
+        service_info.append('Maintenance mode: on')
+
+    if cluster.failover and cluster.failover.scheduled_at:
+        info = 'Switchover scheduled at: ' + cluster.failover.scheduled_at.isoformat()
+        if cluster.failover.leader:
+            info += '\n                    from: ' + cluster.failover.leader
+        if cluster.failover.candidate:
+            info += '\n                      to: ' + cluster.failover.candidate
+        service_info.append(info)
+
+    if service_info:
+        click.echo(' ' + '\n '.join(service_info))
+
 
 @ctl.command('list', help='List the Patroni members for a given Patroni')
 @click.argument('cluster_names', nargs=-1)
 @click.option('--extended', '-e', help='Show some extra information', is_flag=True)
+@click.option('--timestamp', '-t', 'ts', help='Print timestamp', is_flag=True)
 @option_format
 @option_watch
 @option_watchrefresh
 @click.pass_obj
-def members(obj, cluster_names, fmt, watch, w, extended):
+def members(obj, cluster_names, fmt, watch, w, extended, ts):
     if not cluster_names:
-        logging.warning('Listing members: No cluster names were provided')
-        return
+        if 'scope' in obj:
+            cluster_names = [obj['scope']]
+        if not cluster_names:
+            return logging.warning('Listing members: No cluster names were provided')
 
     for cluster_name in cluster_names:
         dcs = get_dcs(obj, cluster_name)
 
         for _ in watching(w, watch):
+            if ts:
+                click.echo(timestamp(0))
+
             cluster = dcs.get_cluster()
             output_members(cluster, cluster_name, extended, fmt)
 
@@ -723,7 +786,7 @@ def touch_member(config, dcs):
         'role': p.role
     }
 
-    return dcs.touch_member(json.dumps(data, separators=(',', ':')), permanent=True)
+    return dcs.touch_member(data, permanent=True)
 
 
 def set_defaults(config, cluster_name):
@@ -780,8 +843,27 @@ def flush(obj, cluster_name, member_names, force, role, target):
                 click.echo('No scheduled restart for member {0}'.format(member.name))
 
 
-def toggle_pause(config, cluster_name, paused):
-    cluster = get_dcs(config, cluster_name).get_cluster()
+def wait_until_pause_is_applied(dcs, paused, old_cluster):
+    click.echo("'{0}' request sent, waiting until it is recognized by all nodes".format(paused and 'pause' or 'resume'))
+    old = {m.name: m.index for m in old_cluster.members if m.api_url}
+    loop_wait = old_cluster.config.data.get('loop_wait', dcs.loop_wait)
+
+    for _ in polling_loop(loop_wait + 1):
+        cluster = dcs.get_cluster()
+        if all(m.data.get('pause', False) == paused for m in cluster.members if m.name in old):
+            break
+    else:
+        remaining = [m.name for m in cluster.members if m.data.get('pause', False) != paused
+                     and m.name in old and old[m.name] != m.index]
+        if remaining:
+            return click.echo("{0} members didn't recognized pause state after {1} seconds"
+                              .format(', '.join(remaining), loop_wait))
+    return click.echo('Success: cluster management is {0}'.format(paused and 'paused' or 'resumed'))
+
+
+def toggle_pause(config, cluster_name, paused, wait):
+    dcs = get_dcs(config, cluster_name)
+    cluster = dcs.get_cluster()
     if cluster.is_paused() == paused:
         raise PatroniCtlException('Cluster is {0} paused'.format(paused and 'already' or 'not'))
 
@@ -798,7 +880,10 @@ def toggle_pause(config, cluster_name, paused):
             continue
 
         if r.status_code == 200:
-            click.echo('Success: cluster management is {0}'.format(paused and 'paused' or 'resumed'))
+            if wait:
+                wait_until_pause_is_applied(dcs, paused, cluster)
+            else:
+                click.echo('Success: cluster management is {0}'.format(paused and 'paused' or 'resumed'))
         else:
             click.echo('Failed: {0} cluster management status code={1}, ({2})'.format(
                        paused and 'pause' or 'resume', r.status_code, r.text))
@@ -808,14 +893,250 @@ def toggle_pause(config, cluster_name, paused):
 
 
 @ctl.command('pause', help='Disable auto failover')
-@click.argument('cluster_name')
+@arg_cluster_name
 @click.pass_obj
-def pause(obj, cluster_name):
-    return toggle_pause(obj, cluster_name, True)
+@click.option('--wait', help='Wait until pause is applied on all nodes', is_flag=True)
+def pause(obj, cluster_name, wait):
+    return toggle_pause(obj, cluster_name, True, wait)
 
 
 @ctl.command('resume', help='Resume auto failover')
-@click.argument('cluster_name')
+@arg_cluster_name
+@click.option('--wait', help='Wait until pause is cleared on all nodes', is_flag=True)
 @click.pass_obj
-def resume(obj, cluster_name):
-    return toggle_pause(obj, cluster_name, False)
+def resume(obj, cluster_name, wait):
+    return toggle_pause(obj, cluster_name, False, wait)
+
+
+@contextmanager
+def temporary_file(contents, suffix='', prefix='tmp'):
+    """Creates a temporary file with specified contents that persists for the context.
+
+    :param contents: binary string that will be written to the file.
+    :param prefix: will be prefixed to the filename.
+    :param suffix: will be appended to the filename.
+    :returns path of the created file.
+    """
+    tmp = tempfile.NamedTemporaryFile(suffix=suffix, prefix=prefix, delete=False)
+    with tmp:
+        tmp.write(contents)
+
+    try:
+        yield tmp.name
+    finally:
+        os.unlink(tmp.name)
+
+
+def show_diff(before_editing, after_editing):
+    """Shows a diff between two strings.
+
+    If the output is to a tty the diff will be colored. Inputs are expected to be unicode strings.
+    """
+    def listify(string):
+        return [l+'\n' for l in string.rstrip('\n').split('\n')]
+
+    unified_diff = difflib.unified_diff(listify(before_editing), listify(after_editing))
+
+    if sys.stdout.isatty():
+        buf = io.StringIO()
+        for line in unified_diff:
+            # Force cast to unicode as difflib on Python 2.7 returns a mix of unicode and str.
+            buf.write(text_type(line))
+        buf.seek(0)
+
+        class opts:
+            side_by_side = False
+            width = 80
+            tab_width = 8
+        cdiff.markup_to_pager(cdiff.PatchStream(buf), opts)
+    else:
+        for line in unified_diff:
+            click.echo(line.rstrip('\n'))
+
+
+def format_config_for_editing(data):
+    """Formats configuration as YAML for human consumption.
+
+    :param data: configuration as nested dictionaries
+    :returns unicode YAML of the configuration"""
+    return yaml.safe_dump(data, default_flow_style=False, encoding=None, allow_unicode=True)
+
+
+def apply_config_changes(before_editing, data, kvpairs):
+    """Applies config changes specified as a list of key-value pairs.
+
+    Keys are interpreted as dotted paths into the configuration data structure. Except for paths beginning with
+    `postgresql.parameters` where rest of the path is used directly to allow for PostgreSQL GUCs containing dots.
+    Values are interpreted as YAML values.
+
+    :param before_editing: human representation before editing
+    :param data: configuration datastructure
+    :param kvpairs: list of strings containing key value pairs separated by =
+    :returns tuple of human readable and parsed datastructure after changes
+    """
+    changed_data = copy.deepcopy(data)
+
+    def set_path_value(config, path, value, prefix=()):
+        # Postgresql GUCs can't be nested, but can contain dots so we re-flatten the structure for this case
+        if prefix == ('postgresql', 'parameters'):
+            path = ['.'.join(path)]
+
+        key = path[0]
+        if len(path) == 1:
+            if value is None:
+                config.pop(key, None)
+            else:
+                config[key] = value
+        else:
+            if not isinstance(config.get(key), dict):
+                config[key] = {}
+            set_path_value(config[key], path[1:], value, prefix + (key,))
+            if config[key] == {}:
+                del config[key]
+
+    for pair in kvpairs:
+        if not pair or "=" not in pair:
+            raise PatroniCtlException("Invalid parameter setting {0}".format(pair))
+        key_path, value = pair.split("=", 1)
+        set_path_value(changed_data, key_path.strip().split("."), yaml.safe_load(value))
+
+    return format_config_for_editing(changed_data), changed_data
+
+
+def apply_yaml_file(data, filename):
+    """Applies changes from a YAML file to configuration
+
+    :param data: configuration datastructure
+    :param filename: name of the YAML file, - is taken to mean standard input
+    :returns tuple of human readable and parsed datastructure after changes
+    """
+    changed_data = copy.deepcopy(data)
+
+    if filename == '-':
+        new_options = yaml.safe_load(sys.stdin)
+    else:
+        with open(filename) as fd:
+            new_options = yaml.safe_load(fd)
+
+    patch_config(changed_data, new_options)
+
+    return format_config_for_editing(changed_data), changed_data
+
+
+def invoke_editor(before_editing, cluster_name):
+    """Starts editor command to edit configuration in human readable format
+
+    :param before_editing: human representation before editing
+    :returns tuple of human readable and parsed datastructure after changes
+    """
+    editor_cmd = os.environ.get('EDITOR')
+    if not editor_cmd:
+        raise PatroniCtlException('EDITOR environment variable is not set')
+
+    with temporary_file(contents=before_editing.encode('utf-8'),
+                        suffix='.yaml',
+                        prefix='{0}-config-'.format(cluster_name)) as tmpfile:
+        ret = subprocess.call([editor_cmd, tmpfile])
+        if ret:
+            raise PatroniCtlException("Editor exited with return code {0}".format(ret))
+
+        with codecs.open(tmpfile, encoding='utf-8') as fd:
+            after_editing = fd.read()
+
+        return after_editing, yaml.safe_load(after_editing)
+
+
+@ctl.command('edit-config', help="Edit cluster configuration")
+@arg_cluster_name
+@click.option('--quiet', '-q', is_flag=True, help='Do not show changes')
+@click.option('--set', '-s', 'kvpairs', multiple=True,
+              help='Set specific configuration value. Can be specified multiple times')
+@click.option('--pg', '-p', 'pgkvpairs', multiple=True,
+              help='Set specific PostgreSQL parameter value. Shorthand for -s postgresql.parameters. '
+                   'Can be specified multiple times')
+@click.option('--apply', 'apply_filename', help='Apply configuration from file. Use - for stdin.')
+@click.option('--replace', 'replace_filename', help='Apply configuration from file, replacing existing configuration.'
+              ' Use - for stdin.')
+@option_force
+@click.pass_obj
+def edit_config(obj, cluster_name, force, quiet, kvpairs, pgkvpairs, apply_filename, replace_filename):
+    dcs = get_dcs(obj, cluster_name)
+    cluster = dcs.get_cluster()
+
+    before_editing = format_config_for_editing(cluster.config.data)
+
+    after_editing = None  # Serves as a flag if any changes were requested
+    changed_data = cluster.config.data
+
+    if replace_filename:
+        after_editing, changed_data = apply_yaml_file({}, replace_filename)
+
+    if apply_filename:
+        after_editing, changed_data = apply_yaml_file(changed_data, apply_filename)
+
+    if kvpairs or pgkvpairs:
+        all_pairs = list(kvpairs) + ['postgresql.parameters.'+v.lstrip() for v in pgkvpairs]
+        after_editing, changed_data = apply_config_changes(before_editing, changed_data, all_pairs)
+
+    # If no changes were specified on the command line invoke editor
+    if after_editing is None:
+        after_editing, changed_data = invoke_editor(before_editing, cluster_name)
+
+    if cluster.config.data == changed_data:
+        if not quiet:
+            click.echo("Not changed")
+        return
+
+    if not quiet:
+        show_diff(before_editing, after_editing)
+
+    if (apply_filename == '-' or replace_filename == '-') and not force:
+        click.echo("Use --force option to apply changes")
+        return
+
+    if force or click.confirm('Apply these changes?'):
+        if not dcs.set_config_value(json.dumps(changed_data), cluster.config.index):
+            raise PatroniCtlException("Config modification aborted due to concurrent changes")
+        click.echo("Configuration changed")
+
+
+@ctl.command('show-config', help="Show cluster configuration")
+@arg_cluster_name
+@click.pass_obj
+def show_config(obj, cluster_name):
+    cluster = get_dcs(obj, cluster_name).get_cluster()
+
+    click.echo(format_config_for_editing(cluster.config.data))
+
+
+@ctl.command('version', help='Output version of patronictl command or a running Patroni instance')
+@click.argument('cluster_name', required=False)
+@click.argument('member_names', nargs=-1)
+@click.pass_obj
+def version(obj, cluster_name, member_names):
+    click.echo("patronictl version {0}".format(__version__))
+
+    if not cluster_name:
+        return
+
+    click.echo("")
+    cluster = get_dcs(obj, cluster_name).get_cluster()
+    for m in cluster.members:
+        if m.api_url:
+            if not member_names or m.name in member_names:
+                try:
+                    response = request_patroni(m, 'get', 'patroni')
+                    data = response.json()
+                    version = data.get('patroni', {}).get('version')
+                    pg_version = data.get('server_version')
+                    pg_version_str = " PostgreSQL {0}".format(format_pg_version(pg_version)) if pg_version else ""
+                    click.echo("{0}: Patroni {1}{2}".format(m.name, version, pg_version_str))
+                except Exception as e:
+                    click.echo("{0}: failed to get version: {1}".format(m.name, e))
+
+
+def format_pg_version(version):
+    if version < 100000:
+        return "{0}.{1}.{2}".format(version // 10000, version // 100 % 100, version % 100)
+    else:
+        return "{0}.{1}".format(version // 10000, version % 100)

patroni/dcs/__init__.py

@@ -3,6 +3,7 @@ import dateutil
 import importlib
 import inspect
 import json
+import logging
 import os
 import pkgutil
 import six
@@ -10,10 +11,13 @@ import sys
 
 from collections import namedtuple
 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
 
+logger = logging.getLogger(__name__)
+
 
 def parse_connection_string(value):
     """Original Governor stores connection strings for each cluster members if a following format:
@@ -51,18 +55,22 @@ def dcs_modules():
 def get_dcs(config):
     available_implementations = set()
     for module_name in dcs_modules():
-        module = importlib.import_module(module_name)
-        for name in filter(lambda name: not name.startswith('__'), dir(module)):  # iterate through module content
-            value = getattr(module, name)
-            name = name.lower()
-            # try to find implementation of AbstractDCS interface, class name must match with module_name
-            if inspect.isclass(value) and issubclass(value, 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 value(config[name])
+        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))
 
@@ -170,40 +178,48 @@ class Leader(namedtuple('Leader', 'index,session,member')):
     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
-    True
+    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
-    True
+    False
     >>> Failover.from_node(1, '{}') is None
-    True
+    False
     >>> 'abc' in Failover.from_node(1, 'abc:def')
     True
     """
     @staticmethod
     def from_node(index, value):
-        if not value:
-            return None
-
-        try:
-            data = json.loads(value)
-            if not data:
-                return None
-        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
+        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'])
@@ -251,8 +267,12 @@ class SyncState(namedtuple('SyncState', 'index,leader,sync_standby')):
         True
         >>> SyncState.from_node(1, '{"leader": "leader"}').leader == "leader"
         True
+        >>> SyncState.from_node(1, {"leader": "leader"}).leader == "leader"
+        True
         """
-        if value:
+        if isinstance(value, dict):
+            data = value
+        elif value:
             try:
                 data = json.loads(value)
                 if not isinstance(data, dict):
@@ -282,7 +302,26 @@ class SyncState(namedtuple('SyncState', 'index,leader,sync_standby')):
         return name is not None and name in (self.leader, self.sync_standby)
 
 
-class Cluster(namedtuple('Cluster', 'initialize,config,leader,last_leader_operation,members,failover,sync')):
+class TimelineHistory(namedtuple('TimelineHistory', 'index,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, 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:
@@ -294,6 +333,7 @@ class Cluster(namedtuple('Cluster', 'initialize,config,leader,last_leader_operat
     :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):
@@ -310,8 +350,14 @@ class Cluster(namedtuple('Cluster', 'initialize,config,leader,last_leader_operat
         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.config and self.config.data.get('pause', False) or False
+        return self.check_mode('pause')
+
+    def is_synchronous_mode(self):
+        return self.check_mode('synchronous_mode')
 
 
 @six.add_metaclass(abc.ABCMeta)
@@ -321,6 +367,7 @@ class AbstractDCS(object):
     _CONFIG = 'config'
     _LEADER = 'leader'
     _FAILOVER = 'failover'
+    _HISTORY = 'history'
     _MEMBERS = 'members/'
     _OPTIME = 'optime'
     _LEADER_OPTIME = _OPTIME + '/' + _LEADER
@@ -332,8 +379,7 @@ class AbstractDCS(object):
             i.e.: `zookeeper` for zookeeper, `etcd` for etcd, etc...
         """
         self._name = config['name']
-        self._namespace = '/{0}'.format(config.get('namespace', '/service/').strip('/'))
-        self._base_path = '/'.join([self._namespace, config['scope']])
+        self._base_path = os.path.join('/', config.get('namespace', '/service/').strip('/'), config['scope'])
         self._set_loop_wait(config.get('loop_wait', 10))
 
         self._ctl = bool(config.get('patronictl', False))
@@ -369,6 +415,10 @@ class AbstractDCS(object):
     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)
@@ -411,7 +461,7 @@ class AbstractDCS(object):
         with self._cluster_thread_lock:
             try:
                 self._load_cluster()
-            except:
+            except Exception:
                 self._cluster = None
                 raise
             return self._cluster
@@ -436,7 +486,7 @@ class AbstractDCS(object):
             self._last_leader_operation = last_operation
 
     @abc.abstractmethod
-    def update_leader(self):
+    def _update_leader(self):
         """Update leader key (or session) ttl
 
         :returns: `!True` if leader key (or session) has been updated successfully.
@@ -445,6 +495,18 @@ class AbstractDCS(object):
         You have to use CAS (Compare And Swap) operation in order to update leader key,
         for example for etcd `prevValue` parameter must be used."""
 
+    def update_leader(self, last_operation):
+        """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
@@ -470,7 +532,6 @@ class AbstractDCS(object):
 
         if scheduled_at:
             failover_value['scheduled_at'] = scheduled_at.isoformat()
-
         return self.set_failover_value(json.dumps(failover_value, separators=(',', ':')), index)
 
     @abc.abstractmethod
@@ -483,7 +544,7 @@ class AbstractDCS(object):
         This method should create or update key with the name = '/members/' + `~self._name`
         and value = data in a given DCS.
 
-        :param data: json serialized information about instance (including connection strings)
+        :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.
@@ -520,8 +581,18 @@ class AbstractDCS(object):
     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):
-        return self.set_sync_state_value(json.dumps({'leader': leader, 'sync_standby': sync_standby}), index=index)
+        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):

patroni/dcs/consul.py

@@ -1,16 +1,18 @@
 from __future__ import absolute_import
+import json
 import logging
 import os
 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
+from patroni.dcs import AbstractDCS, ClusterConfig, Cluster, Failover, Leader, Member, SyncState, TimelineHistory
 from patroni.exceptions import DCSError
-from patroni.utils import Retry, RetryFailedError
+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
+from six.moves.urllib.parse import urlencode, urlparse
 from six.moves.http_client import HTTPException
 
 logger = logging.getLogger(__name__)
@@ -24,21 +26,39 @@ class ConsulInternalError(ConsulException):
     """An internal Consul server error occurred"""
 
 
+class InvalidSessionTTL(ConsulInternalError):
+    """Session TTL is too small or too big"""
+
+
 class HTTPClient(object):
 
-    def __init__(self, host='127.0.0.1', port=8500, scheme='http', verify=True, timeout=10):
-        self.host = host
-        self.port = port
-        self.scheme = scheme
-        self.verify = verify
-        self.set_read_timeout(timeout)
-        self.base_uri = '{0}://{1}:{2}'.format(self.scheme, self.host, self.port)
-        self.http = urllib3.PoolManager(num_pools=10)
+    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
@@ -48,7 +68,11 @@ class HTTPClient(object):
     def response(response):
         data = response.data.decode('utf-8')
         if response.status == 500:
-            raise ConsulInternalError('{0} {1}'.format(response.status, data))
+            msg = '{0} {1}'.format(response.status, data)
+            if data.startswith('Invalid Session TTL'):
+                raise InvalidSessionTTL(msg)
+            else:
+                raise ConsulInternalError(msg)
         return base.Response(response.status, response.headers, data)
 
     def uri(self, path, params=None):
@@ -72,15 +96,30 @@ class HTTPClient(object):
                 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):
 
-    @staticmethod
-    def connect(host, port, scheme, verify=True):
-        return HTTPClient(host, port, scheme, verify)
+    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):
@@ -103,12 +142,37 @@ class Consul(AbstractDCS):
                             retry_exceptions=(ConsulInternalError, HTTPException,
                                               HTTPError, socket.error, socket.timeout))
 
-        self._my_member_data = None
-        host, port = config.get('host', '127.0.0.1:8500').split(':')
-        self._client = ConsulClient(host=host, port=port)
+        self._my_member_data = {}
+        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')
         if not self._ctl:
             self.create_session()
 
@@ -132,6 +196,15 @@ class Consul(AbstractDCS):
         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():
@@ -144,8 +217,15 @@ class Consul(AbstractDCS):
                 self._session = None
         ret = not self._session
         if ret:
-            self._session = self._client.session.create(name=self._scope + '-' + self._name,
-                                                        lock_delay=0.001, behavior='delete')
+            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
 
@@ -184,6 +264,10 @@ class Consul(AbstractDCS):
             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'])
@@ -213,17 +297,17 @@ class Consul(AbstractDCS):
             sync = nodes.get(self._SYNC)
             sync = SyncState.from_node(sync and sync['ModifyIndex'], sync and sync['Value'])
 
-            self._cluster = Cluster(initialize, config, leader, last_leader_operation, members, failover, sync)
+            self._cluster = Cluster(initialize, config, leader, last_leader_operation, members, failover, sync, history)
         except NotFound:
-            self._cluster = Cluster(None, None, None, None, [], None, None)
-        except:
+            self._cluster = Cluster(None, None, None, None, [], None, None, None)
+        except Exception:
             logger.exception('get_cluster')
             raise ConsulError('Consul is not responding properly')
 
-    def touch_member(self, data, **kwargs):
+    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 = self.refresh_session()
+        create_member = not permanent and self.refresh_session()
 
         if member and (create_member or member.session != self._session):
             try:
@@ -232,12 +316,12 @@ class Consul(AbstractDCS):
             except Exception:
                 return False
 
-        if not create_member and member and data == self._my_member_data:
+        if not create_member and member and deep_compare(data, self._my_member_data):
             return True
 
         try:
-            args = {} if kwargs.get('permanent', False) else {'acquire': self._session}
-            self._client.kv.put(self.member_path, data, **args)
+            args = {} if permanent else {'acquire': self._session}
+            self._client.kv.put(self.member_path, json.dumps(data, separators=(',', ':')), **args)
             self._my_member_data = data
             return True
         except Exception:
@@ -245,12 +329,14 @@ class Consul(AbstractDCS):
         return False
 
     @catch_consul_errors
+    def _do_attempt_to_acquire_leader(self, kwargs):
+        return self.retry(self._client.kv.put, self.leader_path, self._name, **kwargs)
+
     def attempt_to_acquire_leader(self, permanent=False):
         if not self._session and not permanent:
             self.refresh_session()
 
-        args = {} if permanent else {'acquire': self._session}
-        ret = self.retry(self._client.kv.put, self.leader_path, self._name, **args)
+        ret = self._do_attempt_to_acquire_leader({} if permanent else {'acquire': self._session})
         if not ret:
             logger.info('Could not take out TTL lock')
         return ret
@@ -271,7 +357,7 @@ class Consul(AbstractDCS):
         return self._client.kv.put(self.leader_optime_path, last_operation)
 
     @catch_consul_errors
-    def update_leader(self):
+    def _update_leader(self):
         if self._session:
             self.retry(self._client.session.renew, self._session)
             self._last_session_refresh = time.time()
@@ -290,6 +376,10 @@ class Consul(AbstractDCS):
     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
@@ -298,11 +388,11 @@ class Consul(AbstractDCS):
 
     @catch_consul_errors
     def set_sync_state_value(self, value, index=None):
-        return self._client.kv.put(self.sync_path, value, cas=index)
+        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._client.kv.delete(self.sync_path, cas=index)
+        return self.retry(self._client.kv.delete, self.sync_path, cas=index)
 
     def watch(self, leader_index, timeout):
         if self.__do_not_watch:

patroni/dcs/etcd.py

@@ -1,18 +1,20 @@
 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
+from patroni.dcs import AbstractDCS, ClusterConfig, Cluster, Failover, Leader, Member, SyncState, TimelineHistory
 from patroni.exceptions import DCSError
-from patroni.utils import Retry, RetryFailedError
+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
@@ -23,6 +25,10 @@ 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
 
@@ -204,8 +210,11 @@ class Client(etcd.Client):
                 self._machines_cache = self.machines
                 if self._base_uri in self._machines_cache:
                     self._machines_cache.remove(self._base_uri)
-        except etcd.EtcdConnectionFailed:
-            self._update_machines_cache = True
+        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)
@@ -227,7 +236,7 @@ class Client(etcd.Client):
             protocol = 'https' if '-ssl' in r else 'http'
             endpoint = '/members' if '-server' in r else ''
             for host, port in self.get_srv_record('_etcd{0}._tcp.{1}'.format(r, srv)):
-                url = '{0}://{1}:{2}{3}'.format(protocol, host, port, endpoint)
+                url = uri(protocol, host, port, endpoint)
                 if endpoint:
                     try:
                         response = requests.get(url, timeout=self.read_timeout, verify=False)
@@ -254,10 +263,10 @@ class Client(etcd.Client):
                 host, port = sa[:2]
                 if af == socket.AF_INET6:
                     host = '[{0}]'.format(host)
-                ret.append('{0}://{1}:{2}'.format(self.protocol, host, port))
+                ret.append(uri(self.protocol, host, port))
             if ret:
                 return list(set(ret))
-        return ['{0}://{1}:{2}'.format(self.protocol, host, port)]
+        return [uri(self.protocol, host, port)]
 
     def _load_machines_cache(self):
         """This method should fill up `_machines_cache` from scratch.
@@ -267,17 +276,20 @@ class Client(etcd.Client):
 
         self._update_machines_cache = True
 
-        if 'srv' not in self._config and 'host' not in self._config:
-            raise Exception('Neither srv nor host url are defined in etcd section of config')
+        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 = ['{0}://{1}:{2}'.format(self.protocol, self._config['host'], self._config['port'])]
+            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'])
 
@@ -347,8 +359,20 @@ class Etcd(AbstractDCS):
             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 = (config['host'] + ':2379').split(':')[:2]
+            host, port = split_host_port(config['host'], 2379)
             config['host'] = host
             if 'port' not in config:
                 config['port'] = int(port)
@@ -434,6 +458,10 @@ class Etcd(AbstractDCS):
             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)
@@ -458,15 +486,16 @@ class Etcd(AbstractDCS):
             sync = nodes.get(self._SYNC)
             sync = SyncState.from_node(sync and sync.modifiedIndex, sync and sync.value)
 
-            self._cluster = Cluster(initialize, config, leader, last_leader_operation, members, failover, sync)
+            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)
+            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.retry(self._client.set, self.member_path, data, None if permanent else ttl or self._ttl)
 
     @catch_etcd_errors
@@ -499,7 +528,7 @@ class Etcd(AbstractDCS):
         return self._client.set(self.leader_optime_path, last_operation)
 
     @catch_etcd_errors
-    def update_leader(self):
+    def _update_leader(self):
         return self.retry(self._client.test_and_set, self.leader_path, self._name, self._name, self._ttl)
 
     @catch_etcd_errors
@@ -518,9 +547,13 @@ class Etcd(AbstractDCS):
     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._client.write(self.sync_path, value, prevIndex=index or 0)
+        return self.retry(self._client.write, self.sync_path, value, prevIndex=index or 0)
 
     @catch_etcd_errors
     def delete_sync_state(self, index=None):

patroni/dcs/kubernetes.py

@@ -0,0 +1,409 @@
+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 '{}')
+
+            # 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):
+        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
+
+        ret = self.patch_or_create(self.leader_path, annotations, self._leader_resource_version, subsets=self.__subsets)
+        if ret:
+            self._leader_resource_version = ret.metadata.resource_version
+        return ret
+
+    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)
+        ret = self.patch_or_create(self.leader_path, annotations, self._leader_resource_version, subsets=self.__subsets)
+        if ret:
+            self._leader_resource_version = ret.metadata.resource_version
+        else:
+            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:
+                    logging.exception('watch')
+
+                timeout = end_time - time.time()
+
+        try:
+            return super(Kubernetes, self).watch(None, timeout)
+        finally:
+            self.event.clear()

patroni/dcs/zookeeper.py

@@ -1,10 +1,13 @@
+import json
 import logging
+import time
 
-from kazoo.client import KazooClient, KazooState
+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
+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__)
 
@@ -51,11 +54,12 @@ class ZooKeeper(AbstractDCS):
             hosts = ','.join(hosts)
 
         self._client = KazooClient(hosts, handler=PatroniSequentialThreadingHandler(config['retry_timeout']),
-                                   timeout=config['ttl'], connection_retry={'max_delay': 1, 'max_tries': -1},
-                                   command_retry={'deadline': config['retry_timeout'], 'max_delay': 1, 'max_tries': -1})
+                                   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._my_member_data = None
+        self._my_member_data = {}
         self._fetch_cluster = True
 
         self._orig_kazoo_connect = self._client._connection._connect
@@ -114,7 +118,8 @@ class ZooKeeper(AbstractDCS):
             return True
 
     def set_retry_timeout(self, retry_timeout):
-        self._client._retry.deadline = 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:
@@ -156,6 +161,10 @@ class ZooKeeper(AbstractDCS):
         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
@@ -188,7 +197,7 @@ class ZooKeeper(AbstractDCS):
         failover = self.get_node(self.failover_path, watch=self.cluster_watcher) if self._FAILOVER in nodes else None
         failover = failover and Failover.from_node(failover[1].version, failover[0])
 
-        self._cluster = Cluster(initialize, config, leader, last_leader_operation, members, failover, sync)
+        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:
@@ -203,7 +212,7 @@ class ZooKeeper(AbstractDCS):
         try:
             self._client.retry(self._client.create, path, value.encode('utf-8'), **kwargs)
             return True
-        except:
+        except Exception:
             return False
 
     def attempt_to_acquire_leader(self, permanent=False):
@@ -212,16 +221,19 @@ class ZooKeeper(AbstractDCS):
             logger.info('Could not take out TTL lock')
         return ret
 
-    def set_failover_value(self, value, index=None):
+    def __set_failover_or_sync_state_value(self, key, value, index=None):
         try:
-            self._client.retry(self._client.set, self.failover_path, value.encode('utf-8'), version=index or -1)
+            self._client.retry(self._client.set, key, value.encode('utf-8'), version=index or -1)
             return True
         except NoNodeError:
-            return value == '' or (index is None and self._create(self.failover_path, value))
-        except:
+            return value == '' or (index is None and self._create(key, value))
+        except Exception:
             logging.exception('set_failover_value')
             return False
 
+    def set_failover_value(self, value, index=None):
+        return self.__set_failover_or_sync_state_value(self.failover_path, value, index)
+
     def set_config_value(self, value, index=None):
         try:
             self._client.retry(self._client.set, self.config_path, value.encode('utf-8'), version=index or -1)
@@ -239,22 +251,23 @@ class ZooKeeper(AbstractDCS):
     def touch_member(self, data, ttl=None, permanent=False):
         cluster = self.cluster
         member = cluster and cluster.get_member(self._name, fallback_to_leader=False)
-        data = data.encode('utf-8')
+        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:
+            except Exception:
                 return False
             member = None
 
         if member:
-            if data == self._my_member_data:
+            if deep_compare(data, self._my_member_data):
                 return True
         else:
             try:
-                self._client.create_async(self.member_path, data, makepath=True, ephemeral=not permanent).get(timeout=1)
+                self._client.create_async(self.member_path, encoded_data, makepath=True,
+                                          ephemeral=not permanent).get(timeout=1)
                 self._my_member_data = data
                 return True
             except Exception as e:
@@ -262,10 +275,10 @@ class ZooKeeper(AbstractDCS):
                     logger.exception('touch_member')
                     return False
         try:
-            self._client.set_async(self.member_path, data).get(timeout=1)
+            self._client.set_async(self.member_path, encoded_data).get(timeout=1)
             self._my_member_data = data
             return True
-        except:
+        except Exception:
             logger.exception('touch_member')
 
         return False
@@ -273,22 +286,25 @@ class ZooKeeper(AbstractDCS):
     def take_leader(self):
         return self.attempt_to_acquire_leader()
 
-    def _write_leader_optime(self, last_operation):
-        last_operation = last_operation.encode('utf-8')
+    def __write_leader_optime_or_history_value(self, key, value):
+        value = value.encode('utf-8')
         try:
-            self._client.set_async(self.leader_optime_path, last_operation).get(timeout=1)
+            self._client.set_async(key, value).get(timeout=1)
             return True
         except NoNodeError:
             try:
-                self._client.create_async(self.leader_optime_path, last_operation, makepath=True).get(timeout=1)
+                self._client.create_async(key, value, makepath=True).get(timeout=1)
                 return True
-            except:
-                logger.exception('Failed to create %s', self.leader_optime_path)
-        except:
-            logger.exception('Failed to update %s', self.leader_optime_path)
+            except Exception:
+                logger.exception('Failed to create %s', key)
+        except Exception:
+            logger.exception('Failed to update %s', key)
         return False
 
-    def update_leader(self):
+    def _write_leader_optime(self, last_operation):
+        return self.__write_leader_optime_or_history_value(self.leader_optime_path, last_operation)
+
+    def _update_leader(self):
         return True
 
     def delete_leader(self):
@@ -304,7 +320,7 @@ class ZooKeeper(AbstractDCS):
     def cancel_initialization(self):
         try:
             self._client.retry(self._cancel_initialization)
-        except:
+        except Exception:
             logger.exception("Unable to delete initialize key")
 
     def delete_cluster(self):
@@ -313,15 +329,11 @@ class ZooKeeper(AbstractDCS):
         except NoNodeError:
             return True
 
+    def set_history_value(self, value):
+        return self.__write_leader_optime_or_history_value(self.history_path, value)
+
     def set_sync_state_value(self, value, index=None):
-        try:
-            self._client.retry(self._client.set, self.sync_path, value.encode('utf-8'), version=index or -1)
-            return True
-        except NoNodeError:
-            return value == '' or (index is None and self._create(self.sync_path, value))
-        except:
-            logging.exception('set_sync_state_value')
-            return False
+        return self.__set_failover_or_sync_state_value(self.sync_path, value, index)
 
     def delete_sync_state(self, index=None):
         return self.set_sync_state_value("{}", index)

patroni/exceptions.py

@@ -23,3 +23,7 @@ class DCSError(PatroniException):
 
 class PostgresConnectionException(PostgresException):
     pass
+
+
+class WatchdogError(PatroniException):
+    pass

patroni/ha.py

@@ -9,8 +9,8 @@ import time
 
 from collections import namedtuple
 from multiprocessing.pool import ThreadPool
-from patroni.async_executor import AsyncExecutor
-from patroni.exceptions import DCSError, PostgresConnectionException
+from patroni.async_executor import AsyncExecutor, CriticalTask
+from patroni.exceptions import DCSError, PostgresConnectionException, PatroniException
 from patroni.postgresql import ACTION_ON_START
 from patroni.utils import polling_loop, tzutc
 from threading import RLock
@@ -18,24 +18,25 @@ from threading import RLock
 logger = logging.getLogger(__name__)
 
 
-class _MemberStatus(namedtuple('_MemberStatus', 'member,reachable,in_recovery,xlog_location,tags')):
+class _MemberStatus(namedtuple('_MemberStatus', 'member,reachable,in_recovery,wal_position,tags,watchdog_failed')):
     """Node status distilled from API response:
 
         member - dcs.Member object of the node
         reachable - `!False` if the node is not reachable or is not responding with correct JSON
         in_recovery - `!True` if pg_is_in_recovery() == true
-        xlog_location - value of `replayed_location` or `location` from JSON, dependin on its role.
+        wal_position - value of `replayed_location` or `location` from JSON, dependin on its role.
         tags - dictionary with values of different tags (i.e. nofailover)
+        watchdog_failed - indicates that watchdog is required by configuration but not available or failed
     """
     @classmethod
     def from_api_response(cls, member, json):
         is_master = json['role'] == 'master'
-        xlog = not is_master and max(json['xlog'].get('received_location', 0), json['xlog'].get('replayed_location', 0))
-        return cls(member, True, not is_master, xlog, json.get('tags', {}))
+        wal = not is_master and max(json['xlog'].get('received_location', 0), json['xlog'].get('replayed_location', 0))
+        return cls(member, True, not is_master, wal, json.get('tags', {}), json.get('watchdog_failed', False))
 
     @classmethod
     def unknown(cls, member):
-        return cls(member, False, None, 0, {})
+        return cls(member, False, None, 0, {}, False)
 
     def failover_limitation(self):
         """Returns reason why this node can't promote or None if everything is ok."""
@@ -43,6 +44,8 @@ class _MemberStatus(namedtuple('_MemberStatus', 'member,reachable,in_recovery,xl
             return 'not reachable'
         if self.tags.get('nofailover', False):
             return 'not allowed to promote'
+        if self.watchdog_failed:
+            return 'not watchdog capable'
         return None
 
 
@@ -54,9 +57,14 @@ class Ha(object):
         self.dcs = patroni.dcs
         self.cluster = None
         self.old_cluster = None
+        self._was_paused = False
+        self._leader_timeline = None
         self.recovering = False
+        self._post_bootstrap_task = None
+        self._crash_recovery_executed = False
         self._start_timeout = None
-        self._async_executor = AsyncExecutor(self.wakeup)
+        self._async_executor = AsyncExecutor(self.state_handler, self.wakeup)
+        self.watchdog = patroni.watchdog
 
         # Each member publishes various pieces of information to the DCS using touch_member. This lock protects
         # the state and publishing procedure to have consistent ordering and avoid publishing stale values.
@@ -65,8 +73,19 @@ class Ha(object):
         # standby. Changes protected by _member_state_lock.
         self._disable_sync = 0
 
+        # We need following property to avoid shutdown of postgres when join of Patroni to the postgres
+        # already running as replica was aborted due to cluster not beeing initialized in DCS.
+        self._join_aborted = False
+
+    def check_mode(self, mode):
+        # Try to protect from the case when DCS was wiped out during pause
+        if self.cluster and self.cluster.config and self.cluster.config.modify_index:
+            return self.cluster.check_mode(mode)
+        else:
+            return self.patroni.config.check_mode(mode)
+
     def is_paused(self):
-        return self.cluster and self.cluster.is_paused()
+        return self.check_mode('pause')
 
     def load_cluster_from_dcs(self):
         cluster = self.dcs.get_cluster()
@@ -76,16 +95,21 @@ class Ha(object):
             self.old_cluster = cluster
         self.cluster = cluster
 
+        self._leader_timeline = None if cluster.is_unlocked() else cluster.leader.timeline
+
     def acquire_lock(self):
         return self.dcs.attempt_to_acquire_leader()
 
     def update_lock(self, write_leader_optime=False):
-        ret = self.dcs.update_leader()
-        if ret and write_leader_optime:
+        last_operation = None
+        if write_leader_optime:
             try:
-                self.dcs.write_leader_optime(self.state_handler.last_operation())
-            except:
-                pass
+                last_operation = self.state_handler.last_operation()
+            except Exception:
+                logger.exception('Exception when called state_handler.last_operation()')
+        ret = self.dcs.update_leader(last_operation)
+        if ret:
+            self.watchdog.keepalive()
         return ret
 
     def has_lock(self):
@@ -114,24 +138,33 @@ class Ha(object):
                 data['tags'] = tags
             if self.state_handler.pending_restart:
                 data['pending_restart'] = True
-            if not self._async_executor.busy and data['state'] in ['running', 'restarting', 'starting']:
+            if self._async_executor.scheduled_action in (None, 'promote') \
+                    and data['state'] in ['running', 'restarting', 'starting']:
                 try:
-                    data['xlog_location'] = self.state_handler.xlog_position(retry=False)
-                except:
+                    timeline, wal_position = self.state_handler.timeline_wal_position()
+                    data['xlog_location'] = wal_position
+                    if not timeline:
+                        timeline = self.state_handler.replica_cached_timeline(self._leader_timeline)
+                    if timeline:
+                        data['timeline'] = timeline
+                except Exception:
                     pass
             if self.patroni.scheduled_restart:
                 scheduled_restart_data = self.patroni.scheduled_restart.copy()
                 scheduled_restart_data['schedule'] = scheduled_restart_data['schedule'].isoformat()
                 data['scheduled_restart'] = scheduled_restart_data
 
-            return self.dcs.touch_member(json.dumps(data, separators=(',', ':')))
+            if self.is_paused():
+                data['pause'] = True
+
+            return self.dcs.touch_member(data)
 
     def clone(self, clone_member=None, msg='(without leader)'):
         if self.state_handler.clone(clone_member):
             logger.info('bootstrapped %s', msg)
             cluster = self.dcs.get_cluster()
             node_to_follow = self._get_node_to_follow(cluster)
-            return self.state_handler.follow(node_to_follow, cluster.leader, True)
+            return self.state_handler.follow(node_to_follow)
         else:
             logger.error('failed to bootstrap %s', msg)
             self.state_handler.remove_data_directory()
@@ -147,44 +180,82 @@ class Ha(object):
         # no initialize key and node is allowed to be master and has 'bootstrap' section in a configuration file
         elif self.cluster.initialize is None and not self.patroni.nofailover and 'bootstrap' in self.patroni.config:
             if self.dcs.initialize(create_new=True):  # race for initialization
-                try:
-                    self.state_handler.bootstrap(self.patroni.config['bootstrap'])
-                    self.dcs.initialize(create_new=False, sysid=self.state_handler.sysid)
-                except:  # initdb or start failed
-                    # remove initialization key and give a chance to other members
-                    logger.info("removing initialize key after failed attempt to initialize the cluster")
-                    self.dcs.cancel_initialization()
-                    self.state_handler.stop('immediate')
-                    self.state_handler.move_data_directory()
-                    raise
-                self.dcs.set_config_value(json.dumps(self.patroni.config.dynamic_configuration, separators=(',', ':')))
-                self.dcs.take_leader()
-                self.load_cluster_from_dcs()
-                return 'initialized a new cluster'
+                self.state_handler.bootstrapping = True
+                self._post_bootstrap_task = CriticalTask()
+                self._async_executor.schedule('bootstrap')
+                self._async_executor.run_async(self.state_handler.bootstrap, args=(self.patroni.config['bootstrap'],))
+                return 'trying to bootstrap a new cluster'
             else:
                 return 'failed to acquire initialize lock'
         else:
             if self.state_handler.can_create_replica_without_replication_connection():
+                msg = 'bootstrap (without leader)'
+                self._async_executor.schedule(msg)
                 self._async_executor.run_async(self.clone)
-                return "trying to bootstrap (without leader)"
+                return 'trying to ' + msg
             return 'waiting for leader to bootstrap'
 
+    def _handle_rewind(self):
+        if self.state_handler.rewind_needed_and_possible(self.cluster.leader):
+            self._async_executor.schedule('running pg_rewind from ' + self.cluster.leader.name)
+            self._async_executor.run_async(self.state_handler.rewind, (self.cluster.leader,))
+            return True
+
+    def _start_crash_recovery(self, msg):
+        self._async_executor.schedule(msg)
+        self._async_executor.run_async(self.state_handler.fix_cluster_state)
+        return msg
+
     def recover(self):
+        # Postgres is not running and we will restart in standby mode. Watchdog is not needed until we promote.
+        self.watchdog.disable()
+
         if self.has_lock() and self.update_lock():
             timeout = self.patroni.config['master_start_timeout']
             if timeout == 0:
                 # We are requested to prefer failing over to restarting master. But see first if there
                 # is anyone to fail over to.
-                if self.is_failover_possible(self.cluster.members):
+                members = self.cluster.members
+                if self.is_synchronous_mode():
+                    members = [m for m in members if self.cluster.sync.matches(m.name)]
+                if self.is_failover_possible(members):
                     logger.info("Master crashed. Failing over.")
                     self.demote('immediate')
                     return 'stopped PostgreSQL to fail over after a crash'
         else:
             timeout = None
 
+        data = self.state_handler.controldata()
+        if data.get('Database cluster state') == 'in production' and not self._crash_recovery_executed and \
+                (self.cluster.is_unlocked() or self.state_handler.can_rewind):
+            self._crash_recovery_executed = True
+            return self._start_crash_recovery('doing crash recovery in a single user mode')
+
+        self.load_cluster_from_dcs()
+
+        if self.has_lock():
+            msg = "starting as readonly because i had the session lock"
+            node_to_follow = None
+        else:
+            if not self.state_handler.rewind_executed:
+                self.state_handler.trigger_check_diverged_lsn()
+            if self._handle_rewind():
+                return self._async_executor.scheduled_action
+            msg = "starting as a secondary"
+            node_to_follow = self._get_node_to_follow(self.cluster)
+
+        # once we already tried to start postgres but failed, single user mode is a rescue in this case
+        if self.recovering and not self.state_handler.rewind_executed \
+                and not self._crash_recovery_executed and self.state_handler.can_rewind \
+                and data.get('Database cluster state') not in ('shut down', 'shut down in recovery'):
+            self.recovering = False
+            return self._start_crash_recovery('fixing cluster state in a single user mode')
+
         self.recovering = True
-        return self.follow("starting as readonly because i had the session lock",
-                           "starting as a secondary", True, True, None, timeout)
+
+        self._async_executor.schedule('restarting after failure')
+        self._async_executor.run_async(self.state_handler.follow, (node_to_follow, timeout))
+        return msg
 
     def _get_node_to_follow(self, cluster):
         # determine the node to follow. If replicatefrom tag is set,
@@ -196,32 +267,39 @@ class Ha(object):
 
         return node_to_follow if node_to_follow and node_to_follow.name != self.state_handler.name else None
 
-    def follow(self, demote_reason, follow_reason, refresh=True, recovery=False, need_rewind=None, timeout=None):
+    def follow(self, demote_reason, follow_reason, refresh=True):
         if refresh:
             self.load_cluster_from_dcs()
 
-        if recovery:
-            ret = demote_reason if self.has_lock() else follow_reason
-        else:
-            is_leader = self.state_handler.is_leader()
-            ret = demote_reason if is_leader else follow_reason
+        is_leader = self.state_handler.is_leader()
 
         node_to_follow = self._get_node_to_follow(self.cluster)
 
-        if self.is_paused() and not (self.state_handler.need_rewind and self.state_handler.can_rewind):
-            self.state_handler.set_role('master' if is_leader else 'replica')
-            if is_leader:
-                return 'continue to run as master without lock'
-            elif not node_to_follow:
-                return 'no action'
+        if self.is_paused():
+            if not (self.state_handler.need_rewind and self.state_handler.can_rewind) or self.cluster.is_unlocked():
+                self.state_handler.set_role('master' if is_leader else 'replica')
+                if is_leader:
+                    return 'continue to run as master without lock'
+                elif not node_to_follow:
+                    return 'no action'
+        elif is_leader:
+            self.demote('immediate-nolock')
+            return demote_reason
 
-        self.state_handler.follow(node_to_follow, self.cluster.leader, recovery,
-                                  self._async_executor, need_rewind, timeout)
+        if self._handle_rewind():
+            return self._async_executor.scheduled_action
 
-        return ret
+        if not self.state_handler.check_recovery_conf(node_to_follow):
+            self._async_executor.schedule('changing primary_conninfo and restarting')
+            self._async_executor.run_async(self.state_handler.follow, (node_to_follow,))
+
+        return follow_reason
 
     def is_synchronous_mode(self):
-        return bool(self.cluster and self.cluster.config and self.cluster.config.data.get('synchronous_mode'))
+        return self.check_mode('synchronous_mode')
+
+    def is_synchronous_mode_strict(self):
+        return self.check_mode('synchronous_mode_strict')
 
     def process_sync_replication(self):
         """Process synchronous standby beahvior.
@@ -242,10 +320,15 @@ class Ha(object):
                     if not self.dcs.write_sync_state(self.state_handler.name, None, index=self.cluster.sync.index):
                         logger.info('Synchronous replication key updated by someone else.')
                         return
+
+                if self.is_synchronous_mode_strict() and picked is None:
+                    picked = '*'
+                    logger.warning("No standbys available!")
+
                 logger.info("Assigning synchronous standby status to %s", picked)
                 self.state_handler.set_synchronous_standby(picked)
 
-                if picked and not allow_promote:
+                if picked and picked != '*' and not allow_promote:
                     # Wait for PostgreSQL to enable synchronous mode and see if we can immediately set sync_standby
                     time.sleep(2)
                     picked, allow_promote = self.state_handler.pick_synchronous_standby(self.cluster)
@@ -304,11 +387,41 @@ class Ha(object):
             with self._member_state_lock:
                 self._disable_sync -= 1
 
+    def update_cluster_history(self):
+        master_timeline = self.state_handler.get_master_timeline()
+        cluster_history = self.cluster.history and self.cluster.history.lines
+        if master_timeline == 1:
+            if cluster_history:
+                self.dcs.set_history_value('[]')
+        elif not cluster_history or cluster_history[-1][0] != master_timeline - 1 or len(cluster_history[-1]) != 4:
+            cluster_history = {l[0]: l for l in cluster_history or []}
+            history = self.state_handler.get_history(master_timeline)
+            if history:
+                for line in history:
+                    # enrich current history with promotion timestamps stored in DCS
+                    if len(line) == 3 and line[0] in cluster_history \
+                            and len(cluster_history[line[0]]) == 4 \
+                            and cluster_history[line[0]][1] == line[1]:
+                        line.append(cluster_history[line[0]][3])
+                self.dcs.set_history_value(json.dumps(history, separators=(',', ':')))
+
     def enforce_master_role(self, message, promote_message):
-        if self.state_handler.is_leader() or self.state_handler.role == 'master':
+        if not self.is_paused() and not self.watchdog.is_running and not self.watchdog.activate():
+            if self.state_handler.is_leader():
+                self.demote('immediate')
+                return 'Demoting self because watchdog could not be activated'
+            else:
+                self.release_leader_key_voluntarily()
+                return 'Not promoting self because watchdog could not be activated'
+
+        if self.state_handler.is_leader():
             # Inform the state handler about its master role.
             # It may be unaware of it if postgres is promoted manually.
             self.state_handler.set_role('master')
+            self.process_sync_replication()
+            self.update_cluster_history()
+            return message
+        elif self.state_handler.role == 'master':
             self.process_sync_replication()
             return message
         else:
@@ -319,8 +432,10 @@ class Ha(object):
                     # Somebody else updated sync state, it may be due to us losing the lock. To be safe, postpone
                     # promotion until next cycle. TODO: trigger immediate retry of run_cycle
                     return 'Postponing promotion because synchronous replication state was updated by somebody else'
-                self.state_handler.set_synchronous_standby(None)
-            self.state_handler.promote()
+                self.state_handler.set_synchronous_standby('*' if self.is_synchronous_mode_strict() else None)
+            if self.state_handler.role != 'master':
+                self._async_executor.schedule('promote')
+                self._async_executor.run_async(self.state_handler.promote, args=(self.dcs.loop_wait,))
             return promote_message
 
     @staticmethod
@@ -344,21 +459,21 @@ class Ha(object):
         pool.join()
         return results
 
-    def is_lagging(self, xlog_location):
-        """Returns if instance with an xlog should consider itself unhealthy to be promoted due to replication lag.
+    def is_lagging(self, wal_position):
+        """Returns if instance with an wal should consider itself unhealthy to be promoted due to replication lag.
 
-        :param xlog_location: Current xlog location.
+        :param wal_position: Current wal position.
         :returns True when node is lagging
         """
-        lag = (self.cluster.last_leader_operation or 0) - xlog_location
+        lag = (self.cluster.last_leader_operation or 0) - wal_position
         return lag > self.state_handler.config.get('maximum_lag_on_failover', 0)
 
     def _is_healthiest_node(self, members, check_replication_lag=True):
         """This method tries to determine whether I am healthy enough to became a new leader candidate or not."""
 
-        my_xlog_location = self.state_handler.xlog_position()
-        if check_replication_lag and self.is_lagging(my_xlog_location):
-            return False  # Too far behind last reported xlog location on master
+        _, my_wal_position = self.state_handler.timeline_wal_position()
+        if check_replication_lag and self.is_lagging(my_wal_position):
+            return False  # Too far behind last reported wal position on master
 
         # Prepare list of nodes to run check against
         members = [m for m in members if m.name != self.state_handler.name and not m.nofailover and m.api_url]
@@ -369,7 +484,7 @@ class Ha(object):
                     if not st.in_recovery:
                         logger.warning('Master (%s) is still alive', st.member.name)
                         return False
-                    if my_xlog_location < st.xlog_location:
+                    if my_wal_position < st.wal_position:
                         return False
         return True
 
@@ -381,7 +496,7 @@ class Ha(object):
                 not_allowed_reason = st.failover_limitation()
                 if not_allowed_reason:
                     logger.info('Member %s is %s', st.member.name, not_allowed_reason)
-                elif self.is_lagging(st.xlog_location):
+                elif self.is_lagging(st.wal_position):
                     logger.info('Member %s exceeds maximum replication lag', st.member.name)
                 else:
                     ret = True
@@ -458,6 +573,9 @@ class Ha(object):
         if self.cluster.failover:
             return self.manual_failover_process_no_leader()
 
+        if not self.watchdog.is_healthy:
+            return False
+
         # When in sync mode, only last known master and sync standby are allowed to promote automatically.
         all_known_members = self.cluster.members + self.old_cluster.members
         if self.is_synchronous_mode() and self.cluster.sync.leader:
@@ -485,31 +603,42 @@ class Ha(object):
             graceful is used when failing over to another node due to user request. May only be called running async.
             immediate is used when we determine that we are not suitable for master and want to failover quickly
                 without regard for data durability. May only be called synchronously.
+            immediate-nolock is used when find out that we have lost the lock to be master. Need to bring down
+                PostgreSQL as quickly as possible without regard for data durability. May only be called synchronously.
         """
-        assert mode in ['offline', 'graceful', 'immediate']
-        if mode != 'offline':
-            if mode == 'immediate':
-                self.state_handler.stop('immediate', checkpoint=False)
-            else:
-                self.state_handler.stop()
-            self.state_handler.set_role('demoted')
+        mode_control = {
+            'offline':          dict(stop='fast', checkpoint=False, release=False, offline=True, async=False),
+            'graceful':         dict(stop='fast', checkpoint=True, release=True, offline=False, async=False),
+            'immediate':        dict(stop='immediate', checkpoint=False, release=True, offline=False, async=True),
+            'immediate-nolock': dict(stop='immediate', checkpoint=False, release=False, offline=False, async=True),
+        }[mode]
+
+        self.state_handler.trigger_check_diverged_lsn()
+        self.state_handler.stop(mode_control['stop'], checkpoint=mode_control['checkpoint'],
+                                on_safepoint=self.watchdog.disable if self.watchdog.is_running else None)
+        self.state_handler.set_role('demoted')
+
+        if mode_control['release']:
             self.release_leader_key_voluntarily()
             time.sleep(2)  # Give a time to somebody to take the leader lock
-            cluster = self.dcs.get_cluster()
-            node_to_follow = self._get_node_to_follow(cluster)
-            if mode == 'immediate':
-                # We will try to start up as a standby now. If no one takes the leader lock before we finish
-                # recovery we will try to promote ourselves.
-                self._async_executor.schedule('waiting for failover to complete')
-                self._async_executor.run_async(self.state_handler.follow,
-                                               (node_to_follow, cluster.leader, True, None, True))
-            else:
-                return self.state_handler.follow(node_to_follow, cluster.leader, recovery=True, need_rewind=True)
+        if mode_control['offline']:
+            node_to_follow, leader = None, None
         else:
-            # Need to become unavailable as soon as possible, so initiate a stop here. However as we can't release
-            # the leader key we don't care about confirming the shutdown quickly and can use a regular stop.
-            self.state_handler.stop(checkpoint=False)
-            self.state_handler.follow(None, None, recovery=True)
+            cluster = self.dcs.get_cluster()
+            node_to_follow, leader = self._get_node_to_follow(cluster), cluster.leader
+
+        # FIXME: with mode offline called from DCS exception handler and handle_long_action_in_progress
+        # there could be an async action already running, calling follow from here will lead
+        # to racy state handler state updates.
+        if mode_control['async']:
+            self._async_executor.schedule('starting after demotion')
+            self._async_executor.run_async(self.state_handler.follow, (node_to_follow,))
+        else:
+            if self.is_synchronous_mode():
+                self.state_handler.set_synchronous_standby(None)
+            if self.state_handler.rewind_needed_and_possible(leader):
+                return False  # do not start postgres, but run pg_rewind on the next iteration
+            self.state_handler.follow(node_to_follow)
 
     def should_run_scheduled_action(self, action_name, scheduled_at, cleanup_fn):
         if scheduled_at and not self.is_paused():
@@ -565,8 +694,16 @@ class Ha(object):
                 if not failover.candidate and self.is_paused():
                     logger.warning('Failover is possible only to a specific candidate in a paused state')
                 else:
-                    members = [m for m in self.cluster.members
-                               if not failover.candidate or m.name == failover.candidate]
+                    if self.is_synchronous_mode():
+                        if failover.candidate and not self.cluster.sync.matches(failover.candidate):
+                            logger.warning('Failover candidate=%s does not match with sync_standby=%s',
+                                           failover.candidate, self.cluster.sync.sync_standby)
+                            members = []
+                        else:
+                            members = [m for m in self.cluster.members if self.cluster.sync.matches(m.name)]
+                    else:
+                        members = [m for m in self.cluster.members
+                                   if not failover.candidate or m.name == failover.candidate]
                     if self.is_failover_possible(members):  # check that there are healthy members
                         self._async_executor.schedule('manual failover: demote')
                         self._async_executor.run_async(self.demote, ('graceful',))
@@ -604,17 +741,15 @@ class Ha(object):
         else:
             # when we are doing manual failover there is no guaranty that new leader is ahead of any other node
             # node tagged as nofailover can be ahead of the new leader either, but it is always excluded from elections
-            need_rewind = bool(self.cluster.failover) or self.patroni.nofailover
-            if need_rewind:
+            if bool(self.cluster.failover) or self.patroni.nofailover:
+                self.state_handler.trigger_check_diverged_lsn()
                 time.sleep(2)  # Give a time to somebody to take the leader lock
 
             if self.patroni.nofailover:
                 return self.follow('demoting self because I am not allowed to become master',
-                                   'following a different leader because I am not allowed to promote',
-                                   need_rewind=need_rewind)
+                                   'following a different leader because I am not allowed to promote')
             return self.follow('demoting self because i am not the healthiest node',
-                               'following a different leader because i am not the healthiest node',
-                               need_rewind=need_rewind)
+                               'following a different leader because i am not the healthiest node')
 
     def process_healthy_cluster(self):
         if self.has_lock():
@@ -637,14 +772,16 @@ class Ha(object):
                 # Either there is no connection to DCS or someone else acquired the lock
                 logger.error('failed to update leader lock')
                 if self.state_handler.is_leader():
-                    self.demote('offline')
+                    if self.is_paused():
+                        return 'continue to run as master after failing to update leader lock in DCS'
+                    self.demote('immediate-nolock')
                     return 'demoted self because failed to update leader lock in DCS'
                 else:
                     return 'not promoting because failed to update leader lock in DCS'
         else:
             logger.info('does not have lock')
         return self.follow('demoting self because i do not have the lock and i was a leader',
-                           'no action.  i am a secondary and i am following a leader', False)
+                           'no action.  i am a secondary and i am following a leader', refresh=False)
 
     def evaluate_scheduled_restart(self):
         if self._async_executor.busy:  # Restart already in progress
@@ -736,13 +873,13 @@ class Ha(object):
             # leader key (if it belong to us) rather than trying to start postgres once again.
             self.recovering = True
 
-        # No that restart is scheduled we can set timeout for startup, it will get reset
+        # Now that restart is scheduled we can set timeout for startup, it will get reset
         # once async executor runs and main loop notices PostgreSQL as up.
         timeout = restart_data.get('timeout', self.patroni.config['master_start_timeout'])
         self.set_start_timeout(timeout)
 
         # For non async cases we want to wait for restart to complete or timeout before returning.
-        do_restart = functools.partial(self.state_handler.restart, timeout)
+        do_restart = functools.partial(self.state_handler.restart, timeout, self._async_executor.critical_task)
         if self.is_synchronous_mode() and not self.has_lock():
             do_restart = functools.partial(self.while_not_sync_standby, do_restart)
 
@@ -766,7 +903,7 @@ class Ha(object):
         member_role = 'leader' if clone_member == self.cluster.leader else 'replica'
         return self.clone(clone_member, "from {0} '{1}'".format(member_role, clone_member.name))
 
-    def reinitialize(self):
+    def reinitialize(self, force=False):
         with self._async_executor:
             self.load_cluster_from_dcs()
 
@@ -776,22 +913,36 @@ class Ha(object):
             if self.cluster.leader.name == self.state_handler.name:
                 return 'I am the leader, can not reinitialize'
 
-            action = self._async_executor.schedule('reinitialize', immediately=True)
+        if force:
+            self._async_executor.cancel()
+
+        with self._async_executor:
+            action = self._async_executor.schedule('reinitialize')
             if action is not None:
                 return '{0} already in progress'.format(action)
 
         self._async_executor.run_async(self._do_reinitialize, args=(self.cluster, ))
 
     def handle_long_action_in_progress(self):
-        if self.has_lock():
-            if self.update_lock():
-                return 'updated leader lock during ' + self._async_executor.scheduled_action
-            else:
-                return 'failed to update leader lock during ' + self._async_executor.scheduled_action
-        elif self.cluster.is_unlocked():
-            return 'not healthy enough for leader race'
-        else:
-            return self._async_executor.scheduled_action + ' in progress'
+        if self.has_lock() and self.update_lock():
+            return 'updated leader lock during ' + self._async_executor.scheduled_action
+        elif not self.state_handler.bootstrapping:
+            # Don't have lock, make sure we are not starting up a master in the background
+            if self.state_handler.role == 'master':
+                logger.info("Demoting master during " + self._async_executor.scheduled_action)
+                if self._async_executor.scheduled_action == 'restart':
+                    # Restart needs a special interlocking cancel because postmaster may be just started in a
+                    # background thread and has not even written a pid file yet.
+                    with self._async_executor.critical_task as task:
+                        if not task.cancel():
+                            self.state_handler.terminate_starting_postmaster(postmaster=task.result)
+                self.demote('immediate-nolock')
+                return 'lost leader lock during ' + self._async_executor.scheduled_action
+
+        if self.cluster.is_unlocked():
+            logger.info('not healthy enough for leader race')
+
+        return self._async_executor.scheduled_action + ' in progress'
 
     @staticmethod
     def sysid_valid(sysid):
@@ -802,14 +953,48 @@ class Ha(object):
 
     def post_recover(self):
         if not self.state_handler.is_running():
+            self.watchdog.disable()
             if self.has_lock():
                 self.state_handler.set_role('demoted')
                 self.dcs.delete_leader()
                 self.dcs.reset_cluster()
                 return 'removed leader key after trying and failing to start postgres'
             return 'failed to start postgres'
+        self._crash_recovery_executed = False
         return None
 
+    def cancel_initialization(self):
+        logger.info('removing initialize key after failed attempt to bootstrap the cluster')
+        self.dcs.cancel_initialization()
+        self.state_handler.stop('immediate')
+        self.state_handler.move_data_directory()
+        raise PatroniException('Failed to bootstrap cluster')
+
+    def post_bootstrap(self):
+        # bootstrap has failed if postgres is not running
+        if not self.state_handler.is_running() or self._post_bootstrap_task.result is False:
+            self.cancel_initialization()
+
+        if self._post_bootstrap_task.result is None:
+            if not self.state_handler.is_leader():
+                return 'waiting for end of recovery after bootstrap'
+
+            self.state_handler.set_role('master')
+            self._async_executor.schedule('post_bootstrap')
+            self._async_executor.run_async(self.state_handler.post_bootstrap,
+                                           args=(self.patroni.config['bootstrap'], self._post_bootstrap_task))
+            return 'running post_bootstrap'
+
+        self.state_handler.bootstrapping = False
+        self.dcs.set_config_value(json.dumps(self.patroni.config.dynamic_configuration, separators=(',', ':')))
+        if not self.watchdog.activate():
+            logger.error('Cancelling bootstrap because watchdog activation failed')
+            self.cancel_initialization()
+        self.dcs.take_leader()
+        self.state_handler.call_nowait(ACTION_ON_START)
+        self.load_cluster_from_dcs()
+        return 'initialized a new cluster'
+
     def handle_starting_instance(self):
         """Starting up PostgreSQL may take a long time. In case we are the leader we may want to
         fail over to."""
@@ -817,13 +1002,15 @@ class Ha(object):
         # Check if we are in startup, when paused defer to main loop for manual failovers.
         if not self.state_handler.check_for_startup() or self.is_paused():
             self.set_start_timeout(None)
+            if self.is_paused():
+                self.state_handler.set_state(self.state_handler.is_running() and 'running' or 'stopped')
             return None
 
         # state_handler.state == 'starting' here
         if self.has_lock():
             if not self.update_lock():
                 logger.info("Lost lock while starting up. Demoting self.")
-                self.demote('immediate')
+                self.demote('immediate-nolock')
                 return 'stopped PostgreSQL while starting up because leader key was lost'
 
             timeout = self._start_timeout or self.patroni.config['master_start_timeout']
@@ -856,8 +1043,17 @@ class Ha(object):
     def _run_cycle(self):
         dcs_failed = False
         try:
+            self.state_handler.reset_cluster_info_state()
             self.load_cluster_from_dcs()
 
+            if self.is_paused():
+                self.watchdog.disable()
+                self._was_paused = True
+            else:
+                if self._was_paused:
+                    self.state_handler.schedule_sanity_checks_after_pause()
+                self._was_paused = False
+
             if not self.cluster.has_member(self.state_handler.name):
                 self.touch_member()
 
@@ -877,6 +1073,9 @@ class Ha(object):
                 return msg
 
             # we've got here, so any async action has finished.
+            if self.state_handler.bootstrapping:
+                return self.post_bootstrap()
+
             if self.recovering and not self.state_handler.need_rewind:
                 self.recovering = False
                 # Check if we tried to recover and failed
@@ -886,6 +1085,11 @@ class Ha(object):
 
             # is data directory empty?
             if self.state_handler.data_directory_empty():
+                self.state_handler.set_role('uninitialized')
+                self.state_handler.stop('immediate')
+                # In case datadir went away while we were master.
+                self.watchdog.disable()
+
                 # is this instance the leader?
                 if self.has_lock():
                     self.release_leader_key_voluntarily()
@@ -894,9 +1098,20 @@ class Ha(object):
                 return self.bootstrap()  # new node
             # "bootstrap", but data directory is not empty
             elif not self.sysid_valid(self.cluster.initialize) and self.cluster.is_unlocked() and not self.is_paused():
+                if not self.state_handler.cb_called and self.state_handler.is_running() \
+                        and not self.state_handler.is_leader():
+                    self._join_aborted = True
+                    logger.error('No initialize key in DCS and PostgreSQL is running as replica, aborting start')
+                    logger.error('Please first start Patroni on the node running as master')
+                    sys.exit(1)
                 self.dcs.initialize(create_new=(self.cluster.initialize is None), sysid=self.state_handler.sysid)
             else:
                 # check if we are allowed to join
+                data_sysid = self.state_handler.sysid
+                if not self.sysid_valid(data_sysid):
+                    # data directory is not empty, but no valid sysid, cluster must be broken, suggest reinit
+                    return "data dir for the cluster is not empty, but system ID is invalid; consider doing reinitalize"
+
                 if self.sysid_valid(self.cluster.initialize) and self.cluster.initialize != self.state_handler.sysid:
                     logger.fatal("system ID mismatch, node %s belongs to a different cluster: %s != %s",
                                  self.state_handler.name, self.cluster.initialize, self.state_handler.sysid)
@@ -908,7 +1123,8 @@ class Ha(object):
                         self.dcs.delete_leader()
                         self.dcs.reset_cluster()
                         return 'removed leader lock because postgres is not running'
-                    elif not (self.state_handler.need_rewind and self.state_handler.can_rewind):
+                    elif not (self.state_handler.rewind_executed or
+                              self.state_handler.need_rewind and self.state_handler.can_rewind):
                         return 'postgres is not running'
 
                 # try to start dead postgres
@@ -926,6 +1142,8 @@ class Ha(object):
                 # asynchronous processes are running (should be always the case for the master)
                 if not self._async_executor.busy and not self.state_handler.is_starting():
                     if not self.state_handler.cb_called:
+                        if not self.state_handler.is_leader():
+                            self.state_handler.trigger_check_diverged_lsn()
                         self.state_handler.call_nowait(ACTION_ON_START)
                     self.state_handler.sync_replication_slots(self.cluster)
         except DCSError:
@@ -946,6 +1164,27 @@ class Ha(object):
             info = self._run_cycle()
             return (self.is_paused() and 'PAUSE: ' or '') + info
 
+    def shutdown(self):
+        if self.is_paused():
+            logger.info('Leader key is not deleted and Postgresql is not stopped due paused state')
+            self.watchdog.disable()
+        elif not self._join_aborted:
+            # FIXME: If stop doesn't reach safepoint quickly enough keepalive is triggered. If shutdown checkpoint
+            # takes longer than ttl, then leader key is lost and replication might not have sent out all xlog.
+            # This might not be the desired behavior of users, as a graceful shutdown of the host can mean lost data.
+            # We probably need to something smarter here.
+            disable_wd = self.watchdog.disable if self.watchdog.is_running else None
+            self.while_not_sync_standby(lambda: self.state_handler.stop(checkpoint=False, on_safepoint=disable_wd))
+            if not self.state_handler.is_running():
+                if self.has_lock():
+                    self.dcs.delete_leader()
+            else:
+                # XXX: what about when Patroni is started as the wrong user that has access to the watchdog device
+                # but cannot shut down PostgreSQL. Root would be the obvious example. Would be nice to not kill the
+                # system due to a bad config.
+                logger.error("PostgreSQL shutdown failed, leader key not removed." +
+                             (" Leaving watchdog running." if self.watchdog.is_running else ""))
+
     def watch(self, timeout):
         cluster = self.cluster
         # watch on leader key changes if the postgres is running and leader is known and current node is not lock owner

patroni/postmaster.py

@@ -0,0 +1,165 @@
+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,
+}
+
+
+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('Too much difference between %s and %s', 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 .. 9.6
+        aux_proc_re = re.compile("(?:postgres:)( .*:)? (?:""(?:startup|logger|checkpointer|writer|wal writer|"
+                                 "autovacuum launcher|autovacuum worker|stats collector|wal receiver|archiver|"
+                                 "wal sender) process|bgworker: )")
+
+        try:
+            user_backends = []
+            user_backends_cmdlines = []
+            for child in self.children():
+                try:
+                    cmdline = child.cmdline()[0]
+                    if not aux_proc_re.match(cmdline):
+                        user_backends.append(child)
+                        user_backends_cmdlines.append(cmdline)
+                except psutil.NoSuchProcess:
+                    pass
+            if user_backends:
+                logger.debug('Waiting for user backends %s to close', ', '.join(user_backends_cmdlines))
+                psutil.wait_procs(user_backends)
+            logger.debug("Backends closed")
+        except psutil.Error:
+            logger.exception('wait_for_user_backends_to_close')
+
+    @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.
+        env = {p: os.environ[p] for p in ('PATH', 'LC_ALL', 'LANG') 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.
+                env['PG_GRANDPARENT_PID'] = str(proc.pid)
+        except psutil.NoSuchProcess:
+            pass
+
+        proc = call_self(['pg_ctl_start', pgcommand, '-D', data_dir,
+                          '--config-file={}'.format(conf)] + options, close_fds=True,
+                         preexec_fn=os.setsid, 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

@@ -10,27 +10,27 @@ from patroni.utils import Retry, RetryFailedError
 
 logger = logging.getLogger(__name__)
 
-retry_timeout = 15
-
 
 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=retry_timeout, max_delay=5, max_tries=-1, retry_exceptions=(boto.exception,))
+        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
 

patroni/scripts/wale_restore.py

@@ -23,19 +23,29 @@
 # 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 time
-import argparse
+
+from collections import namedtuple
 
 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
@@ -50,66 +60,141 @@ def get_major_version(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, no_master, retries):
+    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.no_master = no_master
-        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))
+
+        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:
-            try:
-                ret = self.should_use_s3_to_create_replica()
-                if ret:
-                    return self.create_replica_with_s3()
-                elif ret is None:  # caught an exception, need to retry
-                    return 1
-            except Exception:
-                logger.exception("Exception when running WAL-E restore")
-        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.decode('utf-8').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))
+            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:
+        except KeyError:
             logger.exception("unable to get some of WALE backup parameters")
             return None
 
@@ -124,24 +209,29 @@ class WALERestore(object):
         # construct the LSN from the segment and offset
         backup_start_lsn = '{0}/{1}'.format(lsn_segment, lsn_offset)
 
-        diff_in_bytes = int(backup_size)
+        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_is_in_recovery()
                                                        THEN GREATEST(
-                                                                pg_xlog_location_diff(COALESCE(
-                                                                  pg_last_xlog_receive_location(), '0/0'), %s)::bigint,
-                                                                pg_xlog_location_diff(
-                                                                  pg_last_xlog_replay_location(), %s)::bigint)
-                                                       ELSE pg_xlog_location_diff(
-                                                                pg_current_xlog_location(), %s)::bigint
-                                                   END""", (backup_start_lsn, backup_start_lsn, backup_start_lsn))
+                                                                pg_{0}_{1}_diff(COALESCE(
+                                                                  pg_last_{0}_receive_{1}(), '0/0'), %s)::bigint,
+                                                                pg_{0}_{1}_diff(pg_last_{0}_replay_{1}(), %s)::bigint)
+                                                       ELSE pg_{0}_{1}_diff(pg_current_{0}_{1}(), %s)::bigint
+                                                   END""".format(wal_name, lsn_name),
+                                        (backup_start_lsn, backup_start_lsn, backup_start_lsn))
 
                             diff_in_bytes = int(cur.fetchone()[0])
                 except psycopg2.Error:
@@ -163,8 +253,47 @@ class WALERestore(object):
 
         # 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 < int(threshold_megabytes) * 1048576) and\
-            (diff_in_bytes < int(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
@@ -187,15 +316,19 @@ class WALERestore(object):
     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
 
-        if (ret == 0 and not
-           self.fix_subdirectory_path_if_broken('pg_xlog' if get_major_version(self.data_dir) < 10.0 else 'pg_wal')):
-            return 2
-        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():
@@ -213,6 +346,9 @@ def main():
     parser.add_argument('--no_master', type=int, default=0)
     args = parser.parse_args()
 
+    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
@@ -223,12 +359,13 @@ def main():
                               env_dir=args.envdir, threshold_mb=args.threshold_megabytes,
                               threshold_pct=args.threshold_backup_size_percentage, use_iam=args.use_iam,
                               no_master=args.no_master, retries=args.retries)
-        ret = restore.run()
-        if ret != 1:  # only WAL-E failures lead to the retry
+        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 ret
+    return exit_code
 
 
 if __name__ == '__main__':

patroni/utils.py

@@ -1,6 +1,5 @@
 import random
 import time
-import re
 
 from dateutil import tz
 from patroni.exceptions import PatroniException
@@ -95,17 +94,17 @@ def strtol(value, strict=True):
     True
     """
     value = str(value).strip()
-    l = len(value)
+    ln = len(value)
     i = 0
     # skip sign:
-    if i < l and value[i] in ('-', '+'):
+    if i < ln and value[i] in ('-', '+'):
         i += 1
 
     # we always expect to get digit in the beginning
-    if i < l and value[i].isdigit():
+    if i < ln and value[i].isdigit():
         if value[i] == '0':
             i += 1
-            if i < l and value[i] in ('x', 'X'):  # '0' followed by 'x': HEX
+            if i < ln and value[i] in ('x', 'X'):  # '0' followed by 'x': HEX
                 base = 16
                 i += 1
             else:  # just starts with '0': OCT
@@ -114,7 +113,7 @@ def strtol(value, strict=True):
             base = 10
 
         ret = None
-        while i <= l:
+        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)
@@ -194,10 +193,6 @@ def _sleep(interval):
     time.sleep(interval)
 
 
-def is_valid_pg_version(version):
-    return re.match(r'[1-9][0-9]?(\.(0|([1-9][0-9]?))){2}$', version)
-
-
 class RetryFailedError(PatroniException):
 
     """Raised when retrying an operation ultimately failed, after retrying the maximum number of attempts."""
@@ -280,3 +275,9 @@ def polling_loop(timeout, interval=1):
         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__ = '1.2.4'
+__version__ = '1.4.4'

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

postgres0.yml

@@ -66,6 +66,7 @@ postgresql:
   connect_address: 127.0.0.1:5432
   data_dir: data/postgresql0
 #  bin_dir:
+#  config_dir:
   pgpass: /tmp/pgpass0
   authentication:
     replication:
@@ -76,6 +77,12 @@ postgresql:
       password: zalando
   parameters:
     unix_socket_directories: '.'
+
+#watchdog:
+#  mode: automatic # Allowed values: off, automatic, required
+#  device: /dev/watchdog
+#  safety_margin: 5
+
 tags:
     nofailover: false
     noloadbalance: false

postgres1.yml

@@ -64,6 +64,7 @@ postgresql:
   connect_address: 127.0.0.1:5433
   data_dir: data/postgresql1
 #  bin_dir:
+#  config_dir:
   pgpass: /tmp/pgpass1
   authentication:
     replication:
@@ -74,6 +75,9 @@ postgresql:
       password: zalando
   parameters:
     unix_socket_directories: '.'
+  basebackup:
+      - verbose
+      - max-rate: 100M
 tags:
     nofailover: false
     noloadbalance: false

postgres2.yml

@@ -61,6 +61,7 @@ postgresql:
   connect_address: 127.0.0.1:5434
   data_dir: data/postgresql2
 #  bin_dir:
+#  config_dir:
   pgpass: /tmp/pgpass2
   authentication:
     replication:

requirements.txt

@@ -1,13 +1,16 @@
-urllib3>=1.9
+urllib3>=1.19.1,!=1.21
 boto
-psycopg2>=2.6.1
+psycopg2>=2.5.4
 PyYAML
 requests
 six >= 1.7
-kazoo==2.2.1
-python-etcd==0.4.3
-python-consul==0.7.0
+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==3.0.0

setup.py

@@ -87,7 +87,7 @@ 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:
@@ -115,13 +115,23 @@ 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)
 
-    install_reqs = get_install_requirements('requirements.txt')
+    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:
@@ -145,10 +155,10 @@ def setup_package():
         test_suite='tests',
         packages=find_packages(exclude=['tests', 'tests.*']),
         package_data={MAIN_PACKAGE: ["*.json"]},
-        install_requires=install_reqs,
-        setup_requires=['flake8'],
+        install_requires=install_requires,
+        extras_require=extras_require,
         cmdclass=cmdclass,
-        tests_require=['mock>=2.0.0', '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

@@ -3,9 +3,10 @@ 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 ClusterConfig, Member
+from patroni.ha import _MemberStatus
 from patroni.utils import tzutc
 from six import BytesIO as IO
 from six.moves import BaseHTTPServer
@@ -25,6 +26,8 @@ class MockPostgresql(object):
     sysid = 'dummysysid'
     scope = 'dummy'
     pending_restart = True
+    wal_name = 'wal'
+    lsn_name = 'lsn'
 
     @staticmethod
     def connection():
@@ -34,13 +37,22 @@ class MockPostgresql(object):
     def postmaster_start_time():
         return str(postmaster_start_time)
 
+    @staticmethod
+    def replica_cached_timeline(_):
+        return 2
+
+
+class MockWatchdog(object):
+    is_healthy = False
+
 
 class MockHa(object):
 
     state_handler = MockPostgresql()
+    watchdog = MockWatchdog()
 
     @staticmethod
-    def reinitialize():
+    def reinitialize(_):
         return 'reinitialize'
 
     @staticmethod
@@ -57,14 +69,14 @@ class MockHa(object):
 
     @staticmethod
     def fetch_nodes_statuses(members):
-        return [[None, True, None, None, {}]]
+        return [_MemberStatus(None, True, None, None, {}, False)]
 
     @staticmethod
     def schedule_future_restart(data):
         return True
 
     @staticmethod
-    def is_lagging(xlog):
+    def is_lagging(wal):
         return False
 
     @staticmethod
@@ -75,6 +87,10 @@ class MockHa(object):
     def wakeup():
         pass
 
+    @staticmethod
+    def is_paused():
+        return True
+
 
 class MockPatroni(object):
 
@@ -84,7 +100,7 @@ class MockPatroni(object):
     dcs = Mock()
     tags = {}
     version = '0.00'
-    noloadbalance = Mock(return_value=False)
+    noloadbalance = PropertyMock(return_value=False)
     scheduled_restart = {'schedule': future_restart_time,
                          'postmaster_start_time': postgresql.postmaster_start_time()}
 
@@ -104,6 +120,7 @@ class MockRequest(object):
     def sendall(self, *args, **kwargs):
         pass
 
+
 class MockRestApiServer(RestApiServer):
 
     def __init__(self, Handler, request):
@@ -131,6 +148,12 @@ class TestRestApiHandler(unittest.TestCase):
         with patch.object(RestApiHandler, 'get_postgresql_status', Mock(return_value={'role': 'master'})):
             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'})):
+            MockRestApiServer(RestApiHandler, 'GET /asynchronous')
         MockPatroni.dcs.cluster.leader.name = MockPostgresql.name
         MockRestApiServer(RestApiHandler, 'GET /replica')
         MockPatroni.dcs.cluster = None
@@ -139,10 +162,13 @@ class TestRestApiHandler(unittest.TestCase):
         with patch.object(MockHa, 'restart_scheduled', Mock(return_value=True)):
             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'))
 
     def test_do_OPTIONS(self):
         self.assertIsNotNone(MockRestApiServer(RestApiHandler, 'OPTIONS / HTTP/1.0'))
 
+    @patch.object(MockPostgresql, 'state', PropertyMock(return_value='stopped'))
     def test_do_GET_patroni(self):
         self.assertIsNotNone(MockRestApiServer(RestApiHandler, 'GET /patroni'))
 
@@ -254,7 +280,7 @@ class TestRestApiHandler(unittest.TestCase):
     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
+        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)
@@ -268,11 +294,13 @@ class TestRestApiHandler(unittest.TestCase):
 
     @patch('time.sleep', Mock())
     @patch.object(MockPatroni, 'dcs')
-    def test_do_POST_failover(self, dcs):
+    def test_do_POST_switchover(self, dcs):
         dcs.loop_wait = 10
         cluster = dcs.get_cluster.return_value
+        cluster.is_synchronous_mode.return_value = False
+        cluster.is_paused.return_value = False
 
-        post = 'POST /failover HTTP/1.0' + self._authorization + '\nContent-Length: '
+        post = 'POST /switchover HTTP/1.0' + self._authorization + '\nContent-Length: '
 
         MockRestApiServer(RestApiHandler, post + '7\n\n{"1":2}')
 
@@ -282,14 +310,22 @@ class TestRestApiHandler(unittest.TestCase):
         cluster.leader.name = 'postgresql1'
         MockRestApiServer(RestApiHandler, request)
 
-        MockRestApiServer(RestApiHandler, post + '25\n\n{"leader": "postgresql1"}')
+        request = post + '25\n\n{"leader": "postgresql1"}'
+
+        cluster.is_paused.return_value = True
+        MockRestApiServer(RestApiHandler, request)
+
+        cluster.is_paused.return_value = False
+        for cluster.is_synchronous_mode.return_value in (True, False):
+            MockRestApiServer(RestApiHandler, request)
 
         cluster.leader.name = 'postgresql2'
         request = post + '53\n\n{"leader": "postgresql1", "candidate": "postgresql2"}'
         MockRestApiServer(RestApiHandler, request)
 
         cluster.leader.name = 'postgresql1'
-        MockRestApiServer(RestApiHandler, request)
+        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'})]
@@ -303,6 +339,7 @@ class TestRestApiHandler(unittest.TestCase):
 
         cluster2 = cluster.copy()
         cluster2.leader.name = 'postgresql0'
+        cluster2.is_unlocked.return_value = False
         dcs.get_cluster.side_effect = [cluster, cluster2]
         MockRestApiServer(RestApiHandler, request)
 
@@ -337,3 +374,9 @@ class TestRestApiHandler(unittest.TestCase):
 
         # 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"}')

tests/test_async_executor.py

@@ -1,14 +1,14 @@
 import unittest
 
 from mock import Mock, patch
-from patroni.async_executor import AsyncExecutor
+from patroni.async_executor import AsyncExecutor, CriticalTask
 from threading import Thread
 
 
 class TestAsyncExecutor(unittest.TestCase):
 
     def setUp(self):
-        self.a = AsyncExecutor(Mock())
+        self.a = AsyncExecutor(Mock(), Mock())
 
     @patch.object(Thread, 'start', Mock())
     def test_run_async(self):
@@ -16,3 +16,17 @@ class TestAsyncExecutor(unittest.TestCase):
 
     def test_run(self):
         self.a.run(Mock(side_effect=Exception()))
+
+    def test_cancel(self):
+        self.a.cancel()
+        self.a.schedule('foo')
+        self.a.cancel()
+        self.a.run(Mock())
+
+
+class TestCriticalTask(unittest.TestCase):
+
+    def test_completed_task(self):
+        ct = CriticalTask()
+        ct.complete(1)
+        self.assertFalse(ct.cancel())

tests/test_aws.py

@@ -1,83 +1,69 @@
 import boto.ec2
-import requests
 import sys
 import unittest
 
 from mock import Mock, patch
 from collections import namedtuple
 from patroni.scripts.aws import AWSConnection, main as _main
-from patroni.utils import RetryFailedError
 from requests.exceptions import RequestException
 
 
 class MockEc2Connection(object):
 
-    def __init__(self, error=False):
-        self.error = error
-
-    def get_all_volumes(self, filters):
-        if self.error:
-            raise boto.exception("get_all_volumes")
+    @staticmethod
+    def get_all_volumes(*args, **kwargs):
         oid = namedtuple('Volume', 'id')
         return [oid(id='a'), oid(id='b')]
 
-    def create_tags(self, objects, tags):
-        if self.error or len(objects) == 0:
-            raise boto.exception("create_tags")
+    @staticmethod
+    def create_tags(objects, *args, **kwargs):
+        if len(objects) == 0:
+            raise boto.exception.BotoServerError(503, 'Service Unavailable', 'Request limit exceeded')
         return True
 
 
 class MockResponse(object):
+    ok = True
 
     def __init__(self, content):
         self.content = content
-        self.ok = True
 
     def json(self):
         return self.content
 
 
+def requests_get(url, **kwargs):
+    if url.split('/')[-1] == 'document':
+        result = {"instanceId": "012345", "region": "eu-west-1"}
+    else:
+        result = 'foo'
+    return MockResponse(result)
+
+
+@patch('boto.ec2.connect_to_region', Mock(return_value=MockEc2Connection()))
 class TestAWSConnection(unittest.TestCase):
 
-    def boto_ec2_connect_to_region(self, region):
-        return MockEc2Connection(self.error)
-
-    def requests_get(self, url, **kwargs):
-        if self.error:
-            raise RequestException("foo")
-        result = namedtuple('Request', 'ok content')
-        result.ok = True
-        if url.split('/')[-1] == 'document' and not self.json_error:
-            result = {"instanceId": "012345", "region": "eu-west-1"}
-        else:
-            result = 'foo'
-        return MockResponse(result)
-
+    @patch('requests.get', requests_get)
     def setUp(self):
-        self.error = False
-        self.json_error = False
-        requests.get = self.requests_get
-        boto.ec2.connect_to_region = self.boto_ec2_connect_to_region
         self.conn = AWSConnection('test')
 
-    def test_aws_available(self):
-        self.assertTrue(self.conn.aws_available())
-
     def test_on_role_change(self):
         self.assertTrue(self.conn.on_role_change('master'))
-        self.conn.retry = Mock(side_effect=RetryFailedError("retry failed"))
-        self.assertFalse(self.conn.on_role_change('master'))
+        with patch.object(MockEc2Connection, 'get_all_volumes', Mock(return_value=[])):
+            self.conn._retry.max_tries = 1
+            self.assertFalse(self.conn.on_role_change('master'))
 
+    @patch('requests.get', Mock(side_effect=RequestException('foo')))
     def test_non_aws(self):
-        self.error = True
         conn = AWSConnection('test')
         self.assertFalse(conn.on_role_change("master"))
 
+    @patch('requests.get', Mock(return_value=MockResponse('foo')))
     def test_aws_bizare_response(self):
-        self.json_error = True
         conn = AWSConnection('test')
         self.assertFalse(conn.aws_available())
 
+    @patch('requests.get', requests_get)
     @patch('sys.exit', Mock())
     def test_main(self):
         self.assertIsNone(_main())

tests/test_config.py

@@ -39,6 +39,7 @@ class TestConfig(unittest.TestCase):
             'PATRONI_POSTGRESQL_LISTEN': '0.0.0.0:5432',
             'PATRONI_POSTGRESQL_CONNECT_ADDRESS': '127.0.0.1:5432',
             'PATRONI_POSTGRESQL_DATA_DIR': 'data/postgres0',
+            'PATRONI_POSTGRESQL_CONFIG_DIR': 'data/postgres0',
             'PATRONI_POSTGRESQL_PGPASS': '/tmp/pgpass0',
             'PATRONI_ETCD_HOST': '127.0.0.1:2379',
             'PATRONI_ETCD_URL': 'https://127.0.0.1:2379',
@@ -48,6 +49,9 @@ class TestConfig(unittest.TestCase):
             'PATRONI_ETCD_CERT': '/cert',
             'PATRONI_ETCD_KEY': '/key',
             'PATRONI_CONSUL_HOST': '127.0.0.1:8500',
+            'PATRONI_KUBERNETES_LABELS': 'a:b:c',
+            'PATRONI_KUBERNETES_SCOPE_LABEL': 'a',
+            'PATRONI_KUBERNETES_PORTS': '[{"name": "postgresql"}]',
             'PATRONI_ZOOKEEPER_HOSTS': "'host1:2181','host2:2181'",
             'PATRONI_EXHIBITOR_HOSTS': 'host1,host2',
             'PATRONI_EXHIBITOR_PORT': '8181',

tests/test_consul.py

@@ -3,7 +3,8 @@ import unittest
 
 from consul import ConsulException, NotFound
 from mock import Mock, patch
-from patroni.dcs.consul import AbstractDCS, Cluster, Consul, ConsulInternalError, ConsulError, HTTPClient
+from patroni.dcs.consul import AbstractDCS, Cluster, Consul, ConsulInternalError, \
+                                ConsulError, HTTPClient, InvalidSessionTTL
 from test_etcd import SleepException
 
 
@@ -45,9 +46,12 @@ class TestHTTPClient(unittest.TestCase):
 
     def test_get(self):
         self.client.get(Mock(), '')
-        self.client.get(Mock(), '', {'wait': '1s', 'index': 1})
+        self.client.get(Mock(), '', {'wait': '1s', 'index': 1, 'token': 'foo'})
         self.client.http.request.return_value.status = 500
+        self.client.http.request.return_value.data = b'Foo'
         self.assertRaises(ConsulInternalError, self.client.get, Mock(), '')
+        self.client.http.request.return_value.data = b"Invalid Session TTL '3000000000', must be between [10s=24h0m0s]"
+        self.assertRaises(InvalidSessionTTL, self.client.get, Mock(), '')
 
     def test_unknown_method(self):
         try:
@@ -69,6 +73,10 @@ class TestConsul(unittest.TestCase):
     @patch.object(consul.Consul.KV, 'get', kv_get)
     @patch.object(consul.Consul.KV, 'delete', Mock())
     def setUp(self):
+        Consul({'ttl': 30, 'scope': 't', 'name': 'p', 'url': 'https://l:1', 'retry_timeout': 10,
+                'verify': 'on', 'key': 'foo', 'cert': 'bar', 'cacert': 'buz', 'token': 'asd', 'dc': 'dc1'})
+        Consul({'ttl': 30, 'scope': 't', 'name': 'p', 'url': 'https://l:1', 'retry_timeout': 10,
+                'verify': 'on', 'cert': 'bar', 'cacert': 'buz'})
         self.c = Consul({'ttl': 30, 'scope': 'test', 'name': 'postgresql1', 'host': 'localhost:1', 'retry_timeout': 10})
         self.c._base_path = '/service/good'
         self.c._load_cluster()
@@ -80,7 +88,9 @@ class TestConsul(unittest.TestCase):
         self.assertRaises(SleepException, self.c.create_session)
 
     @patch.object(consul.Consul.Session, 'renew', Mock(side_effect=NotFound))
-    @patch.object(consul.Consul.Session, 'create', Mock(side_effect=ConsulException))
+    @patch.object(consul.Consul.Session, 'create', Mock(side_effect=[InvalidSessionTTL, ConsulException]))
+    @patch.object(consul.Consul.Agent, 'self', Mock(return_value={'Config': {'SessionTTLMin': 0}}))
+    @patch.object(HTTPClient, 'set_ttl', Mock(side_effect=ValueError))
     def test_referesh_session(self):
         self.c._session = '1'
         self.assertFalse(self.c.refresh_session())
@@ -102,11 +112,11 @@ class TestConsul(unittest.TestCase):
     @patch.object(consul.Consul.KV, 'put', Mock(side_effect=[True, ConsulException]))
     def test_touch_member(self):
         self.c.refresh_session = Mock(return_value=True)
-        self.c.touch_member('balbla')
-        self.c.touch_member('balbla')
-        self.c.touch_member('balbla')
+        self.c.touch_member({'balbla': 'blabla'})
+        self.c.touch_member({'balbla': 'blabla'})
+        self.c.touch_member({'balbla': 'blabla'})
         self.c.refresh_session = Mock(return_value=False)
-        self.c.touch_member('balbla')
+        self.c.touch_member({'balbla': 'blabla'})
 
     @patch.object(consul.Consul.KV, 'put', Mock(return_value=False))
     def test_take_leader(self):
@@ -128,7 +138,7 @@ class TestConsul(unittest.TestCase):
 
     @patch.object(consul.Consul.Session, 'renew', Mock())
     def test_update_leader(self):
-        self.c.update_leader()
+        self.c.update_leader(None)
 
     @patch.object(consul.Consul.KV, 'delete', Mock(return_value=True))
     def test_delete_leader(self):
@@ -162,3 +172,7 @@ class TestConsul(unittest.TestCase):
     def test_sync_state(self):
         self.assertTrue(self.c.set_sync_state_value('{}'))
         self.assertTrue(self.c.delete_sync_state())
+
+    @patch.object(consul.Consul.KV, 'put', Mock(return_value=True))
+    def test_set_history_value(self):
+        self.assertTrue(self.c.set_history_value('{}'))

tests/test_ctl.py

@@ -5,14 +5,17 @@ import sys
 import unittest
 
 from click.testing import CliRunner
+from datetime import datetime, timedelta
 from mock import patch, Mock
-from patroni.ctl import ctl, members, store_config, load_config, output_members, request_patroni, get_dcs, parse_dcs, \
-    get_all_members, get_any_member, get_cursor, query_member, configure, PatroniCtlException
-from patroni.dcs.etcd import Client
+from patroni.ctl import ctl, store_config, load_config, output_members, request_patroni, get_dcs, parse_dcs, \
+    get_all_members, get_any_member, get_cursor, query_member, configure, PatroniCtlException, apply_config_changes, \
+    format_config_for_editing, show_diff, invoke_editor, format_pg_version
+from patroni.dcs.etcd import Client, Failover
+from patroni.utils import tzutc
 from psycopg2 import OperationalError
 from test_etcd import etcd_read, requests_get, socket_getaddrinfo, MockResponse
 from test_ha import get_cluster_initialized_without_leader, get_cluster_initialized_with_leader, \
-    get_cluster_initialized_with_only_leader, get_cluster_not_initialized_without_leader
+    get_cluster_initialized_with_only_leader, get_cluster_not_initialized_without_leader, get_cluster, Member
 from test_postgresql import MockConnect, psycopg2_connect
 
 CONFIG_FILE_PATH = './test-ctl.yaml'
@@ -30,7 +33,7 @@ def test_rw_config():
 
 
 @patch('patroni.ctl.load_config',
-       Mock(return_value={'postgresql': {'data_dir': '.', 'parameters': {}, 'retry_timeout': 5},
+       Mock(return_value={'scope': 'alpha', 'postgresql': {'data_dir': '.', 'parameters': {}, 'retry_timeout': 5},
                           'restapi': {'auth': 'u:p', 'listen': ''}, 'etcd': {'host': 'localhost:2379'}}))
 class TestCtl(unittest.TestCase):
 
@@ -63,76 +66,92 @@ class TestCtl(unittest.TestCase):
         self.assertRaises(PatroniCtlException, parse_dcs, 'invalid://test')
 
     def test_output_members(self):
-        cluster = get_cluster_initialized_with_leader()
+        scheduled_at = datetime.now(tzutc) + timedelta(seconds=600)
+        cluster = get_cluster_initialized_with_leader(Failover(1, 'foo', 'bar', scheduled_at))
         self.assertIsNone(output_members(cluster, name='abc', fmt='pretty'))
         self.assertIsNone(output_members(cluster, name='abc', fmt='json'))
+        self.assertIsNone(output_members(cluster, name='abc', fmt='yaml'))
         self.assertIsNone(output_members(cluster, name='abc', fmt='tsv'))
 
+    @patch('patroni.ctl.get_dcs')
+    @patch('patroni.ctl.request_patroni', Mock(return_value=MockResponse()))
+    def test_switchover(self, mock_get_dcs):
+        mock_get_dcs.return_value = self.e
+        mock_get_dcs.return_value.get_cluster = get_cluster_initialized_with_leader
+        mock_get_dcs.return_value.set_failover_value = Mock()
+        result = self.runner.invoke(ctl, ['switchover', 'dummy'], input='leader\nother\n\ny')
+        assert 'leader' in result.output
+
+        result = self.runner.invoke(ctl, ['switchover', 'dummy'], input='leader\nother\n2300-01-01T12:23:00\ny')
+        assert result.exit_code == 0
+
+        with patch('patroni.dcs.Cluster.is_paused', Mock(return_value=True)):
+            result = self.runner.invoke(ctl, ['switchover', 'dummy', '--force', '--scheduled', '2015-01-01T12:00:00'])
+            assert result.exit_code == 1
+
+        # Aborting switchover, as we anser NO to the confirmation
+        result = self.runner.invoke(ctl, ['switchover', 'dummy'], input='leader\nother\n\nN')
+        assert result.exit_code == 1
+
+        # Target and source are equal
+        result = self.runner.invoke(ctl, ['switchover', 'dummy'], input='leader\nleader\n\ny')
+        assert result.exit_code == 1
+
+        # Reality is not part of this cluster
+        result = self.runner.invoke(ctl, ['switchover', 'dummy'], input='leader\nReality\n\ny')
+        assert result.exit_code == 1
+
+        result = self.runner.invoke(ctl, ['switchover', 'dummy', '--force'])
+        assert 'Member' in result.output
+
+        result = self.runner.invoke(ctl, ['switchover', 'dummy', '--force', '--scheduled', '2015-01-01T12:00:00+01:00'])
+        assert result.exit_code == 0
+
+        # Invalid timestamp
+        result = self.runner.invoke(ctl, ['switchover', 'dummy', '--force', '--scheduled', 'invalid'])
+        assert result.exit_code != 0
+
+        # Invalid timestamp
+        result = self.runner.invoke(ctl, ['switchover', 'dummy', '--force', '--scheduled', '2115-02-30T12:00:00+01:00'])
+        assert result.exit_code != 0
+
+        # Specifying wrong leader
+        result = self.runner.invoke(ctl, ['switchover', 'dummy'], input='dummy')
+        assert result.exit_code == 1
+
+        with patch('patroni.ctl.request_patroni', Mock(side_effect=Exception)):
+            # Non-responding patroni
+            result = self.runner.invoke(ctl, ['switchover', 'dummy'], input='leader\nother\n2300-01-01T12:23:00\ny')
+            assert 'falling back to DCS' in result.output
+
+        with patch('patroni.ctl.request_patroni') as mocked:
+            mocked.return_value.status_code = 500
+            result = self.runner.invoke(ctl, ['switchover', 'dummy'], input='leader\nother\n\ny')
+            assert 'Switchover failed' in result.output
+
+            mocked.return_value.status_code = 501
+            mocked.return_value.text = 'Server does not support this operation'
+            result = self.runner.invoke(ctl, ['switchover', 'dummy'], input='leader\nother\n\ny')
+            assert 'Switchover failed' in result.output
+
+        # No members available
+        mock_get_dcs.return_value.get_cluster = get_cluster_initialized_with_only_leader
+        result = self.runner.invoke(ctl, ['switchover', 'dummy'], input='leader\nother\n\ny')
+        assert result.exit_code == 1
+
+        # No master available
+        mock_get_dcs.return_value.get_cluster = get_cluster_initialized_without_leader
+        result = self.runner.invoke(ctl, ['switchover', 'dummy'], input='leader\nother\n\ny')
+        assert result.exit_code == 1
+
     @patch('patroni.ctl.get_dcs')
     @patch('patroni.ctl.request_patroni', Mock(return_value=MockResponse()))
     def test_failover(self, mock_get_dcs):
         mock_get_dcs.return_value = self.e
         mock_get_dcs.return_value.get_cluster = get_cluster_initialized_with_leader
         mock_get_dcs.return_value.set_failover_value = Mock()
-        result = self.runner.invoke(ctl, ['failover', 'dummy'], input='leader\nother\n\ny')
-        assert 'leader' in result.output
-
-        result = self.runner.invoke(ctl, ['failover', 'dummy'], input='leader\nother\n2300-01-01T12:23:00\ny')
-        assert result.exit_code == 0
-
-        with patch('patroni.dcs.Cluster.is_paused', Mock(return_value=True)):
-            result = self.runner.invoke(ctl, ['failover', 'dummy', '--force', '--scheduled', '2015-01-01T12:00:00'])
-            assert result.exit_code == 1
-
-        # Aborting failover,as we anser NO to the confirmation
-        result = self.runner.invoke(ctl, ['failover', 'dummy'], input='leader\nother\n\nN')
-        assert result.exit_code == 1
-
-        # Target and source are equal
-        result = self.runner.invoke(ctl, ['failover', 'dummy'], input='leader\nleader\n\ny')
-        assert result.exit_code == 1
-
-        # Reality is not part of this cluster
-        result = self.runner.invoke(ctl, ['failover', 'dummy'], input='leader\nReality\n\ny')
-        assert result.exit_code == 1
-
-        result = self.runner.invoke(ctl, ['failover', 'dummy', '--force'])
-        assert 'Member' in result.output
-
-        result = self.runner.invoke(ctl, ['failover', 'dummy', '--force', '--scheduled', '2015-01-01T12:00:00+01:00'])
-        assert result.exit_code == 0
-
-        # Invalid timestamp
-        result = self.runner.invoke(ctl, ['failover', 'dummy', '--force', '--scheduled', 'invalid'])
-        assert result.exit_code != 0
-
-        # Invalid timestamp
-        result = self.runner.invoke(ctl, ['failover', 'dummy', '--force', '--scheduled', '2115-02-30T12:00:00+01:00'])
-        assert result.exit_code != 0
-
-        # Specifying wrong leader
-        result = self.runner.invoke(ctl, ['failover', 'dummy'], input='dummy')
-        assert result.exit_code == 1
-
-        with patch('patroni.ctl.request_patroni', Mock(side_effect=Exception)):
-            # Non-responding patroni
-            result = self.runner.invoke(ctl, ['failover', 'dummy'], input='leader\nother\n2300-01-01T12:23:00\ny')
-            assert 'falling back to DCS' in result.output
-
-        with patch('patroni.ctl.request_patroni') as mocked:
-            mocked.return_value.status_code = 500
-            result = self.runner.invoke(ctl, ['failover', 'dummy'], input='leader\nother\n\ny')
-            assert 'Failover failed' in result.output
-
-        # No members available
-        mock_get_dcs.return_value.get_cluster = get_cluster_initialized_with_only_leader
-        result = self.runner.invoke(ctl, ['failover', 'dummy'], input='leader\nother\n\ny')
-        assert result.exit_code == 1
-
-        # No master available
-        mock_get_dcs.return_value.get_cluster = get_cluster_initialized_without_leader
-        result = self.runner.invoke(ctl, ['failover', 'dummy'], input='leader\nother\n\ny')
-        assert result.exit_code == 1
+        result = self.runner.invoke(ctl, ['failover', 'dummy'], input='\n')
+        assert 'Failover could be performed only to a specific candidate' in result.output
 
     def test_get_dcs(self):
         self.assertRaises(PatroniCtlException, get_dcs, {'dummy': {}}, 'dummy')
@@ -218,7 +237,7 @@ class TestCtl(unittest.TestCase):
         assert result.exit_code == 1
 
         # successful reinit
-        result = self.runner.invoke(ctl, ['reinit', 'alpha', 'other'], input='y')
+        result = self.runner.invoke(ctl, ['reinit', 'alpha', 'other'], input='y\ny')
         assert result.exit_code == 0
 
         # Aborted restart
@@ -234,7 +253,7 @@ class TestCtl(unittest.TestCase):
 
         # Wrong pg version
         result = self.runner.invoke(ctl, ['restart', 'alpha', '--any', '--pg-version', '9.1'], input='y')
-        assert 'Error: PostgreSQL version' in result.output
+        assert 'Error: Invalid PostgreSQL version format' in result.output
         assert result.exit_code == 1
 
         result = self.runner.invoke(ctl, ['restart', 'alpha', '--pending', '--force', '--timeout', '10min'])
@@ -260,7 +279,7 @@ class TestCtl(unittest.TestCase):
         with patch('requests.post', Mock(return_value=MockResponse(204))):
             # get restart with the non-200 return code
             # normal restart, the schedule is actually parsed, but not validated in patronictl
-            result = self.runner.invoke(ctl, ['restart', 'alpha', '--pg-version', '42.0.0',
+            result = self.runner.invoke(ctl, ['restart', 'alpha', '--pg-version', '42.0',
                                         '--scheduled', '2300-10-01T14:30'], input='y')
             assert result.exit_code == 0
 
@@ -341,9 +360,11 @@ class TestCtl(unittest.TestCase):
     @patch('patroni.ctl.get_dcs')
     def test_members(self, mock_get_dcs):
         mock_get_dcs.return_value.get_cluster = get_cluster_initialized_with_leader
-        result = self.runner.invoke(members, ['alpha'])
+        result = self.runner.invoke(ctl, ['list'])
         assert '127.0.0.1' in result.output
         assert result.exit_code == 0
+        with patch('patroni.ctl.load_config', Mock(return_value={})):
+            self.runner.invoke(ctl, ['list'])
 
     def test_configure(self):
         result = self.runner.invoke(configure, ['--dcs', 'abc', '-c', 'dummy', '-n', 'bla'])
@@ -379,7 +400,7 @@ class TestCtl(unittest.TestCase):
         cluster = get_cluster_initialized_with_leader(sync=('leader', 'other'))
         mock_get_dcs.return_value.get_cluster = Mock(return_value=cluster)
 
-        result = self.runner.invoke(ctl, ['list', 'dummy', '--extended'])
+        result = self.runner.invoke(ctl, ['list', 'dummy', '--extended', '--timestamp'])
         assert '2100' in result.output
         assert 'Scheduled restart' in result.output
 
@@ -399,14 +420,11 @@ class TestCtl(unittest.TestCase):
             assert 'Failed: flush scheduled restart' in result.output
 
     @patch('patroni.ctl.get_dcs')
+    @patch('patroni.ctl.polling_loop', Mock(return_value=[1]))
     def test_pause_cluster(self, mock_get_dcs):
         mock_get_dcs.return_value = self.e
         mock_get_dcs.return_value.get_cluster = get_cluster_initialized_with_leader
 
-        with patch('requests.patch', Mock(return_value=MockResponse(200))):
-            result = self.runner.invoke(ctl, ['pause', 'dummy'])
-            assert 'Success' in result.output
-
         with patch('requests.patch', Mock(return_value=MockResponse(500))):
             result = self.runner.invoke(ctl, ['pause', 'dummy'])
             assert 'Failed' in result.output
@@ -416,6 +434,17 @@ class TestCtl(unittest.TestCase):
             result = self.runner.invoke(ctl, ['pause', 'dummy'])
             assert 'Cluster is already paused' in result.output
 
+        with patch('requests.patch', Mock(return_value=MockResponse(200))):
+            result = self.runner.invoke(ctl, ['pause', 'dummy', '--wait'])
+            assert "'pause' request sent" in result.output
+            mock_get_dcs.return_value.get_cluster = Mock(side_effect=[get_cluster_initialized_with_leader(),
+                                                                      get_cluster(None, None, [], None, None)])
+            self.runner.invoke(ctl, ['pause', 'dummy', '--wait'])
+            member = Member(1, 'other', 28, {})
+            mock_get_dcs.return_value.get_cluster = Mock(side_effect=[get_cluster_initialized_with_leader(),
+                                                                      get_cluster(None, None, [member], None, None)])
+            self.runner.invoke(ctl, ['pause', 'dummy', '--wait'])
+
     @patch('patroni.ctl.get_dcs')
     def test_resume_cluster(self, mock_get_dcs):
         mock_get_dcs.return_value = self.e
@@ -438,3 +467,88 @@ class TestCtl(unittest.TestCase):
             with patch('requests.patch', Mock(side_effect=Exception)):
                 result = self.runner.invoke(ctl, ['resume', 'dummy'])
                 assert 'Can not find accessible cluster member' in result.output
+
+    def test_apply_config_changes(self):
+        config = {"postgresql": {"parameters": {"work_mem": "4MB"}, "use_pg_rewind": True}, "ttl": 30}
+
+        before_editing = format_config_for_editing(config)
+
+        # Spaces are allowed and stripped, numbers and booleans are interpreted
+        after_editing, changed_config = apply_config_changes(before_editing, config,
+                                                             ["postgresql.parameters.work_mem = 5MB",
+                                                              "ttl=15", "postgresql.use_pg_rewind=off", 'a.b=c'])
+        self.assertEquals(changed_config, {"a": {"b": "c"}, "postgresql": {"parameters": {"work_mem": "5MB"},
+                                                                           "use_pg_rewind": False}, "ttl": 15})
+
+        # postgresql.parameters namespace is flattened
+        after_editing, changed_config = apply_config_changes(before_editing, config,
+                                                             ["postgresql.parameters.work_mem.sub = x"])
+        self.assertEquals(changed_config, {"postgresql": {"parameters": {"work_mem": "4MB", "work_mem.sub": "x"},
+                                                          "use_pg_rewind": True}, "ttl": 30})
+
+        # Setting to null deletes
+        after_editing, changed_config = apply_config_changes(before_editing, config,
+                                                             ["postgresql.parameters.work_mem=null"])
+        self.assertEquals(changed_config, {"postgresql": {"use_pg_rewind": True}, "ttl": 30})
+        after_editing, changed_config = apply_config_changes(before_editing, config,
+                                                             ["postgresql.use_pg_rewind=null",
+                                                              "postgresql.parameters.work_mem=null"])
+        self.assertEquals(changed_config, {"ttl": 30})
+
+        self.assertRaises(PatroniCtlException, apply_config_changes, before_editing, config, ['a'])
+
+    @patch('sys.stdout.isatty', return_value=False)
+    @patch('cdiff.markup_to_pager')
+    def test_show_diff(self, mock_markup_to_pager, mock_isatty):
+        show_diff("foo:\n  bar: 1\n", "foo:\n  bar: 2\n")
+        mock_markup_to_pager.assert_not_called()
+
+        mock_isatty.return_value = True
+        show_diff("foo:\n  bar: 1\n", "foo:\n  bar: 2\n")
+        mock_markup_to_pager.assert_called_once()
+
+        # Test that unicode handling doesn't fail with an exception
+        show_diff(b"foo:\n  bar: \xc3\xb6\xc3\xb6\n".decode('utf-8'),
+                  b"foo:\n  bar: \xc3\xbc\xc3\xbc\n".decode('utf-8'))
+
+    def test_invoke_editor(self):
+        for e in ('', 'false'):
+            os.environ['EDITOR'] = e
+            self.assertRaises(PatroniCtlException, invoke_editor, 'foo: bar\n', 'test')
+
+    @patch('patroni.ctl.get_dcs')
+    def test_show_config(self, mock_get_dcs):
+        mock_get_dcs.return_value = self.e
+        mock_get_dcs.return_value.get_cluster = get_cluster_initialized_with_leader
+        self.runner.invoke(ctl, ['show-config', 'dummy'])
+
+    @patch('patroni.ctl.get_dcs')
+    def test_edit_config(self, mock_get_dcs):
+        mock_get_dcs.return_value = self.e
+        mock_get_dcs.return_value.get_cluster = get_cluster_initialized_with_leader
+        os.environ['EDITOR'] = 'true'
+        self.runner.invoke(ctl, ['edit-config', 'dummy'])
+        self.runner.invoke(ctl, ['edit-config', 'dummy', '-s', 'foo=bar'])
+        self.runner.invoke(ctl, ['edit-config', 'dummy', '--replace', 'postgres0.yml'])
+        self.runner.invoke(ctl, ['edit-config', 'dummy', '--apply', '-'], input='foo: bar')
+        self.runner.invoke(ctl, ['edit-config', 'dummy', '--force', '--apply', '-'], input='foo: bar')
+        mock_get_dcs.return_value.set_config_value = Mock(return_value=True)
+        self.runner.invoke(ctl, ['edit-config', 'dummy', '--force', '--apply', '-'], input='foo: bar')
+
+    @patch('patroni.ctl.get_dcs')
+    def test_version(self, mock_get_dcs):
+        mock_get_dcs.return_value = self.e
+        mock_get_dcs.return_value.get_cluster = get_cluster_initialized_with_leader
+        with patch('patroni.ctl.request_patroni') as mocked:
+            result = self.runner.invoke(ctl, ['version'])
+            assert 'patronictl version' in result.output
+            mocked.return_value.json = lambda: {'patroni': {'version': '1.2.3'}, 'server_version': 100001}
+            result = self.runner.invoke(ctl, ['version', 'dummy'])
+            assert '1.2.3' in result.output
+        with patch('requests.get', Mock(side_effect=Exception)):
+            result = self.runner.invoke(ctl, ['version', 'dummy'])
+            assert 'failed to get version' in result.output
+
+    def test_format_pg_version(self):
+        self.assertEquals(format_pg_version(100001), '10.1')
+        self.assertEquals(format_pg_version(90605), '9.6.5')

tests/test_etcd.py

@@ -18,7 +18,6 @@ class MockResponse(object):
         self.status_code = status_code
         self.content = '{}'
         self.ok = True
-        self.text = ''
 
     def json(self):
         return json.loads(self.content)
@@ -27,6 +26,10 @@ class MockResponse(object):
     def data(self):
         return self.content.encode('utf-8')
 
+    @property
+    def text(self):
+        return self.content
+
     @property
     def status(self):
         return self.status_code
@@ -48,6 +51,12 @@ def requests_get(url, **kwargs):
         response.content = '[{}]' if url.startswith('http://error') else members
     elif url.startswith('http://exhibitor'):
         response.content = '{"servers":["127.0.0.1","127.0.0.2","127.0.0.3"],"port":2181}'
+    elif url.endswith(':8011/reinitialize'):
+        data = kwargs.get('data', '')
+        if ' false}' in data:
+            response.status_code = 503
+            response.ok = False
+            response.content = 'restarting after failure already in progress'
     else:
         response.status_code = 404
         response.ok = False
@@ -81,7 +90,7 @@ def etcd_read(self, key, **kwargs):
         raise etcd.EtcdKeyNotFound
 
     response = {"action": "get", "node": {"key": "/service/batman5", "dir": True, "nodes": [
-                {"key": "/service/batman5/config", "value": '{"foo": "bar"}',
+                {"key": "/service/batman5/config", "value": '{"synchronous_mode": 0}',
                  "modifiedIndex": 1582, "createdIndex": 1582},
                 {"key": "/service/batman5/failover", "value": "",
                  "modifiedIndex": 1582, "createdIndex": 1582},
@@ -198,6 +207,7 @@ class TestClient(unittest.TestCase):
         mock_machines.__get__ = Mock(return_value=['http://localhost:2379'])
         self.client._machines_cache_updated = 0
         self.client.api_execute('/', 'POST', timeout=0)
+        self.client._machines_cache = [self.client._base_uri]
         self.assertRaises(etcd.EtcdWatchTimedOut, self.client.api_execute, '/timeout', 'POST', params={'wait': 'true'})
         self.assertRaises(etcd.EtcdException, self.client.api_execute, '/', '')
         self.client._update_machines_cache = True
@@ -262,9 +272,13 @@ class TestEtcd(unittest.TestCase):
                                   {'url': 'https://test:2379', 'retry_timeout': 10})
                 self.assertRaises(SleepException, self.etcd.get_etcd_client,
                                   {'proxy': 'https://user:password@test:2379', 'retry_timeout': 10})
+                self.assertRaises(SleepException, self.etcd.get_etcd_client,
+                                  {'hosts': 'foo:4001,bar', 'retry_timeout': 10})
 
     def test_get_cluster(self):
-        self.assertIsInstance(self.etcd.get_cluster(), Cluster)
+        cluster = self.etcd.get_cluster()
+        self.assertIsInstance(cluster, Cluster)
+        self.assertFalse(cluster.is_synchronous_mode())
         self.etcd._base_path = '/service/nocluster'
         cluster = self.etcd.get_cluster()
         self.assertIsInstance(cluster, Cluster)
@@ -288,7 +302,7 @@ class TestEtcd(unittest.TestCase):
         self.etcd.write_leader_optime('0')
 
     def test_update_leader(self):
-        self.assertTrue(self.etcd.update_leader())
+        self.assertTrue(self.etcd.update_leader(None))
 
     def test_initialize(self):
         self.assertFalse(self.etcd.initialize())
@@ -324,3 +338,6 @@ class TestEtcd(unittest.TestCase):
     def test_sync_state(self):
         self.assertFalse(self.etcd.write_sync_state('leader', None))
         self.assertFalse(self.etcd.delete_sync_state())
+
+    def test_set_history_value(self):
+        self.assertFalse(self.etcd.set_history_value('{}'))

tests/test_ha.py

@@ -5,16 +5,19 @@ import unittest
 
 from mock import Mock, MagicMock, PropertyMock, patch
 from patroni.config import Config
-from patroni.dcs import Cluster, ClusterConfig, Failover, Leader, Member, get_dcs, SyncState
+from patroni.dcs import Cluster, ClusterConfig, Failover, Leader, Member, get_dcs, SyncState, TimelineHistory
 from patroni.dcs.etcd import Client
-from patroni.exceptions import DCSError, PostgresException
+from patroni.exceptions import DCSError, PostgresConnectionException, PatroniException
 from patroni.ha import Ha, _MemberStatus
 from patroni.postgresql import Postgresql
+from patroni.watchdog import Watchdog
 from patroni.utils import tzutc
 from test_etcd import socket_getaddrinfo, etcd_read, etcd_write, requests_get
-from test_postgresql import psycopg2_connect
+from test_postgresql import psycopg2_connect, MockPostmaster
 
 
+SYSID = '12345678901'
+
 def true(*args, **kwargs):
     return True
 
@@ -24,7 +27,8 @@ def false(*args, **kwargs):
 
 
 def get_cluster(initialize, leader, members, failover, sync):
-    return Cluster(initialize, ClusterConfig(1, {1: 2}, 1), leader, 10, members, failover, sync)
+    history = TimelineHistory(1, [(1, 67197376, 'no recovery target specified', datetime.datetime.now().isoformat())])
+    return Cluster(initialize, ClusterConfig(1, {1: 2}, 1), leader, 10, members, failover, sync, history)
 
 
 def get_cluster_not_initialized_without_leader():
@@ -34,15 +38,16 @@ def get_cluster_not_initialized_without_leader():
 def get_cluster_initialized_without_leader(leader=False, failover=None, sync=None):
     m1 = Member(0, 'leader', 28, {'conn_url': 'postgres://replicator:[email protected]:5435/postgres',
                                   'api_url': 'http://127.0.0.1:8008/patroni', 'xlog_location': 4})
-    l = Leader(0, 0, m1) if leader else None
+    leader = Leader(0, 0, m1) if leader else None
     m2 = Member(0, 'other', 28, {'conn_url': 'postgres://replicator:[email protected]:5436/postgres',
                                  'api_url': 'http://127.0.0.1:8011/patroni',
                                  'state': 'running',
+                                 'pause': True,
                                  'tags': {'clonefrom': True},
                                  'scheduled_restart': {'schedule': "2100-01-01 10:53:07.560445+00:00",
                                                        'postgres_version': '99.0.0'}})
     syncstate = SyncState(0 if sync else None, sync and sync[0], sync and sync[1])
-    return get_cluster(True, l, [m1, m2], failover, syncstate)
+    return get_cluster(SYSID, leader, [m1, m2], failover, syncstate)
 
 
 def get_cluster_initialized_with_leader(failover=None, sync=None):
@@ -50,18 +55,19 @@ def get_cluster_initialized_with_leader(failover=None, sync=None):
 
 
 def get_cluster_initialized_with_only_leader(failover=None):
-    l = get_cluster_initialized_without_leader(leader=True, failover=failover).leader
-    return get_cluster(True, l, [l], failover, None)
+    leader = get_cluster_initialized_without_leader(leader=True, failover=failover).leader
+    return get_cluster(True, leader, [leader], failover, None)
 
 
-def get_node_status(reachable=True, in_recovery=True, xlog_location=10, nofailover=False):
+def get_node_status(reachable=True, in_recovery=True, wal_position=10, nofailover=False, watchdog_failed=False):
     def fetch_node_status(e):
         tags = {}
         if nofailover:
             tags['nofailover'] = True
-        return _MemberStatus(e, reachable, in_recovery, xlog_location, tags)
+        return _MemberStatus(e, reachable, in_recovery, wal_position, tags, watchdog_failed)
     return fetch_node_status
 
+
 future_restart_time = datetime.datetime.now(tzutc) + datetime.timedelta(days=5)
 postmaster_start_time = datetime.datetime.now(tzutc)
 
@@ -84,6 +90,8 @@ postgresql:
   pg_rewind:
     username: postgres
     password: postgres
+watchdog:
+  mode: off
 zookeeper:
   exhibitor:
     hosts: [localhost]
@@ -101,35 +109,41 @@ zookeeper:
         self.nosync = False
         self.scheduled_restart = {'schedule': future_restart_time,
                                   'postmaster_start_time': str(postmaster_start_time)}
+        self.watchdog = Watchdog(self.config)
 
 
 def run_async(self, func, args=()):
     return func(*args) if args else func()
 
 
-@patch.object(Postgresql, 'is_running', Mock(return_value=True))
+@patch.object(Postgresql, 'is_running', Mock(return_value=MockPostmaster()))
 @patch.object(Postgresql, 'is_leader', Mock(return_value=True))
-@patch.object(Postgresql, 'xlog_position', Mock(return_value=10))
+@patch.object(Postgresql, 'timeline_wal_position', Mock(return_value=(1, 10)))
+@patch.object(Postgresql, '_cluster_info_state_get', Mock(return_value=3))
 @patch.object(Postgresql, 'call_nowait', Mock(return_value=True))
 @patch.object(Postgresql, 'data_directory_empty', Mock(return_value=False))
-@patch.object(Postgresql, 'controldata', Mock(return_value={'Database system identifier': '1234567890'}))
+@patch.object(Postgresql, 'controldata', Mock(return_value={'Database system identifier': SYSID}))
 @patch.object(Postgresql, 'sync_replication_slots', Mock())
 @patch.object(Postgresql, 'write_pg_hba', Mock())
-@patch.object(Postgresql, 'write_pgpass', Mock())
+@patch.object(Postgresql, 'write_pgpass', Mock(return_value={}))
 @patch.object(Postgresql, 'write_recovery_conf', Mock())
 @patch.object(Postgresql, 'query', Mock())
 @patch.object(Postgresql, 'checkpoint', Mock())
 @patch.object(Postgresql, 'call_nowait', Mock())
+@patch.object(Postgresql, 'cancellable_subprocess_call', Mock(return_value=0))
 @patch.object(etcd.Client, 'write', etcd_write)
 @patch.object(etcd.Client, 'read', etcd_read)
 @patch.object(etcd.Client, 'delete', Mock(side_effect=etcd.EtcdException))
+@patch('patroni.postgresql.polling_loop', Mock(return_value=range(1)))
 @patch('patroni.async_executor.AsyncExecutor.busy', PropertyMock(return_value=False))
 @patch('patroni.async_executor.AsyncExecutor.run_async', run_async)
 @patch('subprocess.call', Mock(return_value=0))
+@patch('time.sleep', Mock())
 class TestHa(unittest.TestCase):
 
     @patch('socket.getaddrinfo', socket_getaddrinfo)
     @patch('psycopg2.connect', psycopg2_connect)
+    @patch('patroni.dcs.dcs_modules', Mock(return_value=['patroni.dcs.foo', 'patroni.dcs.etcd']))
     @patch.object(etcd.Client, 'read', etcd_read)
     def setUp(self):
         with patch.object(Client, 'machines') as mock_machines:
@@ -149,16 +163,16 @@ class TestHa(unittest.TestCase):
                                        'name': 'foo', 'retry_timeout': 10}})
             self.ha = Ha(MockPatroni(self.p, self.e))
             self.ha.old_cluster = self.e.get_cluster()
-            self.ha.cluster = get_cluster_not_initialized_without_leader()
+            self.ha.cluster = get_cluster_initialized_without_leader()
             self.ha.load_cluster_from_dcs = Mock()
-            self.ha.is_synchronous_mode = false
 
     def test_update_lock(self):
-        self.p.last_operation = Mock(side_effect=PostgresException(''))
+        self.p.last_operation = Mock(side_effect=PostgresConnectionException(''))
         self.assertTrue(self.ha.update_lock(True))
 
     def test_touch_member(self):
-        self.p.xlog_position = Mock(side_effect=Exception)
+        self.p.timeline_wal_position = Mock(return_value=(0, 1))
+        self.p.replica_cached_timeline = Mock(side_effect=Exception)
         self.ha.touch_member()
 
     def test_start_as_replica(self):
@@ -166,26 +180,46 @@ class TestHa(unittest.TestCase):
         self.assertEquals(self.ha.run_cycle(), 'starting as a secondary')
 
     def test_recover_replica_failed(self):
-        self.p.controldata = lambda: {'Database cluster state': 'in production'}
-        self.p.is_healthy = false
+        self.p.controldata = lambda: {'Database cluster state': 'in recovery', 'Database system identifier': SYSID}
         self.p.is_running = false
         self.p.follow = false
         self.assertEquals(self.ha.run_cycle(), 'starting as a secondary')
         self.assertEquals(self.ha.run_cycle(), 'failed to start postgres')
 
-    def test_recover_master_failed(self):
+    def test_recover_former_master(self):
         self.p.follow = false
-        self.p.is_healthy = false
         self.p.is_running = false
         self.p.name = 'leader'
         self.p.set_role('master')
-        self.p.controldata = lambda: {'Database cluster state': 'in production'}
+        self.p.controldata = lambda: {'Database cluster state': 'shut down', 'Database system identifier': SYSID}
         self.ha.cluster = get_cluster_initialized_with_leader()
         self.assertEquals(self.ha.run_cycle(), 'starting as readonly because i had the session lock')
 
+    @patch.object(Postgresql, 'fix_cluster_state', Mock())
+    def test_crash_recovery(self):
+        self.p.is_running = false
+        self.p.controldata = lambda: {'Database cluster state': 'in production', 'Database system identifier': SYSID}
+        self.assertEquals(self.ha.run_cycle(), 'doing crash recovery in a single user mode')
+
+    @patch.object(Postgresql, 'rewind_needed_and_possible', Mock(return_value=True))
+    def test_recover_with_rewind(self):
+        self.p.is_running = false
+        self.ha.cluster = get_cluster_initialized_with_leader()
+        self.assertEquals(self.ha.run_cycle(), 'running pg_rewind from leader')
+
+    @patch.object(Postgresql, 'can_rewind', PropertyMock(return_value=True))
+    @patch.object(Postgresql, 'fix_cluster_state', Mock())
+    def test_single_user_after_recover_failed(self):
+        self.p.controldata = lambda: {'Database cluster state': 'in recovery', 'Database system identifier': SYSID}
+        self.p.is_running = false
+        self.p.follow = false
+        self.assertEquals(self.ha.run_cycle(), 'starting as a secondary')
+        self.assertEquals(self.ha.run_cycle(), 'fixing cluster state in a single user mode')
+
     @patch('sys.exit', return_value=1)
     @patch('patroni.ha.Ha.sysid_valid', MagicMock(return_value=True))
     def test_sysid_no_match(self, exit_mock):
+        self.p.controldata = lambda: {'Database cluster state': 'in recovery', 'Database system identifier': '123'}
         self.ha.run_cycle()
         exit_mock.assert_called_once_with(1)
 
@@ -194,8 +228,10 @@ class TestHa(unittest.TestCase):
         self.p.is_leader = false
         self.p.is_healthy = true
         self.ha.has_lock = true
+        self.p.controldata = lambda: {'Database cluster state': 'in production', 'Database system identifier': SYSID}
         self.assertEquals(self.ha.run_cycle(), 'promoted self to leader because i had the session lock')
 
+    @patch('psycopg2.connect', psycopg2_connect)
     def test_acquire_lock_as_master(self):
         self.assertEquals(self.ha.run_cycle(), 'acquired session lock as a leader')
 
@@ -204,6 +240,13 @@ class TestHa(unittest.TestCase):
         self.p.is_leader = false
         self.assertEquals(self.ha.run_cycle(), 'promoted self to leader by acquiring session lock')
 
+    def test_long_promote(self):
+        self.ha.cluster.is_unlocked = false
+        self.ha.has_lock = true
+        self.p.is_leader = false
+        self.p.set_role('master')
+        self.assertEquals(self.ha.run_cycle(), 'no action.  i am the leader with the lock')
+
     def test_demote_after_failing_to_obtain_lock(self):
         self.ha.acquire_lock = false
         self.assertEquals(self.ha.run_cycle(), 'demoted self after trying and failing to obtain lock')
@@ -229,14 +272,25 @@ class TestHa(unittest.TestCase):
         self.p.is_leader = false
         self.assertEquals(self.ha.run_cycle(), 'promoted self to leader because i had the session lock')
 
+    def test_promote_without_watchdog(self):
+        self.ha.cluster.is_unlocked = false
+        self.ha.has_lock = true
+        self.p.is_leader = true
+        with patch.object(Watchdog, 'activate', Mock(return_value=False)):
+            self.assertEquals(self.ha.run_cycle(), 'Demoting self because watchdog could not be activated')
+            self.p.is_leader = false
+            self.assertEquals(self.ha.run_cycle(), 'Not promoting self because watchdog could not be activated')
+
     def test_leader_with_lock(self):
+        self.ha.cluster = get_cluster_not_initialized_without_leader()
         self.ha.cluster.is_unlocked = false
         self.ha.has_lock = true
         self.assertEquals(self.ha.run_cycle(), 'no action.  i am the leader with the lock')
 
     def test_demote_because_not_having_lock(self):
         self.ha.cluster.is_unlocked = false
-        self.assertEquals(self.ha.run_cycle(), 'demoting self because i do not have the lock and i was a leader')
+        with patch.object(Watchdog, 'is_running', PropertyMock(return_value=True)):
+            self.assertEquals(self.ha.run_cycle(), 'demoting self because i do not have the lock and i was a leader')
 
     def test_demote_because_update_lock_failed(self):
         self.ha.cluster.is_unlocked = false
@@ -260,10 +314,18 @@ class TestHa(unittest.TestCase):
         self.p.is_leader = false
         self.assertEquals(self.ha.run_cycle(), 'PAUSE: no action')
 
+    @patch.object(Postgresql, 'rewind_needed_and_possible', Mock(return_value=True))
+    def test_follow_triggers_rewind(self):
+        self.p.is_leader = false
+        self.p.trigger_check_diverged_lsn()
+        self.ha.cluster = get_cluster_initialized_with_leader()
+        self.assertEquals(self.ha.run_cycle(), 'running pg_rewind from leader')
+
     def test_no_etcd_connection_master_demote(self):
         self.ha.load_cluster_from_dcs = Mock(side_effect=DCSError('Etcd is not responding properly'))
         self.assertEquals(self.ha.run_cycle(), 'demoted self because DCS is not accessible and i was a leader')
 
+    @patch('time.sleep', Mock())
     def test_bootstrap_from_another_member(self):
         self.ha.cluster = get_cluster_initialized_with_leader()
         self.assertEquals(self.ha.bootstrap(), 'trying to bootstrap from replica \'other\'')
@@ -284,25 +346,43 @@ class TestHa(unittest.TestCase):
     def test_bootstrap_initialized_new_cluster(self):
         self.ha.cluster = get_cluster_not_initialized_without_leader()
         self.e.initialize = true
-        self.assertEquals(self.ha.bootstrap(), 'initialized a new cluster')
+        self.assertEquals(self.ha.bootstrap(), 'trying to bootstrap a new cluster')
+        self.p.is_leader = false
+        self.assertEquals(self.ha.run_cycle(), 'waiting for end of recovery after bootstrap')
+        self.p.is_leader = true
+        self.assertEquals(self.ha.run_cycle(), 'running post_bootstrap')
+        self.assertEquals(self.ha.run_cycle(), 'initialized a new cluster')
 
     def test_bootstrap_release_initialize_key_on_failure(self):
         self.ha.cluster = get_cluster_not_initialized_without_leader()
         self.e.initialize = true
-        self.p.bootstrap = Mock(side_effect=PostgresException("Could not bootstrap master PostgreSQL"))
-        self.assertRaises(PostgresException, self.ha.bootstrap)
+        self.ha.bootstrap()
+        self.p.is_running = false
+        self.assertRaises(PatroniException, self.ha.post_bootstrap)
 
+    def test_bootstrap_release_initialize_key_on_watchdog_failure(self):
+        self.ha.cluster = get_cluster_not_initialized_without_leader()
+        self.e.initialize = true
+        self.ha.bootstrap()
+        self.p.is_running.return_value = MockPostmaster()
+        self.p.is_leader = true
+        with patch.object(Watchdog, 'activate', Mock(return_value=False)):
+            self.assertEquals(self.ha.post_bootstrap(), 'running post_bootstrap')
+            self.assertRaises(PatroniException, self.ha.post_bootstrap)
+
+    @patch('psycopg2.connect', psycopg2_connect)
     def test_reinitialize(self):
         self.assertIsNotNone(self.ha.reinitialize())
 
         self.ha.cluster = get_cluster_initialized_with_leader()
-        self.assertIsNone(self.ha.reinitialize())
+        self.assertIsNone(self.ha.reinitialize(True))
 
         self.assertIsNotNone(self.ha.reinitialize())
 
         self.ha.state_handler.name = self.ha.cluster.leader.name
         self.assertIsNotNone(self.ha.reinitialize())
 
+    @patch('time.sleep', Mock())
     def test_restart(self):
         self.assertEquals(self.ha.restart({}), (True, 'restarted successfully'))
         self.p.restart = Mock(return_value=None)
@@ -315,11 +395,12 @@ class TestHa(unittest.TestCase):
         with patch.object(self.ha, "restart_matches", return_value=False):
             self.assertEquals(self.ha.restart({'foo': 'bar'}), (False, "restart conditions are not satisfied"))
 
+    @patch('os.kill', Mock())
     def test_restart_in_progress(self):
         with patch('patroni.async_executor.AsyncExecutor.busy', PropertyMock(return_value=True)):
             self.ha.restart({}, run_async=True)
             self.assertTrue(self.ha.restart_scheduled())
-            self.assertEquals(self.ha.run_cycle(), 'not healthy enough for leader race')
+            self.assertEquals(self.ha.run_cycle(), 'restart in progress')
 
             self.ha.cluster = get_cluster_initialized_with_leader()
             self.assertEquals(self.ha.run_cycle(), 'restart in progress')
@@ -328,10 +409,13 @@ class TestHa(unittest.TestCase):
             self.assertEquals(self.ha.run_cycle(), 'updated leader lock during restart')
 
             self.ha.update_lock = false
-            self.assertEquals(self.ha.run_cycle(), 'failed to update leader lock during restart')
+            self.p.set_role('master')
+            with patch('patroni.async_executor.CriticalTask.cancel', Mock(return_value=False)):
+                with patch('patroni.postgresql.Postgresql.terminate_starting_postmaster') as mock_terminate:
+                    self.assertEquals(self.ha.run_cycle(), 'lost leader lock during restart')
+                    mock_terminate.assert_called()
 
     @patch('requests.get', requests_get)
-    @patch('time.sleep', Mock())
     def test_manual_failover_from_leader(self):
         self.ha.fetch_node_status = get_node_status()
         self.ha.has_lock = true
@@ -344,9 +428,13 @@ class TestHa(unittest.TestCase):
         f = Failover(0, self.p.name, '', None)
         self.ha.cluster = get_cluster_initialized_with_leader(f)
         self.assertEquals(self.ha.run_cycle(), 'manual failover: demoting myself')
+        self.p.rewind_needed_and_possible = true
+        self.assertEquals(self.ha.run_cycle(), 'manual failover: demoting myself')
         self.ha.fetch_node_status = get_node_status(nofailover=True)
         self.assertEquals(self.ha.run_cycle(), 'no action.  i am the leader with the lock')
-        self.ha.fetch_node_status = get_node_status(xlog_location=1)
+        self.ha.fetch_node_status = get_node_status(watchdog_failed=True)
+        self.assertEquals(self.ha.run_cycle(), 'no action.  i am the leader with the lock')
+        self.ha.fetch_node_status = get_node_status(wal_position=1)
         self.assertEquals(self.ha.run_cycle(), 'no action.  i am the leader with the lock')
         # manual failover from the previous leader to us won't happen if we hold the nofailover flag
         self.ha.cluster = get_cluster_initialized_with_leader(Failover(0, 'blabla', self.p.name, None))
@@ -384,7 +472,19 @@ class TestHa(unittest.TestCase):
         self.assertEquals('PAUSE: no action.  i am the leader with the lock', self.ha.run_cycle())
 
     @patch('requests.get', requests_get)
-    @patch('time.sleep', Mock())
+    def test_manual_failover_from_leader_in_synchronous_mode(self):
+        self.p.is_leader = true
+        self.ha.has_lock = true
+        self.ha.is_synchronous_mode = true
+        self.ha.is_failover_possible = false
+        self.ha.process_sync_replication = Mock()
+        self.ha.cluster = get_cluster_initialized_with_leader(Failover(0, self.p.name, 'a', None), (self.p.name, None))
+        self.assertEquals('no action.  i am the leader with the lock', self.ha.run_cycle())
+        self.ha.cluster = get_cluster_initialized_with_leader(Failover(0, self.p.name, 'a', None), (self.p.name, 'a'))
+        self.ha.is_failover_possible = true
+        self.assertEquals('manual failover: demoting myself', self.ha.run_cycle())
+
+    @patch('requests.get', requests_get)
     def test_manual_failover_process_no_leader(self):
         self.p.is_leader = false
         self.ha.cluster = get_cluster_initialized_without_leader(failover=Failover(0, '', self.p.name, None))
@@ -409,7 +509,6 @@ class TestHa(unittest.TestCase):
         self.ha.patroni.nofailover = True
         self.assertEquals(self.ha.run_cycle(), 'following a different leader because I am not allowed to promote')
 
-    @patch('time.sleep', Mock())
     def test_manual_failover_process_no_leader_in_pause(self):
         self.ha.is_paused = true
         self.ha.cluster = get_cluster_initialized_without_leader(failover=Failover(0, '', 'other', None))
@@ -428,6 +527,8 @@ class TestHa(unittest.TestCase):
         self.ha.patroni.nofailover = False
         self.ha.fetch_node_status = get_node_status()
         self.assertTrue(self.ha.is_healthiest_node())
+        with patch.object(Watchdog, 'is_healthy', PropertyMock(return_value=False)):
+            self.assertFalse(self.ha.is_healthiest_node())
         with patch('patroni.postgresql.Postgresql.is_starting', return_value=True):
             self.assertFalse(self.ha.is_healthiest_node())
         self.ha.is_paused = true
@@ -440,9 +541,9 @@ class TestHa(unittest.TestCase):
         self.assertTrue(self.ha._is_healthiest_node(self.ha.old_cluster.members))
         self.ha.fetch_node_status = get_node_status(in_recovery=False)  # accessible, not in_recovery
         self.assertFalse(self.ha._is_healthiest_node(self.ha.old_cluster.members))
-        self.ha.fetch_node_status = get_node_status(xlog_location=11)  # accessible, in_recovery, xlog location ahead
+        self.ha.fetch_node_status = get_node_status(wal_position=11)  # accessible, in_recovery, wal position ahead
         self.assertFalse(self.ha._is_healthiest_node(self.ha.old_cluster.members))
-        with patch('patroni.postgresql.Postgresql.xlog_position', return_value=1):
+        with patch('patroni.postgresql.Postgresql.timeline_wal_position', return_value=(1, 1)):
             self.assertFalse(self.ha._is_healthiest_node(self.ha.old_cluster.members))
         self.ha.patroni.nofailover = True
         self.assertFalse(self.ha._is_healthiest_node(self.ha.old_cluster.members))
@@ -521,6 +622,14 @@ class TestHa(unittest.TestCase):
         self.ha.cluster = get_cluster_initialized_with_leader(Failover(0, '', self.p.name, None))
         self.assertEquals(self.ha.run_cycle(), 'PAUSE: waiting to become master after promote...')
 
+    def test_failed_to_update_lock_in_pause(self):
+        self.ha.update_lock = false
+        self.ha.is_paused = true
+        self.p.name = 'leader'
+        self.ha.cluster = get_cluster_initialized_with_leader()
+        self.assertEquals(self.ha.run_cycle(),
+                          'PAUSE: continue to run as master after failing to update leader lock in DCS')
+
     def test_postgres_unhealthy_in_pause(self):
         self.ha.is_paused = true
         self.p.is_healthy = false
@@ -569,7 +678,6 @@ class TestHa(unittest.TestCase):
         self.assertEquals(self.ha.run_cycle(), 'no action.  i am a secondary and i am following a leader')
         check_calls([(update_lock, False), (demote, False)])
 
-    @patch('time.sleep', Mock())
     def test_manual_failover_while_starting(self):
         self.ha.has_lock = true
         self.p.check_for_startup = true
@@ -581,7 +689,8 @@ class TestHa(unittest.TestCase):
     @patch('patroni.ha.Ha.demote')
     def test_failover_immediately_on_zero_master_start_timeout(self, demote):
         self.p.is_running = false
-        self.ha.cluster = get_cluster_initialized_with_leader()
+        self.ha.cluster = get_cluster_initialized_with_leader(sync=(self.p.name, 'other'))
+        self.ha.cluster.config.data['synchronous_mode'] = True
         self.ha.patroni.config.set_dynamic_configuration({'master_start_timeout': 0})
         self.ha.has_lock = true
         self.ha.update_lock = true
@@ -589,15 +698,13 @@ class TestHa(unittest.TestCase):
         self.assertEquals(self.ha.run_cycle(), 'stopped PostgreSQL to fail over after a crash')
         demote.assert_called_once()
 
-    @patch('time.sleep', Mock())
     @patch('patroni.postgresql.Postgresql.follow')
     def test_demote_immediate(self, follow):
         self.ha.has_lock = true
         self.e.get_cluster = Mock(return_value=get_cluster_initialized_without_leader())
         self.ha.demote('immediate')
-        follow.assert_called_once_with(None, None, True, None, True)
+        follow.assert_called_once_with(None)
 
-    @patch('time.sleep', Mock())
     def test_process_sync_replication(self):
         self.ha.has_lock = true
         mock_set_sync = self.p.set_synchronous_standby = Mock()
@@ -663,6 +770,13 @@ class TestHa(unittest.TestCase):
         self.ha.run_cycle()
         self.assertEquals(self.ha.dcs.write_sync_state.call_count, 1)
 
+        # Test sync set to '*' when synchronous_mode_strict is enabled
+        mock_set_sync.reset_mock()
+        self.ha.is_synchronous_mode_strict = true
+        self.p.pick_synchronous_standby = Mock(return_value=(None, False))
+        self.ha.run_cycle()
+        mock_set_sync.assert_called_once_with('*')
+
     def test_sync_replication_become_master(self):
         self.ha.is_synchronous_mode = true
 
@@ -717,8 +831,7 @@ class TestHa(unittest.TestCase):
         mock_promote.assert_called_once()
         mock_write_sync.assert_called_once_with('other', None, index=0)
 
-    @patch('time.sleep')
-    def test_disable_sync_when_restarting(self, mock_sleep):
+    def test_disable_sync_when_restarting(self):
         self.ha.is_synchronous_mode = true
 
         self.p.name = 'other'
@@ -731,10 +844,10 @@ class TestHa(unittest.TestCase):
             get_cluster_initialized_with_leader(sync=('leader', syncstandby))
             for syncstandby in ['other', None]])
 
-        self.ha.restart({})
-
-        mock_restart.assert_called_once()
-        mock_sleep.assert_called()
+        with patch('time.sleep') as mock_sleep:
+            self.ha.restart({})
+            mock_restart.assert_called_once()
+            mock_sleep.assert_called()
 
         # Restart is still called when DCS connection fails
         mock_restart.reset_mock()
@@ -772,13 +885,41 @@ class TestHa(unittest.TestCase):
     def test_wakup(self):
         self.ha.wakeup()
 
+    def test_shutdown(self):
+        self.p.is_running = false
+        self.ha.has_lock = true
+        self.ha.shutdown()
+
+    @patch('time.sleep', Mock())
     def test_leader_with_empty_directory(self):
         self.ha.cluster = get_cluster_initialized_with_leader()
         self.ha.has_lock = true
         self.p.data_directory_empty = true
         self.assertEquals(self.ha.run_cycle(), 'released leader key voluntarily as data dir empty and currently leader')
+        self.assertEquals(self.p.role, 'uninitialized')
 
         # as has_lock is mocked out, we need to fake the leader key release
         self.ha.has_lock = false
         # will not say bootstrap from leader as replica can't self elect
         self.assertEquals(self.ha.run_cycle(), "trying to bootstrap from replica 'other'")
+
+    def test_update_cluster_history(self):
+        self.p.get_master_timeline = Mock(return_value=1)
+        self.ha.has_lock = true
+        self.ha.cluster.is_unlocked = false
+        self.assertEquals(self.ha.run_cycle(), 'no action.  i am the leader with the lock')
+
+    @patch('sys.exit', return_value=1)
+    def test_abort_join(self, exit_mock):
+        self.ha.cluster = get_cluster_not_initialized_without_leader()
+        self.p.is_leader = false
+        self.ha.run_cycle()
+        exit_mock.assert_called_once_with(1)
+
+    def test_after_pause(self):
+        self.ha.has_lock = true
+        self.ha.cluster.is_unlocked = false
+        self.ha.is_paused = true
+        self.assertEquals(self.ha.run_cycle(), 'PAUSE: no action.  i am the leader with the lock')
+        self.ha.is_paused = false
+        self.assertEquals(self.ha.run_cycle(), 'no action.  i am the leader with the lock')

tests/test_kubernetes.py

@@ -0,0 +1,107 @@
+import unittest
+
+from mock import Mock, patch
+from patroni.dcs.kubernetes import Kubernetes, KubernetesError, k8s_client, k8s_watch, RetryFailedError
+
+
+def mock_list_namespaced_config_map(self, *args, **kwargs):
+    metadata = {'resource_version': '1', 'labels': {'f': 'b'}, 'name': 'test-config',
+                'annotations': {'initialize': '123', 'config': '{}'}}
+    items = [k8s_client.V1ConfigMap(metadata=k8s_client.V1ObjectMeta(**metadata))]
+    metadata.update({'name': 'test-leader', 'annotations': {'optime': '1234', 'leader': 'p-0', 'ttl': '30s'}})
+    items.append(k8s_client.V1ConfigMap(metadata=k8s_client.V1ObjectMeta(**metadata)))
+    metadata.update({'name': 'test-failover', 'annotations': {'leader': 'p-0'}})
+    items.append(k8s_client.V1ConfigMap(metadata=k8s_client.V1ObjectMeta(**metadata)))
+    metadata.update({'name': 'test-sync', 'annotations': {'leader': 'p-0'}})
+    items.append(k8s_client.V1ConfigMap(metadata=k8s_client.V1ObjectMeta(**metadata)))
+    metadata = k8s_client.V1ObjectMeta(resource_version='1')
+    return k8s_client.V1ConfigMapList(metadata=metadata, items=items)
+
+
+def mock_list_namespaced_pod(self, *args, **kwargs):
+    metadata = k8s_client.V1ObjectMeta(resource_version='1', name='p-0', annotations={'status': '{}'})
+    items = [k8s_client.V1Pod(metadata=metadata)]
+    return k8s_client.V1PodList(items=items)
+
+
+@patch.object(k8s_client.CoreV1Api, 'patch_namespaced_config_map', Mock())
+@patch.object(k8s_client.CoreV1Api, 'create_namespaced_config_map', Mock())
+class TestKubernetes(unittest.TestCase):
+
+    @patch('kubernetes.config.load_kube_config', Mock())
+    @patch.object(k8s_client.CoreV1Api, 'list_namespaced_config_map', mock_list_namespaced_config_map)
+    @patch.object(k8s_client.CoreV1Api, 'list_namespaced_pod', mock_list_namespaced_pod)
+    def setUp(self):
+        self.k = Kubernetes({'ttl': 30, 'scope': 'test', 'name': 'p-0', 'retry_timeout': 10, 'labels': {'f': 'b'}})
+        with patch('time.time', Mock(return_value=1)):
+            self.k.get_cluster()
+
+    @patch.object(k8s_client.CoreV1Api, 'list_namespaced_config_map', mock_list_namespaced_config_map)
+    @patch.object(k8s_client.CoreV1Api, 'list_namespaced_pod', mock_list_namespaced_pod)
+    def test_get_cluster(self):
+        self.k.get_cluster()
+        with patch.object(k8s_client.CoreV1Api, 'list_namespaced_pod', Mock(side_effect=Exception)):
+            self.assertRaises(KubernetesError, self.k.get_cluster)
+
+    @patch('kubernetes.config.load_kube_config', Mock())
+    @patch.object(k8s_client.CoreV1Api, 'create_namespaced_endpoints', Mock())
+    def test_update_leader(self):
+        k = Kubernetes({'ttl': 30, 'scope': 'test', 'name': 'p-0', 'retry_timeout': 10,
+                        'labels': {'f': 'b'}, 'use_endpoints': True, 'pod_ip': '10.0.0.0'})
+        self.assertIsNotNone(k.update_leader('123'))
+
+    def test_take_leader(self):
+        self.k.take_leader()
+        self.k._leader_observed_record['leader'] = 'test'
+        self.k.patch_or_create = Mock(return_value=False)
+        self.k.take_leader()
+
+    def test_manual_failover(self):
+        with patch.object(k8s_client.CoreV1Api, 'patch_namespaced_config_map', Mock(side_effect=RetryFailedError(''))):
+            self.k.manual_failover('foo', 'bar')
+
+    def test_set_config_value(self):
+        self.k.set_config_value('{}')
+
+    @patch.object(k8s_client.CoreV1Api, 'patch_namespaced_pod', Mock(return_value=True))
+    def test_touch_member(self):
+        self.k.touch_member({})
+        self.k._name = 'p-1'
+        self.k.touch_member({'state': 'running', 'role': 'replica'})
+        self.k.touch_member({'state': 'stopped', 'role': 'master'})
+
+    def test_initialize(self):
+        self.k.initialize()
+
+    def test_delete_leader(self):
+        self.k.delete_leader()
+
+    def test_cancel_initialization(self):
+        self.k.cancel_initialization()
+
+    @patch.object(k8s_client.CoreV1Api, 'delete_collection_namespaced_config_map',
+                  Mock(side_effect=k8s_client.rest.ApiException(403, '')))
+    def test_delete_cluster(self):
+        self.k.delete_cluster()
+
+    @patch('kubernetes.config.load_kube_config', Mock())
+    @patch.object(k8s_client.CoreV1Api, 'create_namespaced_endpoints',
+                  Mock(side_effect=[k8s_client.rest.ApiException(502, ''), k8s_client.rest.ApiException(500, '')]))
+    def test_delete_sync_state(self):
+        k = Kubernetes({'ttl': 30, 'scope': 'test', 'name': 'p-0', 'retry_timeout': 10,
+                        'labels': {'f': 'b'}, 'use_endpoints': True, 'pod_ip': '10.0.0.0'})
+        self.assertFalse(k.delete_sync_state())
+
+    def test_watch(self):
+        self.k.set_ttl(10)
+        self.k.watch(None, 0)
+        self.k.watch(None, 0)
+        with patch.object(k8s_watch.Watch, 'stream',
+                          Mock(side_effect=[Exception, [], KeyboardInterrupt,
+                                            [{'raw_object': {'metadata': {'resourceVersion': '2'}}}]])):
+            self.assertFalse(self.k.watch('1', 2))
+            self.assertRaises(KeyboardInterrupt, self.k.watch, '1', 2)
+            self.assertTrue(self.k.watch('1', 2))
+
+    def test_set_history_value(self):
+        self.k.set_history_value('{}')

tests/test_patroni.py

@@ -12,7 +12,7 @@ from patroni.exceptions import DCSError
 from patroni import Patroni, main as _main, patroni_main
 from six.moves import BaseHTTPServer
 from test_etcd import SleepException, etcd_read, etcd_write
-from test_postgresql import Postgresql, psycopg2_connect
+from test_postgresql import Postgresql, psycopg2_connect, MockPostmaster
 
 
 class MockFrozenImporter(object):
@@ -26,7 +26,7 @@ class MockFrozenImporter(object):
 @patch.object(Postgresql, 'write_pg_hba', Mock())
 @patch.object(Postgresql, '_write_postgresql_conf', Mock())
 @patch.object(Postgresql, 'write_recovery_conf', Mock())
-@patch.object(Postgresql, 'is_running', Mock(return_value=True))
+@patch.object(Postgresql, 'is_running', Mock(return_value=MockPostmaster()))
 @patch.object(Postgresql, 'call_nowait', Mock())
 @patch.object(BaseHTTPServer.HTTPServer, '__init__', Mock())
 @patch.object(AsyncExecutor, 'run', Mock())
@@ -104,7 +104,9 @@ class TestPatroni(unittest.TestCase):
     @patch('patroni.config.Config.save_cache', Mock())
     @patch('patroni.config.Config.reload_local_configuration', Mock(return_value=True))
     @patch.object(Postgresql, 'state', PropertyMock(return_value='running'))
+    @patch.object(Postgresql, 'data_directory_empty', Mock(return_value=False))
     def test_run(self):
+        self.p.postgresql.set_role('replica')
         self.p.sighup_handler()
         self.p.ha.dcs.watch = Mock(side_effect=SleepException)
         self.p.api.start = Mock()

tests/test_postgresql.py

@@ -1,3 +1,4 @@
+import datetime
 import mock  # for the mock.call method, importing it without a namespace breaks python3
 import os
 import psycopg2
@@ -6,9 +7,11 @@ import subprocess
 import unittest
 
 from mock import Mock, MagicMock, PropertyMock, patch, mock_open
+from patroni.async_executor import CriticalTask
 from patroni.dcs import Cluster, Leader, Member, SyncState
-from patroni.exceptions import PostgresException, PostgresConnectionException
+from patroni.exceptions import PostgresConnectionException, PostgresException
 from patroni.postgresql import Postgresql, STATE_REJECT, STATE_NO_RESPONSE
+from patroni.postmaster import PostmasterProcess
 from patroni.utils import RetryFailedError
 from six.moves import builtins
 from threading import Thread
@@ -23,26 +26,41 @@ class MockCursor(object):
         self.results = []
 
     def execute(self, sql, *params):
-        if sql.startswith('blabla') or sql == 'CHECKPOINT':
+        if sql.startswith('blabla'):
+            raise psycopg2.ProgrammingError()
+        elif sql == 'CHECKPOINT':
             raise psycopg2.OperationalError()
         elif sql.startswith('RetryFailedError'):
             raise RetryFailedError('retry')
         elif sql.startswith('SELECT slot_name'):
             self.results = [('blabla',), ('foobar',)]
         elif sql.startswith('SELECT CASE WHEN pg_is_in_recovery()'):
-            self.results = [(0,)]
-        elif sql == 'SELECT pg_is_in_recovery()':
-            self.results = [(False, )]
+            self.results = [(1, 2)]
+        elif sql.startswith('SELECT pg_is_in_recovery()'):
+            self.results = [(False, 2)]
         elif sql.startswith('WITH replication_info AS ('):
             replication_info = '[{"application_name":"walreceiver","client_addr":"1.2.3.4",' +\
                                '"state":"streaming","sync_state":"async","sync_priority":0}]'
-            self.results = [('', True, '', '', '', '', False, replication_info)]
+            self.results = [('', 0, '', '', '', '', False, replication_info)]
         elif sql.startswith('SELECT name, setting'):
             self.results = [('wal_segment_size', '2048', '8kB', 'integer', 'internal'),
                             ('search_path', 'public', None, 'string', 'user'),
                             ('port', '5433', None, 'integer', 'postmaster'),
                             ('listen_addresses', '*', None, 'string', 'postmaster'),
-                            ('autovacuum', 'on', None, 'bool', 'sighup')]
+                            ('autovacuum', 'on', None, 'bool', 'sighup'),
+                            ('unix_socket_directories', '/tmp', None, 'string', 'postmaster')]
+        elif sql.startswith('IDENTIFY_SYSTEM'):
+            self.results = [('1', 2, '0/402EEC0', '')]
+        elif sql.startswith('SELECT isdir, modification'):
+            self.results = [(False, datetime.datetime.now())]
+        elif sql.startswith('SELECT pg_read_file'):
+            self.results = [('1\t0/40159C0\tno recovery target specified\n\n' +
+                             '2\t1/40159C0\tno recovery target specified\n',)]
+        elif sql.startswith('TIMELINE_HISTORY '):
+            self.results = [('', b'x\t0/40159C0\tno recovery target specified\n\n' +
+                                 b'1\t0/40159C0\tno recovery target specified\n\n' +
+                                 b'2\t0/402DD98\tno recovery target specified\n\n' +
+                                 b'3\t0/403DD98\tno recovery target specified\n')]
         else:
             self.results = [(None, None, None, None, None, None, None, None, None, None)]
 
@@ -65,7 +83,7 @@ class MockCursor(object):
 
 class MockConnect(object):
 
-    server_version = '99999'
+    server_version = 99999
     autocommit = False
     closed = 0
 
@@ -83,6 +101,15 @@ class MockConnect(object):
         pass
 
 
+class MockPostmaster(object):
+    def __init__(self, is_running=True, is_single_master=False):
+        self.is_running = Mock(return_value=is_running)
+        self.is_single_master = Mock(return_value=is_single_master)
+        self.wait_for_user_backends_to_close = Mock()
+        self.signal_stop = Mock(return_value=None)
+        self.wait = Mock()
+
+
 def pg_controldata_string(*args, **kwargs):
     return b"""
 pg_control version number:            942
@@ -138,21 +165,10 @@ Data page checksum version:           0
 """
 
 
-def postmaster_opts_string(*args, **kwargs):
-    return '/usr/local/pgsql/bin/postgres "-D" "data/postgresql0" "--listen_addresses=127.0.0.1" \
-"--port=5432" "--hot_standby=on" "--wal_keep_segments=8" "--wal_level=hot_standby" \
-"--archive_command=mkdir -p ../wal_archive && cp %p ../wal_archive/%f" "--wal_log_hints=on" \
-"--max_wal_senders=5" "--archive_timeout=1800s" "--archive_mode=on" "--max_replication_slots=5"\n'
-
-
 def psycopg2_connect(*args, **kwargs):
     return MockConnect()
 
 
-def fake_listdir(path):
-    return ["a", "b", "c"] if path.endswith('pg_xlog/archive_status') else []
-
-
 @patch('subprocess.call', Mock(return_value=0))
 @patch('psycopg2.connect', psycopg2_connect)
 class TestPostgresql(unittest.TestCase):
@@ -160,30 +176,30 @@ class TestPostgresql(unittest.TestCase):
                    'search_path': 'public', 'hot_standby': 'on', 'max_wal_senders': 5,
                    'wal_keep_segments': 8, 'wal_log_hints': 'on', 'max_locks_per_transaction': 64,
                    'max_worker_processes': 8, 'max_connections': 100, 'max_prepared_transactions': 0,
-                   'track_commit_timestamp': 'off'}
+                   'track_commit_timestamp': 'off', 'unix_socket_directories': '/tmp'}
 
     @patch('subprocess.call', Mock(return_value=0))
     @patch('psycopg2.connect', psycopg2_connect)
     @patch('os.rename', Mock())
-    @patch.object(Postgresql, 'get_major_version', Mock(return_value=9.6))
+    @patch.object(Postgresql, 'get_major_version', Mock(return_value=90600))
     @patch.object(Postgresql, 'is_running', Mock(return_value=True))
     def setUp(self):
         self.data_dir = 'data/test0'
+        self.config_dir = self.data_dir
         if not os.path.exists(self.data_dir):
             os.makedirs(self.data_dir)
-        self.p = Postgresql({'name': 'test0', 'scope': 'batman', 'data_dir': self.data_dir, 'retry_timeout': 10,
-                             'listen': '127.0.0.1, *:5432', 'connect_address': '127.0.0.2:5432',
+        self.p = Postgresql({'name': 'test0', 'scope': 'batman', 'data_dir': self.data_dir,
+                             'config_dir': self.config_dir, 'retry_timeout': 10, 'pgpass': '/tmp/pgpass0',
+                             'listen': '127.0.0.2, 127.0.0.3:5432', 'connect_address': '127.0.0.2:5432',
                              'authentication': {'superuser': {'username': 'test', 'password': 'test'},
                                                 'replication': {'username': 'replicator', 'password': 'rep-pass'}},
                              'remove_data_directory_on_rewind_failure': True,
                              'use_pg_rewind': True, 'pg_ctl_timeout': 'bla',
                              'parameters': self._PARAMETERS,
                              'recovery_conf': {'foo': 'bar'},
-                             'callbacks': {'on_start': 'true', 'on_stop': 'true',
-                                           'on_restart': 'true', 'on_role_change': 'true',
-                                           'on_reload': 'true'
-                                           },
-                             'restore': 'true'})
+                             'pg_hba': ['host all all 0.0.0.0/0 md5'],
+                             'callbacks': {'on_start': 'true', 'on_stop': 'true', 'on_reload': 'true',
+                                           'on_restart': 'true', 'on_role_change': 'true'}})
         self.p._callback_executor = Mock()
         self.leadermem = Member(0, 'leader', 28, {'conn_url': 'postgres://replicator:[email protected]:5435/postgres'})
         self.leader = Leader(-1, 28, self.leadermem)
@@ -194,12 +210,11 @@ class TestPostgresql(unittest.TestCase):
     def tearDown(self):
         shutil.rmtree('data')
 
-    def test_get_initdb_options(self):
-        self.assertEquals(self.p.get_initdb_options([{'encoding': 'UTF8'}, 'data-checksums']),
-                          ['--encoding=UTF8', '--data-checksums'])
-        self.assertRaises(Exception, self.p.get_initdb_options, [{'pgdata': 'bar'}])
-        self.assertRaises(Exception, self.p.get_initdb_options, [{'foo': 'bar', 1: 2}])
-        self.assertRaises(Exception, self.p.get_initdb_options, [1])
+    def test__initdb(self):
+        self.assertRaises(Exception, self.p.bootstrap, {'initdb': [{'pgdata': 'bar'}]})
+        self.assertRaises(Exception, self.p.bootstrap, {'initdb': [{'foo': 'bar', 1: 2}]})
+        self.assertRaises(Exception, self.p.bootstrap, {'initdb': [1]})
+        self.assertRaises(Exception, self.p.bootstrap, {'initdb': 1})
 
     @patch('os.path.exists', Mock(return_value=True))
     @patch('os.unlink', Mock())
@@ -211,66 +226,84 @@ class TestPostgresql(unittest.TestCase):
     @patch.object(Postgresql, 'wait_for_port_open')
     @patch.object(Postgresql, 'is_running')
     def test_start(self, mock_is_running, mock_wait_for_port_open, mock_wait_for_startup, mock_popen):
-        mock_is_running.return_value = True
+        mock_is_running.return_value = MockPostmaster()
         mock_wait_for_port_open.return_value = True
         mock_wait_for_startup.return_value = False
-        mock_popen.stdout.readline.return_value = '123'
+        mock_popen.return_value.stdout.readline.return_value = '123'
         self.assertTrue(self.p.start())
-        mock_is_running.return_value = False
-        open(os.path.join(self.data_dir, 'postmaster.pid'), 'w').close()
-        pg_conf = os.path.join(self.data_dir, 'postgresql.conf')
-        open(pg_conf, 'w').close()
-        self.assertFalse(self.p.start())
-        with open(pg_conf) as f:
-            lines = f.readlines()
-            self.assertTrue("f.oo = 'bar'\n" in lines)
+        mock_is_running.return_value = None
 
-        mock_wait_for_startup.return_value = None
-        self.assertFalse(self.p.start(10))
-        self.assertIsNone(self.p.start())
+        mock_postmaster = MockPostmaster()
+        with patch.object(PostmasterProcess, 'start', return_value=mock_postmaster):
+            pg_conf = os.path.join(self.data_dir, 'postgresql.conf')
+            open(pg_conf, 'w').close()
+            self.assertFalse(self.p.start(task=CriticalTask()))
 
-        mock_wait_for_port_open.return_value = False
+            with open(pg_conf) as f:
+                lines = f.readlines()
+                self.assertTrue("f.oo = 'bar'\n" in lines)
+
+            mock_wait_for_startup.return_value = None
+            self.assertFalse(self.p.start(10))
+            self.assertIsNone(self.p.start())
+
+            mock_wait_for_port_open.return_value = False
+            self.assertFalse(self.p.start())
+            task = CriticalTask()
+            task.cancel()
+            self.assertFalse(self.p.start(task=task))
+
+        self.p.cancel()
         self.assertFalse(self.p.start())
 
     @patch.object(Postgresql, 'pg_isready')
-    @patch.object(Postgresql, 'read_pid_file')
-    @patch.object(Postgresql, 'is_pid_running')
     @patch('patroni.postgresql.polling_loop', Mock(return_value=range(1)))
-    def test_wait_for_port_open(self, mock_is_pid_running, mock_read_pid_file, mock_pg_isready):
-        mock_is_pid_running.return_value = False
+    def test_wait_for_port_open(self, mock_pg_isready):
         mock_pg_isready.return_value = STATE_NO_RESPONSE
+        mock_postmaster = MockPostmaster(is_running=False)
 
         # No pid file and postmaster death
-        mock_read_pid_file.return_value = {}
-        self.assertFalse(self.p.wait_for_port_open(42, 100., 1))
+        self.assertFalse(self.p.wait_for_port_open(mock_postmaster, 1))
 
-        mock_is_pid_running.return_value = True
+        mock_postmaster.is_running.return_value = True
 
         # timeout
-        mock_read_pid_file.return_value = {'pid', 1}
-        self.assertFalse(self.p.wait_for_port_open(42, 100., 1))
-
-        # Garbage pid
-        mock_read_pid_file.return_value = {'pid': 'garbage', 'start_time': '101', 'data_dir': '',
-                                           'socket_dir': '', 'port': '', 'listen_addr': ''}
-        self.assertFalse(self.p.wait_for_port_open(42, 100., 1))
-
-        # Not ready
-        mock_read_pid_file.return_value = {'pid': '42', 'start_time': '101', 'data_dir': '',
-                                           'socket_dir': '', 'port': '', 'listen_addr': ''}
-        self.assertFalse(self.p.wait_for_port_open(42, 100., 1))
+        self.assertFalse(self.p.wait_for_port_open(mock_postmaster, 1))
 
         # pg_isready failure
         mock_pg_isready.return_value = 'garbage'
-        self.assertTrue(self.p.wait_for_port_open(42, 100., 1))
+        self.assertTrue(self.p.wait_for_port_open(mock_postmaster, 1))
 
+        # cancelled
+        self.p.cancel()
+        self.assertFalse(self.p.wait_for_port_open(mock_postmaster, 1))
+
+    @patch('time.sleep', Mock())
     @patch.object(Postgresql, 'is_running')
+    @patch.object(Postgresql, '_wait_for_connection_close', Mock())
     def test_stop(self, mock_is_running):
-        mock_is_running.return_value = True
-        self.assertTrue(self.p.stop())
-        with patch('subprocess.call', Mock(return_value=1)):
-            mock_is_running.return_value = False
-            self.assertTrue(self.p.stop())
+        # Postmaster is not running
+        mock_callback = Mock()
+        mock_is_running.return_value = None
+        self.assertTrue(self.p.stop(on_safepoint=mock_callback))
+        mock_callback.assert_called()
+
+        # Is running, stopped successfully
+        mock_is_running.return_value = mock_postmaster = MockPostmaster()
+        mock_callback.reset_mock()
+        self.assertTrue(self.p.stop(on_safepoint=mock_callback))
+        mock_callback.assert_called()
+        mock_postmaster.signal_stop.assert_called()
+
+        # Stop signal failed
+        mock_postmaster.signal_stop.return_value = False
+        self.assertFalse(self.p.stop())
+
+        # Stop signal failed to find process
+        mock_postmaster.signal_stop.return_value = True
+        mock_callback.reset_mock()
+        self.assertTrue(self.p.stop(on_safepoint=mock_callback))
+        mock_callback.assert_called()
 
     def test_restart(self):
         self.p.start = Mock(return_value=False)
@@ -289,43 +322,86 @@ class TestPostgresql(unittest.TestCase):
             self.assertIsNone(self.p.checkpoint())
         self.assertEquals(self.p.checkpoint(), 'not accessible or not healty')
 
-    @patch('subprocess.call', side_effect=OSError)
+    @patch.object(Postgresql, 'cancellable_subprocess_call')
     @patch('patroni.postgresql.Postgresql.write_pgpass', MagicMock(return_value=dict()))
-    def test_pg_rewind(self, mock_call):
+    def test_pg_rewind(self, mock_cancellable_subprocess_call):
         r = {'user': '', 'host': '', 'port': '', 'database': '', 'password': ''}
-        self.assertTrue(self.p.rewind(r))
-        subprocess.call = mock_call
-        self.assertFalse(self.p.rewind(r))
+        mock_cancellable_subprocess_call.return_value = 0
+        self.assertTrue(self.p.pg_rewind(r))
+        mock_cancellable_subprocess_call.side_effect = OSError
+        self.assertFalse(self.p.pg_rewind(r))
 
-    @patch('os.unlink', Mock(return_value=True))
-    @patch('subprocess.check_output', Mock(return_value=0, side_effect=pg_controldata_string))
-    @patch.object(Postgresql, 'remove_data_directory', Mock(return_value=True))
-    @patch.object(Postgresql, 'single_user_mode', Mock(return_value=1))
-    @patch.object(Postgresql, 'write_pgpass', Mock(return_value={}))
-    @patch.object(Postgresql, 'is_running', Mock(return_value=True))
+    def test_check_recovery_conf(self):
+        self.p.write_recovery_conf({'primary_conninfo': 'foo'})
+        self.assertFalse(self.p.check_recovery_conf(None))
+        self.p.write_recovery_conf({})
+        self.assertTrue(self.p.check_recovery_conf(None))
+
+    @patch.object(Postgresql, 'start', Mock())
     @patch.object(Postgresql, 'can_rewind', PropertyMock(return_value=True))
-    @patch.object(Postgresql, 'rewind', return_value=False)
-    def test_follow(self, mock_pg_rewind):
-        with patch.object(Postgresql, 'check_recovery_conf', Mock(return_value=True)):
-            self.assertTrue(self.p.follow(None, None))  # nothing to do, recovery.conf has good primary_conninfo
+    def test__get_local_timeline_lsn(self):
+        self.p.trigger_check_diverged_lsn()
+        with patch.object(Postgresql, 'controldata',
+                          Mock(return_value={'Database cluster state': 'shut down in recovery',
+                                             'Minimum recovery ending location': '0/0',
+                                             "Min recovery ending loc's timeline": '0'})):
+            self.p.rewind_needed_and_possible(self.leader)
+        with patch.object(Postgresql, 'is_running', Mock(return_value=True)):
+            with patch.object(MockCursor, 'fetchone', Mock(side_effect=[(False, ), Exception])):
+                self.p.rewind_needed_and_possible(self.leader)
 
-        self.p.follow(self.me, self.me)  # follow is called when the node is holding leader lock
+    @patch.object(Postgresql, 'start', Mock())
+    @patch.object(Postgresql, 'can_rewind', PropertyMock(return_value=True))
+    @patch.object(Postgresql, '_get_local_timeline_lsn', Mock(return_value=(2, '40159C1')))
+    @patch.object(Postgresql, 'check_leader_is_not_in_recovery')
+    def test__check_timeline_and_lsn(self, mock_check_leader_is_not_in_recovery):
+        mock_check_leader_is_not_in_recovery.return_value = False
+        self.p.trigger_check_diverged_lsn()
+        self.assertFalse(self.p.rewind_needed_and_possible(self.leader))
+        mock_check_leader_is_not_in_recovery.return_value = True
+        self.assertFalse(self.p.rewind_needed_and_possible(self.leader))
+        self.p.trigger_check_diverged_lsn()
+        with patch('psycopg2.connect', Mock(side_effect=Exception)):
+            self.assertFalse(self.p.rewind_needed_and_possible(self.leader))
+        self.p.trigger_check_diverged_lsn()
+        with patch.object(MockCursor, 'fetchone', Mock(side_effect=[('', 2, '0/0'), ('', b'3\t0/40159C0\tn\n')])):
+            self.assertFalse(self.p.rewind_needed_and_possible(self.leader))
+        self.p.trigger_check_diverged_lsn()
+        with patch.object(MockCursor, 'fetchone', Mock(return_value=('', 1, '0/0'))):
+            with patch.object(Postgresql, '_get_local_timeline_lsn', Mock(return_value=(1, '0/0'))):
+                self.assertFalse(self.p.rewind_needed_and_possible(self.leader))
+            self.p.trigger_check_diverged_lsn()
+            self.assertTrue(self.p.rewind_needed_and_possible(self.leader))
 
-        with patch.object(Postgresql, 'restart', Mock(return_value=False)):
-            self.p.set_role('replica')
-            self.p.follow(None, None)  # restart without rewind
+    @patch.object(MockCursor, 'fetchone', Mock(side_effect=[(True,), Exception]))
+    def test_check_leader_is_not_in_recovery(self):
+        self.p.check_leader_is_not_in_recovery()
+        self.p.check_leader_is_not_in_recovery()
 
-        with patch.object(Postgresql, 'stop', Mock(return_value=False)):
-            self.p.follow(self.leader, self.leader, need_rewind=True)  # failed to stop postgres
+    @patch.object(Postgresql, 'cancellable_subprocess_call', Mock(return_value=0))
+    @patch.object(Postgresql, 'checkpoint', side_effect=['', '1'])
+    @patch.object(Postgresql, 'stop', Mock(return_value=False))
+    @patch.object(Postgresql, 'start', Mock())
+    def test_rewind(self, mock_checkpoint):
+        self.p.rewind(self.leader)
+        with patch.object(Postgresql, 'pg_rewind', Mock(return_value=False)):
+            mock_checkpoint.side_effect = ['1', '', '', '']
+            self.p.rewind(self.leader)
+            self.p.rewind(self.leader)
+            with patch.object(Postgresql, 'check_leader_is_not_in_recovery', Mock(return_value=False)):
+                self.p.rewind(self.leader)
+            self.p.config['remove_data_directory_on_rewind_failure'] = False
+            self.p.trigger_check_diverged_lsn()
+            self.p.rewind(self.leader)
+        with patch.object(Postgresql, 'is_running', Mock(return_value=True)):
+            self.p.rewind(self.leader)
+            self.p.is_leader = Mock(return_value=False)
+            self.p.rewind(self.leader)
 
-        self.p.follow(self.leader, self.leader)  # "leader" is not accessible or is_in_recovery
-
-        with patch.object(Postgresql, 'checkpoint', Mock(return_value=None)):
-            self.p.follow(self.leader, self.leader)
-            mock_pg_rewind.return_value = True
-            self.p.follow(self.leader, self.leader, need_rewind=True)
-
-        self.p.follow(None, None)  # check_recovery_conf...
+    @patch.object(Postgresql, 'is_running', Mock(return_value=False))
+    @patch.object(Postgresql, 'start', Mock())
+    def test_follow(self):
+        self.p.follow(None)
 
     @patch('subprocess.check_output', Mock(return_value=0, side_effect=pg_controldata_string))
     def test_can_rewind(self):
@@ -339,31 +415,64 @@ class TestPostgresql(unittest.TestCase):
         self.assertFalse(self.p.can_rewind)
 
     @patch('time.sleep', Mock())
+    @patch.object(Postgresql, 'cancellable_subprocess_call')
     @patch.object(Postgresql, 'remove_data_directory', Mock(return_value=True))
-    def test_create_replica(self):
+    def test_create_replica(self, mock_cancellable_subprocess_call):
         self.p.delete_trigger_file = Mock(side_effect=OSError)
-        with patch('subprocess.call', Mock(side_effect=[1, 0])):
-            self.assertEquals(self.p.create_replica(self.leader), 0)
-        with patch('subprocess.call', Mock(side_effect=[Exception(), 0])):
-            self.assertEquals(self.p.create_replica(self.leader), 0)
 
         self.p.config['create_replica_method'] = ['wale', 'basebackup']
         self.p.config['wale'] = {'command': 'foo'}
-        with patch('subprocess.call', Mock(return_value=0)):
-            self.assertEquals(self.p.create_replica(self.leader), 0)
-            del self.p.config['wale']
-            self.assertEquals(self.p.create_replica(self.leader), 0)
+        mock_cancellable_subprocess_call.return_value = 0
+        self.assertEquals(self.p.create_replica(self.leader), 0)
+        del self.p.config['wale']
+        self.assertEquals(self.p.create_replica(self.leader), 0)
 
-        with patch('subprocess.call', Mock(side_effect=Exception("foo"))):
-            self.assertEquals(self.p.create_replica(self.leader), 1)
+        self.p.config['create_replica_method'] = ['basebackup']
+        self.p.config['basebackup'] = [{'max_rate': '100M'}, 'no-sync']
+        self.assertEquals(self.p.create_replica(self.leader), 0)
 
-        with patch('subprocess.call', Mock(return_value=1)):
-            self.assertEquals(self.p.create_replica(self.leader), 1)
+        self.p.config['basebackup'] = [{'max_rate': '100M', 'compress': '9'}]
+        with mock.patch('patroni.postgresql.logger.error', new_callable=Mock()) as mock_logger:
+            self.p.create_replica(self.leader)
+            mock_logger.assert_called_once()
+            self.assertTrue("only one key-value is allowed and value should be a string" in mock_logger.call_args[0][0],
+                            "not matching {0}".format(mock_logger.call_args[0][0]))
+
+        self.p.config['basebackup'] = [42]
+        with mock.patch('patroni.postgresql.logger.error', new_callable=Mock()) as mock_logger:
+            self.p.create_replica(self.leader)
+            mock_logger.assert_called_once()
+            self.assertTrue("value should be string value or a single key-value pair" in mock_logger.call_args[0][0],
+                            "not matching {0}".format(mock_logger.call_args[0][0]))
+
+        self.p.config['basebackup'] = {"foo": "bar"}
+        self.assertEquals(self.p.create_replica(self.leader), 0)
+
+        self.p.config['create_replica_method'] = ['wale', 'basebackup']
+        del self.p.config['basebackup']
+        mock_cancellable_subprocess_call.return_value = 1
+        self.assertEquals(self.p.create_replica(self.leader), 1)
+
+        mock_cancellable_subprocess_call.side_effect = Exception('foo')
+        self.assertEquals(self.p.create_replica(self.leader), 1)
+
+        mock_cancellable_subprocess_call.side_effect = [1, 0]
+        self.assertEquals(self.p.create_replica(self.leader), 0)
+
+        mock_cancellable_subprocess_call.side_effect = [Exception(), 0]
+        self.assertEquals(self.p.create_replica(self.leader), 0)
+
+        self.p.cancel()
+        self.assertEquals(self.p.create_replica(self.leader), 1)
+
+    def test_basebackup(self):
+        self.p.cancel()
+        self.p.basebackup(None, None, {'foo': 'bar'})
 
     @patch.object(Postgresql, 'is_running', Mock(return_value=True))
     def test_sync_replication_slots(self):
         self.p.start()
-        cluster = Cluster(True, None, self.leader, 0, [self.me, self.other, self.leadermem], None, None)
+        cluster = Cluster(True, None, self.leader, 0, [self.me, self.other, self.leadermem], None, None, None)
         with mock.patch('patroni.postgresql.Postgresql._query', Mock(side_effect=psycopg2.OperationalError)):
             self.p.sync_replication_slots(cluster)
         self.p.sync_replication_slots(cluster)
@@ -376,10 +485,12 @@ class TestPostgresql(unittest.TestCase):
             cluster.members.extend([alias1, alias2])
             self.p.sync_replication_slots(cluster)
             errorlog_mock.assert_called_once()
-            assert "test-3" in errorlog_mock.call_args[0][1]
-            assert "test.3" in errorlog_mock.call_args[0][1]
+            self.assertTrue("test-3" in errorlog_mock.call_args[0][1],
+                            "non matching {0}".format(errorlog_mock.call_args[0][1]))
+            self.assertTrue("test.3" in errorlog_mock.call_args[0][1],
+                            "non matching {0}".format(errorlog_mock.call_args[0][1]))
 
-    @patch.object(MockConnect, 'closed', 2)
+    @patch.object(MockCursor, 'execute', Mock(side_effect=psycopg2.OperationalError))
     def test__query(self):
         self.assertRaises(PostgresConnectionException, self.p._query, 'blabla')
         self.p._state = 'restarting'
@@ -388,11 +499,12 @@ class TestPostgresql(unittest.TestCase):
     def test_query(self):
         self.p.query('select 1')
         self.assertRaises(PostgresConnectionException, self.p.query, 'RetryFailedError')
-        self.assertRaises(psycopg2.OperationalError, self.p.query, 'blabla')
+        self.assertRaises(psycopg2.ProgrammingError, self.p.query, 'blabla')
 
     @patch.object(Postgresql, 'pg_isready', Mock(return_value=STATE_REJECT))
     def test_is_leader(self):
         self.assertTrue(self.p.is_leader())
+        self.p.reset_cluster_info_state()
         with patch.object(Postgresql, '_query', Mock(side_effect=RetryFailedError(''))):
             self.assertRaises(PostgresConnectionException, self.p.is_leader)
 
@@ -407,31 +519,42 @@ class TestPostgresql(unittest.TestCase):
         self.assertFalse(self.p.is_healthy())
 
     def test_promote(self):
-        self.p._role = 'replica'
-        self.assertTrue(self.p.promote())
-        self.assertTrue(self.p.promote())
+        self.p.set_role('replica')
+        self.assertIsNone(self.p.promote(0))
+        self.assertTrue(self.p.promote(0))
 
-    def test_last_operation(self):
-        self.assertEquals(self.p.last_operation(), '0')
-        Thread(target=self.p.last_operation).start()
+    def test_timeline_wal_position(self):
+        self.assertEquals(self.p.timeline_wal_position(), (1, 2))
+        Thread(target=self.p.timeline_wal_position).start()
 
-    @patch('os.path.isfile', Mock(return_value=True))
-    @patch('os.kill', Mock(side_effect=Exception))
-    @patch('os.getpid', Mock(return_value=2))
-    @patch('os.getppid', Mock(return_value=2))
-    @patch.object(builtins, 'open', mock_open(read_data='-1'))
-    @patch.object(Postgresql, '_version_file_exists', Mock(return_value=True))
-    def test_is_running(self):
-        self.assertFalse(self.p.is_running())
+    @patch.object(PostmasterProcess, 'from_pidfile')
+    def test_is_running(self, mock_frompidfile):
+        # Cached postmaster running
+        mock_postmaster = self.p._postmaster_proc = MockPostmaster()
+        self.assertEquals(self.p.is_running(), mock_postmaster)
+
+        # Cached postmaster not running, no postmaster running
+        mock_postmaster.is_running.return_value = False
+        mock_frompidfile.return_value = None
+        self.assertEquals(self.p.is_running(), None)
+        self.assertEquals(self.p._postmaster_proc, None)
+
+        # No cached postmaster, postmaster running
+        mock_frompidfile.return_value = mock_postmaster2 = MockPostmaster()
+        self.assertEquals(self.p.is_running(), mock_postmaster2)
+        self.assertEquals(self.p._postmaster_proc, mock_postmaster2)
 
     @patch('shlex.split', Mock(side_effect=OSError))
     def test_call_nowait(self):
+        self.p.set_role('replica')
+        self.assertIsNone(self.p.call_nowait('on_start'))
+        self.p.bootstrapping = True
         self.assertIsNone(self.p.call_nowait('on_start'))
 
     def test_non_existing_callback(self):
         self.assertFalse(self.p.call_nowait('foobar'))
 
-    @patch.object(Postgresql, 'is_running', Mock(return_value=True))
+    @patch.object(Postgresql, 'is_running', Mock(return_value=MockPostmaster()))
     def test_is_leader_exception(self):
         self.p.start()
         self.p.query = Mock(side_effect=psycopg2.OperationalError("not supported"))
@@ -447,36 +570,101 @@ class TestPostgresql(unittest.TestCase):
     @patch.object(Postgresql, 'is_running', Mock(return_value=True))
     def test_bootstrap(self):
         with patch('subprocess.call', Mock(return_value=1)):
-            self.assertRaises(PostgresException, self.p.bootstrap, {})
+            self.assertFalse(self.p.bootstrap({}))
 
-        with patch.object(Postgresql, 'run_bootstrap_post_init', Mock(return_value=False)):
-            self.assertRaises(PostgresException, self.p.bootstrap, {})
+        config = {'users': {'replicator': {'password': 'rep-pass', 'options': ['replication']}}}
 
-        self.p.bootstrap({'users': {'replicator': {'password': 'rep-pass', 'options': ['replication']}},
-                          'pg_hba': ['host replication replicator 127.0.0.1/32 md5',
-                                     'hostssl all all 0.0.0.0/0 md5',
-                                     'host all all 0.0.0.0/0 md5'],
-                          'post_init': '/bin/false'})
+        self.p.bootstrap(config)
+        with open(os.path.join(self.config_dir, 'pg_hba.conf')) as f:
+            lines = f.readlines()
+            self.assertTrue('host all all 0.0.0.0/0 md5\n' in lines)
+
+        self.p.config.pop('pg_hba')
+        config.update({'post_init': '/bin/false',
+                       'pg_hba': ['host replication replicator 127.0.0.1/32 md5',
+                                  'hostssl all all 0.0.0.0/0 md5',
+                                  'host all all 0.0.0.0/0 md5']})
+        self.p.bootstrap(config)
         with open(os.path.join(self.data_dir, 'pg_hba.conf')) as f:
             lines = f.readlines()
-            assert 'host replication replicator 127.0.0.1/32 md5\n' in lines
-            assert 'host all all 0.0.0.0/0 md5\n' in lines
+            self.assertTrue('host replication replicator 127.0.0.1/32 md5\n' in lines)
 
-    def test_run_bootstrap_post_init(self):
-        with patch('subprocess.call', Mock(return_value=1)):
-            self.assertFalse(self.p.run_bootstrap_post_init({'post_init': '/bin/false'}))
+    @patch.object(Postgresql, 'cancellable_subprocess_call')
+    def test_custom_bootstrap(self, mock_cancellable_subprocess_call):
+        self.p.config.pop('pg_hba')
+        config = {'method': 'foo', 'foo': {'command': 'bar'}}
 
-        with patch('subprocess.call', Mock(side_effect=OSError)):
-            self.assertFalse(self.p.run_bootstrap_post_init({'post_init': '/bin/false'}))
+        mock_cancellable_subprocess_call.return_value = 1
+        self.assertFalse(self.p.bootstrap(config))
 
-        with patch('subprocess.call', Mock(return_value=0)) as mock_method:
-            self.p._superuser.pop('username')
-            self.assertTrue(self.p.run_bootstrap_post_init({'post_init': '/bin/false'}))
+        mock_cancellable_subprocess_call.return_value = 0
+        with patch('subprocess.Popen', Mock(side_effect=Exception("42"))),\
+                patch('os.path.isfile', Mock(return_value=True)),\
+                patch('os.unlink', Mock()),\
+                patch.object(Postgresql, 'save_configuration_files', Mock()),\
+                patch.object(Postgresql, 'restore_configuration_files', Mock()),\
+                patch.object(Postgresql, 'write_recovery_conf', Mock()):
+            with self.assertRaises(Exception) as e:
+                self.p.bootstrap(config)
+            self.assertEqual(str(e.exception), '42')
 
-        mock_method.assert_called()
-        args, kwargs = mock_method.call_args
-        assert 'PGPASSFILE' in kwargs['env'].keys()
-        self.assertEquals(args[0], ['/bin/false', 'postgres://localhost:5432/postgres'])
+            config['foo']['recovery_conf'] = {'foo': 'bar'}
+
+            with self.assertRaises(Exception) as e:
+                self.p.bootstrap(config)
+            self.assertEqual(str(e.exception), '42')
+
+        mock_cancellable_subprocess_call.side_effect = Exception
+        self.assertFalse(self.p.bootstrap(config))
+
+    @patch('time.sleep', Mock())
+    @patch('os.unlink', Mock())
+    @patch.object(Postgresql, 'run_bootstrap_post_init', Mock(return_value=True))
+    @patch.object(Postgresql, '_custom_bootstrap', Mock(return_value=True))
+    @patch.object(Postgresql, 'start', Mock(return_value=True))
+    def test_post_bootstrap(self):
+        config = {'method': 'foo', 'foo': {'command': 'bar'}}
+        self.p.bootstrap(config)
+
+        task = CriticalTask()
+        with patch.object(Postgresql, 'create_or_update_role', Mock(side_effect=Exception)):
+            self.p.post_bootstrap({}, task)
+            self.assertFalse(task.result)
+
+        self.p.config.pop('pg_hba')
+        self.p.post_bootstrap({}, task)
+        self.assertTrue(task.result)
+
+        self.p.bootstrap(config)
+        self.p.set_state('stopped')
+        self.p.reload_config({'authentication': {'superuser': {'username': 'p', 'password': 'p'},
+                                                 'replication': {'username': 'r', 'password': 'r'}},
+                              'listen': '*', 'retry_timeout': 10, 'parameters': {'hba_file': 'foo'}})
+        with patch.object(Postgresql, 'restart', Mock()) as mock_restart:
+            self.p.post_bootstrap({}, task)
+            mock_restart.assert_called_once()
+
+    @patch.object(Postgresql, 'cancellable_subprocess_call')
+    def test_run_bootstrap_post_init(self, mock_cancellable_subprocess_call):
+        mock_cancellable_subprocess_call.return_value = 1
+        self.assertFalse(self.p.run_bootstrap_post_init({'post_init': '/bin/false'}))
+
+        mock_cancellable_subprocess_call.return_value = 0
+        self.p._superuser.pop('username')
+        self.assertTrue(self.p.run_bootstrap_post_init({'post_init': '/bin/false'}))
+        mock_cancellable_subprocess_call.assert_called()
+        args, kwargs = mock_cancellable_subprocess_call.call_args
+        self.assertTrue('PGPASSFILE' in kwargs['env'])
+        self.assertEquals(args[0], ['/bin/false', 'postgres://127.0.0.2:5432/postgres'])
+
+        mock_cancellable_subprocess_call.reset_mock()
+        self.p._local_address.pop('host')
+        self.assertTrue(self.p.run_bootstrap_post_init({'post_init': '/bin/false'}))
+        mock_cancellable_subprocess_call.assert_called()
+        self.assertEquals(mock_cancellable_subprocess_call.call_args[0][0], ['/bin/false', 'postgres://:5432/postgres'])
+
+        mock_cancellable_subprocess_call.side_effect = OSError
+        self.assertFalse(self.p.run_bootstrap_post_init({'post_init': '/bin/false'}))
 
     @patch('patroni.postgresql.Postgresql.create_replica', Mock(return_value=0))
     def test_clone(self):
@@ -508,65 +696,6 @@ class TestPostgresql(unittest.TestCase):
         with patch('subprocess.check_output', Mock(side_effect=subprocess.CalledProcessError(1, ''))):
             self.assertEquals(self.p.controldata(), {})
 
-    def test_read_postmaster_opts(self):
-        m = mock_open(read_data=postmaster_opts_string())
-        with patch.object(builtins, 'open', m):
-            data = self.p.read_postmaster_opts()
-            self.assertEquals(data['wal_level'], 'hot_standby')
-            self.assertEquals(int(data['max_replication_slots']), 5)
-            self.assertEqual(data.get('D'), None)
-
-            m.side_effect = IOError
-            data = self.p.read_postmaster_opts()
-            self.assertEqual(data, dict())
-
-    @patch('subprocess.Popen')
-    @patch.object(builtins, 'open', MagicMock(return_value=42))
-    def test_single_user_mode(self, subprocess_popen_mock):
-        subprocess_popen_mock.return_value.wait.return_value = 0
-        self.assertEquals(self.p.single_user_mode(options=dict(archive_mode='on', archive_command='false')), 0)
-        subprocess_popen_mock.assert_called_once_with(['postgres', '--single', '-D', self.data_dir,
-                                                      '-c', 'archive_command=false', '-c', 'archive_mode=on',
-                                                       'postgres'], stdin=subprocess.PIPE,
-                                                      stdout=42,
-                                                      stderr=subprocess.STDOUT)
-        subprocess_popen_mock.reset_mock()
-        self.assertEquals(self.p.single_user_mode(command="CHECKPOINT"), 0)
-        subprocess_popen_mock.assert_called_once_with(['postgres', '--single', '-D', self.data_dir,
-                                                      'postgres'], stdin=subprocess.PIPE,
-                                                      stdout=42,
-                                                      stderr=subprocess.STDOUT)
-        subprocess_popen_mock.return_value = None
-        self.assertEquals(self.p.single_user_mode(), 1)
-
-    @patch('os.listdir', MagicMock(side_effect=fake_listdir))
-    @patch('os.unlink', return_value=True)
-    @patch('os.remove', return_value=True)
-    @patch('os.path.islink', return_value=False)
-    @patch('os.path.isfile', return_value=True)
-    def test_cleanup_archive_status(self, mock_file, mock_link, mock_remove, mock_unlink):
-        ap = os.path.join(self.data_dir, 'pg_xlog', 'archive_status/')
-        self.p.cleanup_archive_status()
-        mock_remove.assert_has_calls([mock.call(ap + 'a'), mock.call(ap + 'b'), mock.call(ap + 'c')])
-        mock_unlink.assert_not_called()
-
-        mock_remove.reset_mock()
-
-        mock_file.return_value = False
-        mock_link.return_value = True
-        self.p.cleanup_archive_status()
-        mock_unlink.assert_has_calls([mock.call(ap + 'a'), mock.call(ap + 'b'), mock.call(ap + 'c')])
-        mock_remove.assert_not_called()
-
-        mock_unlink.reset_mock()
-        mock_remove.reset_mock()
-
-        mock_file.side_effect = OSError
-        mock_link.side_effect = OSError
-        self.p.cleanup_archive_status()
-        mock_unlink.assert_not_called()
-        mock_remove.assert_not_called()
-
     @patch('patroni.postgresql.Postgresql._version_file_exists', Mock(return_value=True))
     @patch('subprocess.check_output', MagicMock(return_value=0, side_effect=pg_controldata_string))
     def test_sysid(self):
@@ -601,21 +730,27 @@ class TestPostgresql(unittest.TestCase):
     def test_reload_config(self):
         parameters = self._PARAMETERS.copy()
         parameters.pop('f.oo')
-        self.p.reload_config({'retry_timeout': 10, 'listen': '*', 'parameters': parameters})
+        config = {'pg_hba': [''], 'use_unix_socket': True, 'authentication': {},
+                  'retry_timeout': 10, 'listen': '*', 'parameters': parameters}
+        self.p.reload_config(config)
         parameters['b.ar'] = 'bar'
-        self.p.reload_config({'retry_timeout': 10, 'listen': '*', 'parameters': parameters})
+        self.p.reload_config(config)
         parameters['autovacuum'] = 'on'
-        self.p.reload_config({'retry_timeout': 10, 'listen': '*', 'parameters': parameters})
+        self.p.reload_config(config)
         parameters['autovacuum'] = 'off'
         parameters.pop('search_path')
-        self.p.reload_config({'retry_timeout': 10, 'listen': '*:5433', 'parameters': parameters})
+        config['listen'] = '*:5433'
+        self.p.reload_config(config)
+        parameters['unix_socket_directories'] = '.'
+        self.p.reload_config(config)
+        self.p.resolve_connection_addresses()
 
     @patch.object(Postgresql, '_version_file_exists', Mock(return_value=True))
     def test_get_major_version(self):
         with patch.object(builtins, 'open', mock_open(read_data='9.4')):
-            self.assertEquals(self.p.get_major_version(), 9.4)
+            self.assertEquals(self.p.get_major_version(), 90400)
         with patch.object(builtins, 'open', Mock(side_effect=Exception)):
-            self.assertEquals(self.p.get_major_version(), 0.0)
+            self.assertEquals(self.p.get_major_version(), 0)
 
     def test_postmaster_start_time(self):
         with patch.object(MockCursor, "fetchone", Mock(return_value=('foo', True, '', '', '', '', False))):
@@ -692,22 +827,14 @@ class TestPostgresql(unittest.TestCase):
                 self.assertFalse(self.p.wait_for_startup(timeout=2))
                 self.assertEquals(state['sleeps'], 3)
 
-    def test_read_pid_file(self):
-        pidfile = os.path.join(self.data_dir, 'postmaster.pid')
-        if os.path.exists(pidfile):
-            os.remove(pidfile)
-        self.assertEquals(self.p.read_pid_file(), {})
-
-    @patch('os.kill')
-    def test_is_pid_running(self, mock_kill):
-        mock_kill.return_value = True
-        self.assertTrue(self.p.is_pid_running(-100))
-        self.assertFalse(self.p.is_pid_running(0))
-        self.assertFalse(self.p.is_pid_running(None))
+        with patch.object(Postgresql, 'check_startup_state_changed', Mock(return_value=False)):
+            self.p.cancel()
+            self.p._state = 'starting'
+            self.assertIsNone(self.p.wait_for_startup())
 
     def test_pick_sync_standby(self):
         cluster = Cluster(True, None, self.leader, 0, [self.me, self.other, self.leadermem], None,
-                          SyncState(0, self.me.name, self.leadermem.name))
+                          SyncState(0, self.me.name, self.leadermem.name), None)
 
         with patch.object(Postgresql, "query", return_value=[
                     (self.leadermem.name, 'streaming', 'sync'),
@@ -768,5 +895,77 @@ class TestPostgresql(unittest.TestCase):
     def test_get_server_parameters(self):
         config = {'synchronous_mode': True, 'parameters': {'wal_level': 'hot_standby'}, 'listen': '0'}
         self.p.get_server_parameters(config)
+        config['synchronous_mode_strict'] = True
+        self.p.get_server_parameters(config)
         self.p.set_synchronous_standby('foo')
         self.p.get_server_parameters(config)
+
+    @patch('time.sleep', Mock())
+    def test__wait_for_connection_close(self):
+        mock_postmaster = MockPostmaster()
+        with patch.object(Postgresql, 'is_running', Mock(return_value=mock_postmaster)):
+            mock_postmaster.is_running.side_effect = [True, False, False]
+            mock_callback = Mock()
+            self.p.stop(on_safepoint=mock_callback)
+
+            mock_postmaster.is_running.side_effect = [True, False, False]
+            with patch.object(MockCursor, "execute", Mock(side_effect=psycopg2.Error)):
+                self.p.stop(on_safepoint=mock_callback)
+
+    def test_terminate_starting_postmaster(self):
+        mock_postmaster = MockPostmaster()
+        self.p.terminate_starting_postmaster(mock_postmaster)
+        mock_postmaster.signal_stop.assert_called()
+        mock_postmaster.wait.assert_called()
+
+    def test_read_postmaster_opts(self):
+        m = mock_open(read_data='/usr/lib/postgres/9.6/bin/postgres "-D" "data/postgresql0" \
+"--listen_addresses=127.0.0.1" "--port=5432" "--hot_standby=on" "--wal_level=hot_standby" \
+"--wal_log_hints=on" "--max_wal_senders=5" "--max_replication_slots=5"\n')
+        with patch.object(builtins, 'open', m):
+            data = self.p.read_postmaster_opts()
+            self.assertEquals(data['wal_level'], 'hot_standby')
+            self.assertEquals(int(data['max_replication_slots']), 5)
+            self.assertEqual(data.get('D'), None)
+
+            m.side_effect = IOError
+            data = self.p.read_postmaster_opts()
+            self.assertEqual(data, dict())
+
+    @patch('subprocess.Popen')
+    def test_single_user_mode(self, subprocess_popen_mock):
+        subprocess_popen_mock.return_value.wait.return_value = 0
+        self.assertEquals(self.p.single_user_mode('CHECKPOINT', {'archive_mode': 'on'}), 0)
+
+    @patch('os.listdir', Mock(side_effect=[OSError, ['a', 'b']]))
+    @patch('os.unlink', Mock(side_effect=OSError))
+    @patch('os.remove', Mock())
+    @patch('os.path.islink', Mock(side_effect=[True, False]))
+    @patch('os.path.isfile', Mock(return_value=True))
+    def test_cleanup_archive_status(self):
+        self.p.cleanup_archive_status()
+        self.p.cleanup_archive_status()
+
+    @patch('os.unlink', Mock())
+    @patch('os.path.isfile', Mock(return_value=True))
+    @patch.object(Postgresql, 'single_user_mode', Mock(return_value=0))
+    def test_fix_cluster_state(self):
+        self.assertTrue(self.p.fix_cluster_state())
+
+    def test_replica_cached_timeline(self):
+        self.assertEquals(self.p.replica_cached_timeline(1), 2)
+
+    def test_get_master_timeline(self):
+        self.assertEquals(self.p.get_master_timeline(), 1)
+
+    def test_cancellable_subprocess_call(self):
+        self.p.cancel()
+        self.assertRaises(PostgresException, self.p.cancellable_subprocess_call, communicate_input=None)
+
+    @patch('patroni.postgresql.polling_loop', Mock(return_value=[0, 0]))
+    def test_cancel(self):
+        self.p._cancellable = Mock()
+        self.p._cancellable.returncode = None
+        self.p.cancel()
+        type(self.p._cancellable).returncode = PropertyMock(side_effect=[None, -15])
+        self.p.cancel()

tests/test_postmaster.py

@@ -0,0 +1,103 @@
+import psutil
+import unittest
+
+from mock import Mock, patch, mock_open
+from patroni.postmaster import PostmasterProcess
+from six.moves import builtins
+
+
+class TestPostmasterProcess(unittest.TestCase):
+    @patch('psutil.Process.__init__', Mock())
+    def test_init(self):
+        proc = PostmasterProcess(-123)
+        self.assertTrue(proc.is_single_user)
+
+    @patch('psutil.Process.create_time')
+    @patch('psutil.Process.__init__')
+    @patch('patroni.postmaster.PostmasterProcess._read_postmaster_pidfile')
+    def test_from_pidfile(self, mock_read, mock_init, mock_create_time):
+        mock_init.side_effect = psutil.NoSuchProcess(123)
+        mock_read.return_value = {}
+        self.assertIsNone(PostmasterProcess.from_pidfile(''))
+        mock_read.return_value = {"pid": "foo"}
+        self.assertIsNone(PostmasterProcess.from_pidfile(''))
+        mock_read.return_value = {"pid": "123"}
+        self.assertIsNone(PostmasterProcess.from_pidfile(''))
+
+        mock_init.side_effect = None
+        with patch.object(psutil.Process, 'pid', 123), \
+                patch.object(psutil.Process, 'ppid', return_value=124), \
+                patch('os.getpid', return_value=125) as mock_ospid, \
+                patch('os.getppid', return_value=126):
+
+            self.assertIsNotNone(PostmasterProcess.from_pidfile(''))
+
+            mock_create_time.return_value = 100000
+            mock_read.return_value = {"pid": "123", "start_time": "200000"}
+            self.assertIsNone(PostmasterProcess.from_pidfile(''))
+
+            mock_read.return_value = {"pid": "123", "start_time": "foobar"}
+            self.assertIsNotNone(PostmasterProcess.from_pidfile(''))
+
+            mock_ospid.return_value = 123
+            mock_read.return_value = {"pid": "123", "start_time": "100000"}
+            self.assertIsNone(PostmasterProcess.from_pidfile(''))
+
+    @patch('psutil.Process.__init__')
+    def test_from_pid(self, mock_init):
+        mock_init.side_effect = psutil.NoSuchProcess(123)
+        self.assertEquals(PostmasterProcess.from_pid(123), None)
+        mock_init.side_effect = None
+        self.assertNotEquals(PostmasterProcess.from_pid(123), None)
+
+    @patch('psutil.Process.__init__', Mock())
+    @patch('psutil.Process.send_signal')
+    @patch('psutil.Process.pid', Mock(return_value=123))
+    def test_signal_stop(self, mock_send_signal):
+        proc = PostmasterProcess(-123)
+        self.assertEquals(proc.signal_stop('immediate'), False)
+
+        mock_send_signal.side_effect = [None, psutil.NoSuchProcess(123), psutil.AccessDenied()]
+        proc = PostmasterProcess(123)
+        self.assertEquals(proc.signal_stop('immediate'), None)
+        self.assertEquals(proc.signal_stop('immediate'), True)
+        self.assertEquals(proc.signal_stop('immediate'), False)
+
+    @patch('psutil.Process.__init__', Mock())
+    @patch('psutil.wait_procs')
+    def test_wait_for_user_backends_to_close(self, mock_wait):
+        c1 = Mock()
+        c1.cmdline = Mock(return_value=["postgres: startup process"])
+        c2 = Mock()
+        c2.cmdline = Mock(return_value=["postgres: postgres postgres [local] idle"])
+        c3 = Mock()
+        c3.cmdline = Mock(side_effect=psutil.NoSuchProcess(123))
+        with patch('psutil.Process.children', Mock(return_value=[c1, c2, c3])):
+            proc = PostmasterProcess(123)
+            self.assertIsNone(proc.wait_for_user_backends_to_close())
+            mock_wait.assert_called_with([c2])
+
+        c3.cmdline = Mock(side_effect=psutil.AccessDenied(123))
+        with patch('psutil.Process.children', Mock(return_value=[c3])):
+            proc = PostmasterProcess(123)
+            self.assertIsNone(proc.wait_for_user_backends_to_close())
+
+    @patch('subprocess.Popen')
+    @patch.object(PostmasterProcess, 'from_pid')
+    @patch.object(PostmasterProcess, '_from_pidfile')
+    def test_start(self, mock_frompidfile, mock_frompid, mock_popen):
+        mock_frompidfile.return_value._is_postmaster_process.return_value = False
+        mock_frompid.return_value = "proc 123"
+        mock_popen.return_value.stdout.readline.return_value = '123'
+        self.assertEquals(PostmasterProcess.start('true', '/tmp', '/tmp/test.conf', []), "proc 123")
+        mock_frompid.assert_called_with(123)
+
+        mock_frompidfile.side_effect = psutil.NoSuchProcess(123)
+        self.assertEquals(PostmasterProcess.start('true', '/tmp', '/tmp/test.conf', []), "proc 123")
+
+    @patch('psutil.Process.__init__', Mock(side_effect=psutil.NoSuchProcess(123)))
+    def test_read_postmaster_pidfile(self):
+        with patch.object(builtins, 'open', Mock(side_effect=IOError)):
+            self.assertIsNone(PostmasterProcess.from_pidfile(''))
+        with patch.object(builtins, 'open', mock_open(read_data='123\n')):
+            self.assertIsNone(PostmasterProcess.from_pidfile(''))

tests/test_wale_restore.py

@@ -2,41 +2,67 @@ import psycopg2
 import subprocess
 import unittest
 
-from mock import Mock, MagicMock, patch, mock_open
+from mock import Mock, PropertyMock, patch, mock_open
+from patroni.scripts import wale_restore
 from patroni.scripts.wale_restore import WALERestore, main as _main, get_major_version
 from six.moves import builtins
+from test_postgresql import MockConnect, psycopg2_connect
 
 
-wale_output = b'name last_modified expanded_size_bytes wal_segment_backup_start ' +\
-              b'wal_segment_offset_backup_start wal_segment_backup_stop wal_segment_offset_backup_stop\n' +\
-              b'base_00000001000000000000007F_00000040 2015-05-18T10:13:25.000Z 167772160 ' +\
-              b'00000001000000000000007F 00000040 00000001000000000000007F 00000240\n'
+wale_output_header = (
+    b'name\tlast_modified\t'
+    b'expanded_size_bytes\t'
+    b'wal_segment_backup_start\twal_segment_offset_backup_start\t'
+    b'wal_segment_backup_stop\twal_segment_offset_backup_stop\n'
+)
+
+wale_output_values = (
+    b'base_00000001000000000000007F_00000040\t2015-05-18T10:13:25.000Z\t'
+    b'167772160\t'
+    b'00000001000000000000007F\t00000040\t'
+    b'00000001000000000000007F\t00000240\n'
+)
+
+wale_output = wale_output_header + wale_output_values
+
+wale_restore.RETRY_SLEEP_INTERVAL = 0.001  # Speed up retries
+WALE_TEST_RETRIES = 2
 
 
 @patch('os.access', Mock(return_value=True))
 @patch('os.makedirs', Mock(return_value=True))
 @patch('os.path.exists', Mock(return_value=True))
 @patch('os.path.isdir', Mock(return_value=True))
-@patch('psycopg2.extensions.cursor', Mock(autospec=True))
-@patch('psycopg2.extensions.connection', Mock(autospec=True))
-@patch('psycopg2.connect', MagicMock(autospec=True))
+@patch('psycopg2.connect', psycopg2_connect)
 @patch('subprocess.check_output', Mock(return_value=wale_output))
 class TestWALERestore(unittest.TestCase):
 
     def setUp(self):
-        self.wale_restore = WALERestore("batman", "/data", "host=batman port=5432 user=batman",
-                                        "/etc", 100, 100, 1, 0, 1)
+        self.wale_restore = WALERestore('batman', '/data', 'host=batman port=5432 user=batman',
+                                        '/etc', 100, 100, 1, 0, WALE_TEST_RETRIES)
 
     def test_should_use_s3_to_create_replica(self):
         self.assertTrue(self.wale_restore.should_use_s3_to_create_replica())
+        with patch.object(MockConnect, 'server_version', PropertyMock(return_value=100000)):
+            self.assertTrue(self.wale_restore.should_use_s3_to_create_replica())
+
+        with patch('subprocess.check_output', Mock(return_value=wale_output.replace(b'167772160', b'1'))):
+            self.assertFalse(self.wale_restore.should_use_s3_to_create_replica())
 
         with patch('psycopg2.connect', Mock(side_effect=psycopg2.Error("foo"))):
             save_no_master = self.wale_restore.no_master
             save_master_connection = self.wale_restore.master_connection
 
             self.assertFalse(self.wale_restore.should_use_s3_to_create_replica())
-            self.wale_restore.no_master = 1
-            self.assertTrue(self.wale_restore.should_use_s3_to_create_replica())  # this would do 2 retries 1 sec each
+
+            with patch('time.sleep', Mock(return_value=None)) as mock_sleep:
+                self.wale_restore.no_master = 1
+                assert self.wale_restore.should_use_s3_to_create_replica()
+                # verify retries
+                mock_sleep.assert_has_calls(
+                    [((wale_restore.RETRY_SLEEP_INTERVAL,),)] * WALE_TEST_RETRIES
+                )
+
             self.wale_restore.master_connection = ''
             self.assertTrue(self.wale_restore.should_use_s3_to_create_replica())
 
@@ -45,10 +71,9 @@ class TestWALERestore(unittest.TestCase):
 
         with patch('subprocess.check_output', Mock(side_effect=subprocess.CalledProcessError(1, "cmd", "foo"))):
             self.assertFalse(self.wale_restore.should_use_s3_to_create_replica())
-        with patch('subprocess.check_output', Mock(return_value=wale_output.split(b'\n')[0])):
+        with patch('subprocess.check_output', Mock(return_value=wale_output_header)):
             self.assertFalse(self.wale_restore.should_use_s3_to_create_replica())
-        with patch('subprocess.check_output',
-                   Mock(return_value=wale_output.replace(b' wal_segment_offset_backup_stop', b''))):
+        with patch('subprocess.check_output', Mock(return_value=wale_output + wale_output_values)):
             self.assertFalse(self.wale_restore.should_use_s3_to_create_replica())
         with patch('subprocess.check_output',
                    Mock(return_value=wale_output.replace(b'expanded_size_bytes', b'expanded_size_foo'))):
@@ -70,17 +95,22 @@ class TestWALERestore(unittest.TestCase):
         with patch.object(self.wale_restore, 'should_use_s3_to_create_replica', Mock(return_value=True)):
             with patch.object(self.wale_restore, 'create_replica_with_s3', Mock(return_value=0)):
                 self.assertEqual(self.wale_restore.run(), 0)
-            with patch.object(self.wale_restore, 'should_use_s3_to_create_replica', Mock(return_value=None)):
-                self.assertEqual(self.wale_restore.run(), 1)
-            with patch.object(self.wale_restore, 'should_use_s3_to_create_replica', Mock(side_effect=Exception)):
-                self.assertEqual(self.wale_restore.run(), 2)
+        with patch.object(self.wale_restore, 'should_use_s3_to_create_replica', Mock(return_value=False)):
+            self.assertEqual(self.wale_restore.run(), 2)
+        with patch.object(self.wale_restore, 'should_use_s3_to_create_replica', Mock(return_value=None)):
+            self.assertEqual(self.wale_restore.run(), 1)
+        with patch.object(self.wale_restore, 'should_use_s3_to_create_replica', Mock(side_effect=Exception)):
+            self.assertEqual(self.wale_restore.run(), 2)
 
     @patch('sys.exit', Mock())
     def test_main(self):
         with patch.object(WALERestore, 'run', Mock(return_value=0)):
             self.assertEqual(_main(), 0)
-        with patch.object(WALERestore, 'run', Mock(return_value=1)):
+
+        with patch.object(WALERestore, 'run', Mock(return_value=1)), \
+                patch('time.sleep', Mock(return_value=None)) as mock_sleep:
             self.assertEqual(_main(), 1)
+            assert mock_sleep.call_count == WALE_TEST_RETRIES
 
     @patch('os.path.isfile', Mock(return_value=True))
     def test_get_major_version(self):

tests/test_watchdog.py

@@ -0,0 +1,217 @@
+import ctypes
+import patroni.watchdog.linux as linuxwd
+import sys
+import unittest
+
+from mock import patch, Mock, PropertyMock
+from patroni.watchdog import Watchdog, WatchdogError
+from patroni.watchdog.base import NullWatchdog
+from patroni.watchdog.linux import LinuxWatchdogDevice
+
+
+class MockDevice(object):
+    def __init__(self, fd, filename, flag):
+        self.fd = fd
+        self.filename = filename
+        self.flag = flag
+        self.timeout = 60
+        self.open = True
+        self.writes = []
+
+
+mock_devices = [None]
+
+
+def mock_open(filename, flag):
+    fd = len(mock_devices)
+    mock_devices.append(MockDevice(fd, filename, flag))
+    return fd
+
+
+def mock_ioctl(fd, op, arg=None, mutate_flag=False):
+    assert 0 < fd < len(mock_devices)
+    dev = mock_devices[fd]
+    sys.stderr.write("Ioctl %d %d %r\n" % (fd, op, arg))
+    if op == linuxwd.WDIOC_GETSUPPORT:
+        sys.stderr.write("Get support\n")
+        assert(mutate_flag is True)
+        arg.options = sum(map(linuxwd.WDIOF.get, ['SETTIMEOUT', 'KEEPALIVEPING']))
+        arg.identity = (ctypes.c_ubyte*32)(*map(ord, 'Mock Watchdog'))
+    elif op == linuxwd.WDIOC_GETTIMEOUT:
+        arg.value = dev.timeout
+    elif op == linuxwd.WDIOC_SETTIMEOUT:
+        sys.stderr.write("Set timeout called with %s\n" % arg.value)
+        assert 0 < arg.value < 65535
+        dev.timeout = arg.value - 1
+    else:
+        raise Exception("Unknown op %d", op)
+    return 0
+
+
+def mock_write(fd, string):
+    assert 0 < fd < len(mock_devices)
+    assert len(string) == 1
+    assert mock_devices[fd].open
+    mock_devices[fd].writes.append(string)
+
+
+def mock_close(fd):
+    assert 0 < fd < len(mock_devices)
+    assert mock_devices[fd].open
+    mock_devices[fd].open = False
+
+
+@patch('os.open', mock_open)
+@patch('os.write', mock_write)
+@patch('os.close', mock_close)
+@patch('fcntl.ioctl', mock_ioctl)
+class TestWatchdog(unittest.TestCase):
+    def setUp(self):
+        mock_devices[:] = [None]
+
+    @patch('platform.system', Mock(return_value='Linux'))
+    @patch.object(LinuxWatchdogDevice, 'can_be_disabled', PropertyMock(return_value=True))
+    def test_unsafe_timeout_disable_watchdog_and_exit(self):
+        watchdog = Watchdog({'ttl': 30, 'loop_wait': 15, 'watchdog': {'mode': 'required', 'safety_margin': -1}})
+        self.assertEquals(watchdog.activate(), False)
+        self.assertEquals(watchdog.is_running, False)
+
+    @patch('platform.system', Mock(return_value='Linux'))
+    @patch.object(LinuxWatchdogDevice, 'get_timeout', Mock(return_value=16))
+    def test_timeout_does_not_ensure_safe_termination(self):
+        Watchdog({'ttl': 30, 'loop_wait': 15, 'watchdog': {'mode': 'auto', 'safety_margin': -1}}).activate()
+        self.assertEquals(len(mock_devices), 2)
+
+    @patch('platform.system', Mock(return_value='Linux'))
+    @patch.object(Watchdog, 'is_running', PropertyMock(return_value=False))
+    def test_watchdog_not_activated(self):
+        self.assertFalse(Watchdog({'ttl': 30, 'loop_wait': 10, 'watchdog': {'mode': 'required'}}).activate())
+
+    @patch('platform.system', Mock(return_value='Linux'))
+    @patch.object(LinuxWatchdogDevice, 'is_running', PropertyMock(return_value=False))
+    def test_watchdog_activate(self):
+        with patch.object(LinuxWatchdogDevice, 'open', Mock(side_effect=WatchdogError(''))):
+            self.assertTrue(Watchdog({'ttl': 30, 'loop_wait': 10, 'watchdog': {'mode': 'auto'}}).activate())
+        self.assertFalse(Watchdog({'ttl': 30, 'loop_wait': 10, 'watchdog': {'mode': 'required'}}).activate())
+
+    @patch('platform.system', Mock(return_value='Linux'))
+    def test_basic_operation(self):
+        watchdog = Watchdog({'ttl': 30, 'loop_wait': 10, 'watchdog': {'mode': 'required'}})
+        watchdog.activate()
+
+        self.assertEquals(len(mock_devices), 2)
+        device = mock_devices[-1]
+        self.assertTrue(device.open)
+
+        self.assertEquals(device.timeout, 24)
+
+        watchdog.keepalive()
+        self.assertEquals(len(device.writes), 1)
+
+        watchdog.disable()
+        self.assertFalse(device.open)
+        self.assertEquals(device.writes[-1], b'V')
+
+    def test_invalid_timings(self):
+        watchdog = Watchdog({'ttl': 30, 'loop_wait': 20, 'watchdog': {'mode': 'automatic', 'safety_margin': -1}})
+        watchdog.activate()
+        self.assertEquals(len(mock_devices), 1)
+        self.assertFalse(watchdog.is_running)
+
+    def test_parse_mode(self):
+        with patch('patroni.watchdog.base.logger.warning', new_callable=Mock()) as warning_mock:
+            watchdog = Watchdog({'ttl': 30, 'loop_wait': 10, 'watchdog': {'mode': 'bad'}})
+            self.assertEquals(watchdog.config.mode, 'off')
+            warning_mock.assert_called_once()
+
+    @patch('platform.system', Mock(return_value='Unknown'))
+    def test_unsupported_platform(self):
+        self.assertRaises(SystemExit, Watchdog, {'ttl': 30, 'loop_wait': 10,
+                                                 'watchdog': {'mode': 'required', 'driver': 'bad'}})
+
+    def test_exceptions(self):
+        wd = Watchdog({'ttl': 30, 'loop_wait': 10, 'watchdog': {'mode': 'bad'}})
+        wd.impl.close = wd.impl.keepalive = Mock(side_effect=WatchdogError(''))
+        self.assertTrue(wd.activate())
+        self.assertIsNone(wd.keepalive())
+        self.assertIsNone(wd.disable())
+
+    @patch('platform.system', Mock(return_value='Linux'))
+    def test_config_reload(self):
+        watchdog = Watchdog({'ttl': 30, 'loop_wait': 15, 'watchdog': {'mode': 'required'}})
+        self.assertTrue(watchdog.activate())
+        self.assertTrue(watchdog.is_running)
+
+        watchdog.reload_config({'ttl': 30, 'loop_wait': 15, 'watchdog': {'mode': 'off'}})
+        self.assertFalse(watchdog.is_running)
+
+        watchdog.reload_config({'ttl': 30, 'loop_wait': 15, 'watchdog': {'mode': 'required'}})
+        self.assertFalse(watchdog.is_running)
+        watchdog.keepalive()
+        self.assertTrue(watchdog.is_running)
+
+        watchdog.disable()
+        watchdog.reload_config({'ttl': 30, 'loop_wait': 15, 'watchdog': {'mode': 'required', 'driver': 'unknown'}})
+        self.assertFalse(watchdog.is_healthy)
+
+        self.assertFalse(watchdog.activate())
+        watchdog.reload_config({'ttl': 30, 'loop_wait': 15, 'watchdog': {'mode': 'required'}})
+        self.assertFalse(watchdog.is_running)
+        watchdog.keepalive()
+        self.assertTrue(watchdog.is_running)
+
+        watchdog.reload_config({'ttl': 60, 'loop_wait': 15, 'watchdog': {'mode': 'required'}})
+        watchdog.keepalive()
+
+
+class TestNullWatchdog(unittest.TestCase):
+
+    def test_basics(self):
+        watchdog = NullWatchdog()
+        self.assertTrue(watchdog.can_be_disabled)
+        self.assertRaises(WatchdogError, watchdog.set_timeout, 1)
+        self.assertEquals(watchdog.describe(), 'NullWatchdog')
+        self.assertIsInstance(NullWatchdog.from_config({}), NullWatchdog)
+
+
+class TestLinuxWatchdogDevice(unittest.TestCase):
+
+    def setUp(self):
+        self.impl = LinuxWatchdogDevice.from_config({})
+
+    @patch('os.open', Mock(return_value=3))
+    @patch('os.write', Mock(side_effect=OSError))
+    @patch('fcntl.ioctl', Mock(return_value=0))
+    def test_basics(self):
+        self.impl.open()
+        try:
+            if self.impl.get_support().has_foo:
+                self.assertFail()
+        except Exception as e:
+            self.assertTrue(isinstance(e, AttributeError))
+        self.assertRaises(WatchdogError, self.impl.close)
+        self.assertRaises(WatchdogError, self.impl.keepalive)
+        self.assertRaises(WatchdogError, self.impl.set_timeout, -1)
+
+    @patch('os.open', Mock(return_value=3))
+    @patch('fcntl.ioctl', Mock(side_effect=OSError))
+    def test__ioctl(self):
+        self.assertRaises(WatchdogError, self.impl.get_support)
+        self.impl.open()
+        self.assertRaises(WatchdogError, self.impl.get_support)
+
+    def test_is_healthy(self):
+        self.assertFalse(self.impl.is_healthy)
+
+    @patch('os.open', Mock(return_value=3))
+    @patch('fcntl.ioctl', Mock(side_effect=OSError))
+    def test_error_handling(self):
+        self.impl.open()
+        self.assertRaises(WatchdogError, self.impl.get_timeout)
+        self.assertRaises(WatchdogError, self.impl.set_timeout, 10)
+        # We still try to output a reasonable string even if getting info errors
+        self.assertEquals(self.impl.describe(), "Linux watchdog device")
+
+    @patch('os.open', Mock(side_effect=OSError))
+    def test_open(self):
+        self.assertRaises(WatchdogError, self.impl.open)

tests/test_zookeeper.py

@@ -58,11 +58,11 @@ class MockKazooClient(Mock):
             raise TypeError("Invalid type for 'path' (string expected)")
         if not isinstance(value, (six.binary_type,)):
             raise TypeError("Invalid type for 'value' (must be a byte string)")
-        if value == b'Exception':
+        if b'Exception' in value:
             raise Exception
         if path.endswith('/initialize') or path == '/service/test/optime/leader':
             raise Exception
-        elif value == b'retry' or (value == b'exists' and self.exists):
+        elif b'retry' in value or (b'exists' in value and self.exists):
             raise NodeExistsError
 
     def create_async(self, path, value=b"", acl=None, ephemeral=False, sequence=False, makepath=False):
@@ -76,10 +76,10 @@ class MockKazooClient(Mock):
             raise TypeError("Invalid type for 'value' (must be a byte string)")
         if path == '/service/bla/optime/leader':
             raise Exception
-        if path == '/service/test/members/bar' and value == b'retry':
+        if path == '/service/test/members/bar' and b'retry' in value:
             return
         if path in ('/service/test/failover', '/service/test/config', '/service/test/sync'):
-            if value == b'Exception':
+            if b'Exception' in value:
                 raise Exception
             elif value == b'ok':
                 return
@@ -145,7 +145,7 @@ class TestZooKeeper(unittest.TestCase):
         self.assertRaises(ZooKeeperError, self.zk.get_cluster)
         cluster = self.zk.get_cluster()
         self.assertIsInstance(cluster.leader, Leader)
-        self.zk.touch_member('foo')
+        self.zk.touch_member({'foo': 'foo'})
 
     def test_delete_leader(self):
         self.assertTrue(self.zk.delete_leader())
@@ -169,17 +169,17 @@ class TestZooKeeper(unittest.TestCase):
     def test_touch_member(self):
         self.zk._name = 'buzz'
         self.zk.get_cluster()
-        self.zk.touch_member('new')
+        self.zk.touch_member({'new': 'new'})
         self.zk._name = 'bar'
-        self.zk.touch_member('new')
+        self.zk.touch_member({'new': 'new'})
         self.zk._name = 'na'
         self.zk._client.exists = 1
-        self.zk.touch_member('Exception')
+        self.zk.touch_member({'Exception': 'Exception'})
         self.zk._name = 'bar'
-        self.zk.touch_member('retry')
+        self.zk.touch_member({'retry': 'retry'})
         self.zk._fetch_cluster = True
         self.zk.get_cluster()
-        self.zk.touch_member('retry')
+        self.zk.touch_member({'retry': 'retry'})
 
     def test_take_leader(self):
         self.zk.take_leader()
@@ -187,7 +187,7 @@ class TestZooKeeper(unittest.TestCase):
             self.zk.take_leader()
 
     def test_update_leader(self):
-        self.assertTrue(self.zk.update_leader())
+        self.assertTrue(self.zk.update_leader(None))
 
     def test_write_leader_optime(self):
         self.zk.last_leader_operation = '0'
@@ -217,3 +217,6 @@ class TestZooKeeper(unittest.TestCase):
         self.zk.set_sync_state_value('ok')
         self.zk.set_sync_state_value('Exception')
         self.zk.delete_sync_state()
+
+    def test_set_history_value(self):
+        self.zk.set_history_value('{}')