Skip to content

credativ/ansible-playbook-patroni-debian

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

43 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Ansible Playbook for Patroni on Debian

This Ansible playbook allows to deploy a Patroni cluster using the patroni packages provided by Debian and/or PostgreSQL's apt.postgresql.org repository.

Those packages have been integrated into Debian's postgresql-common framework and will look very similar to a regular stand-alone PostgreSQL install on Debian. This is done by using pg_createconfig_patroni creating setting the manual flag in the /etc/postgresql/$PG_VERSION/$CLUSTER_NAME/start.conf config file. That will cause the postgres init script/systemd unit not to start that cluster automatically. Instead patroni will take care to start and manage that cluster.

For production use it should be hardened (e.g. using Ansible Vault and encrypted communication between the DCS server and the Patroni clients).

The default topology consists of one node (master) acting as the DCS (Distributed Consensus Store) server, and three PostgreSQL/Patroni nodes (pg1, pg2 and pg3), see inventory.

DCS Server

Patroni requires a DCS for leader election and as configuration store. Supported DCS server are Etcd, Consul and Zookeeper. If e.g. an Etcd cluster is already available, then it can be configured either by editing templates/dcs.yml or (if that suffices) by setting the dcs_server Ansible variable to its IP address.

Usage

Assuming password-less SSH access to the four nodes is configured, the playbook can be run as follows:

ansible-playbook -i inventory patroni.yml

Supported Versions

The playbooks have been tested on Debian 9 (stretch), testing (buster) and unstable (sid), as well as Ubuntu LTS 18.04 (bionic). As the apt.postgresql.org PostgreSQL (and Patroni) packages are used, all supported PostgreSQL versions can be installed in principle.

Note that Consul is unsupported as DCS on Debian 9 (stretch).

Variables

The following useful variables can be set:

  • dcs (etcd (default), consul or zookeeper)
  • dcs_server_ips (default: undefined, if set will override dcs_servers - see below)
  • postgresql_cluster_name (default: test)
  • postgresql_major_version (default: 11)
  • postgresql_data_dir_base (default: /var/lib/postgresql)
  • patroni_replication_user (default: replicator)
  • patroni_replication_pass
  • patroni_postgres_pass
  • vip

If dcs_server_ips is set, then it will be used. If not set, then the IPs of hosts of the dcs_servers inventory group will be used.

Example: ansible-playbook -i inventory -e dcs=consul patroni.yml

Running multiple instances of Patroni/PostgreSQL

By default, Patroni uses 8008 as the API REST port. The automatic Patroni configuration generation will increment this for every additional cluster, so running

ansible-playbook -i inventory -e postgresql_cluster_name=test2 --tags=config pgsql-server.yml

will result in a second cluster, 11/test2 using PostgreSQL port 5433 and Patroni API port 8009.

Managing a virtual IP address with vip-manager

The vip-manager package allows to expose a virtual IP address (VIP) for the leader node by monitoring the leader key in the DCS and setting or removing the configured VIP for the local node depending on leader status.

When the vip variable is set to an IP address, an appropriate configuration file for vip-manager will be written and the cluster-specific vip-manager service will be started.

Rewinding/Recloning outdated former primaries

If a failover has occured and the old leader has additional transactions that do not allow for a clean change of timeline, pg_rewind can be used in order to rewind the old leader so that it can reinitiate streaming replication to the new leader. For this to happen, the variable patroni_postgres_pass needs to be set in vars.yml. If this is not the case, a full reclone will be done.

In both cases, the former primary's data directory will be gone, so if you prefer to have the primary stay down for manual inspection instead, you should comment out (or set to false) the parameters use_pg_rewind, remove_data_directory_on_rewind_failure and remove_data_directory_on_diverged_timelines.