gem-hunter
Language: Dockerfile
PostDock - Postgres & Docker - Postgres streaming replication cluster for any docker environment
PostgreSQL cluster with High Availability and Self Healing features for any cloud and docker environment (Amazon, Google Cloud, Kubernetes, Docker Compose, Docker Swarm, Apache Mesos)
This project includes:
postgresql cluster and backup system
Taking into account that PostDock project itself has versioning schema, all docker images produced by the repository have schema - postdock/<component>:<postdock_version>-<component><component_version>-<sub_component><sub_component_version>[...], where:
<postdock_version> - semantic version without bug-fix component (can be 1.1, 1.2, ...)<component>, <component_version> - depends on component:
postgres,postgres-extended - major and minor version without dot in between(can be 95, 96, 10, 11, ...)pgpool - major and minor version of component without dot in between(can be 33, 36, 37, ...)barman - major version only (can be 23, 24, ...)<sub_component>, <sub_component_version> - depends on component:
postgres - repmgr can be 32, 40, ...barman - postgres can be 96, 10, 11, ...pgpool - postgres can be 96, 10, 11, ...Aliases are available (not recommended to use for production):
postdock/<component>:latest-<component><component_version>[-<sub_component><sub_component_version>[...]] - refers to the latest release of the postdock, certain version of the component, certain version of the sub-components(e.g. postdock/postgres:latest-postgres101-repmgr32,postdock/postgres:latest-barman23-postgres101)postdock/<component>:latest - refers to the latest release of the postdock and the latest versions of all the components and sub-components (e.g. postdock/postgres:latest)postdock/<component>:edge - refers to build of postdock from master with the latest version the component, and all sub-components (e.g. postdock/postgres:edge)All available tags (versions and combinations of it) are listed in respective docker-hub repositories:
To start cluster run it as normal docker-compose application docker-compose -f ./docker-compose/latest.yml up -d pgmaster pgslave1 pgslave2 pgslave3 pgslave4 pgpool backup
Schema of the example cluster:
pgmaster (primary node1) --|
|- pgslave1 (node2) --|
| |- pgslave2 (node3) --|----pgpool (master_slave_mode stream)
|- pgslave3 (node4) --|
|- pgslave4 (node5) --|
Each postgres node (pgmaster, pgslaveX) is managed by repmgr/repmgrd. It allows to use automatic failover and check cluster status.
Please check comments for each ENV variable in ./docker-compose/latest.yml file to understand parameter for each cluster node
You can install PostDock with Helm package manager check the README.md of the package for more information
To make it easier repository contains services' objects under k8s dir. Setup PostgreSQL cluster following the steps in the example. It also has information how to check cluster state
You can configure any node of the cluster(postgres.conf) or pgpool(pgpool.conf) with ENV variable CONFIGS (format: variable1:value1[,variable2:value2[,...]], you can redefine delimiter and assignment symbols by using variables CONFIGS_DELIMITER_SYMBOL, CONFIGS_ASSIGNMENT_SYMBOL). Also see the Dockerfiles and docker-compose/latest.yml files in the root of the repository to understand all available and used configurations!
For the rest - you better follow the advise and look into the src/Postgres-latest.Dockerfile file - it full of comments :)
The most important part to configure in Pgpool (apart of general CONFIGS) is backends and users which could access these backends. You can configure backends with ENV variable. You can find good example of setting up pgpool in docker-compose/latest.yml file:
DB_USERS: monkey_user:monkey_pass # in format user:password[,user:password[...]]
BACKENDS: "0:pgmaster:5432:1:/var/lib/postgresql/data:ALLOW_TO_FAILOVER,1:pgslave1::::,3:pgslave3::::,2:pgslave2::::" #,4:pgslaveDOES_NOT_EXIST::::
# in format num:host:port:weight:data_directory:flag[,...]
# defaults:
# port: 5432
# weight: 1
# data_directory: /var/lib/postgresql/data
# flag: ALLOW_TO_FAILOVER
REQUIRE_MIN_BACKENDS: 3 # minimal number of backends to start pgpool (some might be unreachable)
The most important part for barman is to setup access variables. Example can be found in docker-compose/latest.yml file:
REPLICATION_USER: replication_user # default is replication_user
REPLICATION_PASSWORD: replication_pass # default is replication_pass
REPLICATION_HOST: pgmaster
POSTGRES_PASSWORD: monkey_pass
POSTGRES_USER: monkey_user
POSTGRES_DB: monkey_db
See the Dockerfiles and docker-compose/latest.yml files in the root of the repository to understand all available and used configurations!
'Adaptive mode' means that node will be able to decide if instead of acting as a master on it's start or switch to standby role.
That possible if you pass PARTNER_NODES (comma separated list of nodes in the cluster on the same level).
So every time container starts it will check if it was master before and if there is no new master around (from the list PARTNER_NODES),
otherwise it will start as a new standby node with upstream = new master in the cluster.
Keep in mind: this feature does not work for cascade replication and you should not pass PARTNER_NODES to nodes on second level of the cluster.
Instead of it just make sure that all nodes on the first level are running, so after restart any node from second level will be able to follow initial upstream from the first level.
That also can mean - replication from second level potentially can connect to root master... Well not a big deal if you've decided to go with adaptive mode.
But nevertheless you are able to play with NODE_PRIORITY environment variable and make sure entry point for second level of replication will never be elected as a new root master
If you have need to organize your cluster with some tricky logic or less problematic cross checks. You can enable SSH server on each node. Just set ENV variable SSH_ENABLE=1 (disabled by default) in all containers (including pgpool and barman). That will allow you to connect from any to any node by simple command under postgres user: gosu postgres ssh {NODE NETWORK NAME}
You also will have to set identical ssh keys to all containers. For that you need to mount files with your keys in paths /tmp/.ssh/keys/id_rsa, /tmp/.ssh/keys/id_rsa.pub.
If you want to disable the feature of Postgres>=9.4 - replication slots simply set ENV variable USE_REPLICATION_SLOTS=0 (enabled by default). So cluster will rely only on Postgres configuration wal_keep_segments (500 by default). You also should remember that default number for configuration max_replication_slots is 5. You can change it (as any other configuration) with ENV variable CONFIGS.
Component postgres-extended from the section Docker images tags convention should be used if you want to have postgres with extensions and libraries. Each directory here represents extension included in the image.
Barman is used to provide real-time backups and Point In Time Recovery (PITR).. This image requires connection information(host, port) and 2 sets of credentials, as you can see from the Dockerfile:
Barman acts as warm standby and stream WAL from source. Additionaly it periodicaly takes remote physical backups using pg_basebackup.
This allows to make PITR in reasonable time within window of specified size, because you only have to replay WAL from lastest base backup.
Barman automatically deletes old backups and WAL according to retetion policy.
Backup source is static — pgmaster node.
In case of master failover, backuping will continue from standby server
Whole backup procedure is performed remotely, but for recovery SSH access is required.
Before using in production read following documentation:
For Disaster Recovery process see RECOVERY.md
Barman exposes several metrics on :8080/metrics for more information see Barman docs
To make sure you cluster works as expected without 'split-brain' or other issues, you have to setup health-checks and stop container if any health-check returns non-zero result. That is really useful when you use Kubernetes which has livenessProbe (check how to use it in the example)
/usr/local/bin/cluster/healthcheck/is_major_master.sh - detect if node acts as a 'false'-master and there is another master - with more standbys/usr/local/bin/pgpool/has_enough_backends.sh [REQUIRED_NUM_OF_BACKENDS, default=$REQUIRE_MIN_BACKENDS] - check if there are enough backend behind pgpool/usr/local/bin/pgpool/has_write_node.sh - check if one of the backend can be used as a master with write accesspostgres node):
gosu postgres repmgr cluster show - tries to connect to all nodes on request ignore status of node in $(get_repmgr_schema).$REPMGR_NODES_TABLEgosu postgres psql $REPLICATION_DB -c "SELECT * FROM $(get_repmgr_schema).$REPMGR_NODES_TABLE" - just select data from tablespgpool status (on any pgpool node): PGPASSWORD=$CHECK_PASSWORD psql -U $CHECK_USER -h localhost template1 -c "show pool_nodes"pgpool container check if primary node exists: /usr/local/bin/pgpool/has_write_node.shAny command might be wrapped with docker-compose or kubectl - docker-compose exec {NODE} bash -c '{COMMAND}' or kubectl exec {POD_NAME} -- bash -c '{COMMAND}'
Check the document to understand different cases of failover, split-brain resistance and recovery
Check the doc to understand how to contribute
pgslave1) will cause dieing of whole branch