Standby cluster

Source: https://patroni.readthedocs.io/en/latest/standby_cluster.html

Patroni supports cascading replication to a remote datacenter or region through the standby cluster feature. A standby cluster has:

• A standby leader, which behaves like a regular cluster leader but replicates from a remote node.
• Cascade replicas, which replicate from the standby leader.

The standby leader holds and updates a leader lock in the DCS. If the lock expires, the cascade replicas elect another leader from the standbys.

The standby and primary clusters share only replication information. If they use the same DCS, they must use different DCS scopes. The primary cluster does not show the standby cluster in patronictl_list or patronictl_topology output.

Use create_replica_methods in the standby_cluster section to specify replica creation and WAL recovery methods in standby mode. After the cluster detaches and operates normally, replica creation uses create_replica_methods in the postgresql section. Both standby and normal create_replica_methods refer to keys in the postgresql section.

To configure a standby cluster, add a standby_cluster section to the Patroni configuration:

bootstrap:
    dcs:
        standby_cluster:
            host: 1.2.3.4
            port: 5432
            primary_slot_name: patroni
            create_replica_methods:
            - basebackup

These options apply only during cluster bootstrap. After bootstrap, change them through the DCS.

Patroni expects postgresql.conf or postgresql.conf.backup in PGDATA on the remote primary. It will not start if this file is missing after a basebackup. If the primary stores postgresql.conf elsewhere, copy it to PGDATA.

If the standby cluster uses replication slots, create the corresponding slot on the primary cluster. The standby cluster does not create it automatically. You can use permanent replication slots on the primary cluster to maintain it. Use the name from primary_slot_name, or its default if you did not set primary_slot_name.

If the remote site has no single primary endpoint, list all source cluster hosts in standby_cluster.host. For multiple comma-separated hosts, Patroni will:

• Add target_session_attrs=read-write to primary_conninfo on the standby leader.
• Use target_session_attrs=read-write when it checks whether pg_rewind is necessary or runs pg_rewind on any standby cluster node.
• Require data page checksums and/or wal_log_hints set to on for successful pg_rewind operation. Enable checksums with the --data-checksums option for initdb. Otherwise, pg_rewind cannot operate correctly.

To replicate from another standby cluster or a standby member of the primary cluster, specify one host in standby_cluster.host. However, pg_rewind will fail on the standby cluster in this configuration.