Fujitsu Enterprise Postgres allows you to build a cluster system using Patroni and etcd. Data is replicated from the master to multiple replica references. etcd also acts as a Distributed Consensus Store (DCS), managing cluster states such as which nodes are masters, which nodes are replicas, and the health of each node.

Regarding the etcd cluster configuration, a major choice is whether to place Patroni and etcd on different servers or not.

Placing etcd on the same server as Patroni can save on the number of servers, but for example, in the following cases, it is recommended to place Patroni and etcd on different servers to avoid trouble.

• Assuming operation to dynamically change the number of replica nodes

  Placing etcd on all nodes might be excessive. However, placing etcd on only some nodes makes the system configuration a bit more complex.

• The number of replica nodes is large

  This is to avoid resource contention between etcd and the database server.

• Share the same DCS (etcd) with other system.

Regarding the Patroni cluster configuration, various configurations are possible depending on the system requirements, so please decide on your own. However, please note that depending on the configuration method, a single server failure could cause Patroni to demote the primary server to a standby server and halt update operations.

To enable Patroni to manage the configuration of PostgreSQL's streaming replication, all PostgreSQL data directories and settings are described in patroni.yml. In particular, do not directly edit postgresql.conf.