PostgreSQL 12 High Availability Cookbook - Third Edition

There's more…

There may be some debate as to the efficiency of creating replicas now, rather than waiting for the subscription to complete. This is actually a valid and rather astute observation. In cases of extremely large clusters, it may actually be a much better idea to delay the creation of replica nodes, backup nodes, and other dependent servers until the initial synchronization step is complete.

The biggest justification for this is the fact that PostgreSQL physical replication uses write-ahead logging (WAL) traffic to maintain replicas. WAL is generally much larger than the originating data since it contains all page-level operations, including index creation. For a very large databases (VLDB) installation several terabytes in size, ...