At the other end of the spectrum, some replication software does statement-based replication. Rather than sending blocks over, the list of transactions made to each table are extracted from the database, typically with triggers. Then those statements can be saved to a queue, shipped to a number of slave nodes, and then executed there. This introduces some non-trivial overhead on the master server because the overhead of the triggers, queue management, and statement shipping is moderate.

However, the resulting copies are then completely independent of the master server and, unlike WAL shipping approaches, you can pick and choose exactly which tables do and don't get shipped to the standby. In addition, since high-level statements are being shipped, the servers don't even need to match perfectly. This form of replication is therefore useful for...