Introduction

When it comes to maintaining a single PostgreSQL cluster with a single source of WAL files, our job is an easy one. Even a small number of streaming replicas is easily managed manually with PostgreSQL-provided tools. However, what happens when we have a large constellation of PostgreSQL servers, such as this:

This diagram represents seven PostgreSQL servers for a single source of data. The Trading server sends its WAL data to a secondary system for safekeeping. One replica subscribes directly to the Trading database, while two others acquire their data through cascading replication. All clones are attached to the WAL archive in case their respective streams get disconnected.

Further complicating the situation, there's an off-site copy of the entire architecture for disaster recovery. Even though the recovery copy in the alternate data center is reduced in terms of capabilities, it still requires several servers for the client applications to run properly. Worse still, in...