With pgpool installed, we have an abstraction layer above PostgreSQL, which hides the active node from the client. This allows us to change the primary node so that we can perform maintenance, and yet we never have to actually stop the database.

This kind of design will work best when pgpool is not installed on one of the PostgreSQL servers, but it has its own dedicated hardware or virtual machine. This allows us full control over each PostgreSQL server, including the ability to reboot for kernel upgrades, without potentially disrupting pgpool.

Let's discuss the elements involved in switching the primary server with a replica so that we can have high availability in addition to regular maintenance.