Makes sense to me. It seems a lot of features in Postgres optimize for managing a single instance but every setup I've worked on in the last 8 years has been HA

Makes sense to me. It seems a lot of features in Postgres optimize for managing a single instance but every setup I've worked on in the last 8 years has been HA. In that case you'd apply config changes to the replica and fail over versus restart the daemon. Checkpointing/crash recovery/shared memory also seem to be optimized for daemon restarts.