Having a primary and a standby database is the most common PostgreSQL setup on live environments, but the standby has to be able to follow primary’s workload. This is particularly important when using synchronous commit, because when the standby cannot keep up, it affects the application that becomes slower and slower.
In this talk we review how we were able to find the root cause of the synchronous replication wait events by investigating Postgres configuration choices, the interaction with the Linux kernel, and good collaboration with Postgres developers. This process took us a year, and by sharing our experience, we expect that hopefully you don't have to wait that long!
The following slides have been made available for this session: