Document that wal_keep_segments may need to be increased in larger non-Geo environments
Everyone can contribute. Help move this issue forward while earning points, leveling up and collecting rewards.
Problem to solve
If Postgres falls back too far WAL segments may be recycled before replication causing data loss. In the past we have seen this occur with Geo primaries and have increased postgres['wal_keep_segments'] to 50 for those nodes.
Recently, this scenario occurred on an 8,500 user instance (non-Geo) causing a significant outage for the customer.
Proposal
Document that postgres['wal_keep_segments'] should be increased from the default value of 10 on larger instances. Each segment is 16 MB, so there is a cost in disk space
Using 50 as a default in environments using Postgres replication seems reasonable, but perhaps the DB team will have a better recommendation.
/cc @grantyoung