PAF, Pacemaker, Postgres 15 and replication slots about paf HOT 1 CLOSED

Hi @andrecowie ,

PAF doesn't pay attention to the slots, at all, by design. PAF is responsible to keep your service up and running as long as possible. Slots are risky for availability as if a standby fails for a long time, WALs piles up in pg_wal indefinitely, until filling up your disk.

So creating and managing slot imply to monitor them and potentially drop them if the standby fails for a long time. This makes the resource agent code more complex.

Theoretically, we could create/drop them during promote/demote action and forget about them since max_slot_wal_keep_size has been added in v13... But then you have to check for this parameter value.

But the real reason we don't want to deal with slots is that there's a better and expected architecture: WAL archiving. Step one for a reliable and high available cluster are PITR backups. As long as you have PITR backups, you have WAL archives. As long as you have WAL archives reachable from your standby, you just don't need slots.

Regards,

from paf.