Thread: A streaming replica catching up
Dear Colleagues, After a streaming standby has been down for a time, I know of two options for it to catch up with the master: 1. The replica can fetch all the necessary WAL records via the replication protocol from the master (unless max_slot_wal_keep_size is not the default -1, all the required WAL archives should be available from the master). 2. The replica can be provided with a restore_command and will fetch all WAL files from the WAL archive before starting streaming WAL from the master. Which is the preferred way? Of course if max_slot_wal_keep_size is limited, the second options seems the only safe one. OTOH, if the master does not write to a WAL archive, only the first option is left to us. -- Victor Sudakov VAS4-RIPE http://vas.tomsk.ru/ 2:5005/49@fidonet
On Thu, 2021-04-22 at 10:16 +0700, Victor Sudakov wrote: > After a streaming standby has been down for a time, I know of two > options for it to catch up with the master: > > 1. The replica can fetch all the necessary WAL records via the > replication protocol from the master (unless max_slot_wal_keep_size is > not the default -1, all the required WAL archives should be available from > the master). > > 2. The replica can be provided with a restore_command and will fetch > all WAL files from the WAL archive before starting streaming WAL from > the master. > > Which is the preferred way? > > Of course if max_slot_wal_keep_size is limited, the second options seems > the only safe one. > > OTOH, if the master does not write to a WAL archive, only the first > option is left to us. Both methods are equally feasible. Which one you choose depends on your situation, just as you describe in the last two sentences. Yours, Laurenz Albe -- Cybertec | https://www.cybertec-postgresql.com
Laurenz Albe wrote: > On Thu, 2021-04-22 at 10:16 +0700, Victor Sudakov wrote: > > After a streaming standby has been down for a time, I know of two > > options for it to catch up with the master: > > > > 1. The replica can fetch all the necessary WAL records via the > > replication protocol from the master (unless max_slot_wal_keep_size is > > not the default -1, all the required WAL archives should be available from > > the master). > > > > 2. The replica can be provided with a restore_command and will fetch > > all WAL files from the WAL archive before starting streaming WAL from > > the master. > > > > Which is the preferred way? > > > > Of course if max_slot_wal_keep_size is limited, the second options seems > > the only safe one. > > > > OTOH, if the master does not write to a WAL archive, only the first > > option is left to us. > > Both methods are equally feasible. Which one you choose depends on > your situation, just as you describe in the last two sentences. I see. It is also interesting when there is Patroni involved, would it be a good idea to provide Patroni nodes with a restore_command? I'm afraid it can interfere with Patroni failover/switchover operations. -- Victor Sudakov VAS4-RIPE http://vas.tomsk.ru/ 2:5005/49@fidonet