Re: Question regarding failover behavior

[Koen De Groote <kdg.dev@xxxxxxxxx>]

Updating to say I tested and ran into the issue where the timeline switched from 0C to 0D.

Trying to bring the old primary back up as standby fails. It recovers upto a point and then gets stuck asked for a next file that doesn't exist.

Regardless of taking the existing data directory or restoring a new basebackup.

Taking a new basebackup, with the new primary, and restoring that, works. Standby gets in sync and everything is as expected.

Kind regards,

Koen De Groote

On Wed, Aug 24, 2022 at 1:09 AM Koen De Groote <kdg.dev@xxxxxxxxx> wrote:

Hello all,

I have a system that was originally set up on 9.3, a few years ago moved to 11.2

A shared mount is used to host the basebackup and wal archives.

The failover procedure was basically manual and as follow:

1/ Take out the primary, remove the IP from the primary machine/VM

2/ Create the trigger_file on the standby, add the IP to the new primary machine/VM

3/ Create a basebackup from the new primary

4/ Redeploy the new standby, which will unpack this basebackup from step 3

5/ The new standby will have a restore_command to get synced back up.

I see there's more recent tools like pg_rewind these days. Haven't had time to check that out and probably won't for a while.

My question is: is the basebackup actually necessary? Can I not simply modify the configuration old the old primary, set it up like a standby, and then start the process/container, and have it pick up the necessary wal_archives with the restore_command?

Or is it more complicated than that? Something like a timeline jump or something else that makes it impossible to simply set the old primary to standby and start it again?

Kind regards,

Koen De Groote