Re: Long running backup preventing auto vacuum

Hi!

> On 30 Jun 2022, at 14:23, Nikhil Shetty <nikhil(dot)dba04(at)gmail(dot)com> wrote:
>
> Hi,
>
> PostgreSQL version - 11.7
>
> We are using wal-g for backup/restore. We had a duration of high WAL generation due to application usage due to which archive backup was delayed.
>
> Backup was started through cron schedule and it didn't complete because it waits for the last WAL to be archived. Now, I know that long running transactions will prevent vacuuming of dead tuples but in this case we were able to run manual vacuums successfully but auto-vacuum never starts.
>
> As soon as we terminated the long running backup session, the auto-vacuum started kicking in.
>
> Wanted to understand why a backup operation is blocking the auto-vacuum?
>
> Wal-g uses non-exclusive backups and waits for the last wal to be archived.

As far as I know, WAL-G should not prevent running autovacuums on tables. WAL-G can be run against replication standby, where vacuum is not possible at all, so there must be no conflicts...

The only probable problem I can imagine is that we open transaction during pg_stop_backup() [0] to set a statement_timeout. I think this is superflous and unneeded. I think this tx can be safely removed. But anyway, this tx does not have a xid (otherwise it could not run on Standby).

BTW which version of WAL-G do you use?

Best regards, Andrey Borodin.

[0] https://github.com/wal-g/wal-g/blob/8b9b3ca06efb44be4724208dc6f6d11836bd34d9/internal/databases/postgres/queryRunner.go#L243