Hi, that's for digging into this. Turns out I completely missed one of
your emails today morning.
On Wed, 10 Jul 2024 at 10:15, Michael Paquier <michael@paquier.xyz> wrote:
> And then the timestamp of the tests:
> [12:48:23.277](0.058s) not ok 2 - autovacuum worker signaled with
> pg_signal_autovacuum_worker granted
>
> We check for the contents of the logs 1ms before they are generated,
> hence failing the lookup check because the test is faster than the
> backend.
>
> Like what we are doing in 010_pg_basebackup.pl, we could do a
> poll_query_until() until the PID of the autovacuum worker is gone from
> pg_stat_activity before fetching the logs as ProcessInterrupts() stuff
> would be logged before the process exits, say:
> +# Wait for the autovacuum worker to exit before scanning the logs.
> +$node->poll_query_until('postgres',
> + "SELECT count(*) = 0 FROM pg_stat_activity "
> + . "WHERE pid = $av_pid AND backend_type = 'autovacuum worker';");
>
> That gives something like the attached. Does that look correct to
> you?
> --
> Michael
+1.
