Re: Race condition in server-crash testing

Hi,

On 2022-04-04 00:50:27 -0400, Tom Lane wrote:
> My pet dinosaur gaur just failed [1] in
> src/test/recovery/t/022_crash_temp_files.pl, which does this:
>
> -----
> my $ret = PostgreSQL::Test::Utils::system_log('pg_ctl', 'kill', 'KILL', $pid);
> is($ret, 0, 'killed process with KILL');
>
> # Close psql session
> $killme->finish;
> $killme2->finish;
>
> # Wait till server restarts
> $node->poll_query_until('postgres', undef, '');
> -----
>
> It's hard to be totally sure, but I think what happened is that
> gaur hit the in-hindsight-obvious race condition in this code:
> we managed to execute a successful iteration of poll_query_until
> before the postmaster had noticed its dead child and commenced
> the restart. The test lines after these are not prepared to see
> failure-to-connect.
>
> It's not obvious to me how to remove this race condition.
> Thoughts?

Maybe we can use pump_until() with the psql that's not getting killed? With a
non-matching regex? That'd only return once the backend was killed by
postmaster, afaics?

Greetings,

Andres Freund