Re: Add a new BGWORKER_BYPASS_ROLELOGINCHECK flag

On Fri, Oct 06, 2023 at 09:09:10AM +0900, Michael Paquier wrote:
> I am not completely sure what you mean here. We've never supported
> load_session_libraries for background workers, and I'm not quite sure
> that there is a case for it. FWIW, my idea around that would be to
> have two separate sets of flags: one set for the bgworkers and one set
> for PostgresInit() as an effect of load_session_libraries which looks
> like a fuzzy concept for bgworkers.

I have applied v1 a few hours ago as of 991bb0f9653c. Then, the
buildfarm has quickly complained that a bgworkers able to start with
BYPASS_ALLOWCONN while the database has its access restricted could
spawn workers that themselves try to connect to the database
restricted, causing the test to fail. I have applied fd4d93d269c0 as
a quick way to avoid the spawn of workers in this case, and the
buildfarm has turned back to green.

Now, there's been a second type failure on serinus even after all
that:
https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=serinus&dt=2023-10-06%2001%3A04%3A05

The step running a `make check` on worker_spi in the run has worked:
https://buildfarm.postgresql.org/cgi-bin/show_stage_log.pl?nm=serinus&dt=2023-10-06%2001%3A04%3A05&stg=module-worker_spi-check

But the follow-up step doing an installcheck with worker_spi has not.
And this looks like a crash:
https://buildfarm.postgresql.org/cgi-bin/show_stage_log.pl?nm=serinus&dt=2023-10-06%2001%3A04%3A05&stg=testmodules-install-check-C

The logs reported by this step are not really helpful, as they contain
only information about the sql/ tests in the modules. It is the first
time that this stuff is tested, so this could be a race condition
that's been around for some time but we've never seen it until now, or
it could be an issue in the test I fail to see.

Andres, are there logs for this TAP test on serinus? Or perhaps there
is a core file that could be looked at? The other animals are not
showing anything for the moment.
--
Michael