Re: shared memory based stat collector (was: Sharing record typmods between backends)

Robert Haas wrote:

> Actually, now that you mention it, I think it *is* broken already, or
> more to the point, if you configure that value, the autovacuum
> launcher hangs up, because of the AutovacuumWorkItem stuff that Alvaro
> added. When I just tested it, the AV launcher somehow ended up
> waiting for AutovacuumLock and I had to SIGQUIT the server to shut it
> down. That's actually not really entirely the fault of
> dynamic_shared_memory_type = none, though, because the code in
> autovacuum.c does this:
>
> AutoVacuumDSA = dsa_create(AutovacuumLock->tranche);
> /* make sure it doesn't go away even if we do */
> dsa_pin(AutoVacuumDSA);
> dsa_pin_mapping(AutoVacuumDSA);
>
> Now, that's actually really broken because if dsa_create() throws an
> error of any kind, you're going to have already assigned the value to
> AutoVacuumDSA, but you will not have pinned the DSA or the DSA
> mapping. There's evidently some additional bug here because I'd sorta
> expect this code to just go into an infinite loop in this case,
> failing over and over trying to reattach the segment, but evidently
> something even worse happening - perhaps the ERROR isn't releasing
> AutovacuumLock.

Looking into this now.

--
Álvaro Herrera https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services