Re: Parallel query hangs after a smart shutdown is issued

Thomas Munro <thomas(dot)munro(at)gmail(dot)com> writes:
> On Fri, Aug 14, 2020 at 4:45 AM Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>> After some more rethinking and testing, here's a v5 that feels
>> fairly final to me. I realized that the logic in canAcceptConnections
>> was kind of backwards: it's better to check the main pmState restrictions
>> first and then the smart-shutdown restrictions afterwards.

> LGTM. I tested this a bit today and it did what I expected for
> parallel queries and vacuum, on primary and standby.

Thanks for reviewing! I'll do the back-patching and push this today.

>> I'm assuming we want to back-patch this as far as 9.6, where parallel
>> query began to be a thing.

> Yeah. I mean, it's more radical than what I thought we'd be doing for
> this, but you could get into real trouble by running in smart shutdown
> mode without the autovac infrastructure alive.

Right. 99.99% of the time, that early shutdown doesn't really cause
any problems, which is how we've gotten away with it this long. But if
someone did leave session(s) running for a long time after issuing the
SIGTERM, the results could be bad --- and there's no obvious benefit
to the early shutdowns anyway.

regards, tom lane