> 3 июля 2021 г., в 23:44, Jeff Davis <pgsql@j-davis.com> написал(а):
>
> On Sat, 2021-07-03 at 14:06 +0500, Andrey Borodin wrote:
>>> But until you've disabled sync rep, the primary will essentially be
>>> down for writes whether using this new feature or not. Even if you
>>> can
>>> terminate some backends to try to free space, the application will
>>> just
>>> make new connections that will get stuck the same way.
>>
>> Surely I'm talking about terminating postmaster, not individual
>> backends. But postmaster will need to terminate each running query.
>> We surely need to have a way to stop whole instance without making
>> any single query. And I do not like kill -9 for this purpose.
>
> kill -6 would suffice.
SIGABRT is expected to generate a core dump, isn't it? Node failover is somewhat expected state in HA system.
>
> I see the point that you don't want this to interfere with an
> administrative shutdown. But it seems like most shutdowns will need to
> escalate to SIGABRT for cases where things are going badly wrong (low
> memory, etc.) anyway. I don't see a better solution here.
In my experience SIGTERM coped fine so far.
> I don't fully understand why you'd be concerned about cancellation but
> not concerned about similar problems with termination, but if you think
> two GUCs are important I can do that.
I think 2 GUCs is a better solution than 1 GUC disabling both cancelation and termination.
It would be great if some other HA tool developers would chime in.
Thanks!
Best regards, Andrey Borodin.