On 2017-09-17 01:07:52 -0700, Andres Freund wrote:
> On 2017-09-16 13:27:05 -0700, Andres Freund wrote:
> > > This does not seem like a problem that justifies a system-wide change
> > > that's much more delicate than you thought.
> >
> > We need one more initialization call during crash-restart - that doesn't
> > seem particularly hard a fix.
>
> FWIW, attached is that simple fix. Not quite ready for commit - needs
> more comments, thoughts and a glass of wine less to commit.
>
> I'll try to come up with a tap test that tests crash restarts tomorrow -
> not sure if there's a decent way to trigger one on windows without
> writing a C function. Will play around with that tomorrow.
This took a bit longer than I anticipated. Attached. There's
surprisingly many issues with timing that make this not as easy as I
hoped. I think in later commits we should extend this to cover things
like the autovacuum / stats process / ... being killed - but that seems
like material for a separate commit.
One thing that I've noticed for a while, but that I was reminded of
again here. We very frequently allow psql to reconnect in case of crash,
just for postmaster to notice a child has gone and kill that session. I
don't recall that frequently happening, but these days it happens nearly
every time.
Regards,
Andres
--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers