On Tue, Nov 21, 2017 at 03:40:27PM -0800, Andres Freund wrote:
> Hi,
>
> On 2017-11-21 17:09:26 -0600, Justin Pryzby wrote:
> > I'm sorry to report this previously reported problem is happening again,
> > starting shortly after pg_upgrading a customer to PG10.1 from 9.6.5.
> >
> > As $subject: backends are stuck in startup for minutes at a time. I didn't
> > strace this time, but I believe last time I saw one was waiting in a futex.
>
> > - I'm planning do disable sync seqscan for the night (although as I understand
> > that may not matter (?))
>
> I doubt it matters here.
The system still got wedged/stuck after a few hours even with:
synchronize_seqscans=off; and use of pg_buffercache disabled (but extension not
actually dropped), and our maintenance (vacuum and reindex) scripts disabled
(these normally only run at night, so at least REINDEX script wasn't involved
with several earlier incidents of this issue; but, I was still running the
vacuum ANALYZE script post-upgrade).
Justin