On Sun, Aug 14, 2011 at 12:16:39AM -0400, Robert Haas wrote:
> On Fri, Aug 12, 2011 at 7:19 PM, daveg <daveg@sonic.net> wrote:
> > This seems to be bug month for my client. Now there are seeing periods
> > where all new connections fail immediately with the error:
> >
> > FATAL: lock AccessShareLock on object 0/1260/0 is already held
> >
> > This happens on postgresql 8.4.7 on a large (512GB, 32 core) system that has
> > been up for months. It started happening sporadicly a few days ago. It will
> > do this for a period of several minutes to an hour and then go back to
> > normal for hours or days.
> >
> > One complete failing session out of several hundred around that time:
> > -----------------
> > 2011-08-09 00:01:04.446 8823 [unknown] [unknown] LOG: connection received: host=op05.xxx port=34067
> > 2011-08-09 00:01:04.446 8823 c77 apps LOG: connection authorized: user=apps database=c77
> > 2011-08-09 00:01:04.449 8823 c77 apps FATAL: lock AccessShareLock on object 0/1260/0 is already held
> > ------------------
> >
> > What can I do to help track this down?
>
> I've seen that error (though not that exact fact pattern) caused by
> bad RAM. It's unclear to me what else could cause it.
I'll look into that. I think it is only happening on one host, so that might
make sense. On the other hand, these are pretty fancy hosts all ECC and that
so I'd hope they would have squeaked about bad ram.
> In terms of debugging, it seems like it might be sensible to start by
> injecting some debugging code that dumps out the contents of the LOCK
> and LOCALLOCK structures at the point the error occurs.
Hmm, we will update to 9.0 next week on these hosts, so I'll try to hold off
on this part at least until then.
-dg
--
David Gould daveg@sonic.net 510 536 1443 510 282 0869
If simplicity worked, the world would be overrun with insects.