Re: [HACKERS] Re: PD_ALL_VISIBLE flag was incorrectly set happend during repeatable vacuum

On Thu, Mar 03, 2011 at 09:04:04AM -0600, Merlin Moncure wrote:
> On Thu, Mar 3, 2011 at 2:16 AM, Heikki Linnakangas
> <heikki(dot)linnakangas(at)enterprisedb(dot)com> wrote:
> > On 03.03.2011 09:12, daveg wrote:
> >>
> >> Question: what would be the consequence of simply patching out the setting
> >> of this flag? Assuming that the incorrect PD_ALL_VISIBLE flag is the only
> >> problem (big assumption perhaps) then simply never setting it would at
> >> least
> >> avoid the possibility of returning wrong answers, presumably at some
> >> performance cost. We possibly could live with that until we get a handle
> >> on the real cause and fix.
> >
> > Yes. With that assumption.
> >
> > If you really want to do that, I would suggest the attached patch instead.
> > This just disables the optimization in seqscans to trust it, so an
> > incorrectly set flag won't affect correctness of query results, but the
> > flag is still set as usual and you still get the warnings so that we can
> > continue to debug the issue.
>
> This. The mis-set flag can is likely a bug/concurrency issue etc,
> but could also be a symptom of more sinister data corruption. I did
> various vacuum experiments all day yesterday on my windows workstation
> and was not able to produce any mis-flags. I trust iscsi more than
> nfs, but maybe there is a connection here that is hardware based. hm.
> do you think it would be helpful to know what is causing the
> all_visible flag to get flipped? If so, the attached patch shows
> which case is throwing it...

I'll apply your patch and try it. Probably can only do it for a few minutes
tomorrow evening though as the output is huge and we have only limited down
time availability.

-dg

--
David Gould daveg(at)sonic(dot)net 510 536 1443 510 282 0869
If simplicity worked, the world would be overrun with insects.