Re: ERROR: found multixact from before relminmxid

On Fri, Jun 8, 2018 at 1:08 PM Andres Freund <andres(at)anarazel(dot)de> wrote:

> On 2018-06-08 12:38:03 -0500, Jeremy Finzel wrote:
> > On Tue, Jun 5, 2018 at 8:42 PM, Alexandre Arruda <adaldeia(at)gmail(dot)com>
> wrote:
> >
> > > Em seg, 28 de mai de 2018 às 16:44, Andres Freund <andres(at)anarazel(dot)de>
> > > escreveu:
> > > >
> > > > Hi,
> > > >
> > > > I think I found the bug, and am about to post a fix for it belo
> > > >
> https://postgr.es/m/20180525203736.crkbg36muzxrjj5e@alap3.anarazel.de.
> > > >
> > > > Greetings,
> > > >
> > > > Andres Freund
> > >
> > > Hi Andres,
> > >
> > > In end of April we did a complete dump/reload in database to version
> 10.3.
> > > Today, the problem returns:
> > >
> > > production=# vacuum verbose co27t;
> > > INFO: vacuuming "public.co27t"
> > > ERROR: found multixact 81704071 from before relminmxid 107665371
> > > production=# vacuum full verbose co27t;
> > > INFO: vacuuming "public.co27t"
> > > ERROR: found multixact 105476076 from before relminmxid 107665371
> > > production=# cluster co27t;
> > > ERROR: found multixact 105476076 from before relminmxid 107665371
> > >
> > > But this time, regular vacuum versus full/cluster are different in
> > > multixact number.
> > > Your patch is applicable to this issue and is in 10.4 ?
> > >
> > > Best regards,
> > >
> > > Alexandre
> > >
> > >
> > We encountered this issue ourselves for the first time on a busy OLTP
> > system. It is at 9.6.8. We found that patching to 9.6.9 on a snapshot
> of
> > this system did not fix the problem, but I assume that is because the
> patch
> > in 9.6.9 only prevents the problem moving forward. Is that accurate?
>
> Which patch in 9.6.9 are you referring to? The patch I linked to above
> hasn't yet been merged, much less been released.

No I was referring to this from the documentation:

-

Avoid spuriously marking pages as all-visible (Dan Wood, Pavan Deolasee,
Álvaro Herrera)

This could happen if some tuples were locked (but not deleted). While
queries would still function correctly, vacuum would normally ignore such
pages, with the long-term effect that the tuples were never frozen. In
recent releases this would eventually result in errors such as "found
multixact nnnnn from before relminmxid nnnnn".

thanks,

Jeremy

>