Re: Bug: Unreferenced temp tables disables vacuum to update xid

Alvaro Herrera wrote:
> Tom Lane wrote:
>> Andrew - Supernews <andrew+nonews@supernews.com> writes:
>>> On 2008-01-07, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>>>> The real question that Josh's report brings up to me is why the heck was
>>>> there an orphaned temp table?  Especially if it was only a toast table
>>>> and not the linked "regular" temp table?  Something happened there that
>>>> should not have.
>>> The regular table was there too, but the regular table's relfrozenxid
>>> was apparently recent, only the toast table's was old:
>> Hmm, that's even more odd, since AFAICS vacuum will always vacuum a
>> toast table immediately after vacuuming the parent.  I wonder whether
>> we have a bug somewhere that allows a toast table's relfrozenxid to
>> get initially set to something substantially different from the
>> parent's.
> 
> Hmm ... that would be strange.  Off-the-cuff idea: we introduced code to
> advance relfrozenxid in CLUSTER, TRUNCATE and table-rewriting forms of
> ALTER TABLE.  Perhaps the problem is that we're neglecting to update it
> for the toast table there.  AFAIR I analyzed the cases and they were all
> handled, but perhaps I forgot something.

Just to throw another variable into the mix. This machine was a PITR 
slave that was pushed into production about two weeks ago.

Joshua D. Drake