On Thu, Sep 13, 2012 at 1:45 PM, Jeff Davis <pgsql@j-davis.com> wrote:
> On Thu, 2012-09-13 at 12:39 -0400, Robert Haas wrote:
>> On Wed, Sep 12, 2012 at 7:19 PM, Jeff Davis <pgsql@j-davis.com> wrote:
>> > This bug seems particularly troublesome because the right fix would be
>> > to include the relpersistence in the WAL records that need it. But that
>> > can't be backported (right?).
>>
>> No, because if a WAL record was written at all, then by definition the
>> table is RELPERSISTENCE_PERMANENT. So there's probably a localized
>> fix.
>
> Oh, of course (I was worried there for some reason). I suppose we just
> need to set the relpersistence to permanent in CreateFakeRelcacheEntry,
> kind of like ReadBufferWithoutRelcache.
>
> And we should probably assert that both functions are only called during
> recovery (though perhaps redundant for CreateFakeRelcacheEntry, which is
> in xlogutils.c).
>
> Trivial patch attached.
Committed and back-patched to 9.1. Thanks for the report, diagnosis, and fix.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company