On Sun, Dec 01, 2013 at 06:56:10PM +0100, Andres Freund wrote:
> On 2013-12-01 12:49:40 -0500, Noah Misch wrote:
> > This system had multiple problems, a missing pg_subtrans file and a missing
> > TOAST chunk for pg_attribute. I don't see a pg_clog problem connecting it to
> > the freeze bug at hand.
>
> Those all sound like possible problems caused by the bug, no?
pg_subtrans has a lifecycle unrelated to datfrozenxid. I am not aware of a
mechanism connecting that problem to the bug at hand.
The missing TOAST chunk (in pg_statistic, not pg_attribute as I wrote above)
could happen from the XID space wrapping with that TOAST table page marked
all-visible but not frozen. The bug reporter describes the start of that
error coinciding with a minor version upgrade, so that would be an odd
coincidence: 8.4.3 did not have the bug as we know it, so considerable time
would typically pass between the upgrade and that symptom appearing. Can't
rule it out, but this user report fits the known bug symptoms only loosely.
--
Noah Misch
EnterpriseDB http://www.enterprisedb.com