Re: PostgreSQL 8.1.0 catalog corruption

Bob Ippolito <bob(at)redivi(dot)com> writes:
> On Nov 21, 2005, at 5:50 PM, Tom Lane wrote:
>> Hm, do the drop/add constraint functions get executed even when
>> clone_table decides not to make a new table? If so, that would
>> probably explain the pattern I'm seeing in the dump of many updates of the
>> pg_class row.

> Yes, they do. The constraints are there for constraint exclusion.

I dug through the dump more closely and determined that the newest
remaining version of the ping_1132387200 row claims to have been
outdated by transaction 000d585f. However, its ctid points to an item
slot that seems to have been reused by a much later transaction
(000fac5c). So I'm afraid all the evidence is gone about what really
happened :-(. If we had caught the problem earlier maybe we could have
learned more. If you see it happen again, could you get dumps of
pg_class (in both dump formats) as quickly as possible?

regards, tom lane