Darren Reed <darrenr@fastmail.net> writes:
> Tom Lane wrote:
>> BTW, could you send me the contents of file 24579 in that database
>> (dump it with pg_filedump, same as before)? According to your dump
>> that's the reindexed version of pg_class_oid_index, and I'm curious
>> to see whether it contains index entries pointing at block 7 of
>> pg_class. If my theory is right then it will.
> Attached.
Sure enough, there are 26 index entries referencing block 7, which
matches the number of updates into that block that seemed to have
occurred according to pg_class. So those updates definitely happened,
and then *something* made block 7 go away.
What's the platform here --- what kernel, what version exactly,
what hardware and disk subsystem? I think your problem is somewhere
in there ...
regards, tom lane
