Bob Smith <bsmith@h-e.com> writes:
>> I'm rather surprised that restarting the postmaster didn't make the
>> error go away, but it's unlikely that anyone will care to investigate
>> --- unless you can reproduce the problem in 7.1.3 or later.
> So this isn't an error that would be caused by some sort of file
> problem?
AFAICT what you have there is just dangling buffer reference counts;
ie, something didn't release its reference to a disk page buffer in
shared memory. We used to have quite a number of bugs of that ilk,
though I've not heard of such lately. They're fairly harmless
normally.
It's not clear to me why a restart (which would naturally clear shared
memory) didn't fix the problem. Perhaps you had some configuration
on disk that tickled the original buffer-leak bug again each time you
restarted; but I dunno what. VACUUM itself was never prone to such
errors, so it's hard to see how a restart and immediate VACUUM would
show the problem. Were you doing other stuff in between? Have you
got any interesting stuff (like functional indexes, maybe) on this
table?
regards, tom lane
