Bruce Momjian wrote:
> > Is the problem well-understood? Is there a place where I can read up on it? This
> > kind of instability is painful enough to get me thinking about trying to hack my
> > distribution...
>
> I believe it has to do with extra index tuples showing up in the index
> that are not in the heap. When the count's don't match, the problem is
> reported. I believe it only happens when the system crashes during an
> index update.
That is consistent with my crash experiences this evening.
> I think it is harmless. To fix it properly requires a
> very sophisticated write-ahead log that is scheduled for 7.1 in about
> six months.
This problem stops my psql dead in its tracks for related queries even across new
sessions. Requires a rebuild of indices before any queries work with the related
tables/functions, and since I don't know which one to rebuild (die, horsey, die), I
might as well rebuild them all. In production mode, that means stopping user access due
to the possibility of violating unique constraints enforced by unique indices. That
means downtime, which would makes moi persona non grata. But maybe my assumptions are
incorrect or I didn't understand what you mean by harmless?
Cheers,
Ed Loehr