"Matthew T. O'Connor" <matthew@zeut.net> writes:
> I believe this discrepancy has to do with the fact that ANALYZE can
> return some very bogus values for reltuples, where as vacuum always
> returns an accurate count. I'm not sure how to best handle this.
I think 8.0's ANALYZE will do a better estimation job ... at least,
Manfred Koizar rewrote the sampling algorithm in hopes of making it
more robust.
However, given that there are 9334 tuples in 82282 pages, I'd say
that autovacuum has already failed Steve rather badly :-(. There
shouldn't be more than a couple hundred pages given that number of
rows. Perhaps the FSM settings are too small?
regards, tom lane