Tom Lane wrote:
> Dylan Hansen <dhansen@pixpo.com> writes:
>
>> I have been spending some time looking into how auto-vacuum is
>> performing on one of our servers. After putting the PostgreSQL logs
>> in debug I noticed that the threshold for ANALYZE was never being hit
>> for a particular table because the calculated value becomes
>> increasingly negative.
>>
>
> Hmm, it shouldn't ever be negative at all, I would think. The
> calculation in question is
>
> anltuples = tabentry->n_live_tuples + tabentry->n_dead_tuples -
> tabentry->last_anl_tuples;
>
> Apparently somehow last_anl_tuples has managed to get to be bigger than
> n_live_tuples, which maybe could happen after a delete. Should we be
> clamping last_anl_tuples to not exceed n_live_tuples somewhere?
> Alvaro and Matthew, what do you think?
I think I had something in the contrib version that checked this. I
always assumed it would be caused by a stats reset which was more common
in earlier PGSQL releases since stats_reset_on_startup (or whatever the
correct spelling of that is) was enabled by default.