Re: Add pg_stat_autovacuum_priority

> > 1/ + while ((classTup = heap_getnext(relScan, ForwardScanDirection)) != NULL)
> >
> > Missing check_for_interrupts call while scanning the pg_class system catalog.
>
> From a glance I don't see one in the scanning code in do_autovacuum(),
> either. I'm not sure we need to be worried about this.

Yes, I will leave it out. Even for catalogs with thousands of tables, I don't
foresee this being an issue.

> > + values[8] = Float8GetDatum(scores.vac_ins);
> > + values[9] = Float8GetDatum(scores.anl);
> >
> > Nit: It's a matter of taste. How about using something like below
> > instead of hardcoded column numbers? I expect this view to grow in the
> > future, so it helps to keep things simple.
> >
> > values[i++] = Float8GetDatum(scores.anl);
> > Assert(i == NUM_AV_SCORE_COLS);
>
> I don't think either way is substantially better.

I agree. using numbers is more readable IMO.

> > + <row>
> > + <entry role="catalog_table_entry"><para role="column_definition">
> > + <structfield>vacuum_dead_score</structfield> <type>double precision</type>
> > + </para>
> > + <para>
> > + Score component based on the estimated number of dead tuples
> > + needing removal by vacuum.
> > + </para></entry>
> > + </row>
> >
> > I think we should make sure the column names align with the names given to
> > the new parameters [0] and the new "Autovacuum Prioritization" section in
> > the docs [1].

> I will look into this in the next rev.

The field names now match the GUC names (without the _weight). We might as well
also make this name change in the AutoVacuumScore struct.

I attached v4 which includes 4 patches in the set to address earlier comments:

0001:

This modifies relation_needs_vacanalyze() to always compute the score so
the monitoring view can report the score regardless if autovacuum is enabled
on the table. AutoVacuumScores is renamed to AutoVacuumPriority as it
now also tracks needs_analyze, needs_vacuum and is_wraparound which
are different from the dovacuum, doanalyze and wraparound output parameters
which are acted on by autovacuum. This is a clean separation of reporting
the scores/needs for autovacuum/analyze and the eligibility for autovacuum
which is based on the state of autovacuum being enabled and set on the table
level.

0002:

relation_needs_vacanalyze() takes a new elevel argument to control logging.
Callers pass DEBUG3 for autovacuum or 0 to suppress logging.

0003:

Alvaro's earlier suggestion to factor out recheck_relation_needs_vacanalyze()
and introduce compute_autovac_score()

0004:

The view implementation with field names that better match the GUCs as suggested
by Nathan in an earlier comment.

--
Sami