Re: new autovacuum criterion for visible pages

Stephen Frost wrote:

> Tracking the INSERTs as a reason to VACUUM is also very natural when you
> consider the need to update BRIN indexes. I am a bit worried that if we
> focus just on if the VM needs to be updated or not that we might miss
> out on cases where we need to VACUUM because the BRIN indexes are out of
> date.

Actually, I was thinking in a different approach for BRIN -- namely that
brininsert, on its fast path out where it sees an insert on a
non-summarized range, record on the index's metapage what the last
inserted page is, as well as what's the latest summarized pages. When a
range is complete, it saves the index OID in a DSA so that the next
autovacuum worker knows to run (the equivalent of) brinsummarize on the
index.(*)

That's efficient because this only occurs when it is known that a range
can be usefully summarized, instead of using a heuristic based on tuples
inserted, which may or may not correspond to a certain number of filled
pages.

(*) I think it's better if this only summarizes complete ranges, leaving
incomplete ones alone (which is what it does now). That way, if there's
a bulk insert going, we leave it to run at full speed the whole time.

--
Álvaro Herrera https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services