Horton

Re: Optimize join selectivity estimation by not reading MCV stats for unique join attributes

David Geier <geidav.pg@gmail.com>

View thread View on postgr.es
From: David Geier <geidav.pg@gmail.com>
To: Tomas Vondra <tomas.vondra@enterprisedb.com>, Tom Lane <tgl@sss.pgh.pa.us>, Richard Guo <guofenglinux@gmail.com>
Cc: PostgreSQL Developers <pgsql-hackers@lists.postgresql.org>
Date: 2022-11-18T15:59:02Z
Lists: pgsql-hackers

Attachments

On 11/18/22 14:00, Tomas Vondra wrote:
> Seems fine. I wonder if we could/could introduce a new constant for 0,
> similar to ATTSTATSSLOT_NUMBERS/ATTSTATSSLOT_VALUES, instead of using a
> magic constant. Say, ATTSTATSSLOT_NONE or ATTSTATSSLOT_CHECK.
Good idea. I called it ATTSTATSSLOT_EXISTS. New patch attached.
> I don't think you can write a test for this, because there is no change
> to behavior that can be observed by the user. If one side has no MCV,
> the only difference is whether we try to load the other MCV or not.

Yeah. I thought along the lines of checking the number of pages read 
when the pg_stats entry is not in syscache yet. But that seems awfully 
implementation specific. So no test provided.

-- 
David Geier
(ServiceNow)

Commits

  1. Don't read MCV stats needlessly in eqjoinsel().