Horton

Re: [HACKERS] VACUUM and ANALYZE disagreeing on what reltuples means

Tom Lane <tgl@sss.pgh.pa.us>

View thread View on postgr.es
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Tomas Vondra <tomas.vondra@2ndquadrant.com>
Cc: David Steele <david@pgmasters.net>, Stephen Frost <sfrost@snowman.net>, Michael Paquier <michael.paquier@gmail.com>, Haribabu Kommi <kommi.haribabu@gmail.com>, "pgsql-hackers@postgresql.org" <pgsql-hackers@postgresql.org>
Date: 2018-03-22T17:28:52Z
Lists: pgsql-hackers 
Tomas Vondra <tomas.vondra@2ndquadrant.com> writes:
> 0001 fixes this by tracking the number of actually indexed rows in the
> build states, just like in the other index AMs.
> A VACUUM or ANALYZE will fix the estimate, of course, but for tables
> that are not changing very much it may take quite a while. So I think
> this is something we definitely need to back-patch.

Agreed, and done.  I noticed a second bug in contrib/bloom, too: it'd
forget to write out the last index page if that contained only one
tuple, because the new-page code path in bloomBuildCallback failed to
increment the "count" field after clearing it.

> The 0002 part is the main part, unifying the definition of reltuples on
> three main places:

On to this part ...

			regards, tom lane

Commits

  1. Sync up our various ways of estimating pg_class.reltuples.

  2. Fix tuple counting in SP-GiST index build.

  3. Fix errors in contrib/bloom index build.

  4. When updating reltuples after ANALYZE, just extrapolate from our sample.

  5. Avoid holding AutovacuumScheduleLock while rechecking table statistics.