Re: pg_autovacuum misinterpreting reltuples?

From: "Matthew T(dot) O'Connor" <matthew(at)zeut(dot)net>
To: Jeff Boes <jboes(at)nexcerpt(dot)com>
Cc: PostgreSQL-development <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: pg_autovacuum misinterpreting reltuples?
Date: 2004-05-05 18:25:36
Message-ID: 409931A0.4060608@zeut.net
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

Jeff Boes wrote:
> We noticed that one of our high-volume insert tables was being vacuumed
> every time pg_autovacuum woke up. (I"m running it with the default
> threshold values, and a 900-second sleep cycle.) The table has a few
> million rows in it. With "debug = 2" on, here's what the pg_autovacuum
> log reports for this table:
>
> [2004-04-30 09:32:11 AM] table name: nexcerpt."public"."links"
> [2004-04-30 09:32:11 AM] relid: 1014188; relisshared: 0
> [2004-04-30 09:32:11 AM] reltuples: 3; relpages: 118888
>
> Note that "reltuples" is being reported as "3", which seems a bit low. Aha!
>
> # select reltuples from pg_class where relname = 'links';
> reltuples
> -------------
> 3.32127e+06
> (1 row)
>
> Is it possible that pg_autovacuum sees this value as "3"? Or is it just
> mis-formatting the value in the log?

Yes. This is a known bug inside of pg_autovacuum, it is fixed inside of
CVS, and will be released with 7.4.3 (whenever that happens, hopefully
soon since I'm getting lots of reports for this bug). Please downlooad
pg_autovacuum from CVS and compile by hand for the time being.

Matthew O'Connor

In response to

Browse pgsql-hackers by date

  From Date Subject
Next Message Andrew Dunstan 2004-05-05 18:36:11 Re: PostgreSQL pre-fork speedup
Previous Message Richard Huxton 2004-05-05 18:22:46 Re: PostgreSQL pre-fork speedup