Thread: Fwd: Bug#253295: postgresql: VACUUM FULL produce warnings

Fwd: Bug#253295: postgresql: VACUUM FULL produce warnings

From
Martin Pitt
Date:
Hi PostgreSQL developers!

Recently we got the following bug report about VACUUM FULL. Can
anybody comment on this?

If you have questions concerning the submitter's configuration, can
you please ask him directly? (ed@divo.ru)

Thanks in advance and have a nice day!

Martin

----- Forwarded message from Ed <ed@divo.ru> -----

Subject: Bug#253295: postgresql: VACUUM FULL produce warnings
Reply-To: Ed <ed@divo.ru>, 253295@bugs.debian.org
From: Ed <ed@divo.ru>
To: Debian Bug Tracking System <submit@bugs.debian.org>
Date: Tue, 08 Jun 2004 17:19:13 +0400
X-Spam-Status: No, hits=3D0.0 required=3D4.0 tests=3Dnone autolearn=3Dno ve=
rsion=3D2.61

Package: postgresql
Version: 7.4.2-3.backports.org.1
Severity: normal

i have small postgesql database with my own clients (C+libpq). database upd=
ate are rare (several tens per day), transactions common (one or two every =
second)


if i run "VACUUM FULL" in psql i get several messages like this:

WARNING:  index "pg_depend_depender_index" contains 3722 row versions, but =
table contains 3720 row versions

if i do REINDEX DATABASE before VACUUM i don't see any warnings.

it's normal behaviour? or i must look for problem source?

----- End forwarded message -----
----- Forwarded message from Oliver Elphick <olly@lfix.co.uk> -----

Subject: Bug#253295: postgresql: VACUUM FULL produce warnings
Reply-To: olly@lfix.co.uk, 253295@bugs.debian.org
From: Oliver Elphick <olly@lfix.co.uk>
To: Ed <ed@divo.ru>, 253295@bugs.debian.org
Date: Tue, 08 Jun 2004 15:03:03 +0100
X-Spam-Status: No, hits=3D0.0 required=3D4.0 tests=3Dnone autolearn=3Dno ve=
rsion=3D2.61

On Tue, 2004-06-08 at 14:19, Ed wrote:
> Package: postgresql
> Version: 7.4.2-3.backports.org.1
> Severity: normal
>=20
> i have small postgesql database with my own clients (C+libpq). database u=
pdate are rare (several tens per day), transactions common (one or two ever=
y second)
>=20
>=20
> if i run "VACUUM FULL" in psql i get several messages like this:
>=20
> WARNING:  index "pg_depend_depender_index" contains 3722 row versions, bu=
t table contains 3720 row versions
>=20
> if i do REINDEX DATABASE before VACUUM i don't see any warnings.
>=20
> it's normal behaviour? or i must look for problem source?

I wouldn't consider it normal.  Reindexing will remove the problem, of
course.  The question is how the problem arose in the first place (that
is, more row versions in the index than in the table itself).

Do all transactions terminate before you run VACUUM FULL?

Is it the case that this problem keeps recurring?

--=20
Oliver Elphick                                          olly@lfix.co.uk
Isle of Wight                              http://www.lfix.co.uk/oliver
GPG: 1024D/A54310EA  92C8 39E7 280E 3631 3F0E  1EC0 5664 7A2F A543 10EA
                 =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
     "Therefore being justified by faith, we have peace with
      God through our Lord Jesus Christ."       Romans 5:1=20



----- End forwarded message -----
----- Forwarded message from ed@divo.ru -----

Subject: Bug#253295: Re[2]: Bug#253295: postgresql: VACUUM FULL produce
    warnings
Reply-To: ed@divo.ru, 253295@bugs.debian.org
From: ed@divo.ru
Date: Wed, 9 Jun 2004 10:10:25 +0400
To: Oliver Elphick <olly@lfix.co.uk>
Cc: 253295@bugs.debian.org
X-Spam-Status: No, hits=3D0.2 required=3D4.0 tests=3DNO_REAL_NAME,
    SUBJ_BRACKET_BALANCED,SUBJ_BRACKET_OFF,SUBJ_BRACKET_ON autolearn=3Dno=20
    version=3D2.61

OE> Do all transactions terminate before you run VACUUM FULL?

No. but i think it's not problem source:
 - afaik VACUUM FULL must lock tables and can't start before
 write-transactions done
 - if i see this messages once - i get its every time VACUUM FULL
 running, until i execute REINDEX DATABASE

OE> Is it the case that this problem keeps recurring?

i can't say when problem appear, but i see it every day on working
database.

what may be problem source? how can i find it?


 - i found now AUTOVACUUM=3Dyes in postmaster.conf, but
 postgesql-contrib don't installed
 - postgresql work over xfs filesystem on linux 2.6



----- End forwarded message -----

--=20
Martin Pitt                 Debian GNU/Linux Developer
martin@piware.de                      mpitt@debian.org
http://www.piware.de             http://www.debian.org

Re: Fwd: Bug#253295: postgresql: VACUUM FULL produce warnings

From
Tom Lane
Date:
Martin Pitt <martin@piware.de> writes:
> Recently we got the following bug report about VACUUM FULL. Can
> anybody comment on this?

This was reported to the PG lists already.  We asked for a test case
but didn't get one ...

            regards, tom lane

Re: Fwd: Bug#253295: postgresql: VACUUM FULL produce warnings

From
Martin Pitt
Date:
Hi Tom!

On 2004-06-23 11:09 -0400, Tom Lane wrote:
> Martin Pitt <martin@piware.de> writes:
> > Recently we got the following bug report about VACUUM FULL. Can
> > anybody comment on this?
>=20
> This was reported to the PG lists already.  We asked for a test case
> but didn't get one ...

Then maybe Ed (ed@divo.ru) can help you since it appears that he can
reproduce the problem. Since I can't, maybe you can talk with him
directly?

Thanks and have a nice day!

Martin

--=20
Martin Pitt                 Debian GNU/Linux Developer
martin@piware.de                      mpitt@debian.org
http://www.piware.de             http://www.debian.org