Subject: reindexdb dying with SIGPIPE on 8.2.5
From: Torsten Luettgert <t.luettgert@pressestimmen.de>
To: pgsql-admin@postgresql.org
Content-Type: text/plain
Date: Wed, 06 Aug 2008 17:10:21 +0200
Message-Id: <1218035421.19154.24.camel@oberschatzi.cff>
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit

Hello all,

we're using PostgreSQL 8.2.5 with some 300 databases under CentOS 4.6.
To reduce on-disk size of our DBs, we're doing the following after
the nightly dumps:

reindexdb dbname
vacuumdb -f -z dbname

Now, this works fine about 50% of the time, but on some machines, the
reindexdb dies from a SIGPIPE signal. The logs look like this:

Aug  3 04:08:10 prospero postgres[27101]: [1-1] NOTICE:  table
"pg_class" was reindexed
Aug  3 04:08:11 prospero postgres[27101]: [2-1] NOTICE:  table
"sql_sizing" was reindexed
Aug  3 04:08:11 prospero postgres[27101]: [3-1] LOG:  could not send
data to client: Broken pipe
Aug  3 04:08:11 prospero postgres[27101]: [4-1] NOTICE:  table
"sql_sizing_profiles" was reindexed
Aug  3 04:08:11 prospero postgres[27101]: [5-1] NOTICE:  table
"sql_features" was reindexed

...

Aug  3 04:08:18 prospero postgres[27101]: [39-1] NOTICE:  table
"shoprules" was reindexed
Aug  3 04:08:18 prospero postgres[27101]: [40-1] LOG:  unexpected EOF on
client connection

So, it looks to me like the REINDEX command is completed, but the
reindexdb tool dies. This is run via a nightly cronjob; I never saw
those problems if I invoke reindexdb on the command line.

We did get a notice to increase max_fsm_pages at first though, so
we increased it with good margin, but the SIGPIPE problem persists.

What is going wrong here?

Thanks for your help,
Torsten