Date: Mon, 14 May 2012 20:02:38 +0000
From: John Rouillard <rouilj@renesys.com>
To: pgsql-admin@postgresql.org
Subject: Monitoring for failed autovacuum
Message-ID: <20120514200238.GM20157@renesys.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
User-Agent: Mutt/1.5.20 (2009-06-14)

Hello all:

I am running postgres 8.4.7 as supplied by centos 5.8.  I get a number
of messages like:
  
  ERROR:  canceling autovacuum task
  CONTEXT:  automatic vacuum of table "template0.pg_catalog.pg_shdepend"

for a number of different tables in my databases. We use slony to do
replication and that does have a tendency to use some tables very
heavily resulting in the autovacuum being cancelled. So I am not
concerned about these until we have continually failed a certain
number of autovacuums.

I have set up monitoring to detect 6 successive failures in a week
without any intervening successful vacuums (as reported with
log_autovacuum_min_duration = 0).

In looking through the logs I was expecting to see many more attempts
to vacuum a table when the autovacuum had been cancelled. It looks
like it's happening every few days rather than retrying a few times a
day.  I have one ongoing correlation that reports cancelled
autovacuums at:

  Thu May 10 20:59:08 2012
  Sun May 13 08:30:59 2012

I assume my next one will arrive Wednesday the 16th or so.

So what I am wondering is why is the autovacuum not being rescheduled
more often? I assume the fact that the autovacuum wasn't completed
should make it more likely to be scheduled, before things start
bloating.

If this reschedule period is correct operation, what should I set my
thresholds to:

  3 cancelled autovacuums/week?
  6 cancelled/2 weeks?

or something else.

Thanks.

-- 
				-- rouilj

John Rouillard       System Administrator
Renesys Corporation  603-244-9084 (cell)  603-643-9300 x 111