On Fri, 11 Sep 2009, Alan McKay wrote:
> Munin takes a snapshot every 5 minutes, and this graph shows it
> averaged over that timeframe.
The default postgresql.conf puts a checkpoint every 5 minutes as well.
It's not going to be as exact as Munin's time though, they'll be just a
little longer than that. I wonder if you're seeing the checkpoint pattern
anyway. If the checkpoint period is fairly stable (which it could be in
your case), is slighly longer than the monitoring one, and each checkpoint
has the same basic shape (also usually true), each monitoring sample is
going to trace out the usual checkpoint pattern by sampling a slightly
later point from successive ones. The shape of the curve you're seeing
seems way too close to the standard checkpoint one to be just a
coincidence.
In any case, 5 minutes is unfortunately for you a really bad sampling
period for a PG database because of the similarity to the checkpoint
timing. You have to take measurements at least once a minute to see the
checkpoints happening usefully at all. I think you're stuck generating an
activity graph some other way with finer resolution to get to the bottom
of this.
--
* Greg Smith gsmith@gregsmith.com http://www.gregsmith.com Baltimore, MD
