pg.ddx.io  pgsql-performance@postgresql.org mailing list archive  
help / color / mirror / Atom feed
From: Stephan Szabo <sszabo@megazone23.bigpanda.com>
To: Robert John Shepherd <robert@reviewer.co.uk>
Cc: pgsql-performance@postgresql.org
Subject: Re: [BUGS] Rapid deteriation of performance (might be
Date: Fri, 4 Apr 2003 07:29:43 -0800 (PST)
Message-ID: <20030404072355.Q93368-100000@megazone23.bigpanda.com> (raw)
In-Reply-To: <001e01c2fabd$9edfb860$f3b0313e@LAIKA>


On Fri, 4 Apr 2003, Robert John Shepherd wrote:

> > > Up until a few days ago I have been running Postgresl 7.2.3 with
> Tsearch
> > > from the contrib dir, but at various times the performance of the
> > > database would suddenly and rapidly deteriate so that queries which
> > > previously took 500ms then took 8 or 9 seconds.
>
> > Hmm, what are the before and after explain analyze results?  Also,
> what
> > are your conf settings for shared buffers, sort memory and the fsm
> > parameters?
>
> shared_buffers = 40960
> sort_mem = 20480
> #max_fsm_relations = 1000
> #max_fsm_pages = 10000
>
> As you can see I've not uncommented or touched the fsm parameters, I
> have no idea what they do. Optimisation wise I have only played with
> shared_buffers, sort_mem and max_connections.
>
>
> > > The only cure is a backup and restore of the database
>
> > Did you do a regular vacuum or vacuum full?  If only the former, it's
> > possible that you need to either vacuum more frequently and/or raise
> the
> > free space map settings in your configuration file.
>
> I've been running this daily:
>
> 	vacuumdb -h localhost -a -z
>
> Should I be using the full switch then?

Well, you generally shouldn't need to if the fsm settings are high enough.
If you're doing really big updates like update each row of a 1 billion
row table, you may end up having to do one immediately following that.
Of course, if you're doing that, performance is probably not your biggest
concern. ;)

Explain analyze'll tell us if the system is changing plans (presumably to
a worse one) - for example, deciding to move to a sequence scan because it
thinks that the index scan is now to expensive, or conversely moving to an
index scan because it thinks that there'll be too many reads, while those
page reads actually are fairly localized. The vacuum full verbose should
get some idea of how much empty space is there.

> > pgsql-performance is a better list, so I've replied to there.  You'll
> > probably need to join in order to reply to list.
>
> Thanks, especially for not shouting at me heh, this is stressful enough
> as it is.

:)




view thread (6+ messages)  latest in thread

Message-ID: <20030404072355.Q93368-100000@megazone23.bigpanda.com>
Permalink:  ../20030404072355.Q93368-100000@megazone23.bigpanda.com/
Also on:    postgresql.org/message-id/20030404072355.Q93368-100000@megazone23.bigpanda.com

 · 

reply

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Reply to all the recipients from your mail client; the link fills in
  the headers below: reply via email
  Some clients, webmail especially, drop In-Reply-To and References from
  a link, and the reply then starts a new thread: use the mbox below.

  To: pgsql-performance@postgresql.org
  Cc: sszabo@megazone23.bigpanda.com, robert@reviewer.co.uk
  Subject: Re: [BUGS] Rapid deteriation of performance (might be
  In-Reply-To: <20030404072355.Q93368-100000@megazone23.bigpanda.com>
  References: <001e01c2fabd$9edfb860$f3b0313e@LAIKA> <20030404072355.Q93368-100000@megazone23.bigpanda.com>

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox