pg.ddx.io  pgsql-admin@postgresql.org mailing list archive  
help / color / mirror / Atom feed
From: Mike Goldner <mgoldner@agmednet.com>
To: pgsql-admin@postgresql.org
Subject: Re: Vacuum Problems (locking)
Date: Wed, 25 Oct 2006 16:00:29 -0400
Message-ID: <1161806429.2095.157.camel@localhost.localdomain> (raw)
In-Reply-To: <1161806057.2095.154.camel@localhost.localdomain>
References: <1161806057.2095.154.camel@localhost.localdomain>

On Wed, 2006-10-25 at 15:54 -0400, Mike Goldner wrote:
> I have a nightly vacuum scheduled as follows:
> 
> su - postgres -c "/usr/bin/vacuumdb --analyze --dbname=mydb"
> 
> Last night, it appears that the vacuum blocked db access from my
> application server (JBoss).  Here is the logfile snippet:
> 
> [3693-jbossdb-postgres-2006-10-25 06:52:29.488 EDT]NOTICE:  number of
> page slots needed (559472) exceeds max_fsm_pages (200000)
> [3693-jbossdb-postgres-2006-10-25 06:52:29.488 EDT]HINT:  Consider
> increasing the configuration parameter "max_fsm_pages" to a value over
> 559472.
> [3693-jbossdb-postgres-2006-10-25 06:52:29.488 EDT]LOG:  number of page
> slots needed (559472) exceeds max_fsm_pages (200000)
> [3693-jbossdb-postgres-2006-10-25 06:52:29.488 EDT]HINT:  Consider
> increasing the configuration parameter "max_fsm_pages" to a value over
> 559472.
> [3693-jbossdb-postgres-2006-10-25 06:52:29.530 EDT]LOG:  transaction ID
> wrap limit is 1073813592, limited by database "benchpg"
> [3693-jbossdb-postgres-2006-10-25 06:52:29.530 EDT]LOG:  transaction ID
> wrap limit is 1073813592, limited by database "benchpg"
> [3693-jbossdb-postgres-2006-10-25 06:52:29.530 EDT]LOG:  duration:
> 10224738.050 ms  statement: VACUUM ANALYZE;
> 
> First of all, my max_fsm_pages is obviously way off.  However, every
> time I increase my max_fsm_pages the next vacuum says that it requires
> more.  Will there every be a plateau in the requested pages?  What is
> the impact of exceeding the pages slots needed?
> 
> Secondly, in this case, it almost looks like there were two vacuum
> process running based on the duplicated log messages.  There is only one
> cron job, so there shouldn't have been a second vacuum running.  Should
> I just ignore this?
> 
> Finally, and most important is the blocking.  The vacuum duration
> reported in the log converts to about 170 minutes.  I can track
> backwards in the log and the only messages prior to the 6:52am
> completion of the vacuum end at 3:57am (almost exactly 170 minutes
> prior).
> 
> So, all indications point to postgres blocking all access during the
> entire vacuum.  
> 
> The only semi-unusual characteristic of my database is that I have one
> table in particular that grows and shrinks dramatically.  My entire
> database after a vacuum --full is about 8GB.  However, after a day or
> two, the db will grow as large as 75GB.  Truncating that one table
> (since it empties regularly) will shrink the db right back to the 8GB.
> 
> Is it possible that the extreme table size (maybe requiring the high
> number of page slots) is causing postgres to block more than it should
> in order to process that table?
> 
> Thanks,
> 
> Mike
> 
> ---------------------------(end of broadcast)---------------------------
> TIP 6: explain analyze is your friend

I just noticed that my nightly vacuum failed the previous three nights
with the following error:

[2032-jbossdb-postgres-2006-10-24 04:12:30.019 EDT]ERROR:  failed to re-
find parent key in "jms_messages_pkey"
[2032-jbossdb-postgres-2006-10-24 04:12:30.019 EDT]STATEMENT:  VACUUM
ANALYZE;

I'm not sure if it is related, but it seems pertinent.

Thanks,

Mike



view thread (6+ messages)  latest in thread

Message-ID: <1161806429.2095.157.camel@localhost.localdomain>
Permalink:  ../1161806429.2095.157.camel@localhost.localdomain/
Also on:    postgresql.org/message-id/1161806429.2095.157.camel@localhost.localdomain

 · 

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-admin@postgresql.org
  Cc: mgoldner@agmednet.com
  Subject: Re: Vacuum Problems (locking)
  In-Reply-To: <1161806429.2095.157.camel@localhost.localdomain>
  References: <1161806057.2095.154.camel@localhost.localdomain> <1161806429.2095.157.camel@localhost.localdomain>

* 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