DomainKey-Signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma;
	h=mime-version:in-reply-to:references:date:message-id:subject:from:to
	:cc:content-type:content-transfer-encoding;
	b=GKoJ6ta5GrfG2FsFLutBVDw4NooEOuN6DskgFyPDdj19HEcBca+I97hr8vxhGaCrdA
	GUFkcv91MnLJDm5KuIYkRU/O9n9gYPZBCmxYeuSopa/qNXCxqv5+ayzVN2hoLPXXQQ8q
	tafeNSVPuAHqoaChmYtN1tDEjoTrF3PjZ8cVs=
MIME-Version: 1.0
In-Reply-To: <919b7db61003090527k3c8ad4cw84b70e55bd15cd1e@mail.gmail.com>
References: <919b7db61003090138j6da56dffu5e0ddd4276168b48@mail.gmail.com>
	<b42b73151003090434yabba5me4489096769d4bc7@mail.gmail.com>
	<919b7db61003090527k3c8ad4cw84b70e55bd15cd1e@mail.gmail.com>
Date: Tue, 9 Mar 2010 10:03:16 -0500
Message-ID: <b42b73151003090703n2feb05e1r60c9cd359b4f359b@mail.gmail.com>
Subject: Re: Out of shared memory in postgres 8.4.2 and locks
From: Merlin Moncure <mmoncure@gmail.com>
To: Vidhya Bondre <meetvbondre@gmail.com>
Cc: pgsql-performance@postgresql.org
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: quoted-printable

On Tue, Mar 9, 2010 at 8:27 AM, Vidhya Bondre <meetvbondre@gmail.com> wrote=
:
>>
>> are you using the same postgresql.conf? =A0have you created more
>> partitions? =A0using advisory locks?
>
> Yes we are using same conf files. In a week we create around 5 partitions=
.
> We are not using advisory locks
>>
>> In any event, increase the max_locks_per_transaction setting and
>> restart the database.
>
>
> Currently the value of max_locks_per_transaction is 64 modifying it to 96
> works.
> Have a couple of questions
> 1] As and when we add partitions will we have to increase this parameter =
?
> 2] will we have to consider any othe parameter twick while increasing thi=
s
> one?

If you are doing 'in transaction' operations that involve a lot of
tables this figure has to be bumped, sometimes significantly.  You
might be tempted just crank it, and be done with this problem.   If
so, be advised of what happens when you do:

*) more shared memory usage (make sure you have memory and
shared_buffers is appropriately set).  however for what you get the
usage is relatively modest.
*) anything that scans the entire in memory lock table takes longer in
relationship to this .conf value. AFAIK, the only noteworthy thing
that does this is the pg_locks view.

The 'other' big shared memory tradeoff you historically had to deal
with, the fsm map, is gone in 8.4.

merlin