Date: Tue, 8 May 2012 10:55:50 -0600 (MDT)
From: Brian Ferhle <brianf@consistentstate.com>
To: pgsql-admin@postgresql.org
Message-ID: 
 <2012787084.273491.1336496150057.JavaMail.root@mail.consistentstate.com>
In-Reply-To: 
 <448748533.273490.1336496125865.JavaMail.root@mail.consistentstate.com>
Subject: increasing max_pred_locks_per_transaction, what shuold I look for?
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit

Hi all,

I've got a situation where I need to increase max_pred_locks_per_transaction value to allow the addition of a slony node to complete on postgres 9.1. We had this issue before and we increased the default value from 64 to 128, but now we have a sync that takes even longer and the value for 128 isn't enough either.

Looking at the activity I saw in pg_locks, I believe I need to set it to at least 256, but a value this high I want to make sure that I don't cause other issues with the system. The docs mention that it may need to request more system v shared memory, but should I also re-calculate other parameters such as work_mem, shared_buffers, etc?

Current System:
Postgres 9.1.3
252 GB system memory
shared_buffers 40GB
work_mem 200MB
max_connections = 300
max_prepared_transactions = 0

free -g
             total       used       free     shared    buffers     cached
Mem:           252        239         12          0          1        221

I have a lot of cashed memory, but I just want to make sure this isn't something that cascades out of control and I'm suddenly having major 'out of memory' issues.

thanks in advance,
- Brian F