Message-ID: <4D24D7DE.6020507@agliodbs.com>
Date: Wed, 05 Jan 2011 12:43:10 -0800
From: Josh Berkus <josh@agliodbs.com>
User-Agent: Mozilla/5.0 (Macintosh; U; Intel Mac OS X 10.5; en-US;
	rv:1.9.1b3pre) Gecko/20090223 Thunderbird/3.0b2
MIME-Version: 1.0
To: pgsql-performance@postgresql.org
Subject: Wrong docs on wal_buffers?
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit

All,

One of my coworkers just pointed this out:

"The amount of memory used in shared memory for WAL data. The default is
64 kilobytes (64kB). The setting need only be large enough to hold the
amount of WAL data generated by one typical transaction, since the data
is written out to disk at every transaction commit. This parameter can
only be set at server start."
http://www.postgresql.org/docs/9.0/static/runtime-config-wal.html

That's quite incorrect.  The wal_buffers are shared by all concurrent
transactions, so it needs to be sized appropriately for all
*simultaneous* uncommitted transactions, otherwise you'll get
unnecessary flushing.

Certainly performance testing data posted on this list and -hackers.
bears that out.  My suggestion instead:

"The amount of shared memory dedicated to buffering writes to the WAL.
The default is 64 kilobytes (64kB), which is low for a busy production
server.  Users who have high write concurrency, or transactions which
commit individual large data writes, will want to increase it to between
1MB and 16MB. This parameter can only be set at server start."

-- 
                                  -- Josh Berkus
                                     PostgreSQL Experts Inc.
                                     http://www.pgexperts.com