[pgpool-hackers: 440] Re: Proposal: check_unlogged_table parameter

[Tatsuo Ishii]

> Hi, I'd like to suggest a new parameter called check_unlogged_table which is
> very similiar to check_temp_table.
> 
> Our pgpool server with streaming replication backends handles many
> short-lived
> sessions per second, and we've noticed that the master backend is running
> lots
> of system catalog queries. This problem is alleviated when check_temp_table
> is
> set to 0, but the master server still consumes twice as much CPU time as the
> slave, because of session-local checking for unlogged tables.

This is a bug. It should not be session-local. I have already posted
a fix and have been waiting for feedback...

> Subject: Re: [pgpool-general: 2325] Re: thousands of queries to pg_class / second
> From: Nathan Brennan <nathan at healthengine.com.au>
> Date: Wed, 4 Dec 2013 16:48:55 +0800
>
>  Thanks Tatsuo, I will apply the patch in the next couple of days.

> Since we're not planning to use unlogged tables, we'd like to avoid this
> unnecessary cost, as we can do so for temp tables with check_temp_table
> parameter. We've added the parameter to our fork of pgpool2. It would be
> nice
> if it could be merged into the trunk.
> 
> Attached is the patch. The code is almost identical to that for
> check_temp_table. Please take a look.

Sounds like a nice idea. Taking look at your patches, I noticed that
they are missing patches to the manual and
src/utils/pool_process_reporting.c (used to display variables by "show
pool_status" command). Could you please add them?
--
Tatsuo Ishii
SRA OSS, Inc. Japan
English: http://www.sraoss.co.jp/index_en.php
Japanese: http://www.sraoss.co.jp