Received-SPF: pass (google.com: domain of brianf@consistentstate.com
	designates 10.68.216.4 as permitted sender)
	client-ip=10.68.216.4;
Message-ID: <4F500868.8060603@consistentstate.com>
Date: Thu, 01 Mar 2012 16:38:16 -0700
From: Brian Fehrle <brianf@consistentstate.com>
User-Agent: Mozilla/5.0 (X11; U; Linux i686 (x86_64); en-US;
	rv:1.9.2.26) Gecko/20120129 Lightning/1.0b2 Thunderbird/3.1.18
MIME-Version: 1.0
To: Simon Riggs <simon@2ndQuadrant.com>
CC: pgsql-admin@postgresql.org
Subject: Re: Hot standby having high requested checkpoints?
References: <4F4D5AE0.9020204@consistentstate.com>
	<CA+U5nMKcy_wkzJdpwsjyaa76ucyw0FkDhdXnMCypbAwPF0JSEQ@mail.gmail.com>
In-Reply-To: 
 <CA+U5nMKcy_wkzJdpwsjyaa76ucyw0FkDhdXnMCypbAwPF0JSEQ@mail.gmail.com>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

I just now ran the following query a few times after each other on the 
hot standby:

select now(), * from pg_stat_bgwriter;

Here are the results:

               now              | checkpoints_timed | checkpoints_req | 
buffers_checkpoint | buffers_clean | maxwritten_clean | buffers_backend 
| buffers_alloc
-------------------------------+-------------------+-----------------+--------------------+---------------+------------------+-----------------+---------------
  2012-03-01 23:24:49.099194+00 |             11546 |          145300 
|         1000409459 |      38483026 |           170724 |         
5058186 |     438703950
  2012-03-01 23:24:52.139176+00 |             11546 |          145300 
|         1000409459 |      38483027 |           170724 |         
5058191 |     438703975
  2012-03-01 23:24:59.129171+00 |             11546 |          145302 
|         1000409459 |      38483081 |           170724 |         
5058214 |     438704438
  2012-03-01 23:25:05.957532+00 |             11546 |          145304 
|         1000409459 |      38483118 |           170724 |         
5058230 |     438704689
  2012-03-01 23:25:09.519175+00 |             11546 |          145305 
|         1000409459 |      38483160 |           170724 |         
5058239 |     438704968
  2012-03-01 23:25:37.019194+00 |             11546 |          145308 
|         1000409459 |      38483259 |           170724 |         
5058255 |     438705566
  2012-03-01 23:25:40.659164+00 |             11546 |          145308 
|         1000409459 |      38483268 |           170724 |         
5058257 |     438705639
  2012-03-01 23:25:47.239281+00 |             11546 |          145309 
|         1000409459 |      38483283 |           170724 |         
5058266 |     438705815
  2012-03-01 23:26:23.858716+00 |             11546 |          145312 
|         1000409459 |      38483393 |           170724 |         
5058307 |     438706561
  2012-03-01 23:26:46.467493+00 |             11546 |          145317 
|         1000409670 |      38483524 |           170724 |         
5058354 |     438707619

in two minutes, I saw 17 checkpoints_req, and the number of 
buffers_checkpoint didn't budge till the last few and even then not much.

checkpoint_segments = 256

- Brian F

On 03/01/2012 01:35 AM, Simon Riggs wrote:
> On Tue, Feb 28, 2012 at 10:53 PM, Brian Fehrle
> <brianf@consistentstate.com>  wrote:
>
>> I have two PostgreSQL 9.0.5 clusters, one is a master and the other is a hot
>> standby via streaming replication. I'm monitoring some stats on each and I'm
>> noticing something very odd. On the master, I get between 2 and 4 requested
>> checkpoints per hour, but on the hot standby I'm seeing between 200 and 300
>> requested checkpoints per hour.
> We don't do a restartpoint on the standby unless we see a checkpoint
> record, so that result should be impossible.
>
> So I'm guessing you're reading the stats wrong?
>