Date: Tue, 23 Oct 2007 10:00:05 -0700
From: Ron St-Pierre <ron.pgsql@shaw.ca>
Subject: Re: 12 hour table vacuums
In-reply-to: <20071023161204.GK18013@alvh.no-ip.org>
To: Alvaro Herrera <alvherre@commandprompt.com>
Cc: pgsql-performance@postgresql.org
Message-id: <471E2895.4030601@shaw.ca>
MIME-version: 1.0
Content-type: text/plain; charset=ISO-8859-1; format=flowed
Content-transfer-encoding: 7bit
References: <471E18ED.3090704@shaw.ca> <20071023161204.GK18013@alvh.no-ip.org>
User-Agent: Thunderbird 2.0.0.6 (Windows/20070728)

Alvaro Herrera wrote:
> Ron St-Pierre wrote:
>
>   
>> Okay, here's our system:
>>   postgres 8.1.4
>>     
>
> Upgrade to 8.1.10
>   
Any particular fixes in 8.1.10 that would help with this?
>   
>> Here's the table information:
>> The table has 140,000 rows, 130 columns (mostly NUMERIC), 60 indexes.
>>     
>
> 60 indexes?  You gotta be kidding.  You really have 60 columns on which
> to scan?
>
>   
Really. 60 indexes. They're the most commonly requested columns for 
company information (we believe). Any ideas on testing our assumptions 
about that? I would like to know definitively what are the most popular 
columns. Do you think that rules would be a good approach for this? 
(Sorry if I'm getting way off topic here)
>> vacuum_cost_delay = 200
>> vacuum_cost_limit = 100
>>     
>
> Isn't this a bit high?  What happens if you cut the delay to, say, 10?
> (considering you've lowered the limit to half the default)
>
>   
Yes, Tom pointed this out too. I'll lower it and check out the results.

Ron