Message-ID: <4485D9A1.30807@ca.sophos.com>
Date: Tue, 06 Jun 2006 12:38:09 -0700
From: Mischa Sandberg <mischa@ca.sophos.com>
Reply-To: mischa@ca.sophos.com
User-Agent: Thunderbird 1.5.0.4 (X11/20060516)
MIME-Version: 1.0
To: pgsql-performance@postgresql.org
Subject: Re: vacuuming problems continued
References: <92d3a4950606010454y67293765o56001e8996975b45@mail.gmail.com>
	<20060605204151.GC7071@phlogiston.dyndns.org>
	<4485CECC.4040207@ca.sophos.com>
	<4485D5C3.7010803@commandprompt.com>
In-Reply-To: <4485D5C3.7010803@commandprompt.com>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

Joshua D. Drake wrote:
>> - in our env, clients occasionally hit max_connections. This is a 
>> known and (sort of) desired pushback on load. However, that sometimes 
>> knocks pg_autovacuum out.
> 
> That is when you use:
> 
> superuser_reserved_connections

Blush. Good point. Though, when we hit max_connections on 7.4.8 systems,
it's been a lemonade-from-lemons plus that vacuuming didn't fire up on top of 
everything else :-)

>> - db server goes down for any reason: same problem.
> 
> I believe you can use
> stats_reset_on_server_start = on

We do. The problem is not the loss of pg_stat_user_tables.(n_tup_ins,...)
It's the loss of pg_autovacuum's CountAtLastVacuum (and ...Analyze)
numbers, which are kept in process memory. Never considered patching
pg_autovacuum to just sleep and try again, rather than exit, on a failed
db connection.

-- 
Engineers think that equations approximate reality.
Physicists think that reality approximates the equations.
Mathematicians never make the connection.