To: Jakub Ouhrabka <jakub.ouhrabka@comgate.cz>
cc: pgsql-performance@postgresql.org
Subject: Re: Linux/PostgreSQL scalability issue - problem with 8 cores 
In-reply-to: <4782AB4D.8010009@comgate.cz> 
References: <flii57$748$1@news.hub.org> <7083.1199379570@sss.pgh.pa.us>
	<477D1D59.7050002@comgate.cz> <4782063C.7070105@comgate.cz>
	<26502.1199745090@sss.pgh.pa.us> <4782AB4D.8010009@comgate.cz>
Comments: In-reply-to Jakub Ouhrabka <jakub.ouhrabka@comgate.cz>
	message dated "Mon, 07 Jan 2008 23:44:29 +0100"
Date: Mon, 07 Jan 2008 19:01:32 -0500
Message-ID: <27655.1199750492@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

Jakub Ouhrabka <jakub.ouhrabka@comgate.cz> writes:
>>> Does your app create and destroy a tremendous number of temp tables,
>>> or anything else in the way of frequent DDL commands?

> Hmm. I can't think of anything like this. Maybe there are few backends 
> which create temp tables but not tremendous number. I don't think our 
> applications issue DDL statements either. Can LOCK TABLE IN EXCLUSIVE 
> MODE cause this?

No.

I did some experimenting to see exactly how large the sinval message
buffer is in today's terms, and what I find is that about 22 cycles
of

create temp table foo (f1 int, f2 text); drop table foo;

is enough to force a CatchupInterrupt to a sleeping backend.  This case
is a bit more complex than it appears since the text column forces the
temp table to have a toast table; but even with only fixed-width
columns, if you were creating one temp table a second that would be
plenty to explain once-a-minute-or-so CatchupInterrupt processing.
And if you've got a lot of backends that normally *don't* wake up
that often, then they'd be sitting and failing to absorb the sinval
traffic in any more timely fashion.  So I'm betting that we've got
the source of the spike identified.  You could check this theory
out by strace'ing some of the idle backends and seeing if their
activity spikes are triggered by receipt of SIGUSR1 signals.

			regards, tom lane