From: Tom Lane <tgl@sss.pgh.pa.us>
To: =?utf-8?q?Milos_Gajdos?= <gy2kre@centrum.cz>
cc: pgsql-admin@postgresql.org
Subject: Re: =?utf-8?q?Postgres_9=2E1_statistics_in_pg=5Fstat=5Fdatabase?=
In-reply-to: <20130207191641.149E5128@centrum.cz>
References: <20130207191641.149E5128@centrum.cz>
Comments: In-reply-to =?utf-8?q?Milos_Gajdos?= <gy2kre@centrum.cz>
	message dated "Thu, 07 Feb 2013 19:16:41 +0100"
Date: Thu, 07 Feb 2013 14:05:00 -0500
Message-ID: <17450.1360263900@sss.pgh.pa.us>
Precedence: bulk
Sender: pgsql-admin-owner@postgresql.org

=?utf-8?q?Milos_Gajdos?= <gy2kre@centrum.cz> writes:
> What this shows is that xact_commit increases by 3 after just one row insert. tup_inserted does seem to report correct value. After simple SELECT, xact_commit increases by 2 - kind of strange as SELECT shouldn't really be increasing xact_commit ?

Yes, a SELECT does count as an xact_commit.  What I think is confusing
you is that the commit isn't reported to the stats collector right away
unless there are some database-local table access counts to send too.
And fetching from pg_stat_database doesn't result in any such accesses.
So each SELECT results in a local increment of the commit counter but
you won't see it in the stats until after you do an INSERT, or otherwise
do something that involves a countable access to a database-local table.

I do see what seems to be a bug, which is that the "Don't expend a clock
check" short-circuit at the top of pgstat_report_stat() needs to not
fire if "force" is true.  That looks like it could result in never
counting a session's last few transactions at all.  But that's not
what's affecting you here, it's the intentional optimization to delay
sending transaction counts.

			regards, tom lane


-- 
Sent via pgsql-admin mailing list (pgsql-admin@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-admin