From: Bruce Momjian <bruce@momjian.us>
Message-Id: <201008311544.o7VFioj00972@momjian.us>
Subject: Re: Tuple changes from relfilenodes
In-Reply-To: <4C77A7E40200002500034D33@gw.wicourts.gov>
To: Kevin Grittner <Kevin.Grittner@wicourts.gov>
Date: Tue, 31 Aug 2010 11:44:50 -0400 (EDT)
CC: "=?ISO-8859-15?Q?*ukasz=20Brodziak?=" <lukasz.brodziak@hotmail.com>,
	pgsql-admin@postgresql.org
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset="US-ASCII"

Kevin Grittner wrote:
> *ukasz Brodziak<lukasz.brodziak@hotmail.com> wrote:
> > From: Kevin.Grittner@wicourts.gov
>  
> >> Are you looking for statistics or the actual prior versions of
> >> rows?
>  
> > I'm looking for actual versions of row data. What I want to
> > achieve as a final result is a kind of data change history.
>  
> Well, unless you are suppressing the vacuum behavior which normally
> occurs, your history won't go back very far; and if you do suppress
> vacuums, performance will begin to fall of fairly quickly.
>  
> Tuples representing old versions of rows may be eliminated as soon
> as there is no transaction which would still be able to see them.

Well, single-page vacuums are even more likely to remove old data, and
you can't turn that off.  I am working on am MVCC talk for PG/West that
will show how agressive those single-page (HOT and non-HOT) vacuums are.

-- 
  Bruce Momjian  <bruce@momjian.us>        http://momjian.us
  EnterpriseDB                             http://enterprisedb.com

  + It's impossible for everything to be true. +