From: Andres Freund <andres@anarazel.de>
To: pgsql-performance@postgresql.org
Subject: Re: the XID question
Date: Wed, 19 Jan 2011 19:31:51 +0100
User-Agent: KMail/1.13.5 (Linux/2.6.37-andres-03737-g0c21e3a; KDE/4.5.95;
	x86_64; ; )
Cc: Chris Browne <cbbrowne@acm.org>
References: 
 <ca4f167a-0c04-4c59-9dec-a10d8f43437b@i25g2000prd.googlegroups.com>
	<4D36A35302000025000397CC@gw.wicourts.gov>
	<878vyg9399.fsf@cbbrowne.afilias-int.info>
In-Reply-To: <878vyg9399.fsf@cbbrowne.afilias-int.info>
MIME-Version: 1.0
Content-Type: Text/Plain;
  charset="iso-8859-15"
Content-Transfer-Encoding: 7bit
Message-Id: <201101191931.52841.andres@anarazel.de>

On Wednesday, January 19, 2011 07:06:58 PM Chris Browne wrote:
> Kevin.Grittner@wicourts.gov ("Kevin Grittner") writes:
> > Filip Rembia*kowski<plk.zuber@gmail.com> wrote:
> >> 2011/1/19 Charles.Hou <givemeegn@gmail.com>:
> >>> " select * from mybook" SQL command also increase the XID ?
> >> 
> >> Yes. Single SELECT is a transaction. Hence, it needs a transaction
> >> ID.
> > 
> > No, not in recent versions of PostgreSQL.  There's virtual
> > transaction ID, too; which is all that's needed unless the
> > transaction writes something.
> > 
> > Also, as a fine point, if you use explicit database transactions
> > (with BEGIN or START TRANSACTION) then you normally get one XID for
> > the entire transaction, unless you use SAVEPOINTs.
> 
> Erm, "not *necessarily* in recent versions of PostgreSQL."
> 
> A read-only transaction won't consume XIDs, but if you don't expressly
> declare it read-only, they're still liable to get eaten...
No. The Xid is generally only allocated at the first place a real xid is 
needed. See GetCurrentTransactionId, AssignTransactionId in xact.c and the 
caller of the former.

Andres