Date: Fri, 19 Aug 2005 18:01:40 -0700
To: pgsql-interfaces@postgresql.org, Junaili Lie <junaili@gmail.com>,
	pgsql-general@postgresql.org
Cc: elein <elein@varlena.com>
Subject: Re: <IDLE> in transaction
Message-ID: <20050820010140.GC5365@varlena.com>
Mail-Followup-To: pgsql-interfaces@postgresql.org,
	Junaili Lie <junaili@gmail.com>, pgsql-general@postgresql.org
References: <8d04ce990508181616c93fbbc@mail.gmail.com>
	<20050819000218.GS5365@varlena.com>
	<20050819185519.GY5365@varlena.com>
	<20050819230602.GM15622@surnet.cl>
Mime-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <20050819230602.GM15622@surnet.cl>
User-Agent: Mutt/1.5.6+20040907i
From: elein@varlena.com (elein)

On Fri, Aug 19, 2005 at 07:06:02PM -0400, Alvaro Herrera wrote:
> On Fri, Aug 19, 2005 at 11:55:19AM -0700, elein wrote:
> > I'm cross posting to INTERFACES.  Please follow up
> > on INTERFACES and not on general.  Cross posting is evil.
> 
> Well, I'm not on -interfaces, so I'll reply to both :-)  I don't worry
> too much about crossposting, because a) it's commonplace in PostgreSQL
> lists, and b) majordomo can deliver a single copy of the message if you
> configure it to do so.
> 
> 
> > > Some of these I-i-t connections come and go after a while.
> > > Some stick around for DAYS.
> > > 
> > > If ANYONE has any brilliant ideas as to the source and
> > > dare I say correction to this problem, many people, especially
> > > myself would be very very happy.
> 
> While this is a purely client-side problem, which is the client issuing
> a BEGIN right after a COMMIT, we talked about coding around it
> server-side, back in the time when I was doing nested transactions.
> It didn't get done though.  I think if you push hard enough, somebody
> (myself?) may do it for 8.2.

To replicate the situation is psql:
	BEGIN;
	select something;

In another window I see that I have not only shared access locks
but an exclusive access lock.  I do not understand why the exclusive
lock is there.  Am I seeing ghosts?

Also, for some relief we found a piece of code that forgot its
commit.  That helped a lot but I'm not convinced it was the only place
this occurred.

--elein

> 
> Of course, this is no solution if the client started a transaction, did
> some work, and then sat on the connection with the transaction open for
> days.  But this is not a common case and is certainly much more broken,
> if only because other RDBMS behave more reasonably in the COMMIT-BEGIN
> scenario.
> 
> -- 
> Alvaro Herrera (<alvherre[a]alvh.no-ip.org>)
> "La persona que no quería pecar / estaba obligada a sentarse
>  en duras y empinadas sillas    / desprovistas, por cierto
>  de blandos atenuantes"                          (Patricio Vogel)
> 
> ---------------------------(end of broadcast)---------------------------
> TIP 4: Have you searched our list archives?
> 
>                http://archives.postgresql.org
>