Date: Wed, 11 Sep 2013 23:10:08 +0200
From: Andres Freund <andres@2ndquadrant.com>
To: Andrew Dunstan <andrew@dunslane.net>
Cc: Josh Berkus <josh@agliodbs.com>, Jeff Janes <jeff.janes@gmail.com>,
 Amit Kapila <amit.kapila@huawei.com>,
 "pgsql-performance@postgresql.org" <pgsql-performance@postgresql.org>
Subject: Re: Performance bug in prepared statement binding in 9.2?
Message-ID: <20130911211008.GE1138556@alap2.anarazel.de>
References: <51A79512.4020200@agliodbs.com>
 <WM!c84cb8479a8b2ec21d1061c951095da00bc52608e9822124f1438b9de432ec7ebedaf25085801b67d581eed6d368c344!@asav-2.01.com>
 <51FAA1B8.6040809@agliodbs.com>
 <CAMkU=1wn3npjTjEkC8rYMgmoKXYoAgfxosT_Q3v_=wTfriPKdA@mail.gmail.com>
 <522E69F1.8070203@dunslane.net>
 <20130910122036.GF1024477@alap2.anarazel.de>
 <522F146D.2070908@dunslane.net>
 <WM!21a18aa9cbf2098b3ceb8d93a0769806466db819c328e3d732b1667507e087969764ace79b57bc4d6e54a0bce329a6fa!@asav-3.01.com>
 <5230B801.90507@agliodbs.com> <5230BF2F.8070600@dunslane.net>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <5230BF2F.8070600@dunslane.net>
Precedence: bulk
Sender: pgsql-performance-owner@postgresql.org

On 2013-09-11 15:06:23 -0400, Andrew Dunstan wrote:
> 
> On 09/11/2013 02:35 PM, Josh Berkus wrote:
> >All,
> >
> >We've confirmed that this issue is caused by having long-running idle
> >transactions on the server.  When we disabled their queueing system
> >(which prodiced hour-long idle txns), the progressive slowness went away.
> >
> >Why that should affect 9.X far more strongly than 8.4, I'm not sure
> >about.  Does that mean that 8.4 was unsafe, or that this is something
> >which *could* be fixed in later versions?
> >
> >I'm also confused as to why this would affect BIND time rather than
> >EXECUTE time.
> >
> 
> 
> One thing that this made me wonder is why we don't have transaction_timeout,
> or maybe transaction_idle_timeout.

Because it's harder than it sounds, at least if you want to support
idle-in-transactions. Note that we do not support pg_cancel_backend()
for those yet...

Also, I think it might lead to papering over actual issues with
applications leaving transactions open. I don't really see a valid
reason for an application needing cancelling of long idle transactions.

Greetings,

Andres Freund

-- 
 Andres Freund	                   http://www.2ndQuadrant.com/
 PostgreSQL Development, 24x7 Support, Training & Services


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