To: Jan Wieck <JanWieck@Yahoo.com>
Cc: Greg Stark <gsstark@mit.edu>, pgsql-interfaces@postgresql.org
Subject: Re: Lost updates vs resumable connections/transactions
References: <Pine.A41.4.58.0412091641060.68588@zivunix.uni-muenster.de>
	<41C06961.8020406@Yahoo.com> <87is73qrde.fsf@stark.xeocode.com>
	<m2y8fyl58u.fsf@pcwi4002.uni-muenster.de> <41C1CE7A.10405@Yahoo.com>
	<m21xdpkpgy.fsf@pcwi4002.uni-muenster.de> <41C306A1.1070807@Yahoo.com>
	<m2sm64j1pa.fsf@pcwi4002.uni-muenster.de> <41C31A2E.9030107@Yahoo.com>
From: Jens Lechtenboerger <lechten@wi.uni-muenster.de>
Date: Fri, 17 Dec 2004 19:07:08 +0100
In-Reply-To: <41C31A2E.9030107@Yahoo.com> (Jan Wieck's message of "Fri, 17
	Dec 2004 12:41:02 -0500")
Message-ID: <m2hdmk3ik3.fsf@pcwi4002.uni-muenster.de>
User-Agent: Gnus/5.1006 (Gnus v5.10.6) Emacs/21.2 (gnu/linux)
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii

Jan Wieck <JanWieck@Yahoo.com> writes:

> On 12/17/2004 12:04 PM, Jens Lechtenboerger wrote:
>
>> Jan Wieck <JanWieck@Yahoo.com> writes:
>>
>>> [...]
>>> I don't think this idea has much of a chance to make it into the
>>> source tree.
>> I'm disappointed, though, and summarize:
>> PostgreSQL transactions cannot be used naturally with CGI/PHP, and
>> virtually every web application out there is prone to lost updates.
>
> As said, open transactions with DB locks during user interaction are a known
> bad idea for every sort of application. That together with the scaling
> problems is IMHO reason enough not to implement something that is designed to
> avoid proper application side advisory locks.

After writing my last mail, I was riding my bicycle through the rain
and thought about the following additional application areas for an
API extension:

Grid computing: I might want to transfer a transaction from an
overloaded client node to another node.  Currently, this is not
possible.

Web services: I might want to compose an atomic service out of
component services over a single database.  In contrast to my
previous web example, in this scenario there needn't be user
interaction between the individual service invocations.

> Get used to put reasonable amounts of your business logic into stored
> procedures on the database side and you will find that dealing with advisory
> locks is not as painfull as it looks like. Doing it all with PHP coding alone,
> where a single business process is scattered over a input form flow dictated
> number of source files is neither as easy, nor as maintainable.

There is a fundamental problem.  It's not about "scattered" business
processes but about the very simple "look and edit" process that is
not supported.

Jens