From: pgulutzan@ocelot.ca (Peter Gulutzan)
Subject: Re: [GENERAL] Bug with sequence
Date: 26 Nov 2002 07:30:12 -0800
Organization: http://groups.google.com/
Lines: 37
Message-ID: <36c478c6.0211260730.a45279d@posting.google.com>
References: <Pine.LNX.4.33.0211211450100.23804-100000@css120.ihs.com>
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: 8bit
To: pgsql-hackers@postgresql.org

scott.marlowe@ihs.com ("scott.marlowe") wrote in message news:<Pine.LNX.4.33.0211211450100.23804-100000@css120.ihs.com>...
> On 21 Nov 2002, Rod Taylor wrote:
> 
> > On Thu, 2002-11-21 at 15:09, scott.marlowe wrote:
> > > On 21 Nov 2002, Rod Taylor wrote:
> > > 
> > > > On Thu, 2002-11-21 at 14:11, Bruce Momjian wrote:
> > > > > Of course, those would be SQL purists who _don't_ understand
> > > > > concurrency issues.  ;-)
> > > > 
> > > > Or they're the kind that locks the entire table for any given insert.
> > > 
> > > Isn't that what Bruce just said?  ;^)
> > 
> > I suppose so.  I took what Bruce said to be that multiple users could
> > get the same ID.
> > 
> > I keep having developers want to make their own table for a sequence,
> > then use id = id + 1 -- so they hold a lock on it for the duration of
> > the transaction.
> 
> I was just funnin' with ya, but the point behind it was that either way 
> (with or without a lock) that using something other than a sequence is  
> probably a bad idea.  Either way, under parallel load, you have data 
> consistency issues, or you have poor performance issues.
> 
> 
I'm not familiar with these "SQL purists" (perhaps the reference is to
J. Celko?) but the fact is that it's hard to call SEQUENCE
product-specific now that it's in Oracle, DB2, and SQL:2003. The
syntaxes do differ a little, usually due to choice of abbreviation,
but as far as I can tell the internals are similar across
implementations.

Peter Gulutzan
Author of "Sequences And Identity Columns"
(http://dbazine.com/gulutzan4.html)