Yes, this is the problem with GUIDs... you can calculate them by mashing
toghether things like the time, a network address, and some random
numbers, which makes it very unlikely for a collision.... but at the end
of the day that G stand for global, *not* guaranteed.
On Wed, 27 Jul 2005, Scott Marlowe wrote:
> On Wed, 2005-07-27 at 15:32, John DeSoi wrote:
> > On Jul 27, 2005, at 4:49 AM, Tino Wildenhain wrote:
> >
> > > I'd create a sequence:
> > >
> > > CREATE SEQUENCE global_unique_id_seq;
> > >
> > > and a function:
> > >
> > > CREATE OR REPLACE FUNCTION newid()
> > > RETURNS text AS
> > > $BODY$ SELECT nextval('global_unique_id_seq')::text; $BODY$
> > > LANGUAGE 'sql' VOLATILE;
> > >
> > >
> > > now every call to newid() returns a garantied unique id for
> > > say the next 18446744073709551616 calls.
> > > Of course you can obfuscate the ID even more using
> > > md5, include servername and so on, but this will not improve
> > > security in any way (unless you mix data with 2nd database)
> >
> >
> > This is not really a viable replacement for a GUID == globally unique
> > identifier. Here global means that if I use the application in
> > multiple databases, I'm guaranteed that no two identifiers will be
> > the same. Using a sequence will only support uniqueness for a single
> > database.
>
> So, how can two databases, not currently talking to one another,
> guarantee that their GUIDs don't collide? using a large randomly
> generated name space only reduces the chances of collision, it doesn't
> actually guarantee it.
>
> ---------------------------(end of broadcast)---------------------------
> TIP 3: Have you checked our extensive FAQ?
>
> http://www.postgresql.org/docs/faq
>
---------------------------(end of broadcast)---------------------------
TIP 6: explain analyze is your friend
