On Saturday April 5 2003 10:38, nolan@celery.tssi.com wrote:
> I am probably out of my depth here as this seems to be getting into the
> internals of the pgsql replication project. I'm not part of that...
Just fyi, dbmirror is not a part of the official pgsql replication project.
It's an asyncronous replication contribution from Steven Singer. Some may
find it possibly useful between now and when pgsql replication is released,
which may not be for some number of months still. I would expect that
project to be more useful given the considerable effort devoted to it.
> > > If there are INDEPENDENT sequences on the master and the slave,
> > > what's to guarantee uniqueness?
> >
> > Not sure I understand the question. Uniqueness in what respect?
>
> Correct me if I'm wrong, but isn't the primary goal of a sequence to
> ensure uniqueness of the sequence values?
>
> It's atomic and non-transactional (ie, sequences cannot be rolled back),
> and in a data replication environment there must be some method of
> ensuring that the sequence remains atomic, which is to say that there is
> really only ONE sequence that is shared among the replication nodes.
My context is asyncronous master-slave replication. Here, the slaves do not
take updates, except from the master replicator. If someone is accessing
the slave and incrementing the sequences outside the replicator process,
that's a problem for this approach. I think your concern is valid for
multi-master, but irrelevant for my master-slave interests.
Steven had some ideas about how to do a limited multi-master with dbmirror.
Then I suppose the uniqueness concern would come into play.
Ed
