On Tue, 2005-03-29 at 16:06, Caleb Simonyi-Gindele wrote:
> John Burger wrote:
>
> >>> If it were me, and someone proposed a model where two-way replication
> >>> was needed, I would tell them to rethink their model. It's broken.
> >>
> >>
> >> I would respectfully disagree that the requirement for two-way
> >> replication
> >> indicates a broken design.
> >
> >
> > I agree with your disagreement. This design is present in lots of
> > non-RDB systems - CVS, IMAP, PDA syncing, etc. It's clearly more
> > complicated, but can be made to work, and has been many times. I
> > don't see anything about databases in general, or Postgres
> > specifically, that indicates it's a bad idea.
> >
> > - John D. Burger
> > MITRE
> >
> Yes, we use it successfully with the SQL Server edition of our product.
> Does anyone know if this is available with Postgre?
It's important to understand that what you're asking for is MORE than
simple replication, it is replication with ((semi)automatic) conflict
resolution. If you use a simple replication system to try and do this,
you are likely to wind up with inconsistent data.
Just because SQL Server does it doesn't mean it does it right. And the
general philosophy of the PostgreSQL team seems to be do it right or
don't bother.
So, what are the chances that you'll have records on your sales folks
machines that have also been updated back at the home office? What
rules should be applied when conflicts arise? These are the kinds of
questions you need to answer before jumping feet first into the fire and
getting burnt.