[Pgpool-general] Two geographic instances

[Daniel.Crespo at l-3com.com]

> > I suggest you to read this:
> >
> >
http://www.postgresql.org/docs/9.0/interactive/different-replication-
> solutions.html
> >
> >
> >
> > I would look at what Bucardo offers, which is asynchronous multi-
> master
> > replication. In synchronous multi-master, and especially in your
case
> with
> > pgpool-II, if one database goes down, you might end with
inconsistent
> data
> > in each database (some data same, but newly added data on each end
> would
> > cause a conflic in case you wanted to merge somehow the changes).
> 
> Thanks, I will certainly look at Bucardo. However, I don't understand
> why I should have inconsistent data if a site goes down. If a site
> goes down, only the other site will receive updates, so in the end
> when the down site is back up again (before letting users use it) I
> will only have to do a unidirectional synchronization from the working
> site to the recovered site.
> Am I missing something?

That's true for a "database goes down". But remember that each site
might also work "locally" when there is not connection between the
backends (split-brain problem), meaning that each database might end
with equally important data. It would not suffice to synchronize one
way, but both, opening the possibility to conflicts. Bucardo has
conflict resolution.

-Daniel