Le Wednesday 15 April 2009, Greg Smith a écrit :
> On Tue, 14 Apr 2009, William Temperley wrote:
> > I could potentially run a database in each of these countries and
> > provide 100% uptime, obviously raising the issue of version conflicts
> > that would require hand-merging.
Can you partition data on origin?
If that's possible, then do it and use a schema per origin to simplify the
administration thereafter, and "just" replicate some tables from orginin to UK
and central data from UK to editing countries.
> It sounds like you want an asynchronous master-slave database architecture
> where the slaves can also send changes back to the master, but didn't know
> that's what you should be looking for. A quick glance at
> http://wiki.postgresql.org/wiki/Replication%2C_Clustering%2C_and_Connection
>_Pooling suggests Bucardo and Londiste might be useful tools for you to
> investigate; I don't think Slony or Mammoth can handle slaves generating
> their own transactions and feeding them to the master, but given how
> complicated Slony is maybe I just don't know how to do it there.
>
> The Londiste tutorial at
> http://pgsql.tapoueh.org/site/html/londiste/londiste.html even starts out
> with a business situation that sounds similar to yours.
The updated Londiste Tutorial is now to be found on the PostgreSQL wiki, and
I've updated the Replication/Clustering page to add links to it:
http://wiki.postgresql.org/wiki/Skytools
http://wiki.postgresql.org/wiki/Londiste_Tutorial
> I would suggest that if you're new to the area of replication, do not
> assume that just because these tools look complicated that you'd be better
> off rolling your own. The reason they're complicated is because they're
> filled with solutions to hard problems most people never even think they
> need to solve, until they get bit by one.
If you're afraid about their complexity, try londiste and enjoy :)
--
dim