Centuries ago, Nostradamus foresaw when steve@blighty.com (Steve Atkins) would write:
> As a bit of obPostgresql, though... While the registry for .org is
> run on Postgresql, the actual DNS is run on Oracle. That choice was
> driven by the availability of multi-master replication.
>
> Like many of the cases where the problem looks like it needs
> multi-master replication, though, it doesn't really need it. A
> single master at any one time, but with the ability to dub any of
> the slaves a new master at any time would be adequate. If that were
> available for Postgresql I'd choose it over Oracle were I doing a
> big distributed database backed system again.
Well, this is something that actually _IS_ available for PostgreSQL in
the form of Slony-I. Between "MOVE SET" (that does controlled
takeover) and "FAILOVER" (that recovers from the situation where a
'master' node craters), this has indeed become available.
Automating activation of the failover process isn't quite there yet,
though that's mostly a matter that the methodology would involve
considerable tuning of recovery scripts to system behaviour.
--
select 'cbbrowne' || '@' || 'ntlug.org';
http://cbbrowne.com/info/slony.html
Pay no attention to the PDP-11 behind the front panel.
-- PGS, in reference to OZ