Re: Proposal for a cascaded master-slave replication system

Speaking from a non-profit whose enterprise data sits inside postgres,
we would be willing to invest a few thousand dollars into the pot of
synchronous multi-master replication. Postgres-r sounded absolutely
marvelous to us back in the day that it was rumored to be one of the
possible deliverables of 7.4.

Not so much for nine-nines of uptime, but for the case of being able to
take a full hit on a DB box in production yet still remain running w/o
any data loss. Our application servers are JBoss and will be
high-available clustered / fully-mirrored, but even with RAID on the DB
box one bad thing could take it down, and the data between the hourly
backup would go down with it. We have experimented in-house with C-JDBC
[ being 'lucky' enough to have all DB writes to go through JDBC ], but
would feel more confident w/o involving another service in-between the
application and the DB layers, especially since it is not yet fully
high-available -- currently shifts the single-point of failure from the
DB layer to the CJDBC controller single point. It is reported to have
HA via group communication 'soon', but, you never can tell. Read up on
it at http://c-jdbc.objectweb.org/ , but the end feel I got from it was
not nearly so warm and cozy with the problem being solved at the right
place -- the postgres-r way felt much more robust / speedy.

We won't ever have parallel oracle dollars, but we would have dollars
to bring higher-availability to postgres. 'Cause its our butt on the
line hosting our client's data.

----
James Robinson
Socialserve.com