On Thu, Aug 02, 2007 at 08:12:57PM +0000, Christopher Browne wrote:
> 1.  There isn't any synchronous replication system for PostgreSQL to 
> correspond with Slony-I.  

We don't need one, because you can do it in the application with
two-phase commit.  Of course, 2PC sucks for a large number of cases,
web sites being a fairly obvious case, I'd say.  But you can do it in
fact.

Something else is also important, though, in this discussion.  There
_is_ a proposal for asynchronous no-lag replication, which is
Postgres-R.  Postgres-R guarantees that transactions on every node
see the same data, but they don't do multi-phase commit.  This is
possible because it works in serializable mode only, with
transactions processed on all nodes in network total order.  Given
that situation, you're not going to see the wrong data.  But it ain't
cheap, and it is subject to plenty of rollbacks.  Markus
Schiltknecht says he has patches to make it work with something
recent.  He's looking for sponsorship.  If this is an area that is
important to you, I suggest talking to him.

A

-- 
Andrew Sullivan  | ajs at crankycanuck.ca
The whole tendency of modern prose is away from concreteness.
		--George Orwell