Andrew Sullivan ajs
Thu Oct 19 07:27:57 PDT 2006
On Thu, Oct 19, 2006 at 01:22:56PM +0200, Largo wrote:
> (DML changes that are commited only WRITTEN on the Master) ==== (if 
> )====> (They are successfully written on the SLAVE) . I know it is 
> complicated but I do not mean to have a SYNCHRONOUS solution.

Right, that's async.

> DML operations are processed ONLY by the Master. A commit is done . And 
> the records will be written in the Master and SLAVE at the same time. So 
> : the processor is MASTER , and data will be saved in Master only if it 
> also saved at the same time or "before" on the SLAVE.

No.  You can do this with Postgres-R, but it's really just a
prototype these days.  Someone has a working version patched forward
to CVS HEAD as of this summer.  But it's not ready for production, by
any means, AFAIK.  Moreover, you can't use READ COMMITTED with it.

What you want is kinda what Oracle's RAC does.  Not precisely, but
close enough to qualify for what you want, I think.  AFAIK, nobody
else in the world has a technology that does quite this.

You could also try using Ingres (which is now free software) on VMS
(which is not, and is no longer available AFAIK); it has a
replication system in it that is close to what you describe, but it
is basically dependent on the tremendous facilities for this that
were built into VMS.  (Why VMS is not being picked up for this sort
of purpose, given the interest in these problems, is probably an
interesting research paper for historians of technology and
economics.)  Ingres's replication doesn't work on UNIX, at least
according to the presentation Computer Associates staff made on this
topic when I was at OSCON a year or two ago.  You could check to see
whether that has changed.

A

-- 
Andrew Sullivan  | ajs at crankycanuck.ca
A certain description of men are for getting out of debt, yet are
against all taxes for raising money to pay it off.
		--Alexander Hamilton



More information about the Slony1-general mailing list