Re: replication

On Fri, Sep 22, 2000 at 12:03:13PM -0400, Tom Lane wrote:
> Bruce Guenter <bruceg@em.ca> writes:
> > I've been thinking that one way to emulate replication would be to run a
> > pgsql "shim" that would redirect queries to two back-ends
> > simultaneously, creating transactions for update commands, and aborting
> > when necessary.  The shim would be a long-running gateway process, that
> > could check for data consistency when it started up.  This is all off
> > the top of my head, so is this at all a reasonable idea?
>
> Awhile back, someone reported that they were successfully using exactly
> this idea.  Check the PG list archives (sorry, I forget which list).

Cool!  I'll take a look.

> I'm not convinced that this scales real well to multiple concurrent
> clients ... at the very least, you'd need a single gateway that
> serializes all the requests.

Or multiple gateways, one "near" each database server, that try to keep
some synchronization with each other.

> You might still have problems with the
> two databases not executing requests in exactly the same order.

> Also
> it'd be really dangerous to use OIDs as a way of identifying rows in
> application queries.

Oh, definitely agreed.  OIDs go out the window, unless both servers have
have identical queries delivered to them at all times.  This is rather
difficult to guarantee.
--
Bruce Guenter <bruceg@em.ca>                       http://em.ca/~bruceg/