On Fri, Oct 26, 2007 at 04:58:15PM +0100, Simon Riggs wrote:

> We already have a list of the tables we replicate, so why not store the
> list of columns being replicated also? We need only use the list when
> the definitions differ.

It strikes me that this is going to require a buch of extra
maintenance work -- every time you do DDL, you're going also to have
to make sure you get exactly the right order of columns everywhere,
and be able to make sure that doesn't get out of sync in any way.  We
already have problems with cases where we have not relied on the
actual catalogues on the actual nodes (because we fiddled with them),
and they've created all manner of headaches.  Adding another
dependency to some lookaside table that only Slony knows about seems
to me to be a way of addition additional fragility.  Also, it adds
yet more complicated code (are they in the same order?  Ok, do X. 
No?  Then do Y) to a system that is already more complicated than I
think anybody likes.  That wrinkly code will have to be maintained.

So unless this is going to provide a really significant win (on the
order of 20% or more), I can't see the value in chasing it down.  I
think some of the other proposals are more likely to provide similar
benefits without the drawbacks.

-- 
Andrew Sullivan  | ajs at crankycanuck.ca
Everything that happens in the world happens at some place.
		--Jane Jacobs