On Wed, Dec 13, 2006 at 02:15:47AM +0100, Andreas Kostyrka wrote:
> I didn't tell that any DDL statement is safe to apply. Actually, it was
> some experimentation, thinking, and some shoot feet (aka broken
> replication cluster that needed dropping nodes and resubscribing).

All the feet you have have not been shot off yet.  

> Warning, test these before you use them in production:

Warning: do not listen to people who tell you to do things that are
known to break.

> 3.) adding columns is always safe, as long rule 1 is fulfilled. E.g.
> adding a column without a default might involve two steps if necessary.

No, it is not.

> 4.) dropping columns seems to be safe completly outside of slony when
> the table has no foreign key constraints.

No, it is not.

> Just to make sure, I've rechecked it on a test setup, adding columns
> seems safe in practice:

As I keep saying (and you keep assuring me is false), just because
that _seems_ safe, it is not.  Its behaviour is _undefined_.  That
means that it might work sometimes, but under other circumstances you
will get surprises.  They'll be hard to explain for you, because you
won't even know what changed.  Then you will complain that "Slony did
something wrong" when in fact it was your (operator) error.

> hard to automate in Slony. It takes some brain usage to transform a
> given set of DDL statements into something that works well with slony.

Ah, I see, _that's_ what the problem is.  I'm too lazy to use my
brain.  Well, ok, now that it's sorted, I guess I'll go back to
sleep.

A

-- 
Andrew Sullivan  | ajs at crankycanuck.ca
The fact that technology doesn't work is no bar to success in the marketplace.
		--Philip Greenspun