On Wed, Dec 13, 2006 at 06:11:31PM +0100, hubert depesz lubaczewski wrote:
> sorry but making the column with wrong type is by no way safer with execute
> - it's just a problem with called sql. i said that i add the same column to
> both nodes - same type.  same name. same table.

I suppose I wasn't clear enough: the data that got shipped was
_random data from memory_.  We happened to get lucky that there was a
datatype mismatch, or we just wouldn't have found out what we'd done,
which would have meant silent corruption of the replica.  It was
that risk of silent corruption that caused Jan to change the way he
was doing this, so that if someone did it again, it'd be guaranteed
to break instead of corrupting the data.  But that's the only
guarantee, so doing things the way you're doing it will not
automatically cause you problems (at least right away).

> p.s. you seem to be particularu upset about my question - why? i am
> really not implying that "execute" is bad. i am just asking to know
> my limitations (and reasons of them).

I guess I'm getting annoyed about having to keep saying "It's not
working like you think it is, and here's why."  But since somebody
already established in this thread that it's my lack of brain work
that's the issue, perhaps I'll shut up and let people do things the
unsafe way.  The reason it bothers me to do that, though, is that I
know later someone is going to complain that their screwed-up replica
is Slony's fault, when what's actually to blame is that individual
refusing to read the warning written in 46-point Bodoni bold italic
coloured red, with the <blink> tag on, saying "Don't do that."

A

-- 
Andrew Sullivan  | ajs at crankycanuck.ca
When my information changes, I alter my conclusions.  What do you do sir?
		--attr. John Maynard Keynes