On Wed, Dec 13, 2006 at 05:58:11PM +0100, hubert depesz lubaczewski wrote:
> >No, you're _not_ getting good results.  You're getting lucky.  The
> 
> being consistenly lucky on 40+g database with over 600 transactions per
> second - possible, but i doubt i have such luck. and i have did it over 50
> times already.
> 

Uh, that's only 51840000 transactions a day.  And 40 G isn't that
big.  So it's entirely possible that you're consistently lucky.

When we proved this happens, because someone in our office thought
that it ought to be safe, we discovered it by luck: the colliding
data that was overwritten in the place where the added column was
supposed to go happened to be of the wrong datatype.  So the
replication broke.  This happened because we were high enough volume
that we ended up using data that was already in memory, but it didn't
happen right after we made the change.

Honestly, I don't care if you don't believe me that this is
dangerous.  But you can expect me to say "Told you so" if it does
break for you, and you will get exactly no sympathy from me then.

A

-- 
Andrew Sullivan  | ajs at crankycanuck.ca
I remember when computers were frustrating because they *did* exactly what 
you told them to.  That actually seems sort of quaint now.
		--J.D. Baldwin