On Sat, Jun 02, 2007 at 01:30:53AM +0200, Alexander Staubo wrote: > There needs to be a point of synchronization when a DDL transaction > appears that blocks further write transactions from running. As far as > I can tell, the slaves themselves can continue to receive pending > events, but perhaps not. In order to do it automatically, you have to lock everyone, get all the events through, and then perform the DDL, and then come out of lock. Otherwise, what happens when you do DROP COLUMN? If it goes through ahead of data that ought to go into that column, you have just broken your cluster. I suppose you could figure out a way to work around this, but pretty soon you are building an artificial intelligence expert system with event-predicting capabilities. Such systems are not well known for their simplicity and ease of maintenance. > Last I checked, nobody was actually terribly *happy* about having to > pipe schema changes through slonik. Nobody would suggest it's the friendliest arrangement. But this is a field where the details really count, and therefore proposals to make it more friendly have to account for how that friendliness in a lot of cases doesn't lead to complete breakage in others. (I had to be exposed to the multimaster MS SQL stuff, years ago, and I have to say that it was great when it worked; but when things went south, boy did your life suck. Whether it is better now, I don't know.) A -- Andrew Sullivan | ajs@xxxxxxxxxxxxxxx Unfortunately reformatting the Internet is a little more painful than reformatting your hard drive when it gets out of whack. --Scott Morris
