Sorry, this is a repost from the comments on http://www.howtoforge.com/mysql_master_master_replication this is a smart way of doing it.. master-master. BUT, lets say, master 1 goes down. now users are inserting rows in master 2 for lets say 24 hours, before master 1 is back up. Then MASTER 1, comes back online, now Master 1 is 24 hours behind on logs. If your loadbalancer now notices that master 1 is online, it will now send queries over there, but 24 hours worth of logs will take a while to catch up. so if clients make inserts into MASTER1, there will be conflictions? So how do you deal with this? Do you need to have a smarter loadbalancer that does manual switch back?
I haven't tried master-master replication yet. I'm using the standard master-slave replication. But maybe this link helps you: http://www.mysqlperformanceblog.com/2007/04/05/mysql-master-master-replication-manager-released/
