The main problem with MM pairs is caused by non-deterministic values generated by the database engine itself, such as autoincrement fields, random numbers and timestamps. The solution to that is that we don’t let the db generate anything. Every value inserted/modified always comes from the application. This allows us to write to either of the two sides of a MM pair knowing it will get replicated to the other side correctly. I’ve heard that MM pairs don’t make sense since you’re executing everything twice. It’s true that you are executing everything twice, but you’re doing it already if you’re using a master-slave(s) setup, and the benefits that come from MM pairs are huge. In addition to giving you fault tolerance and load balancing, they are the key to being able to do non-disruptive, live schema changes.

Two Sides For Salvation « Code as Craft Tuesday, April 15, 2014 @ 10:34pm