Schema changes on big tables take hours/days depending on their size. You don't do this in production. What you normally do in a MySQL setup is to run a master-

Schema changes on big tables take hours/days depending on their size. You don't do this in production. What you normally do in a MySQL setup is to run a master-master with one active write master. This way you can run optimizations on one master and switch masters without putting your whole production system in jeopardy. Read more about Flickr's setup for more details. There are tons of things that can go wrong in large scale environments (network issues, load balancing, slow code, databases starting to swap, databases starting to use a lot of IO etc.)