A lot of it is actually just configuration of their auto-scaling mechanism. If a web front end stops working it starts new instances according to the pool minimums and CPU load. There are some custom scripts though, mostly for migrating data to another zone.