Though our example used only two schema shards, this process scales well to any number of preallocated shards. It's surprisingly easy to relocate schemas using the method described here, and there's no reason we must limit ourselves to splitting one server into only two. The only real limitation is that we can't effectively recombine servers once they've been split this way.
There is, however, one important caveat we must explain. This type of database sharding works best when the application is designed to accommodate it. In fact, it's even better to create all of the logical shards upfront, before data is inserted into any shard. Why is this?
Consider an existing schema with existing data. Foreign keys, customers, and customer ...
