For the most part, schema changes can be done without breaking the existing code base (add columns, add tables, etc.) These schema changes can just be run on the databases before the code depending on it is pushed. Much more rarely though, there might be code that depends on a certain schema (the kind that say, alter table might do). In this case we can push out code that is compatible with both versions (at the very worst case, using table descriptions to differentiate between the two), run the schema change, and then eliminate the backwards compatibility.

There's not that much different in terms of operating procedure between continuous code deployment and scheduled code deployments; it's just much faster and less prone to merge errors.