This can be annoying to add once there's existing data (since we're using partitioned tables), so we should try to come up with good indices for the initial release.