No mention of sharding by usage pattern, which is the usual pattern at scale, eg partition potentially app and database differently for users with different fanout or scale or other properties that affect scaling.