I'm not entirely sure I understand this paper. Are the frontend servers hitting the auto shard service with say, a key, and it returns a list of backend server

I'm not entirely sure I understand this paper. Are the frontend servers hitting the auto shard service with say, a key, and it returns a list of backend server(s) to hit where it knows that key is stored in memory? Wouldn't the auto sharding service come with the usual challenges of HA, load balancing, integrity etc?