Stateful servers limit load balancing as the server may not be serving requests actively for the current session it is serving, and it cannot be assigned new sessions!

I am a little curious that why not use consistent hashing load balancing strategy in this case. The hashing of sessionId will always redirect to the "nearest" hashing value server. Even if the particular server is down, all request will go to the next "nearest" hashing value server. This will prevent storing session state on each server.

@whitelez I am not very clear about your hashing strategy. If the web servers are created and deleted continuously, how to detect what sessions the newly added server may have processed before without storing the session state?

Is the reason for having sticky sessions to have all the requests for that session be together, and thus they can share information about the session?

Here, every request that a specific user makes is going to the corresponding server. Unfortunately, there's not a good sense of the requests that the users are going to make in the future, so user request imbalance could happen even if sessions are evenly distributed.