We have a proliferation of TaskExecutors across Sharding and Replication which are generally used for their background thread (as well as some limited use around specialized hooks).

It would dramatically simplify the server if these TaskExecutors all used the same NetworkInterface (and shared the same connection pool) where at all possible.

For the thread pool implementation, either a simple unlimited thread pool, or single thread handling callbacks per task executor, would provide much simpler semantics than the current model (re-using the networking thread as the thread pool and requiring a full TaskExecutor + NetworkInterface for each subsystem)