Currently, only a subset of read operations will be canceled if the external client disconnects, since the results of the operation wouldn't be readable anyways. However, there could be some benefit to cancelling operations that have side effects (e.g. writes) on client disconnect, too. In particular, these operations may hold onto resources for a long time after the client has disconnected, such as the connection count itself, storage engine tickets, etc. Worse, a client may initiate a new connection while the old one is still stuck, leading to further buildup.

This would be a behavioral change however, so we have to consider this carefully.