In some cases, a misbehaving client could open up thousands of connections in a short burst, causing a node to reach maxIncomingConnections and preventing both legitimate client operations and internal health ping and monitoring connections from being established.

I'd like to see a new server-side mechanism that kicks in when maxIncomingConnections is reached, somewhat analogous to a JVM garbage collector. This would identify stalled (or unused) connections and quickly force them closed, relieving connection pressure and allowing normal operations.

This would be necessary on both mongos instances as well as all mongod processes.