We had a lengthy discussion with engineering about the behavior of mongos on timeout. It appears that the mongos and drivers rely heavily on the maxTimeMS cursor and connection level setting. Drivers do not monitor this and rely on the server/mongos to terminate connections on erratic server behavior.

This should be documented because developers should care (or at least think) about timeout failure situations. Additionally, we should recommend setting the maxTimeMS setting at the driver level when using sharded clusters.

Details here:
https://jira.mongodb.org/browse/HELP-5417