3.1.8 introduced majority read concern and therefore also config servers as replica sets. When using config servers as replica sets, a mongos will use majority read concern for all reads of the config servers.

It is easy to imagine a scenario where a DBA connects with the shell to the config server replset, maybe to diagnose some problem that's occuring in the sharded cluster. If the DBA doesn't specifically remember to set the read concern, he will use the default read concern and see a different snapshot of the config database than what the mongoses are actually operating on. If there is some network problem that has prevented majority writes from succeeding for the past few minutes, this difference could be significant.

Proposal is for the shell to recognize when it is connecting to a config server replica set, and automatically set read concern majority for the connection. The shell must print a notification to user about doing this.