Mongos currently adds a 30 second timeout to all commands it sends to the config servers. In 3.4, however, we are running a lot more on config servers, and some of the commands we run can be performing multiple network and disk i/o operations, so it can sometimes be reasonable for them to take longer than 30 seconds to complete.
We are already seeing sporadic failures in evergreen related to _configsvrAddShard timing out occasionally