Currently we can test failover of replica set to test that application gracefully handle elections etc. I'd like to see a way to do the same for sharded clusters. Maybe this could turn off a random mongos router node at the same time as trigger an election on one of the shards.