What happens to work for other shards, while you cannot target one shard

Test:

• Set up a two shard cluster
• Run a load of N requests per second that hit each shard through one mongos
• Do a thing:
  • SIGSTOP one shardsvr
  • Trigger an election
• Measure the degradation in throughput and latency for operations that target the other shard. Less than 10% is a decent first target.  Ideally there should be 0 degradation.