These fields in query stats can have misleading names depending on context:

• lastExecutionMicros
• totalExecMicros
  For queries, which are executed one at a time, these names are fine. But for writes performed on the router, we may be executing a batch of write ops, even though we collect query stats for single write ops. We don't have any way of isolating the time for any individual op on the router.

With writes on the router, it makes more sense to use the term "latency." We are reporting the time it takes from the router receiving a batched request to the time it takes when it returns a response. However, because we subtract out execution time from the duration, it may also be misleading. We should come up with a better solution here.

We might consider including both latency (with pauses of execution not subtracted) and execution time (which subtracts pauses in execution).