router_transaction_metrics.js assumes for every commit type that if a commit was successful, the corresponding successfulDurationMicros metric will be greater than 0. The no shards commit type doesn't require any network messages so it's possible to complete in less than a microsecond, which fails this test.

To avoid failures, the test should relax this assertion for just the no shards commit type. The other commit types all require at least one network message, so it should be safe to keep this assumption for them.