The flow_control_replica_set.js test is intended to ensure that, when Flow Control assigns a certain number of tickets in a period of one second, an inserts-based workload executes approximately that number of inserts per second. However, the test is expected to be noisy in its current state, and can fail on a slow machine.
To reduce false positive failures of the test, we should attempt to make it less subject to noise.
One potential solution for this is to perform two calibration runs of the inserts workload. The first calibration run (already present in this test) determines an average rate of inserts without flow control. The second calibration run (to be added in this ticket) runs the same workload where flow control assigns a number of tickets that is equal to the ceiling of the calculated average of the first calibration run. If the rates of writes in both calibration runs are approximately equivalent, the test can proceed. Otherwise, the test run is deemed to be too noisy and it should return early.
Another solution is to move this test into the serial_run suite, which only runs tests serially and therefore reduces the likelihood for noise. Given the comparative ease of this solution, we will pursue it.