We should remove this invariant as the call above to get the currentMillis can go backwards in time, causing us to hit the invariant.
We resolved a similar issue in SERVER-42991 for the ThrottleCursor unit tests.
In my workload where I use the ThrottleCursor I witness that once ~3 minutes the invariant failing because _startMillis is one higher than currentMillis and we only set _startMillis from currentMillis.
- related to
-
SERVER-42991 Throttle Cursor tests should not depend on the systems wall clock
- Closed