Great idea, but I think it works improperly even after adopting all the changes in the most recent version (at kubeless/events.py at f95f94f9c5813736f1f69e4f8e5d754a275f3677 · kubeless/kubeless · GitHub). Per my tests, after an event is processed, control may be given to another co-routine where the process may pend until an event of that only that one type is received if no events are pending. You may see all events eventually, but many are delayed and they arrive in bursts. Is my assessment wrong? If right, is there any easy fix?