We have decided to turn off this feature until we determine the fate of it.
Failure rate tracking should not be done per minute, but per experiment, since for functions with a small number of invocations it can happen that a relative difference becomes too large, which implies per-function driver termination.
It could lead to non-deterministic experiments in the presence of failures.
We have decided to turn off this feature until we determine the fate of it. Failure rate tracking should not be done per minute, but per experiment, since for functions with a small number of invocations it can happen that a relative difference becomes too large, which implies per-function driver termination.
It could lead to non-deterministic experiments in the presence of failures.