|
max.hirschhorn@mongodb.com Parallel testing was not required to reproduce the bug described in SERVER-65814. It was a coincidence that while doing parallel testing we created a plan cache entry with certain properties which we were able to reuse in another test, but due to a programming error the second test would blow up. It's interesting that we didn't create a standalone test for this issue, perhaps the thinking was that we had several BFs in the parallel suite with the same symptom and existing test coverage would be enough.
Similarly, parallel testing wasn't required to cause or reproduce the bug described in SERVER-65897, and an existing test was extended to cover this scenario.
I can't say that we were heavily relying on the parallel suite to test the SBE plan cache, but it was definitely of some interest for us, and luckily we revealed a couple real issues which otherwise could be caught only by the fuzzer. However, I tend to agree that all parallel testing should be planned, designed and implemented using the concurrency suites and be based on well-thought scenarios rather than on randomness and sheer luck (for that we have the fuzzer).
|