|
The reason why SERVER-32606 was committed without coverage is that the repro attached is non-deterministic. It triggers the issue most of the time, but not every time, and it needs to run for a while before triggering it. geert.bosch spent some time trying to use failpoints to make the test deterministic, but was not successful. I spoke with max.hirschhorn about this, and he's okay with the idea of pushing tests that only trigger the issue some of the time. He suggests creating a version of the repro that runs for a fixed number of operations (which has been demonstrated to repro the issue with some regularity) then terminates, and adding it to the noPassthrough suite. It would be important to ensure that the test doesn't fail for unrelated reasons (such as oplog rollover).
|