When tests are run with parallelism, tasks which do not complete for any reason may not upload test results.
When "re-run failed tests only" is run CircleCI retrieves the failed tests, and re-runs those. It does not re-run tests that are missing due to a failed task. This can result in successful tests being reported and tests from failing tasks being ignored, and subsequently potentially causes PRs with missing tests to be merged.
Implementing logic to check for failed tasks and running those tests along with tests reported to have failed would prevent accidentally merging PRs where tests are missing.
Created by Al Clark