GitLab

We have a project that has a large number of tests. Some of them are marked with RUN_SERIAL.

Some of the normal tests take some time, leading to ctest printing out the ***** WAITING message.

However, when only the RUN_SERIAL tests are left, and it's still executing one of the "normal" tests, it will print out System Load: 1, Max Allowed Load: 9, Smallest test requires 8, if it's executed with -l 9 -j 8 for instance.

Now this is of course not quite correct. It's unable to launch any new tests because the remaining ones are RUN_SERIAL, not because the load is too high.

Looking at cmCTestMultiProcessHandler::StartNextTests(), when it does the check in if (allTestsFailedTestLoadCheck), the logic appears to need a further case, like so:

    bool OnlyRunSerialTestsLeft = ...;
    if (this->SerialTestRunning) {
      cmCTestLog(this->CTest, HANDLER_OUTPUT,
                 "Waiting for RUN_SERIAL test to finish.");
    } else if (OnlyRunSerialTestsLeft)  {
      cmCTestLog(this->CTest, HANDLER_OUTPUT,
                 "Waiting for tests to complete, so RUN_SERIAL test can start.");
    } else {
      /* clang-format off */
      cmCTestLog(this->CTest, HANDLER_OUTPUT,
                 "System Load: " << systemLoad << ", "
                 ...)

I'm not entirely sure how to determine this case, I guess it requires looping over all of the tests.

It makes the load balancing look broken when in fact it's working entirely as expected.

Thanks