[Parallelizable] hangs after a few tests

[AqlaSolutions]

This test runs 8 cases and then hangs (but still using my cpu). I'm using the latest R# with the latest nunit console runner available at this time.

[TestFixture]
public class NunitParallelizationTest
{
    [Parallelizable(ParallelScope.All)]
    [Test]
    public void Test([Range(1, 10)] int x)
    {

    }
}

Stack trace: https://www.screencast.com/t/k2hZPk95QBN

[CharliePoole]

Is the test literally empty of code, or are you just not showing it.

You can't use R# "with" the console runner, it's either one or the other. It looks like you are using R#, so please give it a try with the console runner.

[AqlaSolutions]

It's an empty test, right. Just minimized the code as much as possible. Yes, the feature doesn't work correctly regardless of test method code.
Ok, checking console runner.
@CharliePoole , checked, NUnit Console Runner 3.7.0 hangs too.

You can't use R# "with" the console runner, it's either one or the other.

You can specify nunit installation directory in R# options and it will use it (may be not exe but dlls from there).

[CharliePoole]

OK thanks for clarifying. Somebody will take a look.

RE R#, we only have vague info on how they call us, so using the console to check is out best course.

[indy-singh]

Hi,

I can confirm I get the same hanging behaviour as described by @AqlaSolutions . nunit-agent-x86.exe just sits there chewing up my CPU doing nothing.

Cheers,
Indy

[ChrisMaddock]

I had a look into this, as it was getting in the way of solving #2428. I think I know what's going on, but I'm struggling a little understanding things. @CharliePoole - does the below seem sensible to you?

Based on:

• In the test above - the fixture is Non-Parallel (NP), all tests are parallel.
• In this case, each test is assigned to a parallel-shift.
• The parallel shifts all run to end, and trigger OnEndOfShift here.
• One of these shifts then dispatches the OneTimeTearDown for the fixture to the NP queue, which is responsible for marking the Fixture WorkItemState.Complete, and allowing the other workers to shut down (i.e. break out of the OnEndOfShift loop)

The issue, I think, is that OnEndOfShift will call RestoreQueues() if there are any levels of isolation, without first checking that all work on all queues has been completed. I think what is happening, is that the fixture's OneTimeTearDown is being dispatched to an NP queue, but then being lost, as RestoreQueues is being called for the end of another Shift. Adding the below seems to allow the shift to run to completion.

OnEndOfShift() Before

            if (_isolationLevel > 0)
                RestoreQueues();

OnEndOfShift() After

            if (_isolationLevel > 0 && Shifts.All(q => !q.HasWork))
                RestoreQueues();

Does this seem sensible? I'm a little mind-blown by queues and shifts right now, I'll be honest. 😆 I'm out of time for today, so thought I'd post for feedback before continuing - but does the above sound sensible?