Locust: Start on_stop not before on_start has finished

[ThomasSteinbach]

Description of issue / feature request

Wait with running on_stop until on_start has finished. This is because on_stop could depend on variables / statuses set during on_start:

def on_start(self):
    self.myvar = runSomeRequest()

def on_stop(self):
    deleteSomething(self.myvar)

[Edit]

Same case for tasks. The on_stop method would maybe not cleanup resources from the currently running task:

@task
def test(self):
    item = randomString()
    r = self.client.post('...', data = item)
    self.myvar.items.append(item)

def on_stop(self):
    for item in items:
      self.client.delete('...%s' % item)

[cyberw]

I havent validated the behaviour you are seeing, but it definitely looks like a bug, or at least a feature we would want.

[heyman]

This is not supposed to happen within the same Locust I think. If it does happen I'd really like to see a minimal reproducible example.

It's definitely possible that on_stop could be run on Locust User B while the on_start is still executing in Locust User A. That is something that we can'tdo much about since it would requiring synchronization across distributed nodes.

[heyman]

Sorry, I realised what I wrote in the previous message is incorrect. For some reason I thought you meant that on_stop could run before on_start, but I can clearly see that's not what you wrote.

It's correct that with the current implementation we do not wait for on_start to finish when the test is stopped. Instead the Locusts' greenlets are immediately killed which could happen while on_start is still being executed if it's long running, or the test is stopped very quickly after it has been started. I think #1099 might be able to change this behaviour though.

[cyberw]

Yes, this is pretty much solved by #1099 . I added a test case to verify it, and will PR it later (it turns out on_start and on_stop are not tested anywhere...)