ParallelWorkersTest.testParallelWorkersInitFun is flaky

[ezyang]

https://app.circleci.com/jobs/github/pytorch/pytorch/3419913

Oct 31 06:27:59             dummy_worker, init_fun=init_fun
Oct 31 06:27:59         )
Oct 31 06:27:59         worker_coordinator.start()
Oct 31 06:27:59     
Oct 31 06:27:59         for _ in range(10):
Oct 31 06:27:59             value = dequeue_value(queue)
Oct 31 06:27:59             self.assertEqual(
Oct 31 06:27:59                 value, b'initialized', 'Got unexpected value ' + str(value)
Oct 31 06:27:59             )
Oct 31 06:27:59     
Oct 31 06:27:59 >       self.assertTrue(worker_coordinator.stop())
Oct 31 06:27:59 E       AssertionError: False is not true
Oct 31 06:27:59 
Oct 31 06:27:59 ../.local/lib/python3.6/site-packages/caffe2/python/parallel_workers_test.py:98: AssertionError
Oct 31 06:27:59 ----------------------------- Captured stdout call -----------------------------
Oct 31 06:27:59 Wait for workers to die: train
Oct 31 06:27:59 Worker <Thread(parallel_workers worker id 3, started daemon 140220011902720)> failed to close while waiting
Oct 31 06:27:59 All workers terminated: False
Oct 31 06:27:59 - generated xml file: /var/lib/jenkins/workspace/caffe2_tests/python/result.xml -

cc @ezyang @gchanan @zou3519 @jerryzh168

[ezyang]

cc @kevinwilfong (original author of this test) and @akyrola (reviewer of the diff that added the test) in D5538626. How would you feel if we disabled this test?

[akyrola]

I have no objections

[kevinwilfong]

If that assert is the reason it's flaky, it's fine to remove that line, the purpose of the test is really captured in the for loop above. I'm happy to file a diff to make the change.

[ezyang]

@kevinwilfong yes please! I can approve it for you.

[rgommers]

Fix for this was merged in gh-29045, and no new observations since 1 Nov 2019. So closing.