11925: Fix flaky tests in test_computenode_dispatch_slurm.
[arvados.git] / services / nodemanager / tests / test_computenode_dispatch_slurm.py
index c7eb7afc631cd2b06e2c7e6753b2e696ab27b520..d00425e57e55109fa0b6d0e8536f442d228d39ac 100644 (file)
@@ -32,13 +32,20 @@ class SLURMComputeNodeShutdownActorTestCase(ComputeNodeShutdownActorMixin,
             self.timer = testutil.MockTimer(False)
         self.make_actor()
         self.check_success_flag(None, 0)
+        # At this point, 1st try should have happened.
+
         self.timer.deliver()
         self.check_success_flag(None, 0)
-        self.timer.deliver()
+        # At this point, 2nd try should have happened.
+
         # Order is critical here: if the mock gets called when no return value
         # or side effect is set, we may invoke a real subprocess.
         proc_mock.return_value = end_state
         proc_mock.side_effect = None
+
+        # 3rd try
+        self.timer.deliver()
+
         self.check_success_flag(True, 3)
         self.check_slurm_got_args(proc_mock, 'NodeName=compute63')
 
@@ -73,8 +80,9 @@ class SLURMComputeNodeShutdownActorTestCase(ComputeNodeShutdownActorMixin,
             self.make_mocks(arvados_node=testutil.arvados_node_mock(job_uuid=True))
             self.timer = testutil.MockTimer(False)
             self.make_actor()
+            self.shutdown_actor.ping()
             self.busywait(lambda: proc_mock.call_args is not None)
-            self.shutdown_actor.cancel_shutdown("test").get(self.TIMEOUT)
+            self.shutdown_actor.cancel_shutdown("test")
             self.check_success_flag(False, 2)
             self.assertEqual(proc_mock.call_args_list,
                              [mock.call(['scontrol', 'update', 'NodeName=compute99', 'State=DRAIN', 'Reason=Node Manager shutdown']),
@@ -88,10 +96,10 @@ class SLURMComputeNodeShutdownActorTestCase(ComputeNodeShutdownActorMixin,
         proc_mock.side_effect = iter([OSError, 'drain\n', OSError, 'idle\n', 'idle\n'])
         self.make_mocks(arvados_node=testutil.arvados_node_mock(job_uuid=True))
         self.make_actor()
-        self.check_success_flag(False, 2)
+        self.check_success_flag(False, 5)
 
     def test_issue_slurm_drain_retry(self, proc_mock):
-        proc_mock.side_effect = iter([OSError, '', OSError, 'drng\n'])
+        proc_mock.side_effect = iter([OSError, OSError, 'drng\n', 'drain\n'])
         self.check_success_after_reset(proc_mock, timer=False)
 
     def test_arvados_node_cleaned_after_shutdown(self, proc_mock):