Fix race conditions in test_node_undrained_when_shutdown_cancelled
[arvados.git] / services / nodemanager / tests / test_computenode_dispatch_slurm.py
index 212bb3ded3af32a5c51c598905e92804403020c0..85a40ceeb25964e15e3696141b373d9ef87cee4d 100644 (file)
@@ -41,11 +41,11 @@ class SLURMComputeNodeShutdownActorTestCase(ComputeNodeShutdownActorMixin,
             self.check_success_after_reset(proc_mock, end_state)
         return test
 
-    for wait_state in ['alloc\n', 'drng\n', 'idle*\n']:
+    for wait_state in ['alloc\n', 'drng\n']:
         locals()['test_wait_while_' + wait_state.strip()
                  ] = make_wait_state_test(start_state=wait_state)
 
-    for end_state in ['down\n', 'down*\n', 'drain\n', 'fail\n']:
+    for end_state in ['idle*\n', 'down\n', 'down*\n', 'drain\n', 'fail\n']:
         locals()['test_wait_until_' + end_state.strip()
                  ] = make_wait_state_test(end_state=end_state)
 
@@ -55,27 +55,30 @@ class SLURMComputeNodeShutdownActorTestCase(ComputeNodeShutdownActorMixin,
 
     def test_slurm_bypassed_when_no_arvados_node(self, proc_mock):
         # Test we correctly handle a node that failed to bootstrap.
-        proc_mock.return_value = 'idle\n'
+        proc_mock.return_value = 'down\n'
         self.make_actor(start_time=0)
         self.check_success_flag(True)
         self.assertFalse(proc_mock.called)
 
-    def test_node_undrained_when_shutdown_window_closes(self, proc_mock):
-        proc_mock.side_effect = iter(['drng\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_slurm_got_args(proc_mock, 'NodeName=compute99', 'State=RESUME')
-
-    def test_alloc_node_undrained_when_shutdown_window_closes(self, proc_mock):
-        proc_mock.side_effect = iter(['alloc\n'])
-        self.make_mocks(arvados_node=testutil.arvados_node_mock(job_uuid=True))
-        self.make_actor()
-        self.check_success_flag(False, 2)
-        self.check_slurm_got_args(proc_mock, 'sinfo', '--noheader', '-o', '%t', '-n', 'compute99')
+    def test_node_undrained_when_shutdown_cancelled(self, proc_mock):
+        try:
+            proc_mock.side_effect = iter(['', 'drng\n', 'drng\n', ''])
+            self.make_mocks(arvados_node=testutil.arvados_node_mock(job_uuid=True))
+            self.timer = testutil.MockTimer(False)
+            self.make_actor()
+            self.busywait(lambda: proc_mock.call_args is not None)
+            self.shutdown_actor.cancel_shutdown("test").get(self.TIMEOUT)
+            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']),
+                              mock.call(['sinfo', '--noheader', '-o', '%t', '-n', 'compute99']),
+                              mock.call(['sinfo', '--noheader', '-o', '%t', '-n', 'compute99']),
+                              mock.call(['scontrol', 'update', 'NodeName=compute99', 'State=RESUME'])])
+        finally:
+            self.shutdown_actor.actor_ref.stop()
 
     def test_cancel_shutdown_retry(self, proc_mock):
-        proc_mock.side_effect = iter([OSError, 'drain\n', OSError, 'idle\n'])
+        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)
@@ -88,75 +91,3 @@ class SLURMComputeNodeShutdownActorTestCase(ComputeNodeShutdownActorMixin,
         proc_mock.return_value = 'drain\n'
         super(SLURMComputeNodeShutdownActorTestCase,
               self).test_arvados_node_cleaned_after_shutdown()
-
-class SLURMComputeNodeMonitorActorTestCase(testutil.ActorTestMixin,
-                                      unittest.TestCase):
-
-    def make_mocks(self, node_num):
-        self.shutdowns = testutil.MockShutdownTimer()
-        self.shutdowns._set_state(False, 300)
-        self.timer = mock.MagicMock(name='timer_mock')
-        self.updates = mock.MagicMock(name='update_mock')
-        self.cloud_mock = testutil.cloud_node_mock(node_num)
-        self.subscriber = mock.Mock(name='subscriber_mock')
-        self.cloud_client = mock.MagicMock(name='cloud_client')
-        self.cloud_client.broken.return_value = False
-
-    def make_actor(self, node_num=1, arv_node=None, start_time=None):
-        if not hasattr(self, 'cloud_mock'):
-            self.make_mocks(node_num)
-        if start_time is None:
-            start_time = time.time()
-        self.node_actor = slurm_dispatch.ComputeNodeMonitorActor.start(
-            self.cloud_mock, start_time, self.shutdowns,
-            testutil.cloud_node_fqdn, self.timer, self.updates, self.cloud_client,
-            arv_node, boot_fail_after=300).proxy()
-        self.node_actor.subscribe(self.subscriber).get(self.TIMEOUT)
-
-    @mock.patch("subprocess.check_output")
-    def test_resume_node(self, check_output):
-        arv_node = testutil.arvados_node_mock()
-        self.make_actor(arv_node=arv_node)
-        check_output.return_value = "drain\n"
-        self.node_actor.resume_node().get(self.TIMEOUT)
-        check_output.assert_has_calls([mock.call(['sinfo', '--noheader', '-o', '%t', '-n', arv_node['hostname']]),
-                                       mock.call(['sinfo', '--noheader', '-o', '%t', '-n', arv_node['hostname']]),
-                                       mock.call(['sinfo', '--noheader', '-o', '%t', '-n', arv_node['hostname']]),
-                                       mock.call(['scontrol', 'update', 'NodeName=' + arv_node['hostname'], 'State=RESUME'])],
-                                      any_order=True)
-        self.assertEqual(4, check_output.call_count)
-
-    @mock.patch("subprocess.check_output")
-    def test_no_resume_idle_node(self, check_output):
-        arv_node = testutil.arvados_node_mock()
-        self.make_actor(arv_node=arv_node)
-        check_output.return_value = "idle\n"
-        self.node_actor.resume_node().get(self.TIMEOUT)
-        check_output.assert_has_calls([mock.call(['sinfo', '--noheader', '-o', '%t', '-n', arv_node['hostname']]),
-                                       mock.call(['sinfo', '--noheader', '-o', '%t', '-n', arv_node['hostname']]),
-                                       mock.call(['sinfo', '--noheader', '-o', '%t', '-n', arv_node['hostname']])],
-                                      any_order=True)
-        self.assertEqual(3, check_output.call_count)
-
-
-    @mock.patch("subprocess.check_output")
-    def test_resume_node_exception(self, check_output):
-        arv_node = testutil.arvados_node_mock()
-        self.make_actor(arv_node=arv_node)
-        check_output.side_effect = Exception()
-        check_output.return_value = "drain\n"
-        self.node_actor.resume_node().get(self.TIMEOUT)
-        check_output.assert_has_calls([mock.call(['sinfo', '--noheader', '-o', '%t', '-n', arv_node['hostname']]),
-                                       mock.call(['sinfo', '--noheader', '-o', '%t', '-n', arv_node['hostname']]),
-                                       mock.call(['sinfo', '--noheader', '-o', '%t', '-n', arv_node['hostname']])],
-                                      any_order=True)
-        self.assertEqual(3, check_output.call_count)
-
-
-    @mock.patch("subprocess.check_output")
-    def test_shutdown_down_node(self, check_output):
-        check_output.return_value = "down\n"
-        self.make_actor()
-        self.assertEquals('shutdown window is not open.', self.node_actor.shutdown_eligible().get(self.TIMEOUT))
-        self.shutdowns._set_state(True, 600)
-        self.assertTrue(self.node_actor.shutdown_eligible().get(self.TIMEOUT))