8799: shutdown_eligible() returns "node is draining" when in drain state. Add commen...
[arvados.git] / services / nodemanager / tests / test_computenode_dispatch.py
index 707cdc66b2c4c111d7c0efdab0d2bcaa1f780c27..14cd85e414ec271f2d8d1a3a9c283ac3c724fe7c 100644 (file)
@@ -9,6 +9,7 @@ import arvados.errors as arverror
 import httplib2
 import mock
 import pykka
+import threading
 
 import arvnodeman.computenode.dispatch as dispatch
 from . import testutil
@@ -27,24 +28,44 @@ class ComputeNodeSetupActorTestCase(testutil.ActorTestMixin, unittest.TestCase):
 
     def make_actor(self, arv_node=None):
         if not hasattr(self, 'timer'):
-            self.make_mocks(arvados_effect=[arv_node])
+            self.make_mocks(arvados_effect=[arv_node] if arv_node else None)
         self.setup_actor = dispatch.ComputeNodeSetupActor.start(
             self.timer, self.api_client, self.cloud_client,
             testutil.MockSize(1), arv_node).proxy()
 
+    def assert_node_properties_updated(self, uuid=None,
+                                       size=testutil.MockSize(1)):
+        self.api_client.nodes().update.assert_any_call(
+            uuid=(uuid or self.arvados_effect[-1]['uuid']),
+            body={
+                'properties': {
+                    'cloud_node': {
+                        'size': size.id,
+                        'price': size.price}}})
+
     def test_creation_without_arvados_node(self):
         self.make_actor()
+        finished = threading.Event()
+        self.setup_actor.subscribe(lambda _: finished.set())
         self.assertEqual(self.arvados_effect[-1],
                          self.setup_actor.arvados_node.get(self.TIMEOUT))
-        self.assertTrue(self.api_client.nodes().create().execute.called)
+        assert(finished.wait(self.TIMEOUT))
+        self.assertEqual(1, self.api_client.nodes().create().execute.call_count)
+        self.assertEqual(1, self.api_client.nodes().update().execute.call_count)
+        self.assert_node_properties_updated()
         self.assertEqual(self.cloud_client.create_node(),
                          self.setup_actor.cloud_node.get(self.TIMEOUT))
 
     def test_creation_with_arvados_node(self):
+        self.make_mocks(arvados_effect=[testutil.arvados_node_mock()]*2)
         self.make_actor(testutil.arvados_node_mock())
+        finished = threading.Event()
+        self.setup_actor.subscribe(lambda _: finished.set())
         self.assertEqual(self.arvados_effect[-1],
                          self.setup_actor.arvados_node.get(self.TIMEOUT))
-        self.assertTrue(self.api_client.nodes().update().execute.called)
+        assert(finished.wait(self.TIMEOUT))
+        self.assert_node_properties_updated()
+        self.assertEqual(2, self.api_client.nodes().update().execute.call_count)
         self.assertEqual(self.cloud_client.create_node(),
                          self.setup_actor.cloud_node.get(self.TIMEOUT))
 
@@ -98,8 +119,9 @@ class ComputeNodeSetupActorTestCase(testutil.ActorTestMixin, unittest.TestCase):
         self.make_actor()
         subscriber = mock.Mock(name='subscriber_mock')
         self.setup_actor.subscribe(subscriber)
-        self.api_client.nodes().create().execute.side_effect = [
-            testutil.arvados_node_mock()]
+        retry_resp = [testutil.arvados_node_mock()]
+        self.api_client.nodes().create().execute.side_effect = retry_resp
+        self.api_client.nodes().update().execute.side_effect = retry_resp
         self.wait_for_assignment(self.setup_actor, 'cloud_node')
         self.assertEqual(self.setup_actor.actor_ref.actor_urn,
                          subscriber.call_args[0][0].actor_ref.actor_urn)
@@ -116,11 +138,12 @@ class ComputeNodeSetupActorTestCase(testutil.ActorTestMixin, unittest.TestCase):
 
 class ComputeNodeShutdownActorMixin(testutil.ActorTestMixin):
     def make_mocks(self, cloud_node=None, arvados_node=None,
-                   shutdown_open=True):
+                   shutdown_open=True, node_broken=False):
         self.timer = testutil.MockTimer()
         self.shutdowns = testutil.MockShutdownTimer()
         self.shutdowns._set_state(shutdown_open, 300)
         self.cloud_client = mock.MagicMock(name='cloud_client')
+        self.cloud_client.broken.return_value = node_broken
         self.arvados_client = mock.MagicMock(name='arvados_client')
         self.updates = mock.MagicMock(name='update_mock')
         if cloud_node is None:
@@ -201,6 +224,8 @@ class ComputeNodeShutdownActorTestCase(ComputeNodeShutdownActorMixin,
         self.make_actor()
         self.check_success_flag(False, 2)
         self.assertFalse(self.cloud_client.destroy_node.called)
+        self.assertEqual(self.ACTOR_CLASS.WINDOW_CLOSED,
+                         self.shutdown_actor.cancel_reason.get(self.TIMEOUT))
 
     def test_shutdown_retries_when_cloud_fails(self):
         self.make_mocks()
@@ -210,6 +235,15 @@ class ComputeNodeShutdownActorTestCase(ComputeNodeShutdownActorMixin,
         self.cloud_client.destroy_node.return_value = True
         self.check_success_flag(True)
 
+    def test_shutdown_cancelled_when_cloud_fails_on_broken_node(self):
+        self.make_mocks(node_broken=True)
+        self.cloud_client.destroy_node.return_value = False
+        self.make_actor(start_time=0)
+        self.check_success_flag(False, 2)
+        self.assertEqual(1, self.cloud_client.destroy_node.call_count)
+        self.assertEqual(self.ACTOR_CLASS.NODE_BROKEN,
+                         self.shutdown_actor.cancel_reason.get(self.TIMEOUT))
+
     def test_late_subscribe(self):
         self.make_actor()
         subscriber = mock.Mock(name='subscriber_mock')
@@ -233,6 +267,16 @@ class ComputeNodeUpdateActorTestCase(testutil.ActorTestMixin,
         self.updater.sync_node(cloud_node, arv_node).get(self.TIMEOUT)
         self.driver().sync_node.assert_called_with(cloud_node, arv_node)
 
+    @testutil.no_sleep
+    def test_node_sync_error(self):
+        self.make_actor()
+        cloud_node = testutil.cloud_node_mock()
+        arv_node = testutil.arvados_node_mock()
+        self.driver().sync_node.side_effect = (IOError, Exception, True)
+        self.updater.sync_node(cloud_node, arv_node).get(self.TIMEOUT)
+        self.updater.sync_node(cloud_node, arv_node).get(self.TIMEOUT)
+        self.updater.sync_node(cloud_node, arv_node).get(self.TIMEOUT)
+        self.driver().sync_node.assert_called_with(cloud_node, arv_node)
 
 class ComputeNodeMonitorActorTestCase(testutil.ActorTestMixin,
                                       unittest.TestCase):
@@ -244,6 +288,7 @@ class ComputeNodeMonitorActorTestCase(testutil.ActorTestMixin,
         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'):
@@ -311,64 +356,63 @@ class ComputeNodeMonitorActorTestCase(testutil.ActorTestMixin,
     def test_no_shutdown_booting(self):
         self.make_actor()
         self.shutdowns._set_state(True, 600)
-        self.assertFalse(self.node_actor.shutdown_eligible().get(self.TIMEOUT))
+        self.assertTrue(self.node_actor.shutdown_eligible().get(self.TIMEOUT).startswith("node is still booting"))
 
     def test_shutdown_without_arvados_node(self):
         self.make_actor(start_time=0)
         self.shutdowns._set_state(True, 600)
-        self.assertTrue(self.node_actor.shutdown_eligible().get(self.TIMEOUT))
+        self.assertIs(True, self.node_actor.shutdown_eligible().get(self.TIMEOUT))
 
     def test_no_shutdown_missing(self):
         arv_node = testutil.arvados_node_mock(10, job_uuid=None,
                                               crunch_worker_state="down",
-                                              status="missing")
+                                              last_ping_at='1970-01-01T01:02:03.04050607Z')
         self.make_actor(10, arv_node)
         self.shutdowns._set_state(True, 600)
-        self.cloud_client.broken.return_value = False
-        self.assertFalse(self.node_actor.shutdown_eligible().get(self.TIMEOUT))
+        self.assertTrue(self.node_actor.shutdown_eligible().get(self.TIMEOUT).startswith("node is not idle."))
 
     def test_no_shutdown_running_broken(self):
         arv_node = testutil.arvados_node_mock(12, job_uuid=None,
-                                              crunch_worker_state="down",
-                                              status="running")
+                                              crunch_worker_state="down")
         self.make_actor(12, arv_node)
         self.shutdowns._set_state(True, 600)
         self.cloud_client.broken.return_value = True
-        self.assertFalse(self.node_actor.shutdown_eligible().get(self.TIMEOUT))
+        self.assertTrue(self.node_actor.shutdown_eligible().get(self.TIMEOUT).startswith("node is not idle."))
 
     def test_shutdown_missing_broken(self):
         arv_node = testutil.arvados_node_mock(11, job_uuid=None,
                                               crunch_worker_state="down",
-                                              status="missing")
+                                              last_ping_at='1970-01-01T01:02:03.04050607Z')
         self.make_actor(11, arv_node)
         self.shutdowns._set_state(True, 600)
         self.cloud_client.broken.return_value = True
-        self.assertTrue(self.node_actor.shutdown_eligible().get(self.TIMEOUT))
+        self.assertIs(True, self.node_actor.shutdown_eligible().get(self.TIMEOUT))
 
     def test_no_shutdown_when_window_closed(self):
         self.make_actor(3, testutil.arvados_node_mock(3, job_uuid=None))
-        self.assertFalse(self.node_actor.shutdown_eligible().get(self.TIMEOUT))
+        self.assertTrue(self.node_actor.shutdown_eligible().get(self.TIMEOUT).startswith("shutdown window is not open."))
 
     def test_no_shutdown_when_node_running_job(self):
         self.make_actor(4, testutil.arvados_node_mock(4, job_uuid=True))
         self.shutdowns._set_state(True, 600)
-        self.assertFalse(self.node_actor.shutdown_eligible().get(self.TIMEOUT))
+        self.assertTrue(self.node_actor.shutdown_eligible().get(self.TIMEOUT).startswith("node is not idle."))
 
     def test_no_shutdown_when_node_state_unknown(self):
         self.make_actor(5, testutil.arvados_node_mock(
             5, crunch_worker_state=None))
         self.shutdowns._set_state(True, 600)
-        self.assertFalse(self.node_actor.shutdown_eligible().get(self.TIMEOUT))
+        self.assertTrue(self.node_actor.shutdown_eligible().get(self.TIMEOUT).startswith("node is not idle."))
 
     def test_no_shutdown_when_node_state_stale(self):
         self.make_actor(6, testutil.arvados_node_mock(6, age=90000))
         self.shutdowns._set_state(True, 600)
-        self.assertFalse(self.node_actor.shutdown_eligible().get(self.TIMEOUT))
+        self.assertTrue(self.node_actor.shutdown_eligible().get(self.TIMEOUT).startswith("node is not idle."))
 
     def test_arvados_node_match(self):
         self.make_actor(2)
         arv_node = testutil.arvados_node_mock(
             2, hostname='compute-two.zzzzz.arvadosapi.com')
+        self.cloud_client.node_id.return_value = '2'
         pair_id = self.node_actor.offer_arvados_pair(arv_node).get(self.TIMEOUT)
         self.assertEqual(self.cloud_mock.id, pair_id)
         self.stop_proxy(self.node_actor)