8799: shutdown_eligible() returns "node is draining" when in drain state. Add commen...

[arvados.git] / services / nodemanager / tests / test_daemon.py

diff --git a/services/nodemanager/tests/test_daemon.py b/services/nodemanager/tests/test_daemon.py
index e0f051741fbd448cf09451d314a90497fb7994d3..00e05a147ffaf58ea46d345b4b3f4fd84a56a333 100644 (file)
--- a/services/nodemanager/tests/test_daemon.py
+++ b/services/nodemanager/tests/test_daemon.py
@@ -26,6 +26,7 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin,
                                           cloud_size=get_cloud_size,
                                           actor_ref=mock_actor)
         mock_actor.proxy.return_value = mock_proxy
+        mock_actor.tell_proxy.return_value = mock_proxy
 
         self.last_setup = mock_proxy
         return mock_actor
@@ -58,6 +59,7 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin,
         self.cloud_factory().node_start_time.return_value = time.time()
         self.cloud_updates = mock.MagicMock(name='updates_mock')
         self.timer = testutil.MockTimer(deliver_immediately=False)
+        self.cloud_factory().node_id.side_effect = lambda node: node.id
 
         self.node_setup = mock.MagicMock(name='setup_mock')
         self.node_setup.start.side_effect = self.mock_node_start
@@ -122,7 +124,7 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin,
     def test_node_pairing_after_arvados_update(self):
         cloud_node = testutil.cloud_node_mock(2)
         self.make_daemon([cloud_node],
-                         [testutil.arvados_node_mock(2, ip_address=None)])
+                         [testutil.arvados_node_mock(1, ip_address=None)])
         arv_node = testutil.arvados_node_mock(2)
         self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
         self.stop_proxy(self.daemon)
@@ -208,6 +210,7 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin,
         mock_shutdown = self.node_shutdown.start(node_monitor=mock_node_monitor)
 
         self.daemon.shutdowns.get()[cloud_nodes[1].id] = mock_shutdown.proxy()
+        self.daemon.sizes_booting_shutdown.get()[cloud_nodes[1].id] = size
 
         self.assertEqual(2, self.alive_monitor_count())
         for mon_ref in self.monitor_list():
@@ -276,7 +279,7 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin,
         self.last_setup.arvados_node.get.return_value = arv_node
         return self.last_setup
 
-    def test_no_new_node_when_booted_node_not_usable(self):
+    def test_new_node_when_booted_node_not_usable(self):
         cloud_node = testutil.cloud_node_mock(4)
         arv_node = testutil.arvados_node_mock(4, crunch_worker_state='down')
         setup = self.start_node_boot(cloud_node, arv_node)
@@ -287,7 +290,7 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin,
         self.daemon.update_server_wishlist(
             [testutil.MockSize(1)]).get(self.TIMEOUT)
         self.stop_proxy(self.daemon)
-        self.assertEqual(1, self.node_setup.start.call_count)
+        self.assertEqual(2, self.node_setup.start.call_count)
 
     def test_no_duplication_when_booting_node_listed_fast(self):
         # Test that we don't start two ComputeNodeMonitorActors when
@@ -703,10 +706,6 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin,
         booting = self.daemon.booting.get()
         self.stop_proxy(self.daemon)
 
-        # technically something could happen between finishing the busywait and
-        # stopping the daemon so check again
-        self.assertEqual(3, self.node_setup.start.call_count)
-
         sizecounts = {a[0].id: 0 for a in avail_sizes}
         for b in booting.itervalues():
             sizecounts[b.cloud_size.get().id] += 1
@@ -719,3 +718,25 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin,
         # test for that.
         self.assertEqual(2, sizecounts[small.id])
         self.assertEqual(1, sizecounts[big.id])
+
+    @mock.patch("arvnodeman.daemon.NodeManagerDaemonActor._resume_node")
+    def test_resume_drained_nodes(self, resume_node):
+        cloud_node = testutil.cloud_node_mock(1)
+        arv_node = testutil.arvados_node_mock(1, info={"ec2_instance_id": "1", "slurm_state": "down"})
+        self.make_daemon([cloud_node], [arv_node])
+        resume_node.assert_called_with(self.daemon.cloud_nodes.get(self.TIMEOUT).nodes.values()[0])
+        self.stop_proxy(self.daemon)
+
+    @mock.patch("arvnodeman.daemon.NodeManagerDaemonActor._resume_node")
+    def test_no_resume_shutdown_nodes(self, resume_node):
+        cloud_node = testutil.cloud_node_mock(1)
+        arv_node = testutil.arvados_node_mock(1, info={"ec2_instance_id": "1", "slurm_state": "down"})
+
+        self.make_daemon([cloud_node], [])
+
+        self.node_shutdown = mock.MagicMock(name='shutdown_mock')
+        self.daemon.shutdowns.get(self.TIMEOUT)[cloud_node.id] = self.node_shutdown
+
+        self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
+        self.stop_proxy(self.daemon)
+        resume_node.assert_not_called()