[ARVADOS] created: d9325bf35963f7049a2eca281618884484254de4

[Git user]

        at  d9325bf35963f7049a2eca281618884484254de4 (commit)


commit d9325bf35963f7049a2eca281618884484254de4
Author: Lucas Di Pentima <ldipentima at veritasgenetics.com>
Date:   Thu Sep 28 18:09:09 2017 -0300

    12341: If a node_destroy call succesfully destroys the VM but takes too long
    to finish with additional tasks, the cloud_node_update may be called
    in the meantime, removing the node from the cloud_node list. If this happens,
    node_finished_shutdown() won't be able to retrieve the node, and shouldn't
    raise an exception.
    
    Arvados-DCO-1.1-Signed-off-by: Lucas Di Pentima <ldipentima at veritasgenetics.com>

diff --git a/services/nodemanager/arvnodeman/daemon.py b/services/nodemanager/arvnodeman/daemon.py
index e476e5e..170610e 100644
--- a/services/nodemanager/arvnodeman/daemon.py
+++ b/services/nodemanager/arvnodeman/daemon.py
@@ -498,7 +498,14 @@ class NodeManagerDaemonActor(actor_class):
         except pykka.ActorDeadError:
             return
         cloud_node_id = cloud_node.id
-        record = self.cloud_nodes[cloud_node_id]
+
+        try:
+            record = self.cloud_nodes[cloud_node_id]
+        except KeyError:
+            # Cloud node was already removed from the cloud node list
+            # supposedly while the destroy_node call was finishing its
+            # job.
+            return
         shutdown_actor.stop()
         record.shutdown_actor = None
 
diff --git a/services/nodemanager/tests/test_daemon.py b/services/nodemanager/tests/test_daemon.py
index 1efa1ff..ffede75 100644
--- a/services/nodemanager/tests/test_daemon.py
+++ b/services/nodemanager/tests/test_daemon.py
@@ -77,8 +77,9 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin,
 
         self.arv_factory = mock.MagicMock(name='arvados_mock')
         api_client = mock.MagicMock(name='api_client')
-        api_client.nodes().create().execute.side_effect = [testutil.arvados_node_mock(1),
-                                                           testutil.arvados_node_mock(2)]
+        api_client.nodes().create().execute.side_effect = \
+            [testutil.arvados_node_mock(1),
+             testutil.arvados_node_mock(2)]
         self.arv_factory.return_value = api_client
 
         self.cloud_factory = mock.MagicMock(name='cloud_mock')
@@ -400,6 +401,24 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin,
         self.assertTrue(self.node_setup.start.called,
                         "second node not started after booted node stopped")
 
+    def test_booted_node_shut_down_after_being_orphaned(self):
+        cloud_node = testutil.cloud_node_mock(6)
+        setup = self.start_node_boot(cloud_node, id_num=6)
+        self.daemon.node_setup_finished(setup).get(self.TIMEOUT)
+        self.assertEqual(1, self.alive_monitor_count())
+        monitor = self.monitor_list()[0].proxy()
+        self.daemon.update_server_wishlist([])
+        self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
+        self.assertShutdownCancellable(True)
+        shutdown = self.node_shutdown.start().proxy()
+        shutdown.cloud_node.get.return_value = cloud_node
+        # Simulate a successful but slow node destroy call: the cloud node
+        # list gets updated before the ShutdownActor finishes.
+        self.daemon.cloud_nodes.get().nodes.clear()
+        self.daemon.node_finished_shutdown(shutdown).get(self.TIMEOUT)
+        self.assertFalse(shutdown.stop.called,
+                        "shutdown actor shouldn't have been stopped")
+
     def test_booted_node_shut_down_when_never_listed(self):
         setup = self.start_node_boot()
         self.cloud_factory().node_start_time.return_value = time.time() - 3601

-----------------------------------------------------------------------


hooks/post-receive
--