7661: added test with only_pdh (not working yet)

[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 02dec423d8dc0fdecaf943d9d65501871d4a9645..bbfbe4b7452504ad9935729b1654821bddbd3903 100644 (file)
--- a/services/nodemanager/tests/test_daemon.py
+++ b/services/nodemanager/tests/test_daemon.py
@@ -146,6 +146,31 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin,
         self.stop_proxy(self.daemon)
         self.assertFalse(self.node_setup.start.called)
 
+    def test_excess_counts_missing(self):
+        size = testutil.MockSize(1)
+        cloud_nodes = [testutil.cloud_node_mock(1), testutil.cloud_node_mock(2)]
+        self.make_daemon(cloud_nodes=cloud_nodes,
+                         arvados_nodes=[testutil.arvados_node_mock(1),
+                                        testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
+                         want_sizes=[size])
+        self.assertEqual(2, self.alive_monitor_count())
+        for mon_ref in self.monitor_list():
+            self.daemon.node_can_shutdown(mon_ref.proxy()).get(self.TIMEOUT)
+        self.assertEqual(1, self.node_shutdown.start.call_count)
+
+    def test_missing_shutdown_not_excess(self):
+        size = testutil.MockSize(1)
+        cloud_nodes = [testutil.cloud_node_mock(1), testutil.cloud_node_mock(2)]
+        self.make_daemon(cloud_nodes=cloud_nodes,
+                         arvados_nodes=[testutil.arvados_node_mock(1),
+                                        testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
+                         want_sizes=[size])
+        self.daemon.shutdowns.get()[cloud_nodes[1].id] = True
+        self.assertEqual(2, self.alive_monitor_count())
+        for mon_ref in self.monitor_list():
+            self.daemon.node_can_shutdown(mon_ref.proxy()).get(self.TIMEOUT)
+        self.assertEqual(0, self.node_shutdown.start.call_count)
+
     def test_booting_nodes_counted(self):
         cloud_node = testutil.cloud_node_mock(1)
         arv_node = testutil.arvados_node_mock(1)
@@ -424,6 +449,26 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin,
         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
         self.assertTrue(shutdown_proxy.called)
 
+    def test_broken_node_blackholed_after_cancelled_shutdown(self):
+        cloud_node = testutil.cloud_node_mock(8)
+        wishlist = [testutil.MockSize(8)]
+        self.make_daemon([cloud_node], [testutil.arvados_node_mock(8)],
+                         wishlist)
+        self.assertEqual(1, self.alive_monitor_count())
+        self.assertFalse(self.node_setup.start.called)
+        monitor = self.monitor_list()[0].proxy()
+        shutdown_proxy = self.node_shutdown.start().proxy
+        shutdown_proxy().cloud_node.get.return_value = cloud_node
+        shutdown_proxy().success.get.return_value = False
+        shutdown_proxy().cancel_reason.get.return_value = self.node_shutdown.NODE_BROKEN
+        self.daemon.update_server_wishlist([]).get(self.TIMEOUT)
+        self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
+        self.daemon.node_finished_shutdown(shutdown_proxy()).get(self.TIMEOUT)
+        self.daemon.update_cloud_nodes([cloud_node]).get(self.TIMEOUT)
+        self.daemon.update_server_wishlist(wishlist).get(self.TIMEOUT)
+        self.stop_proxy(self.daemon)
+        self.assertEqual(1, self.node_setup.start.call_count)
+
     def test_nodes_shutting_down_replaced_below_max_nodes(self):
         cloud_node = testutil.cloud_node_mock(6)
         self.make_daemon([cloud_node], [testutil.arvados_node_mock(6)])
@@ -480,3 +525,26 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin,
         self.timer.deliver()
         self.stop_proxy(self.daemon)
         self.assertEqual(1, self.node_setup.start.call_count)
+
+    def test_shutdown_actor_stopped_when_cloud_node_delisted(self):
+        self.make_daemon(cloud_nodes=[testutil.cloud_node_mock()])
+        self.assertEqual(1, self.alive_monitor_count())
+        monitor = self.monitor_list()[0].proxy()
+        self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
+        self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
+        self.stop_proxy(self.daemon)
+        self.assertEqual(
+            1, self.node_shutdown.start().proxy().stop().get.call_count)
+
+    def test_shutdown_actor_cleanup_copes_with_dead_actors(self):
+        self.make_daemon(cloud_nodes=[testutil.cloud_node_mock()])
+        self.assertEqual(1, self.alive_monitor_count())
+        monitor = self.monitor_list()[0].proxy()
+        self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
+        # We're mainly testing that update_cloud_nodes catches and handles
+        # the ActorDeadError.
+        stop_method = self.node_shutdown.start().proxy().stop().get
+        stop_method.side_effect = pykka.ActorDeadError
+        self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
+        self.stop_proxy(self.daemon)
+        self.assertEqual(1, stop_method.call_count)