X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/55b017847fe87e00e7fb9c6bfd0444b83f8ca12c..2ab1b540a710fd454f33e9ba22db601e67f3d217:/services/nodemanager/tests/test_daemon.py diff --git a/services/nodemanager/tests/test_daemon.py b/services/nodemanager/tests/test_daemon.py index 00e05a147f..f1d168c1e6 100644 --- a/services/nodemanager/tests/test_daemon.py +++ b/services/nodemanager/tests/test_daemon.py @@ -48,18 +48,31 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin, return mock_actor def make_daemon(self, cloud_nodes=[], arvados_nodes=[], want_sizes=[], - avail_sizes=[(testutil.MockSize(1), {"cores": 1})], + avail_sizes=None, min_nodes=0, max_nodes=8, shutdown_windows=[54, 5, 1], max_total_price=None): for name in ['cloud_nodes', 'arvados_nodes', 'server_wishlist']: setattr(self, name + '_poller', mock.MagicMock(name=name + '_mock')) + + if not avail_sizes: + if cloud_nodes or want_sizes: + avail_sizes=[(c.size, {"cores": int(c.id)}) for c in cloud_nodes] + [(s, {"cores": 1}) for s in want_sizes] + else: + avail_sizes=[(testutil.MockSize(1), {"cores": 1})] + 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)] + self.arv_factory.return_value = api_client + self.cloud_factory = mock.MagicMock(name='cloud_mock') 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.cloud_factory().broken.return_value = False self.node_setup = mock.MagicMock(name='setup_mock') self.node_setup.start.side_effect = self.mock_node_start @@ -76,10 +89,10 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin, min_nodes, max_nodes, 600, 1800, 3600, self.node_setup, self.node_shutdown, max_total_price=max_total_price).proxy() - if cloud_nodes is not None: - self.daemon.update_cloud_nodes(cloud_nodes).get(self.TIMEOUT) if arvados_nodes is not None: self.daemon.update_arvados_nodes(arvados_nodes).get(self.TIMEOUT) + if cloud_nodes is not None: + self.daemon.update_cloud_nodes(cloud_nodes).get(self.TIMEOUT) if want_sizes is not None: self.daemon.update_server_wishlist(want_sizes).get(self.TIMEOUT) @@ -133,7 +146,8 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin, def test_arvados_node_un_and_re_paired(self): # We need to create the Arvados node mock after spinning up the daemon # to make sure it's new enough to pair with the cloud node. - self.make_daemon([testutil.cloud_node_mock(3)], arvados_nodes=None) + self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(3)], + arvados_nodes=None) arv_node = testutil.arvados_node_mock(3) self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT) self.check_monitors_arvados_nodes(arv_node) @@ -146,7 +160,8 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin, def test_old_arvados_node_not_double_assigned(self): arv_node = testutil.arvados_node_mock(3, age=9000) size = testutil.MockSize(3) - self.make_daemon(arvados_nodes=[arv_node], avail_sizes=[(size, {"cores":1})]) + self.make_daemon(arvados_nodes=[arv_node], + avail_sizes=[(size, {"cores":1})]) self.daemon.update_server_wishlist([size]).get(self.TIMEOUT) self.daemon.update_server_wishlist([size, size]).get(self.TIMEOUT) self.stop_proxy(self.daemon) @@ -157,25 +172,27 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin, self.assertIn(None, used_nodes) def test_node_count_satisfied(self): - self.make_daemon([testutil.cloud_node_mock()], + self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(1)], want_sizes=[testutil.MockSize(1)]) self.stop_proxy(self.daemon) self.assertFalse(self.node_setup.start.called) def test_dont_count_missing_as_busy(self): size = testutil.MockSize(1) - self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(1), - testutil.cloud_node_mock(2)], + self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(1, size=size), + testutil.cloud_node_mock(2, size=size)], arvados_nodes=[testutil.arvados_node_mock(1), - testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')], + testutil.arvados_node_mock( + 2, + last_ping_at='1970-01-01T01:02:03.04050607Z')], want_sizes=[size, size]) self.stop_proxy(self.daemon) self.assertTrue(self.node_setup.start.called) def test_missing_counts_towards_max(self): size = testutil.MockSize(1) - self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(1), - testutil.cloud_node_mock(2)], + self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(1, size=size), + testutil.cloud_node_mock(2, size=size)], arvados_nodes=[testutil.arvados_node_mock(1), testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')], want_sizes=[size, size], @@ -185,7 +202,7 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin, def test_excess_counts_missing(self): size = testutil.MockSize(1) - cloud_nodes = [testutil.cloud_node_mock(1), testutil.cloud_node_mock(2)] + cloud_nodes = [testutil.cloud_node_mock(1, size=size), testutil.cloud_node_mock(2, size=size)] 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')], @@ -197,7 +214,7 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin, def test_missing_shutdown_not_excess(self): size = testutil.MockSize(1) - cloud_nodes = [testutil.cloud_node_mock(1), testutil.cloud_node_mock(2)] + cloud_nodes = [testutil.cloud_node_mock(1, size=size), testutil.cloud_node_mock(2, size=size)] 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')], @@ -209,8 +226,7 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin, mock_node_monitor.proxy.return_value = mock.NonCallableMock(cloud_node=get_cloud_node) 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.daemon.cloud_nodes.get()[cloud_nodes[1].id].shutdown_actor = mock_shutdown.proxy() self.assertEqual(2, self.alive_monitor_count()) for mon_ref in self.monitor_list(): @@ -233,6 +249,7 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin, arv_node = testutil.arvados_node_mock(2, job_uuid=True) self.make_daemon([testutil.cloud_node_mock(2, size=size)], [arv_node], [size], avail_sizes=[(size, {"cores":1})]) + self.busywait(lambda: self.node_setup.start.called) self.stop_proxy(self.daemon) self.assertTrue(self.node_setup.start.called) @@ -251,7 +268,8 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin, for call in self.node_setup.start.call_args_list]) def test_no_new_node_when_ge_min_nodes_busy(self): - cloud_nodes = [testutil.cloud_node_mock(n) for n in range(1, 4)] + size = testutil.MockSize(2) + cloud_nodes = [testutil.cloud_node_mock(n, size=size) for n in range(1, 4)] arv_nodes = [testutil.arvados_node_mock(n, job_uuid=True) for n in range(1, 4)] self.make_daemon(cloud_nodes, arv_nodes, [], min_nodes=2) @@ -259,9 +277,10 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin, self.assertEqual(0, self.node_setup.start.call_count) def test_no_new_node_when_max_nodes_busy(self): - self.make_daemon([testutil.cloud_node_mock(3)], - [testutil.arvados_node_mock(3, job_uuid=True)], - [testutil.MockSize(3)], + size = testutil.MockSize(3) + self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(3)], + arvados_nodes=[testutil.arvados_node_mock(3, job_uuid=True)], + want_sizes=[size], max_nodes=1) self.stop_proxy(self.daemon) self.assertFalse(self.node_setup.start.called) @@ -269,6 +288,7 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin, def start_node_boot(self, cloud_node=None, arv_node=None, id_num=1): if cloud_node is None: cloud_node = testutil.cloud_node_mock(id_num) + id_num = int(cloud_node.id) if arv_node is None: arv_node = testutil.arvados_node_mock(id_num) self.make_daemon(want_sizes=[testutil.MockSize(id_num)], @@ -285,10 +305,11 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin, setup = self.start_node_boot(cloud_node, arv_node) self.daemon.node_up(setup).get(self.TIMEOUT) self.assertEqual(1, self.alive_monitor_count()) - self.daemon.update_cloud_nodes([cloud_node]) self.daemon.update_arvados_nodes([arv_node]) + self.daemon.update_cloud_nodes([cloud_node]) + self.monitor_list()[0].proxy().cloud_node_start_time = time.time()-1801 self.daemon.update_server_wishlist( - [testutil.MockSize(1)]).get(self.TIMEOUT) + [testutil.MockSize(4)]).get(self.TIMEOUT) self.stop_proxy(self.daemon) self.assertEqual(2, self.node_setup.start.call_count) @@ -350,6 +371,7 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin, shutdown = self.node_shutdown.start().proxy() shutdown.cloud_node.get.return_value = cloud_node self.daemon.node_finished_shutdown(shutdown).get(self.TIMEOUT) + self.daemon.update_cloud_nodes([]) self.assertTrue(shutdown.stop.called, "shutdown actor not stopped after finishing") self.assertTrue(monitor.actor_ref.actor_stopped.wait(self.TIMEOUT), @@ -362,20 +384,25 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin, 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 self.daemon.node_up(setup).get(self.TIMEOUT) self.assertEqual(1, self.alive_monitor_count()) self.assertFalse(self.node_shutdown.start.called) - self.timer.deliver() + now = time.time() + self.monitor_list()[0].tell_proxy().consider_shutdown() + self.busywait(lambda: self.node_shutdown.start.called) self.stop_proxy(self.daemon) self.assertShutdownCancellable(False) def test_booted_node_shut_down_when_never_paired(self): cloud_node = testutil.cloud_node_mock(2) setup = self.start_node_boot(cloud_node) + self.cloud_factory().node_start_time.return_value = time.time() - 3601 self.daemon.node_up(setup).get(self.TIMEOUT) self.assertEqual(1, self.alive_monitor_count()) self.daemon.update_cloud_nodes([cloud_node]) - self.timer.deliver() + self.monitor_list()[0].tell_proxy().consider_shutdown() + self.busywait(lambda: self.node_shutdown.start.called) self.stop_proxy(self.daemon) self.assertShutdownCancellable(False) @@ -383,11 +410,12 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin, 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) + self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT) self.daemon.node_up(setup).get(self.TIMEOUT) self.assertEqual(1, self.alive_monitor_count()) + self.monitor_list()[0].proxy().cloud_node_start_time = time.time()-3601 self.daemon.update_cloud_nodes([cloud_node]) - self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT) - self.timer.deliver() + self.busywait(lambda: self.node_shutdown.start.called) self.stop_proxy(self.daemon) self.assertShutdownCancellable(False) @@ -441,8 +469,9 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin, def test_shutdown_declined_at_wishlist_capacity(self): cloud_node = testutil.cloud_node_mock(1) + arv_node = testutil.arvados_node_mock(1) size = testutil.MockSize(1) - self.make_daemon(cloud_nodes=[cloud_node], want_sizes=[size]) + self.make_daemon(cloud_nodes=[cloud_node], arvados_nodes=[arv_node], want_sizes=[size]) self.assertEqual(1, self.alive_monitor_count()) monitor = self.monitor_list()[0].proxy() self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT) @@ -451,7 +480,8 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin, def test_shutdown_declined_below_min_nodes(self): cloud_node = testutil.cloud_node_mock(1) - self.make_daemon(cloud_nodes=[cloud_node], min_nodes=1) + arv_node = testutil.arvados_node_mock(1) + self.make_daemon(cloud_nodes=[cloud_node], arvados_nodes=[arv_node], min_nodes=1) self.assertEqual(1, self.alive_monitor_count()) monitor = self.monitor_list()[0].proxy() self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT) @@ -467,10 +497,11 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin, self.assertTrue(self.node_shutdown.start.called) def test_shutdown_declined_when_idle_and_job_queued(self): - cloud_nodes = [testutil.cloud_node_mock(n) for n in [3, 4]] + size = testutil.MockSize(1) + cloud_nodes = [testutil.cloud_node_mock(n, size=size) for n in [3, 4]] arv_nodes = [testutil.arvados_node_mock(3, job_uuid=True), testutil.arvados_node_mock(4, job_uuid=None)] - self.make_daemon(cloud_nodes, arv_nodes, [testutil.MockSize(1)]) + self.make_daemon(cloud_nodes, arv_nodes, [size]) self.assertEqual(2, self.alive_monitor_count()) for mon_ref in self.monitor_list(): monitor = mon_ref.proxy() @@ -498,7 +529,7 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin, self.daemon.node_finished_shutdown(self.last_shutdown).get(self.TIMEOUT) self.assertEqual(0, self.alive_monitor_count()) - def test_broken_node_blackholed_after_cancelled_shutdown(self): + def test_broken_node_not_counted(self): size = testutil.MockSize(8) cloud_node = testutil.cloud_node_mock(8, size=size) wishlist = [size] @@ -510,7 +541,7 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin, 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.cloud_factory().broken.return_value = True 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) @@ -522,7 +553,7 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin, def test_nodes_shutting_down_replaced_below_max_nodes(self): size = testutil.MockSize(6) cloud_node = testutil.cloud_node_mock(6, size=size) - self.make_daemon([cloud_node], [testutil.arvados_node_mock(6)], + self.make_daemon([cloud_node], [testutil.arvados_node_mock(6, crunch_worker_state='down')], avail_sizes=[(size, {"cores":1})]) self.assertEqual(1, self.alive_monitor_count()) monitor = self.monitor_list()[0].proxy() @@ -602,13 +633,6 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin, self.stop_proxy(self.daemon) self.assertEqual(1, self.last_shutdown.stop.call_count) - def busywait(self, f): - n = 0 - while not f() and n < 10: - time.sleep(.1) - n += 1 - self.assertTrue(f()) - def test_node_create_two_sizes(self): small = testutil.MockSize(1) big = testutil.MockSize(2) @@ -670,7 +694,7 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin, self.daemon.node_can_shutdown(c.actor) booting = self.daemon.booting.get() - shutdowns = self.daemon.shutdowns.get() + cloud_nodes = self.daemon.cloud_nodes.get() self.stop_proxy(self.daemon) @@ -686,8 +710,9 @@ class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin, # shutting down a small node sizecounts = {a[0].id: 0 for a in avail_sizes} - for b in shutdowns.itervalues(): - sizecounts[b.cloud_node.get().size.id] += 1 + for b in cloud_nodes.nodes.itervalues(): + if b.shutdown_actor is not None: + sizecounts[b.cloud_node.size.id] += 1 self.assertEqual(1, sizecounts[small.id]) self.assertEqual(0, sizecounts[big.id]) @@ -718,25 +743,3 @@ 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()