services/nodemanager/tests/test_computenode_dispatch.py

   1 #!/usr/bin/env python
   2
   3 from __future__ import absolute_import, print_function
   4
   5 import time
   6 import unittest
   7
   8 import arvados.errors as arverror
   9 import httplib2
  10 import mock
  11 import pykka
  12
  13 import arvnodeman.computenode.dispatch as dispatch
  14 from . import testutil
  15
  16 class ComputeNodeSetupActorTestCase(testutil.ActorTestMixin, unittest.TestCase):
  17     def make_mocks(self, arvados_effect=None):
  18         if arvados_effect is None:
  19             arvados_effect = [testutil.arvados_node_mock()]
  20         self.arvados_effect = arvados_effect
  21         self.timer = testutil.MockTimer()
  22         self.api_client = mock.MagicMock(name='api_client')
  23         self.api_client.nodes().create().execute.side_effect = arvados_effect
  24         self.api_client.nodes().update().execute.side_effect = arvados_effect
  25         self.cloud_client = mock.MagicMock(name='cloud_client')
  26         self.cloud_client.create_node.return_value = testutil.cloud_node_mock(1)
  27
  28     def make_actor(self, arv_node=None):
  29         if not hasattr(self, 'timer'):
  30             self.make_mocks(arvados_effect=[arv_node])
  31         self.setup_actor = dispatch.ComputeNodeSetupActor.start(
  32             self.timer, self.api_client, self.cloud_client,
  33             testutil.MockSize(1), arv_node).proxy()
  34
  35     def test_creation_without_arvados_node(self):
  36         self.make_actor()
  37         self.assertEqual(self.arvados_effect[-1],
  38                          self.setup_actor.arvados_node.get(self.TIMEOUT))
  39         self.assertTrue(self.api_client.nodes().create().execute.called)
  40         self.assertEqual(self.cloud_client.create_node(),
  41                          self.setup_actor.cloud_node.get(self.TIMEOUT))
  42
  43     def test_creation_with_arvados_node(self):
  44         self.make_actor(testutil.arvados_node_mock())
  45         self.assertEqual(self.arvados_effect[-1],
  46                          self.setup_actor.arvados_node.get(self.TIMEOUT))
  47         self.assertTrue(self.api_client.nodes().update().execute.called)
  48         self.assertEqual(self.cloud_client.create_node(),
  49                          self.setup_actor.cloud_node.get(self.TIMEOUT))
  50
  51     def test_failed_arvados_calls_retried(self):
  52         self.make_mocks([
  53                 arverror.ApiError(httplib2.Response({'status': '500'}), ""),
  54                 testutil.arvados_node_mock(),
  55                 ])
  56         self.make_actor()
  57         self.wait_for_assignment(self.setup_actor, 'arvados_node')
  58
  59     def test_failed_cloud_calls_retried(self):
  60         self.make_mocks()
  61         self.cloud_client.create_node.side_effect = [
  62             Exception("test cloud creation error"),
  63             self.cloud_client.create_node.return_value,
  64             ]
  65         self.make_actor()
  66         self.wait_for_assignment(self.setup_actor, 'cloud_node')
  67
  68     def test_failed_post_create_retried(self):
  69         self.make_mocks()
  70         self.cloud_client.post_create_node.side_effect = [
  71             Exception("test cloud post-create error"), None]
  72         self.make_actor()
  73         done = self.FUTURE_CLASS()
  74         self.setup_actor.subscribe(done.set)
  75         done.get(self.TIMEOUT)
  76         self.assertEqual(2, self.cloud_client.post_create_node.call_count)
  77
  78     def test_stop_when_no_cloud_node(self):
  79         self.make_mocks(
  80             arverror.ApiError(httplib2.Response({'status': '500'}), ""))
  81         self.make_actor()
  82         self.assertTrue(
  83             self.setup_actor.stop_if_no_cloud_node().get(self.TIMEOUT))
  84         self.assertTrue(
  85             self.setup_actor.actor_ref.actor_stopped.wait(self.TIMEOUT))
  86
  87     def test_no_stop_when_cloud_node(self):
  88         self.make_actor()
  89         self.wait_for_assignment(self.setup_actor, 'cloud_node')
  90         self.assertFalse(
  91             self.setup_actor.stop_if_no_cloud_node().get(self.TIMEOUT))
  92         self.assertTrue(self.stop_proxy(self.setup_actor),
  93                         "actor was stopped by stop_if_no_cloud_node")
  94
  95     def test_subscribe(self):
  96         self.make_mocks(
  97             arverror.ApiError(httplib2.Response({'status': '500'}), ""))
  98         self.make_actor()
  99         subscriber = mock.Mock(name='subscriber_mock')
 100         self.setup_actor.subscribe(subscriber)
 101         self.api_client.nodes().create().execute.side_effect = [
 102             testutil.arvados_node_mock()]
 103         self.wait_for_assignment(self.setup_actor, 'cloud_node')
 104         self.assertEqual(self.setup_actor.actor_ref.actor_urn,
 105                          subscriber.call_args[0][0].actor_ref.actor_urn)
 106
 107     def test_late_subscribe(self):
 108         self.make_actor()
 109         subscriber = mock.Mock(name='subscriber_mock')
 110         self.wait_for_assignment(self.setup_actor, 'cloud_node')
 111         self.setup_actor.subscribe(subscriber).get(self.TIMEOUT)
 112         self.stop_proxy(self.setup_actor)
 113         self.assertEqual(self.setup_actor.actor_ref.actor_urn,
 114                          subscriber.call_args[0][0].actor_ref.actor_urn)
 115
 116
 117 class ComputeNodeShutdownActorMixin(testutil.ActorTestMixin):
 118     def make_mocks(self, cloud_node=None, arvados_node=None,
 119                    shutdown_open=True):
 120         self.timer = testutil.MockTimer()
 121         self.shutdowns = testutil.MockShutdownTimer()
 122         self.shutdowns._set_state(shutdown_open, 300)
 123         self.cloud_client = mock.MagicMock(name='cloud_client')
 124         self.updates = mock.MagicMock(name='update_mock')
 125         if cloud_node is None:
 126             cloud_node = testutil.cloud_node_mock()
 127         self.cloud_node = cloud_node
 128         self.arvados_node = arvados_node
 129
 130     def make_actor(self, cancellable=True):
 131         if not hasattr(self, 'timer'):
 132             self.make_mocks()
 133         monitor_actor = dispatch.ComputeNodeMonitorActor.start(
 134             self.cloud_node, time.time(), self.shutdowns,
 135             testutil.cloud_node_fqdn, self.timer, self.updates,
 136             self.arvados_node)
 137         self.shutdown_actor = self.ACTOR_CLASS.start(
 138             self.timer, self.cloud_client, monitor_actor, cancellable).proxy()
 139         self.monitor_actor = monitor_actor.proxy()
 140
 141     def check_success_flag(self, expected, allow_msg_count=1):
 142         # allow_msg_count is the number of internal messages that may
 143         # need to be handled for shutdown to finish.
 144         for try_num in range(1 + allow_msg_count):
 145             last_flag = self.shutdown_actor.success.get(self.TIMEOUT)
 146             if last_flag is expected:
 147                 break
 148         else:
 149             self.fail("success flag {} is not {}".format(last_flag, expected))
 150
 151     def test_uncancellable_shutdown(self, *mocks):
 152         self.make_mocks(shutdown_open=False)
 153         self.cloud_client.destroy_node.return_value = False
 154         self.make_actor(cancellable=False)
 155         self.check_success_flag(None, 0)
 156         self.shutdowns._set_state(True, 600)
 157         self.cloud_client.destroy_node.return_value = True
 158         self.check_success_flag(True)
 159
 160
 161 class ComputeNodeShutdownActorTestCase(ComputeNodeShutdownActorMixin,
 162                                        unittest.TestCase):
 163     ACTOR_CLASS = dispatch.ComputeNodeShutdownActor
 164
 165     def test_easy_shutdown(self):
 166         self.make_actor()
 167         self.check_success_flag(True)
 168         self.assertTrue(self.cloud_client.destroy_node.called)
 169
 170     def test_shutdown_cancelled_when_window_closes(self):
 171         self.make_mocks(shutdown_open=False)
 172         self.make_actor()
 173         self.check_success_flag(False, 2)
 174         self.assertFalse(self.cloud_client.destroy_node.called)
 175
 176     def test_shutdown_retries_when_cloud_fails(self):
 177         self.make_mocks()
 178         self.cloud_client.destroy_node.return_value = False
 179         self.make_actor()
 180         self.assertIsNone(self.shutdown_actor.success.get(self.TIMEOUT))
 181         self.cloud_client.destroy_node.return_value = True
 182         self.check_success_flag(True)
 183
 184     def test_late_subscribe(self):
 185         self.make_actor()
 186         subscriber = mock.Mock(name='subscriber_mock')
 187         self.shutdown_actor.subscribe(subscriber).get(self.TIMEOUT)
 188         self.stop_proxy(self.shutdown_actor)
 189         self.assertTrue(subscriber.called)
 190         self.assertEqual(self.shutdown_actor.actor_ref.actor_urn,
 191                          subscriber.call_args[0][0].actor_ref.actor_urn)
 192
 193
 194 class ComputeNodeUpdateActorTestCase(testutil.ActorTestMixin,
 195                                      unittest.TestCase):
 196     def make_actor(self):
 197         self.driver = mock.MagicMock(name='driver_mock')
 198         self.updater = dispatch.ComputeNodeUpdateActor.start(self.driver).proxy()
 199
 200     def test_node_sync(self):
 201         self.make_actor()
 202         cloud_node = testutil.cloud_node_mock()
 203         arv_node = testutil.arvados_node_mock()
 204         self.updater.sync_node(cloud_node, arv_node).get(self.TIMEOUT)
 205         self.driver().sync_node.assert_called_with(cloud_node, arv_node)
 206
 207
 208 class ComputeNodeMonitorActorTestCase(testutil.ActorTestMixin,
 209                                       unittest.TestCase):
 210     def make_mocks(self, node_num):
 211         self.shutdowns = testutil.MockShutdownTimer()
 212         self.shutdowns._set_state(False, 300)
 213         self.timer = mock.MagicMock(name='timer_mock')
 214         self.updates = mock.MagicMock(name='update_mock')
 215         self.cloud_mock = testutil.cloud_node_mock(node_num)
 216         self.subscriber = mock.Mock(name='subscriber_mock')
 217
 218     def make_actor(self, node_num=1, arv_node=None, start_time=None):
 219         if not hasattr(self, 'cloud_mock'):
 220             self.make_mocks(node_num)
 221         if start_time is None:
 222             start_time = time.time()
 223         self.node_actor = dispatch.ComputeNodeMonitorActor.start(
 224             self.cloud_mock, start_time, self.shutdowns,
 225             testutil.cloud_node_fqdn, self.timer, self.updates,
 226             arv_node).proxy()
 227         self.node_actor.subscribe(self.subscriber).get(self.TIMEOUT)
 228
 229     def node_state(self, *states):
 230         return self.node_actor.in_state(*states).get(self.TIMEOUT)
 231
 232     def test_in_state_when_unpaired(self):
 233         self.make_actor()
 234         self.assertIsNone(self.node_state('idle', 'busy'))
 235
 236     def test_in_state_when_pairing_stale(self):
 237         self.make_actor(arv_node=testutil.arvados_node_mock(
 238                 job_uuid=None, age=90000))
 239         self.assertIsNone(self.node_state('idle', 'busy'))
 240
 241     def test_in_state_when_no_state_available(self):
 242         self.make_actor(arv_node=testutil.arvados_node_mock(
 243                 crunch_worker_state=None))
 244         self.assertIsNone(self.node_state('idle', 'busy'))
 245
 246     def test_in_idle_state(self):
 247         self.make_actor(2, arv_node=testutil.arvados_node_mock(job_uuid=None))
 248         self.assertTrue(self.node_state('idle'))
 249         self.assertFalse(self.node_state('busy'))
 250         self.assertTrue(self.node_state('idle', 'busy'))
 251
 252     def test_in_busy_state(self):
 253         self.make_actor(3, arv_node=testutil.arvados_node_mock(job_uuid=True))
 254         self.assertFalse(self.node_state('idle'))
 255         self.assertTrue(self.node_state('busy'))
 256         self.assertTrue(self.node_state('idle', 'busy'))
 257
 258     def test_init_shutdown_scheduling(self):
 259         self.make_actor()
 260         self.assertTrue(self.timer.schedule.called)
 261         self.assertEqual(300, self.timer.schedule.call_args[0][0])
 262
 263     def test_shutdown_window_close_scheduling(self):
 264         self.make_actor()
 265         self.shutdowns._set_state(False, 600)
 266         self.timer.schedule.reset_mock()
 267         self.node_actor.consider_shutdown().get(self.TIMEOUT)
 268         self.stop_proxy(self.node_actor)
 269         self.assertTrue(self.timer.schedule.called)
 270         self.assertEqual(600, self.timer.schedule.call_args[0][0])
 271         self.assertFalse(self.subscriber.called)
 272
 273     def test_shutdown_subscription(self):
 274         self.make_actor()
 275         self.shutdowns._set_state(True, 600)
 276         self.node_actor.consider_shutdown().get(self.TIMEOUT)
 277         self.assertTrue(self.subscriber.called)
 278         self.assertEqual(self.node_actor.actor_ref.actor_urn,
 279                          self.subscriber.call_args[0][0].actor_ref.actor_urn)
 280
 281     def test_shutdown_without_arvados_node(self):
 282         self.make_actor()
 283         self.shutdowns._set_state(True, 600)
 284         self.assertTrue(self.node_actor.shutdown_eligible().get(self.TIMEOUT))
 285
 286     def test_no_shutdown_without_arvados_node_and_old_cloud_node(self):
 287         self.make_actor(start_time=0)
 288         self.shutdowns._set_state(True, 600)
 289         self.assertFalse(self.node_actor.shutdown_eligible().get(self.TIMEOUT))
 290
 291     def test_no_shutdown_when_window_closed(self):
 292         self.make_actor(3, testutil.arvados_node_mock(3, job_uuid=None))
 293         self.assertFalse(self.node_actor.shutdown_eligible().get(self.TIMEOUT))
 294
 295     def test_no_shutdown_when_node_running_job(self):
 296         self.make_actor(4, testutil.arvados_node_mock(4, job_uuid=True))
 297         self.shutdowns._set_state(True, 600)
 298         self.assertFalse(self.node_actor.shutdown_eligible().get(self.TIMEOUT))
 299
 300     def test_no_shutdown_when_node_state_unknown(self):
 301         self.make_actor(5, testutil.arvados_node_mock(
 302             5, crunch_worker_state=None))
 303         self.shutdowns._set_state(True, 600)
 304         self.assertFalse(self.node_actor.shutdown_eligible().get(self.TIMEOUT))
 305
 306     def test_no_shutdown_when_node_state_stale(self):
 307         self.make_actor(6, testutil.arvados_node_mock(6, age=90000))
 308         self.shutdowns._set_state(True, 600)
 309         self.assertFalse(self.node_actor.shutdown_eligible().get(self.TIMEOUT))
 310
 311     def test_arvados_node_match(self):
 312         self.make_actor(2)
 313         arv_node = testutil.arvados_node_mock(
 314             2, hostname='compute-two.zzzzz.arvadosapi.com')
 315         pair_id = self.node_actor.offer_arvados_pair(arv_node).get(self.TIMEOUT)
 316         self.assertEqual(self.cloud_mock.id, pair_id)
 317         self.stop_proxy(self.node_actor)
 318         self.updates.sync_node.assert_called_with(self.cloud_mock, arv_node)
 319
 320     def test_arvados_node_mismatch(self):
 321         self.make_actor(3)
 322         arv_node = testutil.arvados_node_mock(1)
 323         self.assertIsNone(
 324             self.node_actor.offer_arvados_pair(arv_node).get(self.TIMEOUT))
 325
 326     def test_arvados_node_mismatch_first_ping_too_early(self):
 327         self.make_actor(4)
 328         arv_node = testutil.arvados_node_mock(
 329             4, first_ping_at='1971-03-02T14:15:16.1717282Z')
 330         self.assertIsNone(
 331             self.node_actor.offer_arvados_pair(arv_node).get(self.TIMEOUT))
 332
 333     def test_update_cloud_node(self):
 334         self.make_actor(1)
 335         self.make_mocks(2)
 336         self.cloud_mock.id = '1'
 337         self.node_actor.update_cloud_node(self.cloud_mock)
 338         current_cloud = self.node_actor.cloud_node.get(self.TIMEOUT)
 339         self.assertEqual([testutil.ip_address_mock(2)],
 340                          current_cloud.private_ips)
 341
 342     def test_missing_cloud_node_update(self):
 343         self.make_actor(1)
 344         self.node_actor.update_cloud_node(None)
 345         current_cloud = self.node_actor.cloud_node.get(self.TIMEOUT)
 346         self.assertEqual([testutil.ip_address_mock(1)],
 347                          current_cloud.private_ips)
 348
 349     def test_update_arvados_node(self):
 350         self.make_actor(3)
 351         job_uuid = 'zzzzz-jjjjj-updatejobnode00'
 352         new_arvados = testutil.arvados_node_mock(3, job_uuid)
 353         self.node_actor.update_arvados_node(new_arvados)
 354         current_arvados = self.node_actor.arvados_node.get(self.TIMEOUT)
 355         self.assertEqual(job_uuid, current_arvados['job_uuid'])
 356
 357     def test_missing_arvados_node_update(self):
 358         self.make_actor(4, testutil.arvados_node_mock(4))
 359         self.node_actor.update_arvados_node(None)
 360         current_arvados = self.node_actor.arvados_node.get(self.TIMEOUT)
 361         self.assertEqual(testutil.ip_address_mock(4),
 362                          current_arvados['ip_address'])
 363
 364     def test_update_arvados_node_syncs_when_fqdn_mismatch(self):
 365         self.make_mocks(5)
 366         self.cloud_mock.extra['testname'] = 'cloudfqdn.zzzzz.arvadosapi.com'
 367         self.make_actor()
 368         arv_node = testutil.arvados_node_mock(5)
 369         self.node_actor.update_arvados_node(arv_node).get(self.TIMEOUT)
 370         self.assertEqual(1, self.updates.sync_node.call_count)
 371
 372     def test_update_arvados_node_skips_sync_when_fqdn_match(self):
 373         self.make_mocks(6)
 374         arv_node = testutil.arvados_node_mock(6)
 375         self.cloud_mock.extra['testname'] ='{n[hostname]}.{n[domain]}'.format(
 376             n=arv_node)
 377         self.make_actor()
 378         self.node_actor.update_arvados_node(arv_node).get(self.TIMEOUT)
 379         self.assertEqual(0, self.updates.sync_node.call_count)