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.setup_actor.stop_if_no_cloud_node()
  83         self.assertTrue(
  84             self.setup_actor.actor_ref.actor_stopped.wait(self.TIMEOUT))
  85
  86     def test_no_stop_when_cloud_node(self):
  87         self.make_actor()
  88         self.wait_for_assignment(self.setup_actor, 'cloud_node')
  89         self.setup_actor.stop_if_no_cloud_node().get(self.TIMEOUT)
  90         self.assertTrue(self.stop_proxy(self.setup_actor),
  91                         "actor was stopped by stop_if_no_cloud_node")
  92
  93     def test_subscribe(self):
  94         self.make_mocks(
  95             arverror.ApiError(httplib2.Response({'status': '500'}), ""))
  96         self.make_actor()
  97         subscriber = mock.Mock(name='subscriber_mock')
  98         self.setup_actor.subscribe(subscriber)
  99         self.api_client.nodes().create().execute.side_effect = [
 100             testutil.arvados_node_mock()]
 101         self.wait_for_assignment(self.setup_actor, 'cloud_node')
 102         self.assertEqual(self.setup_actor.actor_ref.actor_urn,
 103                          subscriber.call_args[0][0].actor_ref.actor_urn)
 104
 105     def test_late_subscribe(self):
 106         self.make_actor()
 107         subscriber = mock.Mock(name='subscriber_mock')
 108         self.wait_for_assignment(self.setup_actor, 'cloud_node')
 109         self.setup_actor.subscribe(subscriber).get(self.TIMEOUT)
 110         self.stop_proxy(self.setup_actor)
 111         self.assertEqual(self.setup_actor.actor_ref.actor_urn,
 112                          subscriber.call_args[0][0].actor_ref.actor_urn)
 113
 114
 115 class ComputeNodeShutdownActorMixin(testutil.ActorTestMixin):
 116     def make_mocks(self, cloud_node=None, arvados_node=None,
 117                    shutdown_open=True):
 118         self.timer = testutil.MockTimer()
 119         self.shutdowns = testutil.MockShutdownTimer()
 120         self.shutdowns._set_state(shutdown_open, 300)
 121         self.cloud_client = mock.MagicMock(name='cloud_client')
 122         self.updates = mock.MagicMock(name='update_mock')
 123         if cloud_node is None:
 124             cloud_node = testutil.cloud_node_mock()
 125         self.cloud_node = cloud_node
 126         self.arvados_node = arvados_node
 127
 128     def make_actor(self, cancellable=True):
 129         if not hasattr(self, 'timer'):
 130             self.make_mocks()
 131         monitor_actor = dispatch.ComputeNodeMonitorActor.start(
 132             self.cloud_node, time.time(), self.shutdowns, self.timer,
 133             self.updates, self.arvados_node)
 134         self.shutdown_actor = self.ACTOR_CLASS.start(
 135             self.timer, self.cloud_client, monitor_actor, cancellable).proxy()
 136         self.monitor_actor = monitor_actor.proxy()
 137
 138     def check_success_flag(self, expected, allow_msg_count=1):
 139         # allow_msg_count is the number of internal messages that may
 140         # need to be handled for shutdown to finish.
 141         for try_num in range(1 + allow_msg_count):
 142             last_flag = self.shutdown_actor.success.get(self.TIMEOUT)
 143             if last_flag is expected:
 144                 break
 145         else:
 146             self.fail("success flag {} is not {}".format(last_flag, expected))
 147
 148     def test_uncancellable_shutdown(self, *mocks):
 149         self.make_mocks(shutdown_open=False)
 150         self.cloud_client.destroy_node.return_value = False
 151         self.make_actor(cancellable=False)
 152         self.check_success_flag(None, 0)
 153         self.shutdowns._set_state(True, 600)
 154         self.cloud_client.destroy_node.return_value = True
 155         self.check_success_flag(True)
 156
 157
 158 class ComputeNodeShutdownActorTestCase(ComputeNodeShutdownActorMixin,
 159                                        unittest.TestCase):
 160     ACTOR_CLASS = dispatch.ComputeNodeShutdownActor
 161
 162     def test_easy_shutdown(self):
 163         self.make_actor()
 164         self.check_success_flag(True)
 165         self.assertTrue(self.cloud_client.destroy_node.called)
 166
 167     def test_shutdown_cancelled_when_window_closes(self):
 168         self.make_mocks(shutdown_open=False)
 169         self.make_actor()
 170         self.check_success_flag(False, 2)
 171         self.assertFalse(self.cloud_client.destroy_node.called)
 172
 173     def test_shutdown_retries_when_cloud_fails(self):
 174         self.make_mocks()
 175         self.cloud_client.destroy_node.return_value = False
 176         self.make_actor()
 177         self.assertIsNone(self.shutdown_actor.success.get(self.TIMEOUT))
 178         self.cloud_client.destroy_node.return_value = True
 179         self.check_success_flag(True)
 180
 181     def test_late_subscribe(self):
 182         self.make_actor()
 183         subscriber = mock.Mock(name='subscriber_mock')
 184         self.shutdown_actor.subscribe(subscriber).get(self.TIMEOUT)
 185         self.stop_proxy(self.shutdown_actor)
 186         self.assertTrue(subscriber.called)
 187         self.assertEqual(self.shutdown_actor.actor_ref.actor_urn,
 188                          subscriber.call_args[0][0].actor_ref.actor_urn)
 189
 190
 191 class ComputeNodeUpdateActorTestCase(testutil.ActorTestMixin,
 192                                      unittest.TestCase):
 193     def make_actor(self):
 194         self.driver = mock.MagicMock(name='driver_mock')
 195         self.updater = dispatch.ComputeNodeUpdateActor.start(self.driver).proxy()
 196
 197     def test_node_sync(self):
 198         self.make_actor()
 199         cloud_node = testutil.cloud_node_mock()
 200         arv_node = testutil.arvados_node_mock()
 201         self.updater.sync_node(cloud_node, arv_node).get(self.TIMEOUT)
 202         self.driver().sync_node.assert_called_with(cloud_node, arv_node)
 203
 204
 205 class ComputeNodeMonitorActorTestCase(testutil.ActorTestMixin,
 206                                       unittest.TestCase):
 207     def make_mocks(self, node_num):
 208         self.shutdowns = testutil.MockShutdownTimer()
 209         self.shutdowns._set_state(False, 300)
 210         self.timer = mock.MagicMock(name='timer_mock')
 211         self.updates = mock.MagicMock(name='update_mock')
 212         self.cloud_mock = testutil.cloud_node_mock(node_num)
 213         self.subscriber = mock.Mock(name='subscriber_mock')
 214
 215     def make_actor(self, node_num=1, arv_node=None, start_time=None):
 216         if not hasattr(self, 'cloud_mock'):
 217             self.make_mocks(node_num)
 218         if start_time is None:
 219             start_time = time.time()
 220         self.node_actor = dispatch.ComputeNodeMonitorActor.start(
 221             self.cloud_mock, start_time, self.shutdowns, self.timer,
 222             self.updates, arv_node).proxy()
 223         self.node_actor.subscribe(self.subscriber).get(self.TIMEOUT)
 224
 225     def node_state(self, *states):
 226         return self.node_actor.in_state(*states).get(self.TIMEOUT)
 227
 228     def test_in_state_when_unpaired(self):
 229         self.make_actor()
 230         self.assertIsNone(self.node_state('idle', 'alloc'))
 231
 232     def test_in_state_when_pairing_stale(self):
 233         self.make_actor(arv_node=testutil.arvados_node_mock(
 234                 job_uuid=None, age=90000))
 235         self.assertIsNone(self.node_state('idle', 'alloc'))
 236
 237     def test_in_state_when_no_state_available(self):
 238         self.make_actor(arv_node=testutil.arvados_node_mock(info={}))
 239         self.assertIsNone(self.node_state('idle', 'alloc'))
 240
 241     def test_in_idle_state(self):
 242         self.make_actor(2, arv_node=testutil.arvados_node_mock(job_uuid=None))
 243         self.assertTrue(self.node_state('idle'))
 244         self.assertFalse(self.node_state('alloc'))
 245         self.assertTrue(self.node_state('idle', 'alloc'))
 246
 247     def test_in_alloc_state(self):
 248         self.make_actor(3, arv_node=testutil.arvados_node_mock(job_uuid=True))
 249         self.assertFalse(self.node_state('idle'))
 250         self.assertTrue(self.node_state('alloc'))
 251         self.assertTrue(self.node_state('idle', 'alloc'))
 252
 253     def test_init_shutdown_scheduling(self):
 254         self.make_actor()
 255         self.assertTrue(self.timer.schedule.called)
 256         self.assertEqual(300, self.timer.schedule.call_args[0][0])
 257
 258     def test_shutdown_window_close_scheduling(self):
 259         self.make_actor()
 260         self.shutdowns._set_state(False, 600)
 261         self.timer.schedule.reset_mock()
 262         self.node_actor.consider_shutdown().get(self.TIMEOUT)
 263         self.stop_proxy(self.node_actor)
 264         self.assertTrue(self.timer.schedule.called)
 265         self.assertEqual(600, self.timer.schedule.call_args[0][0])
 266         self.assertFalse(self.subscriber.called)
 267
 268     def test_shutdown_subscription(self):
 269         self.make_actor()
 270         self.shutdowns._set_state(True, 600)
 271         self.node_actor.consider_shutdown().get(self.TIMEOUT)
 272         self.assertTrue(self.subscriber.called)
 273         self.assertEqual(self.node_actor.actor_ref.actor_urn,
 274                          self.subscriber.call_args[0][0].actor_ref.actor_urn)
 275
 276     def test_shutdown_without_arvados_node(self):
 277         self.make_actor()
 278         self.shutdowns._set_state(True, 600)
 279         self.assertTrue(self.node_actor.shutdown_eligible().get(self.TIMEOUT))
 280
 281     def test_no_shutdown_without_arvados_node_and_old_cloud_node(self):
 282         self.make_actor(start_time=0)
 283         self.shutdowns._set_state(True, 600)
 284         self.assertFalse(self.node_actor.shutdown_eligible().get(self.TIMEOUT))
 285
 286     def test_no_shutdown_when_window_closed(self):
 287         self.make_actor(3, testutil.arvados_node_mock(3, job_uuid=None))
 288         self.assertFalse(self.node_actor.shutdown_eligible().get(self.TIMEOUT))
 289
 290     def test_no_shutdown_when_node_running_job(self):
 291         self.make_actor(4, testutil.arvados_node_mock(4, job_uuid=True))
 292         self.shutdowns._set_state(True, 600)
 293         self.assertFalse(self.node_actor.shutdown_eligible().get(self.TIMEOUT))
 294
 295     def test_no_shutdown_when_node_state_unknown(self):
 296         self.make_actor(5, testutil.arvados_node_mock(5, info={}))
 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_stale(self):
 301         self.make_actor(6, testutil.arvados_node_mock(6, age=90000))
 302         self.shutdowns._set_state(True, 600)
 303         self.assertFalse(self.node_actor.shutdown_eligible().get(self.TIMEOUT))
 304
 305     def test_arvados_node_match(self):
 306         self.make_actor(2)
 307         arv_node = testutil.arvados_node_mock(
 308             2, hostname='compute-two.zzzzz.arvadosapi.com')
 309         pair_id = self.node_actor.offer_arvados_pair(arv_node).get(self.TIMEOUT)
 310         self.assertEqual(self.cloud_mock.id, pair_id)
 311         self.stop_proxy(self.node_actor)
 312         self.updates.sync_node.assert_called_with(self.cloud_mock, arv_node)
 313
 314     def test_arvados_node_mismatch(self):
 315         self.make_actor(3)
 316         arv_node = testutil.arvados_node_mock(1)
 317         self.assertIsNone(
 318             self.node_actor.offer_arvados_pair(arv_node).get(self.TIMEOUT))
 319
 320     def test_update_cloud_node(self):
 321         self.make_actor(1)
 322         self.make_mocks(2)
 323         self.cloud_mock.id = '1'
 324         self.node_actor.update_cloud_node(self.cloud_mock)
 325         current_cloud = self.node_actor.cloud_node.get(self.TIMEOUT)
 326         self.assertEqual([testutil.ip_address_mock(2)],
 327                          current_cloud.private_ips)
 328
 329     def test_missing_cloud_node_update(self):
 330         self.make_actor(1)
 331         self.node_actor.update_cloud_node(None)
 332         current_cloud = self.node_actor.cloud_node.get(self.TIMEOUT)
 333         self.assertEqual([testutil.ip_address_mock(1)],
 334                          current_cloud.private_ips)
 335
 336     def test_update_arvados_node(self):
 337         self.make_actor(3)
 338         job_uuid = 'zzzzz-jjjjj-updatejobnode00'
 339         new_arvados = testutil.arvados_node_mock(3, job_uuid)
 340         self.node_actor.update_arvados_node(new_arvados)
 341         current_arvados = self.node_actor.arvados_node.get(self.TIMEOUT)
 342         self.assertEqual(job_uuid, current_arvados['job_uuid'])
 343
 344     def test_missing_arvados_node_update(self):
 345         self.make_actor(4, testutil.arvados_node_mock(4))
 346         self.node_actor.update_arvados_node(None)
 347         current_arvados = self.node_actor.arvados_node.get(self.TIMEOUT)
 348         self.assertEqual(testutil.ip_address_mock(4),
 349                          current_arvados['ip_address'])