services/nodemanager/tests/test_daemon.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 mock
   9 import pykka
  10
  11 import arvnodeman.daemon as nmdaemon
  12 from arvnodeman.jobqueue import ServerCalculator
  13 from arvnodeman.computenode.dispatch import ComputeNodeMonitorActor
  14 from . import testutil
  15 import logging
  16
  17 class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin,
  18                                      unittest.TestCase):
  19     def mock_node_start(self, **kwargs):
  20         # Make sure that every time the daemon starts a setup actor,
  21         # it gets a new mock object back.
  22         get_cloud_size = mock.MagicMock()
  23         get_cloud_size.get.return_value = kwargs["cloud_size"]
  24         mock_actor = mock.MagicMock()
  25         mock_proxy = mock.NonCallableMock(name='setup_mock_proxy',
  26                                           cloud_size=get_cloud_size,
  27                                           actor_ref=mock_actor)
  28         mock_actor.proxy.return_value = mock_proxy
  29
  30         self.last_setup = mock_proxy
  31         return mock_actor
  32
  33     def mock_node_shutdown(self, **kwargs):
  34         # Make sure that every time the daemon starts a shutdown actor,
  35         # it gets a new mock object back.
  36         get_cloud_node = mock.MagicMock()
  37         if "node_monitor" in kwargs:
  38             get_cloud_node.get.return_value = kwargs["node_monitor"].proxy().cloud_node.get()
  39         mock_actor = mock.MagicMock()
  40         mock_proxy = mock.NonCallableMock(name='shutdown_mock_proxy',
  41                                           cloud_node=get_cloud_node,
  42                                           actor_ref=mock_actor)
  43
  44         mock_actor.proxy.return_value = mock_proxy
  45         self.last_shutdown = mock_proxy
  46
  47         return mock_actor
  48
  49     def make_daemon(self, cloud_nodes=[], arvados_nodes=[], want_sizes=[],
  50                     avail_sizes=[(testutil.MockSize(1), {"cores": 1})],
  51                     min_nodes=0, max_nodes=8,
  52                     shutdown_windows=[54, 5, 1],
  53                     max_total_price=None):
  54         for name in ['cloud_nodes', 'arvados_nodes', 'server_wishlist']:
  55             setattr(self, name + '_poller', mock.MagicMock(name=name + '_mock'))
  56         self.arv_factory = mock.MagicMock(name='arvados_mock')
  57         self.cloud_factory = mock.MagicMock(name='cloud_mock')
  58         self.cloud_factory().node_start_time.return_value = time.time()
  59         self.cloud_updates = mock.MagicMock(name='updates_mock')
  60         self.timer = testutil.MockTimer(deliver_immediately=False)
  61
  62         self.node_setup = mock.MagicMock(name='setup_mock')
  63         self.node_setup.start.side_effect = self.mock_node_start
  64         self.node_setup.reset_mock()
  65
  66         self.node_shutdown = mock.MagicMock(name='shutdown_mock')
  67         self.node_shutdown.start.side_effect = self.mock_node_shutdown
  68
  69         self.daemon = nmdaemon.NodeManagerDaemonActor.start(
  70             self.server_wishlist_poller, self.arvados_nodes_poller,
  71             self.cloud_nodes_poller, self.cloud_updates, self.timer,
  72             self.arv_factory, self.cloud_factory,
  73             shutdown_windows, ServerCalculator(avail_sizes),
  74             min_nodes, max_nodes, 600, 1800, 3600,
  75             self.node_setup, self.node_shutdown,
  76             max_total_price=max_total_price).proxy()
  77         if cloud_nodes is not None:
  78             self.daemon.update_cloud_nodes(cloud_nodes).get(self.TIMEOUT)
  79         if arvados_nodes is not None:
  80             self.daemon.update_arvados_nodes(arvados_nodes).get(self.TIMEOUT)
  81         if want_sizes is not None:
  82             self.daemon.update_server_wishlist(want_sizes).get(self.TIMEOUT)
  83
  84     def monitor_list(self):
  85         return pykka.ActorRegistry.get_by_class(ComputeNodeMonitorActor)
  86
  87     def monitored_arvados_nodes(self):
  88         pairings = []
  89         for future in [actor.proxy().arvados_node
  90                        for actor in self.monitor_list()]:
  91             try:
  92                 pairings.append(future.get(self.TIMEOUT))
  93             except pykka.ActorDeadError:
  94                 pass
  95         return pairings
  96
  97     def alive_monitor_count(self):
  98         return len(self.monitored_arvados_nodes())
  99
 100     def assertShutdownCancellable(self, expected=True):
 101         self.assertTrue(self.node_shutdown.start.called)
 102         self.assertIs(expected,
 103                       self.node_shutdown.start.call_args[1]['cancellable'],
 104                       "ComputeNodeShutdownActor incorrectly cancellable")
 105
 106     def test_easy_node_creation(self):
 107         size = testutil.MockSize(1)
 108         self.make_daemon(want_sizes=[size])
 109         self.stop_proxy(self.daemon)
 110         self.assertTrue(self.node_setup.start.called)
 111
 112     def check_monitors_arvados_nodes(self, *arv_nodes):
 113         self.assertItemsEqual(arv_nodes, self.monitored_arvados_nodes())
 114
 115     def test_node_pairing(self):
 116         cloud_node = testutil.cloud_node_mock(1)
 117         arv_node = testutil.arvados_node_mock(1)
 118         self.make_daemon([cloud_node], [arv_node])
 119         self.stop_proxy(self.daemon)
 120         self.check_monitors_arvados_nodes(arv_node)
 121
 122     def test_node_pairing_after_arvados_update(self):
 123         cloud_node = testutil.cloud_node_mock(2)
 124         self.make_daemon([cloud_node],
 125                          [testutil.arvados_node_mock(2, ip_address=None)])
 126         arv_node = testutil.arvados_node_mock(2)
 127         self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
 128         self.stop_proxy(self.daemon)
 129         self.check_monitors_arvados_nodes(arv_node)
 130
 131     def test_arvados_node_un_and_re_paired(self):
 132         # We need to create the Arvados node mock after spinning up the daemon
 133         # to make sure it's new enough to pair with the cloud node.
 134         self.make_daemon([testutil.cloud_node_mock(3)], arvados_nodes=None)
 135         arv_node = testutil.arvados_node_mock(3)
 136         self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
 137         self.check_monitors_arvados_nodes(arv_node)
 138         self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
 139         self.assertEqual(0, self.alive_monitor_count())
 140         self.daemon.update_cloud_nodes([testutil.cloud_node_mock(3)])
 141         self.stop_proxy(self.daemon)
 142         self.check_monitors_arvados_nodes(arv_node)
 143
 144     def test_old_arvados_node_not_double_assigned(self):
 145         arv_node = testutil.arvados_node_mock(3, age=9000)
 146         size = testutil.MockSize(3)
 147         self.make_daemon(arvados_nodes=[arv_node], avail_sizes=[(size, {"cores":1})])
 148         self.daemon.update_server_wishlist([size]).get(self.TIMEOUT)
 149         self.daemon.update_server_wishlist([size, size]).get(self.TIMEOUT)
 150         self.stop_proxy(self.daemon)
 151         used_nodes = [call[1].get('arvados_node')
 152                       for call in self.node_setup.start.call_args_list]
 153         self.assertEqual(2, len(used_nodes))
 154         self.assertIn(arv_node, used_nodes)
 155         self.assertIn(None, used_nodes)
 156
 157     def test_node_count_satisfied(self):
 158         self.make_daemon([testutil.cloud_node_mock()],
 159                          want_sizes=[testutil.MockSize(1)])
 160         self.stop_proxy(self.daemon)
 161         self.assertFalse(self.node_setup.start.called)
 162
 163     def test_dont_count_missing_as_busy(self):
 164         size = testutil.MockSize(1)
 165         self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(1),
 166                                       testutil.cloud_node_mock(2)],
 167                          arvados_nodes=[testutil.arvados_node_mock(1),
 168                                       testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
 169                          want_sizes=[size, size])
 170         self.stop_proxy(self.daemon)
 171         self.assertTrue(self.node_setup.start.called)
 172
 173     def test_missing_counts_towards_max(self):
 174         size = testutil.MockSize(1)
 175         self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(1),
 176                                       testutil.cloud_node_mock(2)],
 177                          arvados_nodes=[testutil.arvados_node_mock(1),
 178                                         testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
 179                          want_sizes=[size, size],
 180                          max_nodes=2)
 181         self.stop_proxy(self.daemon)
 182         self.assertFalse(self.node_setup.start.called)
 183
 184     def test_excess_counts_missing(self):
 185         size = testutil.MockSize(1)
 186         cloud_nodes = [testutil.cloud_node_mock(1), testutil.cloud_node_mock(2)]
 187         self.make_daemon(cloud_nodes=cloud_nodes,
 188                          arvados_nodes=[testutil.arvados_node_mock(1),
 189                                         testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
 190                          want_sizes=[size])
 191         self.assertEqual(2, self.alive_monitor_count())
 192         for mon_ref in self.monitor_list():
 193             self.daemon.node_can_shutdown(mon_ref.proxy()).get(self.TIMEOUT)
 194         self.assertEqual(1, self.node_shutdown.start.call_count)
 195
 196     def test_missing_shutdown_not_excess(self):
 197         size = testutil.MockSize(1)
 198         cloud_nodes = [testutil.cloud_node_mock(1), testutil.cloud_node_mock(2)]
 199         self.make_daemon(cloud_nodes=cloud_nodes,
 200                          arvados_nodes=[testutil.arvados_node_mock(1),
 201                                         testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
 202                          want_sizes=[size])
 203
 204         get_cloud_node = mock.MagicMock(name="get_cloud_node")
 205         get_cloud_node.get.return_value = cloud_nodes[1]
 206         mock_node_monitor = mock.MagicMock()
 207         mock_node_monitor.proxy.return_value = mock.NonCallableMock(cloud_node=get_cloud_node)
 208         mock_shutdown = self.node_shutdown.start(node_monitor=mock_node_monitor)
 209
 210         self.daemon.shutdowns.get()[cloud_nodes[1].id] = mock_shutdown.proxy()
 211         self.daemon.sizes_booting_shutdown.get()[cloud_nodes[1].id] = size
 212
 213         self.assertEqual(2, self.alive_monitor_count())
 214         for mon_ref in self.monitor_list():
 215             self.daemon.node_can_shutdown(mon_ref.proxy()).get(self.TIMEOUT)
 216         self.assertEqual(1, self.node_shutdown.start.call_count)
 217
 218     def test_booting_nodes_counted(self):
 219         cloud_node = testutil.cloud_node_mock(1)
 220         arv_node = testutil.arvados_node_mock(1)
 221         server_wishlist = [testutil.MockSize(1)] * 2
 222         self.make_daemon([cloud_node], [arv_node], server_wishlist)
 223         self.daemon.max_nodes.get(self.TIMEOUT)
 224         self.assertTrue(self.node_setup.start.called)
 225         self.daemon.update_server_wishlist(server_wishlist).get(self.TIMEOUT)
 226         self.stop_proxy(self.daemon)
 227         self.assertEqual(1, self.node_setup.start.call_count)
 228
 229     def test_boot_new_node_when_all_nodes_busy(self):
 230         size = testutil.MockSize(2)
 231         arv_node = testutil.arvados_node_mock(2, job_uuid=True)
 232         self.make_daemon([testutil.cloud_node_mock(2, size=size)], [arv_node],
 233                          [size], avail_sizes=[(size, {"cores":1})])
 234         self.stop_proxy(self.daemon)
 235         self.assertTrue(self.node_setup.start.called)
 236
 237     def test_boot_new_node_below_min_nodes(self):
 238         min_size = testutil.MockSize(1)
 239         wish_size = testutil.MockSize(3)
 240         avail_sizes = [(min_size, {"cores": 1}),
 241                        (wish_size, {"cores": 3})]
 242         self.make_daemon([], [], None, avail_sizes=avail_sizes, min_nodes=2)
 243         self.daemon.update_server_wishlist([wish_size]).get(self.TIMEOUT)
 244         self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
 245         self.daemon.update_server_wishlist([wish_size]).get(self.TIMEOUT)
 246         self.stop_proxy(self.daemon)
 247         self.assertEqual([wish_size, min_size],
 248                          [call[1].get('cloud_size')
 249                           for call in self.node_setup.start.call_args_list])
 250
 251     def test_no_new_node_when_ge_min_nodes_busy(self):
 252         cloud_nodes = [testutil.cloud_node_mock(n) for n in range(1, 4)]
 253         arv_nodes = [testutil.arvados_node_mock(n, job_uuid=True)
 254                      for n in range(1, 4)]
 255         self.make_daemon(cloud_nodes, arv_nodes, [], min_nodes=2)
 256         self.stop_proxy(self.daemon)
 257         self.assertEqual(0, self.node_setup.start.call_count)
 258
 259     def test_no_new_node_when_max_nodes_busy(self):
 260         self.make_daemon([testutil.cloud_node_mock(3)],
 261                          [testutil.arvados_node_mock(3, job_uuid=True)],
 262                          [testutil.MockSize(3)],
 263                          max_nodes=1)
 264         self.stop_proxy(self.daemon)
 265         self.assertFalse(self.node_setup.start.called)
 266
 267     def start_node_boot(self, cloud_node=None, arv_node=None, id_num=1):
 268         if cloud_node is None:
 269             cloud_node = testutil.cloud_node_mock(id_num)
 270         if arv_node is None:
 271             arv_node = testutil.arvados_node_mock(id_num)
 272         self.make_daemon(want_sizes=[testutil.MockSize(id_num)],
 273                          avail_sizes=[(testutil.MockSize(id_num), {"cores":1})])
 274         self.daemon.max_nodes.get(self.TIMEOUT)
 275         self.assertEqual(1, self.node_setup.start.call_count)
 276         self.last_setup.cloud_node.get.return_value = cloud_node
 277         self.last_setup.arvados_node.get.return_value = arv_node
 278         return self.last_setup
 279
 280     def test_no_new_node_when_booted_node_not_usable(self):
 281         cloud_node = testutil.cloud_node_mock(4)
 282         arv_node = testutil.arvados_node_mock(4, crunch_worker_state='down')
 283         setup = self.start_node_boot(cloud_node, arv_node)
 284         self.daemon.node_up(setup).get(self.TIMEOUT)
 285         self.assertEqual(1, self.alive_monitor_count())
 286         self.daemon.update_cloud_nodes([cloud_node])
 287         self.daemon.update_arvados_nodes([arv_node])
 288         self.daemon.update_server_wishlist(
 289             [testutil.MockSize(1)]).get(self.TIMEOUT)
 290         self.stop_proxy(self.daemon)
 291         self.assertEqual(1, self.node_setup.start.call_count)
 292
 293     def test_no_duplication_when_booting_node_listed_fast(self):
 294         # Test that we don't start two ComputeNodeMonitorActors when
 295         # we learn about a booting node through a listing before we
 296         # get the "node up" message from CloudNodeSetupActor.
 297         cloud_node = testutil.cloud_node_mock(1)
 298         setup = self.start_node_boot(cloud_node)
 299         self.daemon.update_cloud_nodes([cloud_node])
 300         self.daemon.node_up(setup).get(self.TIMEOUT)
 301         self.assertEqual(1, self.alive_monitor_count())
 302
 303     def test_no_duplication_when_booted_node_listed(self):
 304         cloud_node = testutil.cloud_node_mock(2)
 305         setup = self.start_node_boot(cloud_node, id_num=2)
 306         self.daemon.node_up(setup)
 307         self.daemon.update_cloud_nodes([cloud_node]).get(self.TIMEOUT)
 308         self.assertEqual(1, self.alive_monitor_count())
 309
 310     def test_node_counted_after_boot_with_slow_listing(self):
 311         # Test that, after we boot a compute node, we assume it exists
 312         # even it doesn't appear in the listing (e.g., because of delays
 313         # propagating tags).
 314         setup = self.start_node_boot()
 315         self.daemon.node_up(setup).get(self.TIMEOUT)
 316         self.assertEqual(1, self.alive_monitor_count())
 317         self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
 318         self.assertEqual(1, self.alive_monitor_count())
 319
 320     def test_booted_unlisted_node_counted(self):
 321         setup = self.start_node_boot(id_num=1)
 322         self.daemon.node_up(setup)
 323         self.daemon.update_server_wishlist(
 324             [testutil.MockSize(1)]).get(self.TIMEOUT)
 325         self.stop_proxy(self.daemon)
 326         self.assertEqual(1, self.node_setup.start.call_count)
 327
 328     def test_booted_node_can_shutdown(self):
 329         setup = self.start_node_boot()
 330         self.daemon.node_up(setup).get(self.TIMEOUT)
 331         self.assertEqual(1, self.alive_monitor_count())
 332         monitor = self.monitor_list()[0].proxy()
 333         self.daemon.update_server_wishlist([])
 334         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
 335         self.stop_proxy(self.daemon)
 336         self.assertTrue(self.node_shutdown.start.called,
 337                         "daemon did not shut down booted node on offer")
 338
 339     def test_booted_node_lifecycle(self):
 340         cloud_node = testutil.cloud_node_mock(6)
 341         setup = self.start_node_boot(cloud_node, id_num=6)
 342         self.daemon.node_up(setup).get(self.TIMEOUT)
 343         self.assertEqual(1, self.alive_monitor_count())
 344         monitor = self.monitor_list()[0].proxy()
 345         self.daemon.update_server_wishlist([])
 346         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
 347         self.assertShutdownCancellable(True)
 348         shutdown = self.node_shutdown.start().proxy()
 349         shutdown.cloud_node.get.return_value = cloud_node
 350         self.daemon.node_finished_shutdown(shutdown).get(self.TIMEOUT)
 351         self.assertTrue(shutdown.stop.called,
 352                         "shutdown actor not stopped after finishing")
 353         self.assertTrue(monitor.actor_ref.actor_stopped.wait(self.TIMEOUT),
 354                         "monitor for booted node not stopped after shutdown")
 355         self.daemon.update_server_wishlist(
 356             [testutil.MockSize(2)]).get(self.TIMEOUT)
 357         self.stop_proxy(self.daemon)
 358         self.assertTrue(self.node_setup.start.called,
 359                         "second node not started after booted node stopped")
 360
 361     def test_booted_node_shut_down_when_never_listed(self):
 362         setup = self.start_node_boot()
 363         self.daemon.node_up(setup).get(self.TIMEOUT)
 364         self.assertEqual(1, self.alive_monitor_count())
 365         self.assertFalse(self.node_shutdown.start.called)
 366         self.timer.deliver()
 367         self.stop_proxy(self.daemon)
 368         self.assertShutdownCancellable(False)
 369
 370     def test_booted_node_shut_down_when_never_paired(self):
 371         cloud_node = testutil.cloud_node_mock(2)
 372         setup = self.start_node_boot(cloud_node)
 373         self.daemon.node_up(setup).get(self.TIMEOUT)
 374         self.assertEqual(1, self.alive_monitor_count())
 375         self.daemon.update_cloud_nodes([cloud_node])
 376         self.timer.deliver()
 377         self.stop_proxy(self.daemon)
 378         self.assertShutdownCancellable(False)
 379
 380     def test_booted_node_shut_down_when_never_working(self):
 381         cloud_node = testutil.cloud_node_mock(4)
 382         arv_node = testutil.arvados_node_mock(4, crunch_worker_state='down')
 383         setup = self.start_node_boot(cloud_node, arv_node)
 384         self.daemon.node_up(setup).get(self.TIMEOUT)
 385         self.assertEqual(1, self.alive_monitor_count())
 386         self.daemon.update_cloud_nodes([cloud_node])
 387         self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
 388         self.timer.deliver()
 389         self.stop_proxy(self.daemon)
 390         self.assertShutdownCancellable(False)
 391
 392     def test_node_that_pairs_not_considered_failed_boot(self):
 393         cloud_node = testutil.cloud_node_mock(3)
 394         arv_node = testutil.arvados_node_mock(3)
 395         setup = self.start_node_boot(cloud_node, arv_node)
 396         self.daemon.node_up(setup).get(self.TIMEOUT)
 397         self.assertEqual(1, self.alive_monitor_count())
 398         self.daemon.update_cloud_nodes([cloud_node])
 399         self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
 400         self.timer.deliver()
 401         self.stop_proxy(self.daemon)
 402         self.assertFalse(self.node_shutdown.start.called)
 403
 404     def test_node_that_pairs_busy_not_considered_failed_boot(self):
 405         cloud_node = testutil.cloud_node_mock(5)
 406         arv_node = testutil.arvados_node_mock(5, job_uuid=True)
 407         setup = self.start_node_boot(cloud_node, arv_node)
 408         self.daemon.node_up(setup).get(self.TIMEOUT)
 409         self.assertEqual(1, self.alive_monitor_count())
 410         self.daemon.update_cloud_nodes([cloud_node])
 411         self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
 412         self.timer.deliver()
 413         self.stop_proxy(self.daemon)
 414         self.assertFalse(self.node_shutdown.start.called)
 415
 416     def test_booting_nodes_shut_down(self):
 417         self.make_daemon(want_sizes=[testutil.MockSize(1)])
 418         self.daemon.update_server_wishlist([]).get(self.TIMEOUT)
 419         self.stop_proxy(self.daemon)
 420         self.assertTrue(self.last_setup.stop_if_no_cloud_node.called)
 421
 422     def test_all_booting_nodes_tried_to_shut_down(self):
 423         size = testutil.MockSize(2)
 424         self.make_daemon(want_sizes=[size], avail_sizes=[(size, {"cores":1})])
 425         self.daemon.max_nodes.get(self.TIMEOUT)
 426         setup1 = self.last_setup
 427         setup1.stop_if_no_cloud_node().get.return_value = False
 428         setup1.stop_if_no_cloud_node.reset_mock()
 429         self.daemon.update_server_wishlist([size, size]).get(self.TIMEOUT)
 430         self.daemon.max_nodes.get(self.TIMEOUT)
 431         self.assertIsNot(setup1, self.last_setup)
 432         self.last_setup.stop_if_no_cloud_node().get.return_value = True
 433         self.last_setup.stop_if_no_cloud_node.reset_mock()
 434         self.daemon.update_server_wishlist([]).get(self.TIMEOUT)
 435         self.daemon.max_nodes.get(self.TIMEOUT)
 436         self.stop_proxy(self.daemon)
 437         self.assertEqual(1, self.last_setup.stop_if_no_cloud_node.call_count)
 438         self.assertTrue(setup1.stop_if_no_cloud_node.called)
 439
 440     def test_shutdown_declined_at_wishlist_capacity(self):
 441         cloud_node = testutil.cloud_node_mock(1)
 442         size = testutil.MockSize(1)
 443         self.make_daemon(cloud_nodes=[cloud_node], want_sizes=[size])
 444         self.assertEqual(1, self.alive_monitor_count())
 445         monitor = self.monitor_list()[0].proxy()
 446         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
 447         self.stop_proxy(self.daemon)
 448         self.assertFalse(self.node_shutdown.start.called)
 449
 450     def test_shutdown_declined_below_min_nodes(self):
 451         cloud_node = testutil.cloud_node_mock(1)
 452         self.make_daemon(cloud_nodes=[cloud_node], min_nodes=1)
 453         self.assertEqual(1, self.alive_monitor_count())
 454         monitor = self.monitor_list()[0].proxy()
 455         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
 456         self.stop_proxy(self.daemon)
 457         self.assertFalse(self.node_shutdown.start.called)
 458
 459     def test_shutdown_accepted_below_capacity(self):
 460         self.make_daemon(cloud_nodes=[testutil.cloud_node_mock()])
 461         self.assertEqual(1, self.alive_monitor_count())
 462         monitor = self.monitor_list()[0].proxy()
 463         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
 464         self.stop_proxy(self.daemon)
 465         self.assertTrue(self.node_shutdown.start.called)
 466
 467     def test_shutdown_declined_when_idle_and_job_queued(self):
 468         cloud_nodes = [testutil.cloud_node_mock(n) for n in [3, 4]]
 469         arv_nodes = [testutil.arvados_node_mock(3, job_uuid=True),
 470                      testutil.arvados_node_mock(4, job_uuid=None)]
 471         self.make_daemon(cloud_nodes, arv_nodes, [testutil.MockSize(1)])
 472         self.assertEqual(2, self.alive_monitor_count())
 473         for mon_ref in self.monitor_list():
 474             monitor = mon_ref.proxy()
 475             if monitor.cloud_node.get(self.TIMEOUT) is cloud_nodes[-1]:
 476                 break
 477         else:
 478             self.fail("monitor for idle node not found")
 479         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
 480         self.stop_proxy(self.daemon)
 481         self.assertFalse(self.node_shutdown.start.called)
 482
 483     def test_node_shutdown_after_cancelled_shutdown(self):
 484         cloud_node = testutil.cloud_node_mock(5)
 485         self.make_daemon([cloud_node], [testutil.arvados_node_mock(5)])
 486         self.assertEqual(1, self.alive_monitor_count())
 487         monitor = self.monitor_list()[0].proxy()
 488         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
 489         self.last_shutdown.success.get.return_value = False
 490         self.daemon.node_finished_shutdown(self.last_shutdown).get(self.TIMEOUT)
 491         self.assertEqual(1, self.alive_monitor_count())
 492
 493         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
 494         self.last_shutdown.success.get.return_value = True
 495         self.last_shutdown.stop.side_effect = lambda: monitor.stop()
 496         self.daemon.node_finished_shutdown(self.last_shutdown).get(self.TIMEOUT)
 497         self.assertEqual(0, self.alive_monitor_count())
 498
 499     def test_broken_node_blackholed_after_cancelled_shutdown(self):
 500         size = testutil.MockSize(8)
 501         cloud_node = testutil.cloud_node_mock(8, size=size)
 502         wishlist = [size]
 503         self.make_daemon([cloud_node], [testutil.arvados_node_mock(8)],
 504                          wishlist, avail_sizes=[(size, {"cores":1})])
 505         self.assertEqual(1, self.alive_monitor_count())
 506         self.assertFalse(self.node_setup.start.called)
 507         monitor = self.monitor_list()[0].proxy()
 508         shutdown_proxy = self.node_shutdown.start().proxy
 509         shutdown_proxy().cloud_node.get.return_value = cloud_node
 510         shutdown_proxy().success.get.return_value = False
 511         shutdown_proxy().cancel_reason.get.return_value = self.node_shutdown.NODE_BROKEN
 512         self.daemon.update_server_wishlist([]).get(self.TIMEOUT)
 513         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
 514         self.daemon.node_finished_shutdown(shutdown_proxy()).get(self.TIMEOUT)
 515         self.daemon.update_cloud_nodes([cloud_node]).get(self.TIMEOUT)
 516         self.daemon.update_server_wishlist(wishlist).get(self.TIMEOUT)
 517         self.stop_proxy(self.daemon)
 518         self.assertEqual(1, self.node_setup.start.call_count)
 519
 520     def test_nodes_shutting_down_replaced_below_max_nodes(self):
 521         size = testutil.MockSize(6)
 522         cloud_node = testutil.cloud_node_mock(6, size=size)
 523         self.make_daemon([cloud_node], [testutil.arvados_node_mock(6)],
 524                          avail_sizes=[(size, {"cores":1})])
 525         self.assertEqual(1, self.alive_monitor_count())
 526         monitor = self.monitor_list()[0].proxy()
 527         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
 528         self.assertTrue(self.node_shutdown.start.called)
 529         self.daemon.update_server_wishlist(
 530             [testutil.MockSize(6)]).get(self.TIMEOUT)
 531         self.stop_proxy(self.daemon)
 532         self.assertTrue(self.node_setup.start.called)
 533
 534     def test_nodes_shutting_down_not_replaced_at_max_nodes(self):
 535         cloud_node = testutil.cloud_node_mock(7)
 536         self.make_daemon([cloud_node], [testutil.arvados_node_mock(7)],
 537                          max_nodes=1)
 538         self.assertEqual(1, self.alive_monitor_count())
 539         monitor = self.monitor_list()[0].proxy()
 540         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
 541         self.assertTrue(self.node_shutdown.start.called)
 542         self.daemon.update_server_wishlist(
 543             [testutil.MockSize(7)]).get(self.TIMEOUT)
 544         self.stop_proxy(self.daemon)
 545         self.assertFalse(self.node_setup.start.called)
 546
 547     def test_nodes_shutting_down_count_against_excess(self):
 548         size = testutil.MockSize(8)
 549         cloud_nodes = [testutil.cloud_node_mock(n, size=size) for n in [8, 9]]
 550         arv_nodes = [testutil.arvados_node_mock(n, size=size) for n in [8, 9]]
 551         self.make_daemon(cloud_nodes, arv_nodes, [size],
 552                          avail_sizes=[(size, {"cores":1})])
 553         self.assertEqual(2, self.alive_monitor_count())
 554         for mon_ref in self.monitor_list():
 555             self.daemon.node_can_shutdown(mon_ref.proxy()).get(self.TIMEOUT)
 556         self.assertEqual(1, self.node_shutdown.start.call_count)
 557
 558     def test_clean_shutdown_waits_for_node_setup_finish(self):
 559         new_node = self.start_node_boot()
 560         new_node.stop_if_no_cloud_node().get.return_value = False
 561         new_node.stop_if_no_cloud_node.reset_mock()
 562         self.daemon.shutdown().get(self.TIMEOUT)
 563         self.assertTrue(new_node.stop_if_no_cloud_node.called)
 564         self.daemon.node_up(new_node).get(self.TIMEOUT)
 565         self.assertTrue(new_node.stop.called)
 566         self.timer.deliver()
 567         self.assertTrue(
 568             self.daemon.actor_ref.actor_stopped.wait(self.TIMEOUT))
 569
 570     def test_wishlist_ignored_after_shutdown(self):
 571         new_node = self.start_node_boot()
 572         new_node.stop_if_no_cloud_node().get.return_value = False
 573         new_node.stop_if_no_cloud_node.reset_mock()
 574         self.daemon.shutdown().get(self.TIMEOUT)
 575         size = testutil.MockSize(2)
 576         self.daemon.update_server_wishlist([size] * 2).get(self.TIMEOUT)
 577         self.timer.deliver()
 578         self.stop_proxy(self.daemon)
 579         self.assertEqual(1, self.node_setup.start.call_count)
 580
 581     def test_shutdown_actor_stopped_when_cloud_node_delisted(self):
 582         self.make_daemon(cloud_nodes=[testutil.cloud_node_mock()])
 583         self.assertEqual(1, self.alive_monitor_count())
 584         monitor = self.monitor_list()[0].proxy()
 585         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
 586         self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
 587         self.stop_proxy(self.daemon)
 588         self.assertEqual(
 589             1, self.last_shutdown.stop.call_count)
 590
 591     def test_shutdown_actor_cleanup_copes_with_dead_actors(self):
 592         self.make_daemon(cloud_nodes=[testutil.cloud_node_mock()])
 593         self.assertEqual(1, self.alive_monitor_count())
 594         monitor = self.monitor_list()[0].proxy()
 595         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
 596         # We're mainly testing that update_cloud_nodes catches and handles
 597         # the ActorDeadError.
 598         self.last_shutdown.stop.side_effect = pykka.ActorDeadError
 599         self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
 600         self.stop_proxy(self.daemon)
 601         self.assertEqual(1, self.last_shutdown.stop.call_count)
 602
 603     def busywait(self, f):
 604         n = 0
 605         while not f() and n < 10:
 606             time.sleep(.1)
 607             n += 1
 608         self.assertTrue(f())
 609
 610     def test_node_create_two_sizes(self):
 611         small = testutil.MockSize(1)
 612         big = testutil.MockSize(2)
 613         avail_sizes = [(testutil.MockSize(1), {"cores":1}),
 614                         (testutil.MockSize(2), {"cores":2})]
 615         self.make_daemon(want_sizes=[small, small, small, big],
 616                          avail_sizes=avail_sizes, max_nodes=4)
 617
 618         # the daemon runs in another thread, so we need to wait and see
 619         # if it does all the work we're expecting it to do before stopping it.
 620         self.busywait(lambda: self.node_setup.start.call_count == 4)
 621         booting = self.daemon.booting.get(self.TIMEOUT)
 622         self.stop_proxy(self.daemon)
 623         sizecounts = {a[0].id: 0 for a in avail_sizes}
 624         for b in booting.itervalues():
 625             sizecounts[b.cloud_size.get().id] += 1
 626         logging.info(sizecounts)
 627         self.assertEqual(3, sizecounts[small.id])
 628         self.assertEqual(1, sizecounts[big.id])
 629
 630     def test_node_max_nodes_two_sizes(self):
 631         small = testutil.MockSize(1)
 632         big = testutil.MockSize(2)
 633         avail_sizes = [(testutil.MockSize(1), {"cores":1}),
 634                         (testutil.MockSize(2), {"cores":2})]
 635         self.make_daemon(want_sizes=[small, small, small, big],
 636                          avail_sizes=avail_sizes, max_nodes=3)
 637
 638         # the daemon runs in another thread, so we need to wait and see
 639         # if it does all the work we're expecting it to do before stopping it.
 640         self.busywait(lambda: self.node_setup.start.call_count == 3)
 641         booting = self.daemon.booting.get(self.TIMEOUT)
 642         self.stop_proxy(self.daemon)
 643         sizecounts = {a[0].id: 0 for a in avail_sizes}
 644         for b in booting.itervalues():
 645             sizecounts[b.cloud_size.get().id] += 1
 646         self.assertEqual(2, sizecounts[small.id])
 647         self.assertEqual(1, sizecounts[big.id])
 648
 649     def test_wishlist_reconfigure(self):
 650         small = testutil.MockSize(1)
 651         big = testutil.MockSize(2)
 652         avail_sizes = [(small, {"cores":1}), (big, {"cores":2})]
 653
 654         self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(1, small),
 655                                       testutil.cloud_node_mock(2, small),
 656                                       testutil.cloud_node_mock(3, big)],
 657                          arvados_nodes=[testutil.arvados_node_mock(1),
 658                                         testutil.arvados_node_mock(2),
 659                                         testutil.arvados_node_mock(3)],
 660                          want_sizes=[small, small, big],
 661                          avail_sizes=avail_sizes)
 662
 663         self.daemon.update_server_wishlist([small, big, big]).get(self.TIMEOUT)
 664
 665         self.assertEqual(0, self.node_shutdown.start.call_count)
 666
 667         for c in self.daemon.cloud_nodes.get().nodes.itervalues():
 668             self.daemon.node_can_shutdown(c.actor)
 669
 670         booting = self.daemon.booting.get()
 671         shutdowns = self.daemon.shutdowns.get()
 672
 673         self.stop_proxy(self.daemon)
 674
 675         self.assertEqual(1, self.node_setup.start.call_count)
 676         self.assertEqual(1, self.node_shutdown.start.call_count)
 677
 678         # booting a new big node
 679         sizecounts = {a[0].id: 0 for a in avail_sizes}
 680         for b in booting.itervalues():
 681             sizecounts[b.cloud_size.get().id] += 1
 682         self.assertEqual(0, sizecounts[small.id])
 683         self.assertEqual(1, sizecounts[big.id])
 684
 685         # shutting down a small node
 686         sizecounts = {a[0].id: 0 for a in avail_sizes}
 687         for b in shutdowns.itervalues():
 688             sizecounts[b.cloud_node.get().size.id] += 1
 689         self.assertEqual(1, sizecounts[small.id])
 690         self.assertEqual(0, sizecounts[big.id])
 691
 692     def test_node_max_price(self):
 693         small = testutil.MockSize(1)
 694         big = testutil.MockSize(2)
 695         avail_sizes = [(testutil.MockSize(1), {"cores":1, "price":1}),
 696                         (testutil.MockSize(2), {"cores":2, "price":2})]
 697         self.make_daemon(want_sizes=[small, small, small, big],
 698                          avail_sizes=avail_sizes,
 699                          max_nodes=4,
 700                          max_total_price=4)
 701         # the daemon runs in another thread, so we need to wait and see
 702         # if it does all the work we're expecting it to do before stopping it.
 703         self.busywait(lambda: self.node_setup.start.call_count == 3)
 704         booting = self.daemon.booting.get()
 705         self.stop_proxy(self.daemon)
 706
 707         sizecounts = {a[0].id: 0 for a in avail_sizes}
 708         for b in booting.itervalues():
 709             sizecounts[b.cloud_size.get().id] += 1
 710         logging.info(sizecounts)
 711
 712         # Booting 3 small nodes and not booting a big node would also partially
 713         # satisfy the wishlist and come in under the price cap, however the way
 714         # the update_server_wishlist() currently works effectively results in a
 715         # round-robin creation of one node of each size in the wishlist, so
 716         # test for that.
 717         self.assertEqual(2, sizecounts[small.id])
 718         self.assertEqual(1, sizecounts[big.id])