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         mock_actor.tell_proxy.return_value = mock_proxy
  30
  31         self.last_setup = mock_proxy
  32         return mock_actor
  33
  34     def mock_node_shutdown(self, **kwargs):
  35         # Make sure that every time the daemon starts a shutdown actor,
  36         # it gets a new mock object back.
  37         get_cloud_node = mock.MagicMock()
  38         if "node_monitor" in kwargs:
  39             get_cloud_node.get.return_value = kwargs["node_monitor"].proxy().cloud_node.get()
  40         mock_actor = mock.MagicMock()
  41         mock_proxy = mock.NonCallableMock(name='shutdown_mock_proxy',
  42                                           cloud_node=get_cloud_node,
  43                                           actor_ref=mock_actor)
  44
  45         mock_actor.proxy.return_value = mock_proxy
  46         self.last_shutdown = mock_proxy
  47
  48         return mock_actor
  49
  50     def make_daemon(self, cloud_nodes=[], arvados_nodes=[], want_sizes=[],
  51                     avail_sizes=[(testutil.MockSize(1), {"cores": 1})],
  52                     min_nodes=0, max_nodes=8,
  53                     shutdown_windows=[54, 5, 1],
  54                     max_total_price=None):
  55         for name in ['cloud_nodes', 'arvados_nodes', 'server_wishlist']:
  56             setattr(self, name + '_poller', mock.MagicMock(name=name + '_mock'))
  57         self.arv_factory = mock.MagicMock(name='arvados_mock')
  58         api_client = mock.MagicMock(name='api_client')
  59         api_client.nodes().create().execute.side_effect = [testutil.arvados_node_mock(1),
  60                                                            testutil.arvados_node_mock(2)]
  61         self.arv_factory.return_value = api_client
  62
  63         self.cloud_factory = mock.MagicMock(name='cloud_mock')
  64         self.cloud_factory().node_start_time.return_value = time.time()
  65         self.cloud_updates = mock.MagicMock(name='updates_mock')
  66         self.timer = testutil.MockTimer(deliver_immediately=False)
  67         self.cloud_factory().node_id.side_effect = lambda node: node.id
  68
  69         self.node_setup = mock.MagicMock(name='setup_mock')
  70         self.node_setup.start.side_effect = self.mock_node_start
  71         self.node_setup.reset_mock()
  72
  73         self.node_shutdown = mock.MagicMock(name='shutdown_mock')
  74         self.node_shutdown.start.side_effect = self.mock_node_shutdown
  75
  76         self.daemon = nmdaemon.NodeManagerDaemonActor.start(
  77             self.server_wishlist_poller, self.arvados_nodes_poller,
  78             self.cloud_nodes_poller, self.cloud_updates, self.timer,
  79             self.arv_factory, self.cloud_factory,
  80             shutdown_windows, ServerCalculator(avail_sizes),
  81             min_nodes, max_nodes, 600, 1800, 3600,
  82             self.node_setup, self.node_shutdown,
  83             max_total_price=max_total_price).proxy()
  84         if arvados_nodes is not None:
  85             self.daemon.update_arvados_nodes(arvados_nodes).get(self.TIMEOUT)
  86         if cloud_nodes is not None:
  87             self.daemon.update_cloud_nodes(cloud_nodes).get(self.TIMEOUT)
  88         if want_sizes is not None:
  89             self.daemon.update_server_wishlist(want_sizes).get(self.TIMEOUT)
  90
  91     def monitor_list(self):
  92         return pykka.ActorRegistry.get_by_class(ComputeNodeMonitorActor)
  93
  94     def monitored_arvados_nodes(self):
  95         pairings = []
  96         for future in [actor.proxy().arvados_node
  97                        for actor in self.monitor_list()]:
  98             try:
  99                 pairings.append(future.get(self.TIMEOUT))
 100             except pykka.ActorDeadError:
 101                 pass
 102         return pairings
 103
 104     def alive_monitor_count(self):
 105         return len(self.monitored_arvados_nodes())
 106
 107     def assertShutdownCancellable(self, expected=True):
 108         self.assertTrue(self.node_shutdown.start.called)
 109         self.assertIs(expected,
 110                       self.node_shutdown.start.call_args[1]['cancellable'],
 111                       "ComputeNodeShutdownActor incorrectly cancellable")
 112
 113     def test_easy_node_creation(self):
 114         size = testutil.MockSize(1)
 115         self.make_daemon(want_sizes=[size])
 116         self.stop_proxy(self.daemon)
 117         self.assertTrue(self.node_setup.start.called)
 118
 119     def check_monitors_arvados_nodes(self, *arv_nodes):
 120         self.assertItemsEqual(arv_nodes, self.monitored_arvados_nodes())
 121
 122     def test_node_pairing(self):
 123         cloud_node = testutil.cloud_node_mock(1)
 124         arv_node = testutil.arvados_node_mock(1)
 125         self.make_daemon([cloud_node], [arv_node])
 126         self.stop_proxy(self.daemon)
 127         self.check_monitors_arvados_nodes(arv_node)
 128
 129     def test_node_pairing_after_arvados_update(self):
 130         cloud_node = testutil.cloud_node_mock(2)
 131         self.make_daemon([cloud_node],
 132                          [testutil.arvados_node_mock(1, ip_address=None)])
 133         arv_node = testutil.arvados_node_mock(2)
 134         self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
 135         self.stop_proxy(self.daemon)
 136         self.check_monitors_arvados_nodes(arv_node)
 137
 138     def test_arvados_node_un_and_re_paired(self):
 139         # We need to create the Arvados node mock after spinning up the daemon
 140         # to make sure it's new enough to pair with the cloud node.
 141         self.make_daemon([testutil.cloud_node_mock(3)], arvados_nodes=None)
 142         arv_node = testutil.arvados_node_mock(3)
 143         self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
 144         self.check_monitors_arvados_nodes(arv_node)
 145         self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
 146         self.assertEqual(0, self.alive_monitor_count())
 147         self.daemon.update_cloud_nodes([testutil.cloud_node_mock(3)])
 148         self.stop_proxy(self.daemon)
 149         self.check_monitors_arvados_nodes(arv_node)
 150
 151     def test_old_arvados_node_not_double_assigned(self):
 152         arv_node = testutil.arvados_node_mock(3, age=9000)
 153         size = testutil.MockSize(3)
 154         self.make_daemon(arvados_nodes=[arv_node], avail_sizes=[(size, {"cores":1})])
 155         self.daemon.update_server_wishlist([size]).get(self.TIMEOUT)
 156         self.daemon.update_server_wishlist([size, size]).get(self.TIMEOUT)
 157         self.stop_proxy(self.daemon)
 158         used_nodes = [call[1].get('arvados_node')
 159                       for call in self.node_setup.start.call_args_list]
 160         self.assertEqual(2, len(used_nodes))
 161         self.assertIn(arv_node, used_nodes)
 162         self.assertIn(None, used_nodes)
 163
 164     def test_node_count_satisfied(self):
 165         self.make_daemon([testutil.cloud_node_mock()],
 166                          want_sizes=[testutil.MockSize(1)])
 167         self.stop_proxy(self.daemon)
 168         self.assertFalse(self.node_setup.start.called)
 169
 170     def test_dont_count_missing_as_busy(self):
 171         size = testutil.MockSize(1)
 172         self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(1),
 173                                       testutil.cloud_node_mock(2)],
 174                          arvados_nodes=[testutil.arvados_node_mock(1),
 175                                       testutil.arvados_node_mock(2,
 176                                                                  last_ping_at='1970-01-01T01:02:03.04050607Z')],
 177                          want_sizes=[size, size])
 178         self.stop_proxy(self.daemon)
 179         self.assertTrue(self.node_setup.start.called)
 180
 181     def test_missing_counts_towards_max(self):
 182         size = testutil.MockSize(1)
 183         self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(1),
 184                                       testutil.cloud_node_mock(2)],
 185                          arvados_nodes=[testutil.arvados_node_mock(1),
 186                                         testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
 187                          want_sizes=[size, size],
 188                          max_nodes=2)
 189         self.stop_proxy(self.daemon)
 190         self.assertFalse(self.node_setup.start.called)
 191
 192     def test_excess_counts_missing(self):
 193         size = testutil.MockSize(1)
 194         cloud_nodes = [testutil.cloud_node_mock(1), testutil.cloud_node_mock(2)]
 195         self.make_daemon(cloud_nodes=cloud_nodes,
 196                          arvados_nodes=[testutil.arvados_node_mock(1),
 197                                         testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
 198                          want_sizes=[size])
 199         self.assertEqual(2, self.alive_monitor_count())
 200         for mon_ref in self.monitor_list():
 201             self.daemon.node_can_shutdown(mon_ref.proxy()).get(self.TIMEOUT)
 202         self.assertEqual(1, self.node_shutdown.start.call_count)
 203
 204     def test_missing_shutdown_not_excess(self):
 205         size = testutil.MockSize(1)
 206         cloud_nodes = [testutil.cloud_node_mock(1), testutil.cloud_node_mock(2)]
 207         self.make_daemon(cloud_nodes=cloud_nodes,
 208                          arvados_nodes=[testutil.arvados_node_mock(1),
 209                                         testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
 210                          want_sizes=[size])
 211
 212         get_cloud_node = mock.MagicMock(name="get_cloud_node")
 213         get_cloud_node.get.return_value = cloud_nodes[1]
 214         mock_node_monitor = mock.MagicMock()
 215         mock_node_monitor.proxy.return_value = mock.NonCallableMock(cloud_node=get_cloud_node)
 216         mock_shutdown = self.node_shutdown.start(node_monitor=mock_node_monitor)
 217
 218         self.daemon.cloud_nodes.get()[cloud_nodes[1].id].shutdown_actor = mock_shutdown.proxy()
 219
 220         self.assertEqual(2, self.alive_monitor_count())
 221         for mon_ref in self.monitor_list():
 222             self.daemon.node_can_shutdown(mon_ref.proxy()).get(self.TIMEOUT)
 223         self.assertEqual(1, self.node_shutdown.start.call_count)
 224
 225     def test_booting_nodes_counted(self):
 226         cloud_node = testutil.cloud_node_mock(1)
 227         arv_node = testutil.arvados_node_mock(1)
 228         server_wishlist = [testutil.MockSize(1)] * 2
 229         self.make_daemon([cloud_node], [arv_node], server_wishlist)
 230         self.daemon.max_nodes.get(self.TIMEOUT)
 231         self.assertTrue(self.node_setup.start.called)
 232         self.daemon.update_server_wishlist(server_wishlist).get(self.TIMEOUT)
 233         self.stop_proxy(self.daemon)
 234         self.assertEqual(1, self.node_setup.start.call_count)
 235
 236     def test_boot_new_node_when_all_nodes_busy(self):
 237         size = testutil.MockSize(2)
 238         arv_node = testutil.arvados_node_mock(2, job_uuid=True)
 239         self.make_daemon([testutil.cloud_node_mock(2, size=size)], [arv_node],
 240                          [size], avail_sizes=[(size, {"cores":1})])
 241         self.busywait(lambda: self.node_setup.start.called)
 242         self.stop_proxy(self.daemon)
 243         self.assertTrue(self.node_setup.start.called)
 244
 245     def test_boot_new_node_below_min_nodes(self):
 246         min_size = testutil.MockSize(1)
 247         wish_size = testutil.MockSize(3)
 248         avail_sizes = [(min_size, {"cores": 1}),
 249                        (wish_size, {"cores": 3})]
 250         self.make_daemon([], [], None, avail_sizes=avail_sizes, min_nodes=2)
 251         self.daemon.update_server_wishlist([wish_size]).get(self.TIMEOUT)
 252         self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
 253         self.daemon.update_server_wishlist([wish_size]).get(self.TIMEOUT)
 254         self.stop_proxy(self.daemon)
 255         self.assertEqual([wish_size, min_size],
 256                          [call[1].get('cloud_size')
 257                           for call in self.node_setup.start.call_args_list])
 258
 259     def test_no_new_node_when_ge_min_nodes_busy(self):
 260         cloud_nodes = [testutil.cloud_node_mock(n) for n in range(1, 4)]
 261         arv_nodes = [testutil.arvados_node_mock(n, job_uuid=True)
 262                      for n in range(1, 4)]
 263         self.make_daemon(cloud_nodes, arv_nodes, [], min_nodes=2)
 264         self.stop_proxy(self.daemon)
 265         self.assertEqual(0, self.node_setup.start.call_count)
 266
 267     def test_no_new_node_when_max_nodes_busy(self):
 268         self.make_daemon([testutil.cloud_node_mock(3)],
 269                          [testutil.arvados_node_mock(3, job_uuid=True)],
 270                          [testutil.MockSize(3)],
 271                          max_nodes=1)
 272         self.stop_proxy(self.daemon)
 273         self.assertFalse(self.node_setup.start.called)
 274
 275     def start_node_boot(self, cloud_node=None, arv_node=None, id_num=1):
 276         if cloud_node is None:
 277             cloud_node = testutil.cloud_node_mock(id_num)
 278         if arv_node is None:
 279             arv_node = testutil.arvados_node_mock(id_num)
 280         self.make_daemon(want_sizes=[testutil.MockSize(id_num)],
 281                          avail_sizes=[(testutil.MockSize(id_num), {"cores":1})])
 282         self.daemon.max_nodes.get(self.TIMEOUT)
 283         self.assertEqual(1, self.node_setup.start.call_count)
 284         self.last_setup.cloud_node.get.return_value = cloud_node
 285         self.last_setup.arvados_node.get.return_value = arv_node
 286         return self.last_setup
 287
 288     def test_new_node_when_booted_node_not_usable(self):
 289         cloud_node = testutil.cloud_node_mock(4)
 290         arv_node = testutil.arvados_node_mock(4, crunch_worker_state='down')
 291         setup = self.start_node_boot(cloud_node, arv_node)
 292         self.daemon.node_up(setup).get(self.TIMEOUT)
 293         self.assertEqual(1, self.alive_monitor_count())
 294         self.daemon.update_arvados_nodes([arv_node])
 295         self.daemon.update_cloud_nodes([cloud_node])
 296         self.monitor_list()[0].proxy().cloud_node_start_time = time.time()-1801
 297         self.daemon.update_server_wishlist(
 298             [testutil.MockSize(1)]).get(self.TIMEOUT)
 299         self.stop_proxy(self.daemon)
 300         self.assertEqual(2, self.node_setup.start.call_count)
 301
 302     def test_no_duplication_when_booting_node_listed_fast(self):
 303         # Test that we don't start two ComputeNodeMonitorActors when
 304         # we learn about a booting node through a listing before we
 305         # get the "node up" message from CloudNodeSetupActor.
 306         cloud_node = testutil.cloud_node_mock(1)
 307         setup = self.start_node_boot(cloud_node)
 308         self.daemon.update_cloud_nodes([cloud_node])
 309         self.daemon.node_up(setup).get(self.TIMEOUT)
 310         self.assertEqual(1, self.alive_monitor_count())
 311
 312     def test_no_duplication_when_booted_node_listed(self):
 313         cloud_node = testutil.cloud_node_mock(2)
 314         setup = self.start_node_boot(cloud_node, id_num=2)
 315         self.daemon.node_up(setup)
 316         self.daemon.update_cloud_nodes([cloud_node]).get(self.TIMEOUT)
 317         self.assertEqual(1, self.alive_monitor_count())
 318
 319     def test_node_counted_after_boot_with_slow_listing(self):
 320         # Test that, after we boot a compute node, we assume it exists
 321         # even it doesn't appear in the listing (e.g., because of delays
 322         # propagating tags).
 323         setup = self.start_node_boot()
 324         self.daemon.node_up(setup).get(self.TIMEOUT)
 325         self.assertEqual(1, self.alive_monitor_count())
 326         self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
 327         self.assertEqual(1, self.alive_monitor_count())
 328
 329     def test_booted_unlisted_node_counted(self):
 330         setup = self.start_node_boot(id_num=1)
 331         self.daemon.node_up(setup)
 332         self.daemon.update_server_wishlist(
 333             [testutil.MockSize(1)]).get(self.TIMEOUT)
 334         self.stop_proxy(self.daemon)
 335         self.assertEqual(1, self.node_setup.start.call_count)
 336
 337     def test_booted_node_can_shutdown(self):
 338         setup = self.start_node_boot()
 339         self.daemon.node_up(setup).get(self.TIMEOUT)
 340         self.assertEqual(1, self.alive_monitor_count())
 341         monitor = self.monitor_list()[0].proxy()
 342         self.daemon.update_server_wishlist([])
 343         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
 344         self.stop_proxy(self.daemon)
 345         self.assertTrue(self.node_shutdown.start.called,
 346                         "daemon did not shut down booted node on offer")
 347
 348     def test_booted_node_lifecycle(self):
 349         cloud_node = testutil.cloud_node_mock(6)
 350         setup = self.start_node_boot(cloud_node, id_num=6)
 351         self.daemon.node_up(setup).get(self.TIMEOUT)
 352         self.assertEqual(1, self.alive_monitor_count())
 353         monitor = self.monitor_list()[0].proxy()
 354         self.daemon.update_server_wishlist([])
 355         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
 356         self.assertShutdownCancellable(True)
 357         shutdown = self.node_shutdown.start().proxy()
 358         shutdown.cloud_node.get.return_value = cloud_node
 359         self.daemon.node_finished_shutdown(shutdown).get(self.TIMEOUT)
 360         self.daemon.update_cloud_nodes([])
 361         self.assertTrue(shutdown.stop.called,
 362                         "shutdown actor not stopped after finishing")
 363         self.assertTrue(monitor.actor_ref.actor_stopped.wait(self.TIMEOUT),
 364                         "monitor for booted node not stopped after shutdown")
 365         self.daemon.update_server_wishlist(
 366             [testutil.MockSize(2)]).get(self.TIMEOUT)
 367         self.stop_proxy(self.daemon)
 368         self.assertTrue(self.node_setup.start.called,
 369                         "second node not started after booted node stopped")
 370
 371     def test_booted_node_shut_down_when_never_listed(self):
 372         setup = self.start_node_boot()
 373         self.cloud_factory().node_start_time.return_value = time.time() - 3601
 374         self.daemon.node_up(setup).get(self.TIMEOUT)
 375         self.assertEqual(1, self.alive_monitor_count())
 376         self.assertFalse(self.node_shutdown.start.called)
 377         now = time.time()
 378         self.monitor_list()[0].tell_proxy().consider_shutdown()
 379         self.busywait(lambda: self.node_shutdown.start.called)
 380         self.stop_proxy(self.daemon)
 381         self.assertShutdownCancellable(False)
 382
 383     def test_booted_node_shut_down_when_never_paired(self):
 384         cloud_node = testutil.cloud_node_mock(2)
 385         setup = self.start_node_boot(cloud_node)
 386         self.cloud_factory().node_start_time.return_value = time.time() - 3601
 387         self.daemon.node_up(setup).get(self.TIMEOUT)
 388         self.assertEqual(1, self.alive_monitor_count())
 389         self.daemon.update_cloud_nodes([cloud_node])
 390         self.monitor_list()[0].tell_proxy().consider_shutdown()
 391         self.busywait(lambda: self.node_shutdown.start.called)
 392         self.stop_proxy(self.daemon)
 393         self.assertShutdownCancellable(False)
 394
 395     def test_booted_node_shut_down_when_never_working(self):
 396         cloud_node = testutil.cloud_node_mock(4)
 397         arv_node = testutil.arvados_node_mock(4, crunch_worker_state='down')
 398         setup = self.start_node_boot(cloud_node, arv_node)
 399         self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
 400         self.daemon.node_up(setup).get(self.TIMEOUT)
 401         self.assertEqual(1, self.alive_monitor_count())
 402         self.monitor_list()[0].proxy().cloud_node_start_time = time.time()-3601
 403         self.daemon.update_cloud_nodes([cloud_node])
 404         self.busywait(lambda: self.node_shutdown.start.called)
 405         self.stop_proxy(self.daemon)
 406         self.assertShutdownCancellable(False)
 407
 408     def test_node_that_pairs_not_considered_failed_boot(self):
 409         cloud_node = testutil.cloud_node_mock(3)
 410         arv_node = testutil.arvados_node_mock(3)
 411         setup = self.start_node_boot(cloud_node, arv_node)
 412         self.daemon.node_up(setup).get(self.TIMEOUT)
 413         self.assertEqual(1, self.alive_monitor_count())
 414         self.daemon.update_cloud_nodes([cloud_node])
 415         self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
 416         self.timer.deliver()
 417         self.stop_proxy(self.daemon)
 418         self.assertFalse(self.node_shutdown.start.called)
 419
 420     def test_node_that_pairs_busy_not_considered_failed_boot(self):
 421         cloud_node = testutil.cloud_node_mock(5)
 422         arv_node = testutil.arvados_node_mock(5, job_uuid=True)
 423         setup = self.start_node_boot(cloud_node, arv_node)
 424         self.daemon.node_up(setup).get(self.TIMEOUT)
 425         self.assertEqual(1, self.alive_monitor_count())
 426         self.daemon.update_cloud_nodes([cloud_node])
 427         self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
 428         self.timer.deliver()
 429         self.stop_proxy(self.daemon)
 430         self.assertFalse(self.node_shutdown.start.called)
 431
 432     def test_booting_nodes_shut_down(self):
 433         self.make_daemon(want_sizes=[testutil.MockSize(1)])
 434         self.daemon.update_server_wishlist([]).get(self.TIMEOUT)
 435         self.stop_proxy(self.daemon)
 436         self.assertTrue(self.last_setup.stop_if_no_cloud_node.called)
 437
 438     def test_all_booting_nodes_tried_to_shut_down(self):
 439         size = testutil.MockSize(2)
 440         self.make_daemon(want_sizes=[size], avail_sizes=[(size, {"cores":1})])
 441         self.daemon.max_nodes.get(self.TIMEOUT)
 442         setup1 = self.last_setup
 443         setup1.stop_if_no_cloud_node().get.return_value = False
 444         setup1.stop_if_no_cloud_node.reset_mock()
 445         self.daemon.update_server_wishlist([size, size]).get(self.TIMEOUT)
 446         self.daemon.max_nodes.get(self.TIMEOUT)
 447         self.assertIsNot(setup1, self.last_setup)
 448         self.last_setup.stop_if_no_cloud_node().get.return_value = True
 449         self.last_setup.stop_if_no_cloud_node.reset_mock()
 450         self.daemon.update_server_wishlist([]).get(self.TIMEOUT)
 451         self.daemon.max_nodes.get(self.TIMEOUT)
 452         self.stop_proxy(self.daemon)
 453         self.assertEqual(1, self.last_setup.stop_if_no_cloud_node.call_count)
 454         self.assertTrue(setup1.stop_if_no_cloud_node.called)
 455
 456     def test_shutdown_declined_at_wishlist_capacity(self):
 457         cloud_node = testutil.cloud_node_mock(1)
 458         arv_node = testutil.arvados_node_mock(1)
 459         size = testutil.MockSize(1)
 460         self.make_daemon(cloud_nodes=[cloud_node], arvados_nodes=[arv_node], want_sizes=[size])
 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.assertFalse(self.node_shutdown.start.called)
 466
 467     def test_shutdown_declined_below_min_nodes(self):
 468         cloud_node = testutil.cloud_node_mock(1)
 469         arv_node = testutil.arvados_node_mock(1)
 470         self.make_daemon(cloud_nodes=[cloud_node], arvados_nodes=[arv_node], min_nodes=1)
 471         self.assertEqual(1, self.alive_monitor_count())
 472         monitor = self.monitor_list()[0].proxy()
 473         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
 474         self.stop_proxy(self.daemon)
 475         self.assertFalse(self.node_shutdown.start.called)
 476
 477     def test_shutdown_accepted_below_capacity(self):
 478         self.make_daemon(cloud_nodes=[testutil.cloud_node_mock()])
 479         self.assertEqual(1, self.alive_monitor_count())
 480         monitor = self.monitor_list()[0].proxy()
 481         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
 482         self.stop_proxy(self.daemon)
 483         self.assertTrue(self.node_shutdown.start.called)
 484
 485     def test_shutdown_declined_when_idle_and_job_queued(self):
 486         cloud_nodes = [testutil.cloud_node_mock(n) for n in [3, 4]]
 487         arv_nodes = [testutil.arvados_node_mock(3, job_uuid=True),
 488                      testutil.arvados_node_mock(4, job_uuid=None)]
 489         self.make_daemon(cloud_nodes, arv_nodes, [testutil.MockSize(1)])
 490         self.assertEqual(2, self.alive_monitor_count())
 491         for mon_ref in self.monitor_list():
 492             monitor = mon_ref.proxy()
 493             if monitor.cloud_node.get(self.TIMEOUT) is cloud_nodes[-1]:
 494                 break
 495         else:
 496             self.fail("monitor for idle node not found")
 497         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
 498         self.stop_proxy(self.daemon)
 499         self.assertFalse(self.node_shutdown.start.called)
 500
 501     def test_node_shutdown_after_cancelled_shutdown(self):
 502         cloud_node = testutil.cloud_node_mock(5)
 503         self.make_daemon([cloud_node], [testutil.arvados_node_mock(5)])
 504         self.assertEqual(1, self.alive_monitor_count())
 505         monitor = self.monitor_list()[0].proxy()
 506         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
 507         self.last_shutdown.success.get.return_value = False
 508         self.daemon.node_finished_shutdown(self.last_shutdown).get(self.TIMEOUT)
 509         self.assertEqual(1, self.alive_monitor_count())
 510
 511         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
 512         self.last_shutdown.success.get.return_value = True
 513         self.last_shutdown.stop.side_effect = lambda: monitor.stop()
 514         self.daemon.node_finished_shutdown(self.last_shutdown).get(self.TIMEOUT)
 515         self.assertEqual(0, self.alive_monitor_count())
 516
 517     def test_broken_node_blackholed_after_cancelled_shutdown(self):
 518         size = testutil.MockSize(8)
 519         cloud_node = testutil.cloud_node_mock(8, size=size)
 520         wishlist = [size]
 521         self.make_daemon([cloud_node], [testutil.arvados_node_mock(8)],
 522                          wishlist, avail_sizes=[(size, {"cores":1})])
 523         self.assertEqual(1, self.alive_monitor_count())
 524         self.assertFalse(self.node_setup.start.called)
 525         monitor = self.monitor_list()[0].proxy()
 526         shutdown_proxy = self.node_shutdown.start().proxy
 527         shutdown_proxy().cloud_node.get.return_value = cloud_node
 528         shutdown_proxy().success.get.return_value = False
 529         shutdown_proxy().cancel_reason.get.return_value = self.node_shutdown.NODE_BROKEN
 530         self.daemon.update_server_wishlist([]).get(self.TIMEOUT)
 531         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
 532         self.daemon.node_finished_shutdown(shutdown_proxy()).get(self.TIMEOUT)
 533         self.daemon.update_cloud_nodes([cloud_node]).get(self.TIMEOUT)
 534         self.daemon.update_server_wishlist(wishlist).get(self.TIMEOUT)
 535         self.stop_proxy(self.daemon)
 536         self.assertEqual(1, self.node_setup.start.call_count)
 537
 538     def test_nodes_shutting_down_replaced_below_max_nodes(self):
 539         size = testutil.MockSize(6)
 540         cloud_node = testutil.cloud_node_mock(6, size=size)
 541         self.make_daemon([cloud_node], [testutil.arvados_node_mock(6, crunch_worker_state='down')],
 542                          avail_sizes=[(size, {"cores":1})])
 543         self.assertEqual(1, self.alive_monitor_count())
 544         monitor = self.monitor_list()[0].proxy()
 545         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
 546         self.assertTrue(self.node_shutdown.start.called)
 547         self.daemon.update_server_wishlist(
 548             [testutil.MockSize(6)]).get(self.TIMEOUT)
 549         self.stop_proxy(self.daemon)
 550         self.assertTrue(self.node_setup.start.called)
 551
 552     def test_nodes_shutting_down_not_replaced_at_max_nodes(self):
 553         cloud_node = testutil.cloud_node_mock(7)
 554         self.make_daemon([cloud_node], [testutil.arvados_node_mock(7)],
 555                          max_nodes=1)
 556         self.assertEqual(1, self.alive_monitor_count())
 557         monitor = self.monitor_list()[0].proxy()
 558         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
 559         self.assertTrue(self.node_shutdown.start.called)
 560         self.daemon.update_server_wishlist(
 561             [testutil.MockSize(7)]).get(self.TIMEOUT)
 562         self.stop_proxy(self.daemon)
 563         self.assertFalse(self.node_setup.start.called)
 564
 565     def test_nodes_shutting_down_count_against_excess(self):
 566         size = testutil.MockSize(8)
 567         cloud_nodes = [testutil.cloud_node_mock(n, size=size) for n in [8, 9]]
 568         arv_nodes = [testutil.arvados_node_mock(n, size=size) for n in [8, 9]]
 569         self.make_daemon(cloud_nodes, arv_nodes, [size],
 570                          avail_sizes=[(size, {"cores":1})])
 571         self.assertEqual(2, self.alive_monitor_count())
 572         for mon_ref in self.monitor_list():
 573             self.daemon.node_can_shutdown(mon_ref.proxy()).get(self.TIMEOUT)
 574         self.assertEqual(1, self.node_shutdown.start.call_count)
 575
 576     def test_clean_shutdown_waits_for_node_setup_finish(self):
 577         new_node = self.start_node_boot()
 578         new_node.stop_if_no_cloud_node().get.return_value = False
 579         new_node.stop_if_no_cloud_node.reset_mock()
 580         self.daemon.shutdown().get(self.TIMEOUT)
 581         self.assertTrue(new_node.stop_if_no_cloud_node.called)
 582         self.daemon.node_up(new_node).get(self.TIMEOUT)
 583         self.assertTrue(new_node.stop.called)
 584         self.timer.deliver()
 585         self.assertTrue(
 586             self.daemon.actor_ref.actor_stopped.wait(self.TIMEOUT))
 587
 588     def test_wishlist_ignored_after_shutdown(self):
 589         new_node = self.start_node_boot()
 590         new_node.stop_if_no_cloud_node().get.return_value = False
 591         new_node.stop_if_no_cloud_node.reset_mock()
 592         self.daemon.shutdown().get(self.TIMEOUT)
 593         size = testutil.MockSize(2)
 594         self.daemon.update_server_wishlist([size] * 2).get(self.TIMEOUT)
 595         self.timer.deliver()
 596         self.stop_proxy(self.daemon)
 597         self.assertEqual(1, self.node_setup.start.call_count)
 598
 599     def test_shutdown_actor_stopped_when_cloud_node_delisted(self):
 600         self.make_daemon(cloud_nodes=[testutil.cloud_node_mock()])
 601         self.assertEqual(1, self.alive_monitor_count())
 602         monitor = self.monitor_list()[0].proxy()
 603         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
 604         self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
 605         self.stop_proxy(self.daemon)
 606         self.assertEqual(
 607             1, self.last_shutdown.stop.call_count)
 608
 609     def test_shutdown_actor_cleanup_copes_with_dead_actors(self):
 610         self.make_daemon(cloud_nodes=[testutil.cloud_node_mock()])
 611         self.assertEqual(1, self.alive_monitor_count())
 612         monitor = self.monitor_list()[0].proxy()
 613         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
 614         # We're mainly testing that update_cloud_nodes catches and handles
 615         # the ActorDeadError.
 616         self.last_shutdown.stop.side_effect = pykka.ActorDeadError
 617         self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
 618         self.stop_proxy(self.daemon)
 619         self.assertEqual(1, self.last_shutdown.stop.call_count)
 620
 621     def test_node_create_two_sizes(self):
 622         small = testutil.MockSize(1)
 623         big = testutil.MockSize(2)
 624         avail_sizes = [(testutil.MockSize(1), {"cores":1}),
 625                         (testutil.MockSize(2), {"cores":2})]
 626         self.make_daemon(want_sizes=[small, small, small, big],
 627                          avail_sizes=avail_sizes, max_nodes=4)
 628
 629         # the daemon runs in another thread, so we need to wait and see
 630         # if it does all the work we're expecting it to do before stopping it.
 631         self.busywait(lambda: self.node_setup.start.call_count == 4)
 632         booting = self.daemon.booting.get(self.TIMEOUT)
 633         self.stop_proxy(self.daemon)
 634         sizecounts = {a[0].id: 0 for a in avail_sizes}
 635         for b in booting.itervalues():
 636             sizecounts[b.cloud_size.get().id] += 1
 637         logging.info(sizecounts)
 638         self.assertEqual(3, sizecounts[small.id])
 639         self.assertEqual(1, sizecounts[big.id])
 640
 641     def test_node_max_nodes_two_sizes(self):
 642         small = testutil.MockSize(1)
 643         big = testutil.MockSize(2)
 644         avail_sizes = [(testutil.MockSize(1), {"cores":1}),
 645                         (testutil.MockSize(2), {"cores":2})]
 646         self.make_daemon(want_sizes=[small, small, small, big],
 647                          avail_sizes=avail_sizes, max_nodes=3)
 648
 649         # the daemon runs in another thread, so we need to wait and see
 650         # if it does all the work we're expecting it to do before stopping it.
 651         self.busywait(lambda: self.node_setup.start.call_count == 3)
 652         booting = self.daemon.booting.get(self.TIMEOUT)
 653         self.stop_proxy(self.daemon)
 654         sizecounts = {a[0].id: 0 for a in avail_sizes}
 655         for b in booting.itervalues():
 656             sizecounts[b.cloud_size.get().id] += 1
 657         self.assertEqual(2, sizecounts[small.id])
 658         self.assertEqual(1, sizecounts[big.id])
 659
 660     def test_wishlist_reconfigure(self):
 661         small = testutil.MockSize(1)
 662         big = testutil.MockSize(2)
 663         avail_sizes = [(small, {"cores":1}), (big, {"cores":2})]
 664
 665         self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(1, small),
 666                                       testutil.cloud_node_mock(2, small),
 667                                       testutil.cloud_node_mock(3, big)],
 668                          arvados_nodes=[testutil.arvados_node_mock(1),
 669                                         testutil.arvados_node_mock(2),
 670                                         testutil.arvados_node_mock(3)],
 671                          want_sizes=[small, small, big],
 672                          avail_sizes=avail_sizes)
 673
 674         self.daemon.update_server_wishlist([small, big, big]).get(self.TIMEOUT)
 675
 676         self.assertEqual(0, self.node_shutdown.start.call_count)
 677
 678         for c in self.daemon.cloud_nodes.get().nodes.itervalues():
 679             self.daemon.node_can_shutdown(c.actor)
 680
 681         booting = self.daemon.booting.get()
 682         cloud_nodes = self.daemon.cloud_nodes.get()
 683
 684         self.stop_proxy(self.daemon)
 685
 686         self.assertEqual(1, self.node_setup.start.call_count)
 687         self.assertEqual(1, self.node_shutdown.start.call_count)
 688
 689         # booting a new big node
 690         sizecounts = {a[0].id: 0 for a in avail_sizes}
 691         for b in booting.itervalues():
 692             sizecounts[b.cloud_size.get().id] += 1
 693         self.assertEqual(0, sizecounts[small.id])
 694         self.assertEqual(1, sizecounts[big.id])
 695
 696         # shutting down a small node
 697         sizecounts = {a[0].id: 0 for a in avail_sizes}
 698         for b in cloud_nodes.nodes.itervalues():
 699             if b.shutdown_actor is not None:
 700                 sizecounts[b.cloud_node.size.id] += 1
 701         self.assertEqual(1, sizecounts[small.id])
 702         self.assertEqual(0, sizecounts[big.id])
 703
 704     def test_node_max_price(self):
 705         small = testutil.MockSize(1)
 706         big = testutil.MockSize(2)
 707         avail_sizes = [(testutil.MockSize(1), {"cores":1, "price":1}),
 708                         (testutil.MockSize(2), {"cores":2, "price":2})]
 709         self.make_daemon(want_sizes=[small, small, small, big],
 710                          avail_sizes=avail_sizes,
 711                          max_nodes=4,
 712                          max_total_price=4)
 713         # the daemon runs in another thread, so we need to wait and see
 714         # if it does all the work we're expecting it to do before stopping it.
 715         self.busywait(lambda: self.node_setup.start.call_count == 3)
 716         booting = self.daemon.booting.get()
 717         self.stop_proxy(self.daemon)
 718
 719         sizecounts = {a[0].id: 0 for a in avail_sizes}
 720         for b in booting.itervalues():
 721             sizecounts[b.cloud_size.get().id] += 1
 722         logging.info(sizecounts)
 723
 724         # Booting 3 small nodes and not booting a big node would also partially
 725         # satisfy the wishlist and come in under the price cap, however the way
 726         # the update_server_wishlist() currently works effectively results in a
 727         # round-robin creation of one node of each size in the wishlist, so
 728         # test for that.
 729         self.assertEqual(2, sizecounts[small.id])
 730         self.assertEqual(1, sizecounts[big.id])