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