3 from __future__ import absolute_import, print_function
11 import arvnodeman.daemon as nmdaemon
12 from arvnodeman.jobqueue import ServerCalculator
13 from arvnodeman.computenode.dispatch import ComputeNodeMonitorActor
14 from . import testutil
17 class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin,
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,
28 mock_actor.proxy.return_value = mock_proxy
30 self.last_setup = mock_proxy
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,
44 mock_actor.proxy.return_value = mock_proxy
45 self.last_shutdown = mock_proxy
49 def make_daemon(self, cloud_nodes=[], arvados_nodes=[], want_sizes=[],
50 avail_sizes=[(testutil.MockSize(1), {"cores": 1})], min_nodes=0, max_nodes=8):
51 for name in ['cloud_nodes', 'arvados_nodes', 'server_wishlist']:
52 setattr(self, name + '_poller', mock.MagicMock(name=name + '_mock'))
53 self.arv_factory = mock.MagicMock(name='arvados_mock')
54 self.cloud_factory = mock.MagicMock(name='cloud_mock')
55 self.cloud_factory().node_start_time.return_value = time.time()
56 self.cloud_updates = mock.MagicMock(name='updates_mock')
57 self.timer = testutil.MockTimer(deliver_immediately=False)
59 self.node_setup = mock.MagicMock(name='setup_mock')
60 self.node_setup.start.side_effect = self.mock_node_start
61 self.node_setup.reset_mock()
63 self.node_shutdown = mock.MagicMock(name='shutdown_mock')
64 self.node_shutdown.start.side_effect = self.mock_node_shutdown
66 self.daemon = nmdaemon.NodeManagerDaemonActor.start(
67 self.server_wishlist_poller, self.arvados_nodes_poller,
68 self.cloud_nodes_poller, self.cloud_updates, self.timer,
69 self.arv_factory, self.cloud_factory,
70 [54, 5, 1], ServerCalculator(avail_sizes),
71 min_nodes, max_nodes, 600, 1800, 3600,
72 self.node_setup, self.node_shutdown).proxy()
73 if cloud_nodes is not None:
74 self.daemon.update_cloud_nodes(cloud_nodes).get(self.TIMEOUT)
75 if arvados_nodes is not None:
76 self.daemon.update_arvados_nodes(arvados_nodes).get(self.TIMEOUT)
77 if want_sizes is not None:
78 self.daemon.update_server_wishlist(want_sizes).get(self.TIMEOUT)
80 def monitor_list(self):
81 return pykka.ActorRegistry.get_by_class(ComputeNodeMonitorActor)
83 def monitored_arvados_nodes(self):
85 for future in [actor.proxy().arvados_node
86 for actor in self.monitor_list()]:
88 pairings.append(future.get(self.TIMEOUT))
89 except pykka.ActorDeadError:
93 def alive_monitor_count(self):
94 return len(self.monitored_arvados_nodes())
96 def assertShutdownCancellable(self, expected=True):
97 self.assertTrue(self.node_shutdown.start.called)
98 self.assertIs(expected,
99 self.node_shutdown.start.call_args[1]['cancellable'],
100 "ComputeNodeShutdownActor incorrectly cancellable")
102 def test_easy_node_creation(self):
103 size = testutil.MockSize(1)
104 self.make_daemon(want_sizes=[size])
105 self.stop_proxy(self.daemon)
106 self.assertTrue(self.node_setup.start.called)
108 def check_monitors_arvados_nodes(self, *arv_nodes):
109 self.assertItemsEqual(arv_nodes, self.monitored_arvados_nodes())
111 def test_node_pairing(self):
112 cloud_node = testutil.cloud_node_mock(1)
113 arv_node = testutil.arvados_node_mock(1)
114 self.make_daemon([cloud_node], [arv_node])
115 self.stop_proxy(self.daemon)
116 self.check_monitors_arvados_nodes(arv_node)
118 def test_node_pairing_after_arvados_update(self):
119 cloud_node = testutil.cloud_node_mock(2)
120 self.make_daemon([cloud_node],
121 [testutil.arvados_node_mock(2, ip_address=None)])
122 arv_node = testutil.arvados_node_mock(2)
123 self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
124 self.stop_proxy(self.daemon)
125 self.check_monitors_arvados_nodes(arv_node)
127 def test_arvados_node_un_and_re_paired(self):
128 # We need to create the Arvados node mock after spinning up the daemon
129 # to make sure it's new enough to pair with the cloud node.
130 self.make_daemon([testutil.cloud_node_mock(3)], arvados_nodes=None)
131 arv_node = testutil.arvados_node_mock(3)
132 self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
133 self.check_monitors_arvados_nodes(arv_node)
134 self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
135 self.assertEqual(0, self.alive_monitor_count())
136 self.daemon.update_cloud_nodes([testutil.cloud_node_mock(3)])
137 self.stop_proxy(self.daemon)
138 self.check_monitors_arvados_nodes(arv_node)
140 def test_old_arvados_node_not_double_assigned(self):
141 arv_node = testutil.arvados_node_mock(3, age=9000)
142 size = testutil.MockSize(3)
143 self.make_daemon(arvados_nodes=[arv_node], avail_sizes=[(size, {"cores":1})])
144 self.daemon.update_server_wishlist([size]).get(self.TIMEOUT)
145 self.daemon.update_server_wishlist([size, size]).get(self.TIMEOUT)
146 self.stop_proxy(self.daemon)
147 used_nodes = [call[1].get('arvados_node')
148 for call in self.node_setup.start.call_args_list]
149 self.assertEqual(2, len(used_nodes))
150 self.assertIn(arv_node, used_nodes)
151 self.assertIn(None, used_nodes)
153 def test_node_count_satisfied(self):
154 self.make_daemon([testutil.cloud_node_mock()],
155 want_sizes=[testutil.MockSize(1)])
156 self.stop_proxy(self.daemon)
157 self.assertFalse(self.node_setup.start.called)
159 def test_dont_count_missing_as_busy(self):
160 size = testutil.MockSize(1)
161 self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(1),
162 testutil.cloud_node_mock(2)],
163 arvados_nodes=[testutil.arvados_node_mock(1),
164 testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
165 want_sizes=[size, size])
166 self.stop_proxy(self.daemon)
167 self.assertTrue(self.node_setup.start.called)
169 def test_missing_counts_towards_max(self):
170 size = testutil.MockSize(1)
171 self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(1),
172 testutil.cloud_node_mock(2)],
173 arvados_nodes=[testutil.arvados_node_mock(1),
174 testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
175 want_sizes=[size, size],
177 self.stop_proxy(self.daemon)
178 self.assertFalse(self.node_setup.start.called)
180 def test_excess_counts_missing(self):
181 size = testutil.MockSize(1)
182 cloud_nodes = [testutil.cloud_node_mock(1), testutil.cloud_node_mock(2)]
183 self.make_daemon(cloud_nodes=cloud_nodes,
184 arvados_nodes=[testutil.arvados_node_mock(1),
185 testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
187 self.assertEqual(2, self.alive_monitor_count())
188 for mon_ref in self.monitor_list():
189 self.daemon.node_can_shutdown(mon_ref.proxy()).get(self.TIMEOUT)
190 self.assertEqual(1, self.node_shutdown.start.call_count)
192 def test_missing_shutdown_not_excess(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')],
200 get_cloud_node = mock.MagicMock(name="get_cloud_node")
201 get_cloud_node.get.return_value = cloud_nodes[1]
202 mock_node_monitor = mock.MagicMock()
203 mock_node_monitor.proxy.return_value = mock.NonCallableMock(cloud_node=get_cloud_node)
204 mock_shutdown = self.node_shutdown.start(node_monitor=mock_node_monitor)
206 self.daemon.shutdowns.get()[cloud_nodes[1].id] = mock_shutdown.proxy()
208 self.assertEqual(2, self.alive_monitor_count())
209 for mon_ref in self.monitor_list():
210 self.daemon.node_can_shutdown(mon_ref.proxy()).get(self.TIMEOUT)
211 self.assertEqual(1, self.node_shutdown.start.call_count)
213 def test_booting_nodes_counted(self):
214 cloud_node = testutil.cloud_node_mock(1)
215 arv_node = testutil.arvados_node_mock(1)
216 server_wishlist = [testutil.MockSize(1)] * 2
217 self.make_daemon([cloud_node], [arv_node], server_wishlist)
218 self.daemon.max_nodes.get(self.TIMEOUT)
219 self.assertTrue(self.node_setup.start.called)
220 self.daemon.update_server_wishlist(server_wishlist).get(self.TIMEOUT)
221 self.stop_proxy(self.daemon)
222 self.assertEqual(1, self.node_setup.start.call_count)
224 def test_boot_new_node_when_all_nodes_busy(self):
225 size = testutil.MockSize(2)
226 arv_node = testutil.arvados_node_mock(2, job_uuid=True)
227 self.make_daemon([testutil.cloud_node_mock(2, size=size)], [arv_node],
228 [size], avail_sizes=[(size, {"cores":1})])
229 self.stop_proxy(self.daemon)
230 self.assertTrue(self.node_setup.start.called)
232 def test_boot_new_node_below_min_nodes(self):
233 min_size = testutil.MockSize(1)
234 wish_size = testutil.MockSize(3)
235 avail_sizes = [(min_size, {"cores": 1}),
236 (wish_size, {"cores": 3})]
237 self.make_daemon([], [], None, avail_sizes=avail_sizes, min_nodes=2)
238 self.daemon.update_server_wishlist([wish_size]).get(self.TIMEOUT)
239 self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
240 self.daemon.update_server_wishlist([wish_size]).get(self.TIMEOUT)
241 self.stop_proxy(self.daemon)
242 self.assertEqual([wish_size, min_size],
243 [call[1].get('cloud_size')
244 for call in self.node_setup.start.call_args_list])
246 def test_no_new_node_when_ge_min_nodes_busy(self):
247 cloud_nodes = [testutil.cloud_node_mock(n) for n in range(1, 4)]
248 arv_nodes = [testutil.arvados_node_mock(n, job_uuid=True)
249 for n in range(1, 4)]
250 self.make_daemon(cloud_nodes, arv_nodes, [], min_nodes=2)
251 self.stop_proxy(self.daemon)
252 self.assertEqual(0, self.node_setup.start.call_count)
254 def test_no_new_node_when_max_nodes_busy(self):
255 self.make_daemon([testutil.cloud_node_mock(3)],
256 [testutil.arvados_node_mock(3, job_uuid=True)],
257 [testutil.MockSize(3)],
259 self.stop_proxy(self.daemon)
260 self.assertFalse(self.node_setup.start.called)
262 def start_node_boot(self, cloud_node=None, arv_node=None, id_num=1):
263 if cloud_node is None:
264 cloud_node = testutil.cloud_node_mock(id_num)
266 arv_node = testutil.arvados_node_mock(id_num)
267 self.make_daemon(want_sizes=[testutil.MockSize(id_num)],
268 avail_sizes=[(testutil.MockSize(id_num), {"cores":1})])
269 self.daemon.max_nodes.get(self.TIMEOUT)
270 self.assertEqual(1, self.node_setup.start.call_count)
271 self.last_setup.cloud_node.get.return_value = cloud_node
272 self.last_setup.arvados_node.get.return_value = arv_node
273 return self.last_setup
275 def test_no_new_node_when_booted_node_not_usable(self):
276 cloud_node = testutil.cloud_node_mock(4)
277 arv_node = testutil.arvados_node_mock(4, crunch_worker_state='down')
278 setup = self.start_node_boot(cloud_node, arv_node)
279 self.daemon.node_up(setup).get(self.TIMEOUT)
280 self.assertEqual(1, self.alive_monitor_count())
281 self.daemon.update_cloud_nodes([cloud_node])
282 self.daemon.update_arvados_nodes([arv_node])
283 self.daemon.update_server_wishlist(
284 [testutil.MockSize(1)]).get(self.TIMEOUT)
285 self.stop_proxy(self.daemon)
286 self.assertEqual(1, self.node_setup.start.call_count)
288 def test_no_duplication_when_booting_node_listed_fast(self):
289 # Test that we don't start two ComputeNodeMonitorActors when
290 # we learn about a booting node through a listing before we
291 # get the "node up" message from CloudNodeSetupActor.
292 cloud_node = testutil.cloud_node_mock(1)
293 setup = self.start_node_boot(cloud_node)
294 self.daemon.update_cloud_nodes([cloud_node])
295 self.daemon.node_up(setup).get(self.TIMEOUT)
296 self.assertEqual(1, self.alive_monitor_count())
298 def test_no_duplication_when_booted_node_listed(self):
299 cloud_node = testutil.cloud_node_mock(2)
300 setup = self.start_node_boot(cloud_node, id_num=2)
301 self.daemon.node_up(setup)
302 self.daemon.update_cloud_nodes([cloud_node]).get(self.TIMEOUT)
303 self.assertEqual(1, self.alive_monitor_count())
305 def test_node_counted_after_boot_with_slow_listing(self):
306 # Test that, after we boot a compute node, we assume it exists
307 # even it doesn't appear in the listing (e.g., because of delays
309 setup = self.start_node_boot()
310 self.daemon.node_up(setup).get(self.TIMEOUT)
311 self.assertEqual(1, self.alive_monitor_count())
312 self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
313 self.assertEqual(1, self.alive_monitor_count())
315 def test_booted_unlisted_node_counted(self):
316 setup = self.start_node_boot(id_num=1)
317 self.daemon.node_up(setup)
318 self.daemon.update_server_wishlist(
319 [testutil.MockSize(1)]).get(self.TIMEOUT)
320 self.stop_proxy(self.daemon)
321 self.assertEqual(1, self.node_setup.start.call_count)
323 def test_booted_node_can_shutdown(self):
324 setup = self.start_node_boot()
325 self.daemon.node_up(setup).get(self.TIMEOUT)
326 self.assertEqual(1, self.alive_monitor_count())
327 monitor = self.monitor_list()[0].proxy()
328 self.daemon.update_server_wishlist([])
329 self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
330 self.stop_proxy(self.daemon)
331 self.assertTrue(self.node_shutdown.start.called,
332 "daemon did not shut down booted node on offer")
334 def test_booted_node_lifecycle(self):
335 cloud_node = testutil.cloud_node_mock(6)
336 setup = self.start_node_boot(cloud_node, id_num=6)
337 self.daemon.node_up(setup).get(self.TIMEOUT)
338 self.assertEqual(1, self.alive_monitor_count())
339 monitor = self.monitor_list()[0].proxy()
340 self.daemon.update_server_wishlist([])
341 self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
342 self.assertShutdownCancellable(True)
343 shutdown = self.node_shutdown.start().proxy()
344 shutdown.cloud_node.get.return_value = cloud_node
345 self.daemon.node_finished_shutdown(shutdown).get(self.TIMEOUT)
346 self.assertTrue(shutdown.stop.called,
347 "shutdown actor not stopped after finishing")
348 self.assertTrue(monitor.actor_ref.actor_stopped.wait(self.TIMEOUT),
349 "monitor for booted node not stopped after shutdown")
350 self.daemon.update_server_wishlist(
351 [testutil.MockSize(2)]).get(self.TIMEOUT)
352 self.stop_proxy(self.daemon)
353 self.assertTrue(self.node_setup.start.called,
354 "second node not started after booted node stopped")
356 def test_booted_node_shut_down_when_never_listed(self):
357 setup = self.start_node_boot()
358 self.daemon.node_up(setup).get(self.TIMEOUT)
359 self.assertEqual(1, self.alive_monitor_count())
360 self.assertFalse(self.node_shutdown.start.called)
362 self.stop_proxy(self.daemon)
363 self.assertShutdownCancellable(False)
365 def test_booted_node_shut_down_when_never_paired(self):
366 cloud_node = testutil.cloud_node_mock(2)
367 setup = self.start_node_boot(cloud_node)
368 self.daemon.node_up(setup).get(self.TIMEOUT)
369 self.assertEqual(1, self.alive_monitor_count())
370 self.daemon.update_cloud_nodes([cloud_node])
372 self.stop_proxy(self.daemon)
373 self.assertShutdownCancellable(False)
375 def test_booted_node_shut_down_when_never_working(self):
376 cloud_node = testutil.cloud_node_mock(4)
377 arv_node = testutil.arvados_node_mock(4, crunch_worker_state='down')
378 setup = self.start_node_boot(cloud_node, arv_node)
379 self.daemon.node_up(setup).get(self.TIMEOUT)
380 self.assertEqual(1, self.alive_monitor_count())
381 self.daemon.update_cloud_nodes([cloud_node])
382 self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
384 self.stop_proxy(self.daemon)
385 self.assertShutdownCancellable(False)
387 def test_node_that_pairs_not_considered_failed_boot(self):
388 cloud_node = testutil.cloud_node_mock(3)
389 arv_node = testutil.arvados_node_mock(3)
390 setup = self.start_node_boot(cloud_node, arv_node)
391 self.daemon.node_up(setup).get(self.TIMEOUT)
392 self.assertEqual(1, self.alive_monitor_count())
393 self.daemon.update_cloud_nodes([cloud_node])
394 self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
396 self.stop_proxy(self.daemon)
397 self.assertFalse(self.node_shutdown.start.called)
399 def test_node_that_pairs_busy_not_considered_failed_boot(self):
400 cloud_node = testutil.cloud_node_mock(5)
401 arv_node = testutil.arvados_node_mock(5, job_uuid=True)
402 setup = self.start_node_boot(cloud_node, arv_node)
403 self.daemon.node_up(setup).get(self.TIMEOUT)
404 self.assertEqual(1, self.alive_monitor_count())
405 self.daemon.update_cloud_nodes([cloud_node])
406 self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
408 self.stop_proxy(self.daemon)
409 self.assertFalse(self.node_shutdown.start.called)
411 def test_booting_nodes_shut_down(self):
412 self.make_daemon(want_sizes=[testutil.MockSize(1)])
413 self.daemon.update_server_wishlist([]).get(self.TIMEOUT)
414 self.stop_proxy(self.daemon)
415 self.assertTrue(self.last_setup.stop_if_no_cloud_node.called)
417 def test_all_booting_nodes_tried_to_shut_down(self):
418 size = testutil.MockSize(2)
419 self.make_daemon(want_sizes=[size], avail_sizes=[(size, {"cores":1})])
420 self.daemon.max_nodes.get(self.TIMEOUT)
421 setup1 = self.last_setup
422 setup1.stop_if_no_cloud_node().get.return_value = False
423 setup1.stop_if_no_cloud_node.reset_mock()
424 self.daemon.update_server_wishlist([size, size]).get(self.TIMEOUT)
425 self.daemon.max_nodes.get(self.TIMEOUT)
426 self.assertIsNot(setup1, self.last_setup)
427 self.last_setup.stop_if_no_cloud_node().get.return_value = True
428 self.last_setup.stop_if_no_cloud_node.reset_mock()
429 self.daemon.update_server_wishlist([]).get(self.TIMEOUT)
430 self.daemon.max_nodes.get(self.TIMEOUT)
431 self.stop_proxy(self.daemon)
432 self.assertEqual(1, self.last_setup.stop_if_no_cloud_node.call_count)
433 self.assertTrue(setup1.stop_if_no_cloud_node.called)
435 def test_shutdown_declined_at_wishlist_capacity(self):
436 cloud_node = testutil.cloud_node_mock(1)
437 size = testutil.MockSize(1)
438 self.make_daemon(cloud_nodes=[cloud_node], want_sizes=[size])
439 self.assertEqual(1, self.alive_monitor_count())
440 monitor = self.monitor_list()[0].proxy()
441 self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
442 self.stop_proxy(self.daemon)
443 self.assertFalse(self.node_shutdown.start.called)
445 def test_shutdown_declined_below_min_nodes(self):
446 cloud_node = testutil.cloud_node_mock(1)
447 self.make_daemon(cloud_nodes=[cloud_node], min_nodes=1)
448 self.assertEqual(1, self.alive_monitor_count())
449 monitor = self.monitor_list()[0].proxy()
450 self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
451 self.stop_proxy(self.daemon)
452 self.assertFalse(self.node_shutdown.start.called)
454 def test_shutdown_accepted_below_capacity(self):
455 self.make_daemon(cloud_nodes=[testutil.cloud_node_mock()])
456 self.assertEqual(1, self.alive_monitor_count())
457 monitor = self.monitor_list()[0].proxy()
458 self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
459 self.stop_proxy(self.daemon)
460 self.assertTrue(self.node_shutdown.start.called)
462 def test_shutdown_declined_when_idle_and_job_queued(self):
463 cloud_nodes = [testutil.cloud_node_mock(n) for n in [3, 4]]
464 arv_nodes = [testutil.arvados_node_mock(3, job_uuid=True),
465 testutil.arvados_node_mock(4, job_uuid=None)]
466 self.make_daemon(cloud_nodes, arv_nodes, [testutil.MockSize(1)])
467 self.assertEqual(2, self.alive_monitor_count())
468 for mon_ref in self.monitor_list():
469 monitor = mon_ref.proxy()
470 if monitor.cloud_node.get(self.TIMEOUT) is cloud_nodes[-1]:
473 self.fail("monitor for idle node not found")
474 self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
475 self.stop_proxy(self.daemon)
476 self.assertFalse(self.node_shutdown.start.called)
478 def test_node_shutdown_after_cancelled_shutdown(self):
479 cloud_node = testutil.cloud_node_mock(5)
480 self.make_daemon([cloud_node], [testutil.arvados_node_mock(5)])
481 self.assertEqual(1, self.alive_monitor_count())
482 monitor = self.monitor_list()[0].proxy()
483 self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
484 self.last_shutdown.success.get.return_value = False
485 self.daemon.node_finished_shutdown(self.last_shutdown).get(self.TIMEOUT)
486 self.assertEqual(1, self.alive_monitor_count())
488 self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
489 self.last_shutdown.success.get.return_value = True
490 self.last_shutdown.stop.side_effect = lambda: monitor.stop()
491 self.daemon.node_finished_shutdown(self.last_shutdown).get(self.TIMEOUT)
492 self.assertEqual(0, self.alive_monitor_count())
494 def test_broken_node_blackholed_after_cancelled_shutdown(self):
495 size = testutil.MockSize(8)
496 cloud_node = testutil.cloud_node_mock(8, size=size)
498 self.make_daemon([cloud_node], [testutil.arvados_node_mock(8)],
499 wishlist, avail_sizes=[(size, {"cores":1})])
500 self.assertEqual(1, self.alive_monitor_count())
501 self.assertFalse(self.node_setup.start.called)
502 monitor = self.monitor_list()[0].proxy()
503 shutdown_proxy = self.node_shutdown.start().proxy
504 shutdown_proxy().cloud_node.get.return_value = cloud_node
505 shutdown_proxy().success.get.return_value = False
506 shutdown_proxy().cancel_reason.get.return_value = self.node_shutdown.NODE_BROKEN
507 self.daemon.update_server_wishlist([]).get(self.TIMEOUT)
508 self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
509 self.daemon.node_finished_shutdown(shutdown_proxy()).get(self.TIMEOUT)
510 self.daemon.update_cloud_nodes([cloud_node]).get(self.TIMEOUT)
511 self.daemon.update_server_wishlist(wishlist).get(self.TIMEOUT)
512 self.stop_proxy(self.daemon)
513 self.assertEqual(1, self.node_setup.start.call_count)
515 def test_nodes_shutting_down_replaced_below_max_nodes(self):
516 size = testutil.MockSize(6)
517 cloud_node = testutil.cloud_node_mock(6, size=size)
518 self.make_daemon([cloud_node], [testutil.arvados_node_mock(6)],
519 avail_sizes=[(size, {"cores":1})])
520 self.assertEqual(1, self.alive_monitor_count())
521 monitor = self.monitor_list()[0].proxy()
522 self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
523 self.assertTrue(self.node_shutdown.start.called)
524 self.daemon.update_server_wishlist(
525 [testutil.MockSize(6)]).get(self.TIMEOUT)
526 self.stop_proxy(self.daemon)
527 self.assertTrue(self.node_setup.start.called)
529 def test_nodes_shutting_down_not_replaced_at_max_nodes(self):
530 cloud_node = testutil.cloud_node_mock(7)
531 self.make_daemon([cloud_node], [testutil.arvados_node_mock(7)],
533 self.assertEqual(1, self.alive_monitor_count())
534 monitor = self.monitor_list()[0].proxy()
535 self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
536 self.assertTrue(self.node_shutdown.start.called)
537 self.daemon.update_server_wishlist(
538 [testutil.MockSize(7)]).get(self.TIMEOUT)
539 self.stop_proxy(self.daemon)
540 self.assertFalse(self.node_setup.start.called)
542 def test_nodes_shutting_down_count_against_excess(self):
543 size = testutil.MockSize(8)
544 cloud_nodes = [testutil.cloud_node_mock(n, size=size) for n in [8, 9]]
545 arv_nodes = [testutil.arvados_node_mock(n, size=size) for n in [8, 9]]
546 self.make_daemon(cloud_nodes, arv_nodes, [size],
547 avail_sizes=[(size, {"cores":1})])
548 self.assertEqual(2, self.alive_monitor_count())
549 for mon_ref in self.monitor_list():
550 self.daemon.node_can_shutdown(mon_ref.proxy()).get(self.TIMEOUT)
551 self.assertEqual(1, self.node_shutdown.start.call_count)
553 def test_clean_shutdown_waits_for_node_setup_finish(self):
554 new_node = self.start_node_boot()
555 new_node.stop_if_no_cloud_node().get.return_value = False
556 new_node.stop_if_no_cloud_node.reset_mock()
557 self.daemon.shutdown().get(self.TIMEOUT)
558 self.assertTrue(new_node.stop_if_no_cloud_node.called)
559 self.daemon.node_up(new_node).get(self.TIMEOUT)
560 self.assertTrue(new_node.stop.called)
563 self.daemon.actor_ref.actor_stopped.wait(self.TIMEOUT))
565 def test_wishlist_ignored_after_shutdown(self):
566 new_node = self.start_node_boot()
567 new_node.stop_if_no_cloud_node().get.return_value = False
568 new_node.stop_if_no_cloud_node.reset_mock()
569 self.daemon.shutdown().get(self.TIMEOUT)
570 size = testutil.MockSize(2)
571 self.daemon.update_server_wishlist([size] * 2).get(self.TIMEOUT)
573 self.stop_proxy(self.daemon)
574 self.assertEqual(1, self.node_setup.start.call_count)
576 def test_shutdown_actor_stopped_when_cloud_node_delisted(self):
577 self.make_daemon(cloud_nodes=[testutil.cloud_node_mock()])
578 self.assertEqual(1, self.alive_monitor_count())
579 monitor = self.monitor_list()[0].proxy()
580 self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
581 self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
582 self.stop_proxy(self.daemon)
584 1, self.last_shutdown.stop.call_count)
586 def test_shutdown_actor_cleanup_copes_with_dead_actors(self):
587 self.make_daemon(cloud_nodes=[testutil.cloud_node_mock()])
588 self.assertEqual(1, self.alive_monitor_count())
589 monitor = self.monitor_list()[0].proxy()
590 self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
591 # We're mainly testing that update_cloud_nodes catches and handles
592 # the ActorDeadError.
593 self.last_shutdown.stop.side_effect = pykka.ActorDeadError
594 self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
595 self.stop_proxy(self.daemon)
596 self.assertEqual(1, self.last_shutdown.stop.call_count)