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