3 from __future__ import absolute_import, print_function
11 import arvnodeman.daemon as nmdaemon
12 from arvnodeman.computenode.dispatch import ComputeNodeMonitorActor
13 from . import testutil
15 class NodeManagerDaemonActorTestCase(testutil.ActorTestMixin,
17 def new_setup_proxy(self):
18 # Make sure that every time the daemon starts a setup actor,
19 # it gets a new mock object back.
20 self.last_setup = mock.MagicMock(name='setup_proxy_mock')
21 return self.last_setup
23 def make_daemon(self, cloud_nodes=[], arvados_nodes=[], want_sizes=[],
24 min_size=testutil.MockSize(1), min_nodes=0, max_nodes=8):
25 for name in ['cloud_nodes', 'arvados_nodes', 'server_wishlist']:
26 setattr(self, name + '_poller', mock.MagicMock(name=name + '_mock'))
27 self.arv_factory = mock.MagicMock(name='arvados_mock')
28 self.cloud_factory = mock.MagicMock(name='cloud_mock')
29 self.cloud_factory().node_start_time.return_value = time.time()
30 self.cloud_updates = mock.MagicMock(name='updates_mock')
31 self.timer = testutil.MockTimer(deliver_immediately=False)
32 self.node_setup = mock.MagicMock(name='setup_mock')
33 self.node_setup.start().proxy.side_effect = self.new_setup_proxy
34 self.node_setup.reset_mock()
35 self.node_shutdown = mock.MagicMock(name='shutdown_mock')
36 self.daemon = nmdaemon.NodeManagerDaemonActor.start(
37 self.server_wishlist_poller, self.arvados_nodes_poller,
38 self.cloud_nodes_poller, self.cloud_updates, self.timer,
39 self.arv_factory, self.cloud_factory,
40 [54, 5, 1], min_size, min_nodes, max_nodes, 600, 1800, 3600,
41 self.node_setup, self.node_shutdown).proxy()
42 if cloud_nodes is not None:
43 self.daemon.update_cloud_nodes(cloud_nodes).get(self.TIMEOUT)
44 if arvados_nodes is not None:
45 self.daemon.update_arvados_nodes(arvados_nodes).get(self.TIMEOUT)
46 if want_sizes is not None:
47 self.daemon.update_server_wishlist(want_sizes).get(self.TIMEOUT)
49 def monitor_list(self):
50 return pykka.ActorRegistry.get_by_class(ComputeNodeMonitorActor)
52 def monitored_arvados_nodes(self):
54 for future in [actor.proxy().arvados_node
55 for actor in self.monitor_list()]:
57 pairings.append(future.get(self.TIMEOUT))
58 except pykka.ActorDeadError:
62 def alive_monitor_count(self):
63 return len(self.monitored_arvados_nodes())
65 def assertShutdownCancellable(self, expected=True):
66 self.assertTrue(self.node_shutdown.start.called)
67 self.assertIs(expected,
68 self.node_shutdown.start.call_args[1]['cancellable'],
69 "ComputeNodeShutdownActor incorrectly cancellable")
71 def test_easy_node_creation(self):
72 size = testutil.MockSize(1)
73 self.make_daemon(want_sizes=[size])
74 self.stop_proxy(self.daemon)
75 self.assertTrue(self.node_setup.start.called)
77 def check_monitors_arvados_nodes(self, *arv_nodes):
78 self.assertItemsEqual(arv_nodes, self.monitored_arvados_nodes())
80 def test_node_pairing(self):
81 cloud_node = testutil.cloud_node_mock(1)
82 arv_node = testutil.arvados_node_mock(1)
83 self.make_daemon([cloud_node], [arv_node])
84 self.stop_proxy(self.daemon)
85 self.check_monitors_arvados_nodes(arv_node)
87 def test_node_pairing_after_arvados_update(self):
88 cloud_node = testutil.cloud_node_mock(2)
89 self.make_daemon([cloud_node],
90 [testutil.arvados_node_mock(2, ip_address=None)])
91 arv_node = testutil.arvados_node_mock(2)
92 self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
93 self.stop_proxy(self.daemon)
94 self.check_monitors_arvados_nodes(arv_node)
96 def test_arvados_node_un_and_re_paired(self):
97 # We need to create the Arvados node mock after spinning up the daemon
98 # to make sure it's new enough to pair with the cloud node.
99 self.make_daemon([testutil.cloud_node_mock(3)], arvados_nodes=None)
100 arv_node = testutil.arvados_node_mock(3)
101 self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
102 self.check_monitors_arvados_nodes(arv_node)
103 self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
104 self.assertEqual(0, self.alive_monitor_count())
105 self.daemon.update_cloud_nodes([testutil.cloud_node_mock(3)])
106 self.stop_proxy(self.daemon)
107 self.check_monitors_arvados_nodes(arv_node)
109 def test_old_arvados_node_not_double_assigned(self):
110 arv_node = testutil.arvados_node_mock(3, age=9000)
111 size = testutil.MockSize(3)
112 self.make_daemon(arvados_nodes=[arv_node])
113 self.daemon.update_server_wishlist([size]).get(self.TIMEOUT)
114 self.daemon.update_server_wishlist([size, size]).get(self.TIMEOUT)
115 self.stop_proxy(self.daemon)
116 used_nodes = [call[1].get('arvados_node')
117 for call in self.node_setup.start.call_args_list]
118 self.assertEqual(2, len(used_nodes))
119 self.assertIn(arv_node, used_nodes)
120 self.assertIn(None, used_nodes)
122 def test_node_count_satisfied(self):
123 self.make_daemon([testutil.cloud_node_mock()],
124 want_sizes=[testutil.MockSize(1)])
125 self.stop_proxy(self.daemon)
126 self.assertFalse(self.node_setup.start.called)
128 def test_dont_count_missing_as_busy(self):
129 size = testutil.MockSize(1)
130 self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(1),
131 testutil.cloud_node_mock(2)],
132 arvados_nodes=[testutil.arvados_node_mock(1),
133 testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
134 want_sizes=[size, size])
135 self.stop_proxy(self.daemon)
136 self.assertTrue(self.node_setup.start.called)
138 def test_missing_counts_towards_max(self):
139 size = testutil.MockSize(1)
140 self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(1),
141 testutil.cloud_node_mock(2)],
142 arvados_nodes=[testutil.arvados_node_mock(1),
143 testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
144 want_sizes=[size, size],
146 self.stop_proxy(self.daemon)
147 self.assertFalse(self.node_setup.start.called)
149 def test_booting_nodes_counted(self):
150 cloud_node = testutil.cloud_node_mock(1)
151 arv_node = testutil.arvados_node_mock(1)
152 server_wishlist = [testutil.MockSize(1)] * 2
153 self.make_daemon([cloud_node], [arv_node], server_wishlist)
154 self.daemon.max_nodes.get(self.TIMEOUT)
155 self.assertTrue(self.node_setup.start.called)
156 self.daemon.update_server_wishlist(server_wishlist).get(self.TIMEOUT)
157 self.stop_proxy(self.daemon)
158 self.assertEqual(1, self.node_setup.start.call_count)
160 def test_boot_new_node_when_all_nodes_busy(self):
161 arv_node = testutil.arvados_node_mock(2, job_uuid=True)
162 self.make_daemon([testutil.cloud_node_mock(2)], [arv_node],
163 [testutil.MockSize(2)])
164 self.stop_proxy(self.daemon)
165 self.assertTrue(self.node_setup.start.called)
167 def test_boot_new_node_below_min_nodes(self):
168 min_size = testutil.MockSize(1)
169 wish_size = testutil.MockSize(3)
170 self.make_daemon([], [], None, min_size=min_size, min_nodes=2)
171 self.daemon.update_server_wishlist([wish_size]).get(self.TIMEOUT)
172 self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
173 self.daemon.update_server_wishlist([wish_size]).get(self.TIMEOUT)
174 self.stop_proxy(self.daemon)
175 self.assertEqual([wish_size, min_size],
176 [call[1].get('cloud_size')
177 for call in self.node_setup.start.call_args_list])
179 def test_no_new_node_when_ge_min_nodes_busy(self):
180 cloud_nodes = [testutil.cloud_node_mock(n) for n in range(1, 4)]
181 arv_nodes = [testutil.arvados_node_mock(n, job_uuid=True)
182 for n in range(1, 4)]
183 self.make_daemon(cloud_nodes, arv_nodes, [], min_nodes=2)
184 self.stop_proxy(self.daemon)
185 self.assertEqual(0, self.node_setup.start.call_count)
187 def test_no_new_node_when_max_nodes_busy(self):
188 self.make_daemon([testutil.cloud_node_mock(3)],
189 [testutil.arvados_node_mock(3, job_uuid=True)],
190 [testutil.MockSize(3)],
192 self.stop_proxy(self.daemon)
193 self.assertFalse(self.node_setup.start.called)
195 def start_node_boot(self, cloud_node=None, arv_node=None, id_num=1):
196 if cloud_node is None:
197 cloud_node = testutil.cloud_node_mock(id_num)
199 arv_node = testutil.arvados_node_mock(id_num)
200 self.make_daemon(want_sizes=[testutil.MockSize(id_num)])
201 self.daemon.max_nodes.get(self.TIMEOUT)
202 self.assertEqual(1, self.node_setup.start.call_count)
203 self.last_setup.cloud_node.get.return_value = cloud_node
204 self.last_setup.arvados_node.get.return_value = arv_node
205 return self.last_setup
207 def test_no_new_node_when_booted_node_not_usable(self):
208 cloud_node = testutil.cloud_node_mock(4)
209 arv_node = testutil.arvados_node_mock(4, crunch_worker_state='down')
210 setup = self.start_node_boot(cloud_node, arv_node)
211 self.daemon.node_up(setup).get(self.TIMEOUT)
212 self.assertEqual(1, self.alive_monitor_count())
213 self.daemon.update_cloud_nodes([cloud_node])
214 self.daemon.update_arvados_nodes([arv_node])
215 self.daemon.update_server_wishlist(
216 [testutil.MockSize(1)]).get(self.TIMEOUT)
217 self.stop_proxy(self.daemon)
218 self.assertEqual(1, self.node_setup.start.call_count)
220 def test_no_duplication_when_booting_node_listed_fast(self):
221 # Test that we don't start two ComputeNodeMonitorActors when
222 # we learn about a booting node through a listing before we
223 # get the "node up" message from CloudNodeSetupActor.
224 cloud_node = testutil.cloud_node_mock(1)
225 setup = self.start_node_boot(cloud_node)
226 self.daemon.update_cloud_nodes([cloud_node])
227 self.daemon.node_up(setup).get(self.TIMEOUT)
228 self.assertEqual(1, self.alive_monitor_count())
230 def test_no_duplication_when_booted_node_listed(self):
231 cloud_node = testutil.cloud_node_mock(2)
232 setup = self.start_node_boot(cloud_node, id_num=2)
233 self.daemon.node_up(setup)
234 self.daemon.update_cloud_nodes([cloud_node]).get(self.TIMEOUT)
235 self.assertEqual(1, self.alive_monitor_count())
237 def test_node_counted_after_boot_with_slow_listing(self):
238 # Test that, after we boot a compute node, we assume it exists
239 # even it doesn't appear in the listing (e.g., because of delays
241 setup = self.start_node_boot()
242 self.daemon.node_up(setup).get(self.TIMEOUT)
243 self.assertEqual(1, self.alive_monitor_count())
244 self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
245 self.assertEqual(1, self.alive_monitor_count())
247 def test_booted_unlisted_node_counted(self):
248 setup = self.start_node_boot(id_num=1)
249 self.daemon.node_up(setup)
250 self.daemon.update_server_wishlist(
251 [testutil.MockSize(1)]).get(self.TIMEOUT)
252 self.stop_proxy(self.daemon)
253 self.assertEqual(1, self.node_setup.start.call_count)
255 def test_booted_node_can_shutdown(self):
256 setup = self.start_node_boot()
257 self.daemon.node_up(setup).get(self.TIMEOUT)
258 self.assertEqual(1, self.alive_monitor_count())
259 monitor = self.monitor_list()[0].proxy()
260 self.daemon.update_server_wishlist([])
261 self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
262 self.stop_proxy(self.daemon)
263 self.assertTrue(self.node_shutdown.start.called,
264 "daemon did not shut down booted node on offer")
266 def test_booted_node_lifecycle(self):
267 cloud_node = testutil.cloud_node_mock(6)
268 setup = self.start_node_boot(cloud_node, id_num=6)
269 self.daemon.node_up(setup).get(self.TIMEOUT)
270 self.assertEqual(1, self.alive_monitor_count())
271 monitor = self.monitor_list()[0].proxy()
272 self.daemon.update_server_wishlist([])
273 self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
274 self.assertShutdownCancellable(True)
275 shutdown = self.node_shutdown.start().proxy()
276 shutdown.cloud_node.get.return_value = cloud_node
277 self.daemon.node_finished_shutdown(shutdown).get(self.TIMEOUT)
278 self.assertTrue(shutdown.stop.called,
279 "shutdown actor not stopped after finishing")
280 self.assertTrue(monitor.actor_ref.actor_stopped.wait(self.TIMEOUT),
281 "monitor for booted node not stopped after shutdown")
282 self.daemon.update_server_wishlist(
283 [testutil.MockSize(2)]).get(self.TIMEOUT)
284 self.stop_proxy(self.daemon)
285 self.assertTrue(self.node_setup.start.called,
286 "second node not started after booted node stopped")
288 def test_booted_node_shut_down_when_never_listed(self):
289 setup = self.start_node_boot()
290 self.daemon.node_up(setup).get(self.TIMEOUT)
291 self.assertEqual(1, self.alive_monitor_count())
292 self.assertFalse(self.node_shutdown.start.called)
294 self.stop_proxy(self.daemon)
295 self.assertShutdownCancellable(False)
297 def test_booted_node_shut_down_when_never_paired(self):
298 cloud_node = testutil.cloud_node_mock(2)
299 setup = self.start_node_boot(cloud_node)
300 self.daemon.node_up(setup).get(self.TIMEOUT)
301 self.assertEqual(1, self.alive_monitor_count())
302 self.daemon.update_cloud_nodes([cloud_node])
304 self.stop_proxy(self.daemon)
305 self.assertShutdownCancellable(False)
307 def test_booted_node_shut_down_when_never_working(self):
308 cloud_node = testutil.cloud_node_mock(4)
309 arv_node = testutil.arvados_node_mock(4, crunch_worker_state='down')
310 setup = self.start_node_boot(cloud_node, arv_node)
311 self.daemon.node_up(setup).get(self.TIMEOUT)
312 self.assertEqual(1, self.alive_monitor_count())
313 self.daemon.update_cloud_nodes([cloud_node])
314 self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
316 self.stop_proxy(self.daemon)
317 self.assertShutdownCancellable(False)
319 def test_node_that_pairs_not_considered_failed_boot(self):
320 cloud_node = testutil.cloud_node_mock(3)
321 arv_node = testutil.arvados_node_mock(3)
322 setup = self.start_node_boot(cloud_node, arv_node)
323 self.daemon.node_up(setup).get(self.TIMEOUT)
324 self.assertEqual(1, self.alive_monitor_count())
325 self.daemon.update_cloud_nodes([cloud_node])
326 self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
328 self.stop_proxy(self.daemon)
329 self.assertFalse(self.node_shutdown.start.called)
331 def test_node_that_pairs_busy_not_considered_failed_boot(self):
332 cloud_node = testutil.cloud_node_mock(5)
333 arv_node = testutil.arvados_node_mock(5, job_uuid=True)
334 setup = self.start_node_boot(cloud_node, arv_node)
335 self.daemon.node_up(setup).get(self.TIMEOUT)
336 self.assertEqual(1, self.alive_monitor_count())
337 self.daemon.update_cloud_nodes([cloud_node])
338 self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
340 self.stop_proxy(self.daemon)
341 self.assertFalse(self.node_shutdown.start.called)
343 def test_booting_nodes_shut_down(self):
344 self.make_daemon(want_sizes=[testutil.MockSize(1)])
345 self.daemon.update_server_wishlist([]).get(self.TIMEOUT)
346 self.stop_proxy(self.daemon)
347 self.assertTrue(self.last_setup.stop_if_no_cloud_node.called)
349 def test_all_booting_nodes_tried_to_shut_down(self):
350 size = testutil.MockSize(2)
351 self.make_daemon(want_sizes=[size])
352 self.daemon.max_nodes.get(self.TIMEOUT)
353 setup1 = self.last_setup
354 setup1.stop_if_no_cloud_node().get.return_value = False
355 setup1.stop_if_no_cloud_node.reset_mock()
356 self.daemon.update_server_wishlist([size, size]).get(self.TIMEOUT)
357 self.daemon.max_nodes.get(self.TIMEOUT)
358 self.assertIsNot(setup1, self.last_setup)
359 self.last_setup.stop_if_no_cloud_node().get.return_value = True
360 self.last_setup.stop_if_no_cloud_node.reset_mock()
361 self.daemon.update_server_wishlist([]).get(self.TIMEOUT)
362 self.daemon.max_nodes.get(self.TIMEOUT)
363 self.stop_proxy(self.daemon)
364 self.assertEqual(1, self.last_setup.stop_if_no_cloud_node.call_count)
365 self.assertTrue(setup1.stop_if_no_cloud_node.called)
367 def test_shutdown_declined_at_wishlist_capacity(self):
368 cloud_node = testutil.cloud_node_mock(1)
369 size = testutil.MockSize(1)
370 self.make_daemon(cloud_nodes=[cloud_node], want_sizes=[size])
371 self.assertEqual(1, self.alive_monitor_count())
372 monitor = self.monitor_list()[0].proxy()
373 self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
374 self.stop_proxy(self.daemon)
375 self.assertFalse(self.node_shutdown.start.called)
377 def test_shutdown_declined_below_min_nodes(self):
378 cloud_node = testutil.cloud_node_mock(1)
379 self.make_daemon(cloud_nodes=[cloud_node], min_nodes=1)
380 self.assertEqual(1, self.alive_monitor_count())
381 monitor = self.monitor_list()[0].proxy()
382 self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
383 self.stop_proxy(self.daemon)
384 self.assertFalse(self.node_shutdown.start.called)
386 def test_shutdown_accepted_below_capacity(self):
387 self.make_daemon(cloud_nodes=[testutil.cloud_node_mock()])
388 self.assertEqual(1, self.alive_monitor_count())
389 monitor = self.monitor_list()[0].proxy()
390 self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
391 self.stop_proxy(self.daemon)
392 self.assertTrue(self.node_shutdown.start.called)
394 def test_shutdown_declined_when_idle_and_job_queued(self):
395 cloud_nodes = [testutil.cloud_node_mock(n) for n in [3, 4]]
396 arv_nodes = [testutil.arvados_node_mock(3, job_uuid=True),
397 testutil.arvados_node_mock(4, job_uuid=None)]
398 self.make_daemon(cloud_nodes, arv_nodes, [testutil.MockSize(1)])
399 self.assertEqual(2, self.alive_monitor_count())
400 for mon_ref in self.monitor_list():
401 monitor = mon_ref.proxy()
402 if monitor.cloud_node.get(self.TIMEOUT) is cloud_nodes[-1]:
405 self.fail("monitor for idle node not found")
406 self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
407 self.stop_proxy(self.daemon)
408 self.assertFalse(self.node_shutdown.start.called)
410 def test_node_shutdown_after_cancelled_shutdown(self):
411 cloud_node = testutil.cloud_node_mock(5)
412 self.make_daemon([cloud_node], [testutil.arvados_node_mock(5)])
413 self.assertEqual(1, self.alive_monitor_count())
414 monitor = self.monitor_list()[0].proxy()
415 shutdown_proxy = self.node_shutdown.start().proxy
416 shutdown_proxy().cloud_node.get.return_value = cloud_node
417 shutdown_proxy().success.get.return_value = False
418 shutdown_proxy.reset_mock()
419 self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
420 self.assertTrue(shutdown_proxy.called)
421 self.daemon.node_finished_shutdown(shutdown_proxy()).get(self.TIMEOUT)
422 shutdown_proxy().success.get.return_value = True
423 shutdown_proxy.reset_mock()
424 self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
425 self.assertTrue(shutdown_proxy.called)
427 def test_nodes_shutting_down_replaced_below_max_nodes(self):
428 cloud_node = testutil.cloud_node_mock(6)
429 self.make_daemon([cloud_node], [testutil.arvados_node_mock(6)])
430 self.assertEqual(1, self.alive_monitor_count())
431 monitor = self.monitor_list()[0].proxy()
432 self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
433 self.assertTrue(self.node_shutdown.start.called)
434 self.daemon.update_server_wishlist(
435 [testutil.MockSize(6)]).get(self.TIMEOUT)
436 self.stop_proxy(self.daemon)
437 self.assertTrue(self.node_setup.start.called)
439 def test_nodes_shutting_down_not_replaced_at_max_nodes(self):
440 cloud_node = testutil.cloud_node_mock(7)
441 self.make_daemon([cloud_node], [testutil.arvados_node_mock(7)],
443 self.assertEqual(1, self.alive_monitor_count())
444 monitor = self.monitor_list()[0].proxy()
445 self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
446 self.assertTrue(self.node_shutdown.start.called)
447 self.daemon.update_server_wishlist(
448 [testutil.MockSize(7)]).get(self.TIMEOUT)
449 self.stop_proxy(self.daemon)
450 self.assertFalse(self.node_setup.start.called)
452 def test_nodes_shutting_down_count_against_excess(self):
453 cloud_nodes = [testutil.cloud_node_mock(n) for n in [8, 9]]
454 arv_nodes = [testutil.arvados_node_mock(n) for n in [8, 9]]
455 self.make_daemon(cloud_nodes, arv_nodes, [testutil.MockSize(8)])
456 self.assertEqual(2, self.alive_monitor_count())
457 for mon_ref in self.monitor_list():
458 self.daemon.node_can_shutdown(mon_ref.proxy()).get(self.TIMEOUT)
459 self.assertEqual(1, self.node_shutdown.start.call_count)
461 def test_clean_shutdown_waits_for_node_setup_finish(self):
462 new_node = self.start_node_boot()
463 new_node.stop_if_no_cloud_node().get.return_value = False
464 new_node.stop_if_no_cloud_node.reset_mock()
465 self.daemon.shutdown().get(self.TIMEOUT)
466 self.assertTrue(new_node.stop_if_no_cloud_node.called)
467 self.daemon.node_up(new_node).get(self.TIMEOUT)
468 self.assertTrue(new_node.stop.called)
471 self.daemon.actor_ref.actor_stopped.wait(self.TIMEOUT))
473 def test_wishlist_ignored_after_shutdown(self):
474 new_node = self.start_node_boot()
475 new_node.stop_if_no_cloud_node().get.return_value = False
476 new_node.stop_if_no_cloud_node.reset_mock()
477 self.daemon.shutdown().get(self.TIMEOUT)
478 size = testutil.MockSize(2)
479 self.daemon.update_server_wishlist([size] * 2).get(self.TIMEOUT)
481 self.stop_proxy(self.daemon)
482 self.assertEqual(1, self.node_setup.start.call_count)