5353: Parameterize the following methods on node size: _nodes_up, _nodes_busy,
[arvados.git] / 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 new_setup(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_node = mock.MagicMock()
23         get_cloud_node.get.return_value = mock.NonCallableMock(size=kwargs["cloud_size"])
24
25         self.last_setup = mock.NonCallableMock(name='setup_mock',
26                                                cloud_node=get_cloud_node)
27         self.last_setup.proxy = mock.MagicMock(return_value = self.last_setup)
28
29         return self.last_setup
30
31     def make_daemon(self, cloud_nodes=[], arvados_nodes=[], want_sizes=[],
32                     avail_sizes=[(testutil.MockSize(1), {"cores": 1})], min_nodes=0, max_nodes=8):
33         for name in ['cloud_nodes', 'arvados_nodes', 'server_wishlist']:
34             setattr(self, name + '_poller', mock.MagicMock(name=name + '_mock'))
35         self.arv_factory = mock.MagicMock(name='arvados_mock')
36         self.cloud_factory = mock.MagicMock(name='cloud_mock')
37         self.cloud_factory().node_start_time.return_value = time.time()
38         self.cloud_updates = mock.MagicMock(name='updates_mock')
39         self.timer = testutil.MockTimer(deliver_immediately=False)
40
41         self.node_setup = mock.MagicMock(name='setup_mock')
42         self.node_setup.start.side_effect = self.new_setup
43         self.node_setup.reset_mock()
44         self.node_shutdown = mock.MagicMock(name='shutdown_mock')
45         self.daemon = nmdaemon.NodeManagerDaemonActor.start(
46             self.server_wishlist_poller, self.arvados_nodes_poller,
47             self.cloud_nodes_poller, self.cloud_updates, self.timer,
48             self.arv_factory, self.cloud_factory,
49             [54, 5, 1], ServerCalculator(avail_sizes),
50             min_nodes, max_nodes, 600, 1800, 3600,
51             self.node_setup, self.node_shutdown).proxy()
52         if cloud_nodes is not None:
53             self.daemon.update_cloud_nodes(cloud_nodes).get(self.TIMEOUT)
54         if arvados_nodes is not None:
55             self.daemon.update_arvados_nodes(arvados_nodes).get(self.TIMEOUT)
56         if want_sizes is not None:
57             self.daemon.update_server_wishlist(want_sizes).get(self.TIMEOUT)
58
59     def monitor_list(self):
60         return pykka.ActorRegistry.get_by_class(ComputeNodeMonitorActor)
61
62     def monitored_arvados_nodes(self):
63         pairings = []
64         for future in [actor.proxy().arvados_node
65                        for actor in self.monitor_list()]:
66             try:
67                 pairings.append(future.get(self.TIMEOUT))
68             except pykka.ActorDeadError:
69                 pass
70         return pairings
71
72     def alive_monitor_count(self):
73         return len(self.monitored_arvados_nodes())
74
75     def assertShutdownCancellable(self, expected=True):
76         self.assertTrue(self.node_shutdown.start.called)
77         self.assertIs(expected,
78                       self.node_shutdown.start.call_args[1]['cancellable'],
79                       "ComputeNodeShutdownActor incorrectly cancellable")
80
81     def test_easy_node_creation(self):
82         size = testutil.MockSize(1)
83         self.make_daemon(want_sizes=[size])
84         self.stop_proxy(self.daemon)
85         self.assertTrue(self.node_setup.start.called)
86
87     def check_monitors_arvados_nodes(self, *arv_nodes):
88         self.assertItemsEqual(arv_nodes, self.monitored_arvados_nodes())
89
90     def test_node_pairing(self):
91         cloud_node = testutil.cloud_node_mock(1)
92         arv_node = testutil.arvados_node_mock(1)
93         self.make_daemon([cloud_node], [arv_node])
94         self.stop_proxy(self.daemon)
95         self.check_monitors_arvados_nodes(arv_node)
96
97     def test_node_pairing_after_arvados_update(self):
98         cloud_node = testutil.cloud_node_mock(2)
99         self.make_daemon([cloud_node],
100                          [testutil.arvados_node_mock(2, ip_address=None)])
101         arv_node = testutil.arvados_node_mock(2)
102         self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
103         self.stop_proxy(self.daemon)
104         self.check_monitors_arvados_nodes(arv_node)
105
106     def test_arvados_node_un_and_re_paired(self):
107         # We need to create the Arvados node mock after spinning up the daemon
108         # to make sure it's new enough to pair with the cloud node.
109         self.make_daemon([testutil.cloud_node_mock(3)], arvados_nodes=None)
110         arv_node = testutil.arvados_node_mock(3)
111         self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
112         self.check_monitors_arvados_nodes(arv_node)
113         self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
114         self.assertEqual(0, self.alive_monitor_count())
115         self.daemon.update_cloud_nodes([testutil.cloud_node_mock(3)])
116         self.stop_proxy(self.daemon)
117         self.check_monitors_arvados_nodes(arv_node)
118
119     def test_old_arvados_node_not_double_assigned(self):
120         arv_node = testutil.arvados_node_mock(3, age=9000)
121         size = testutil.MockSize(3)
122         self.make_daemon(arvados_nodes=[arv_node])
123         self.daemon.update_server_wishlist([size]).get(self.TIMEOUT)
124         self.daemon.update_server_wishlist([size, size]).get(self.TIMEOUT)
125         self.stop_proxy(self.daemon)
126         used_nodes = [call[1].get('arvados_node')
127                       for call in self.node_setup.start.call_args_list]
128         self.assertEqual(2, len(used_nodes))
129         self.assertIn(arv_node, used_nodes)
130         self.assertIn(None, used_nodes)
131
132     def test_node_count_satisfied(self):
133         self.make_daemon([testutil.cloud_node_mock()],
134                          want_sizes=[testutil.MockSize(1)])
135         self.stop_proxy(self.daemon)
136         self.assertFalse(self.node_setup.start.called)
137
138     def test_dont_count_missing_as_busy(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])
145         self.stop_proxy(self.daemon)
146         self.assertTrue(self.node_setup.start.called)
147
148     def test_missing_counts_towards_max(self):
149         size = testutil.MockSize(1)
150         self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(1),
151                                       testutil.cloud_node_mock(2)],
152                          arvados_nodes=[testutil.arvados_node_mock(1),
153                                         testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
154                          want_sizes=[size, size],
155                          max_nodes=2)
156         self.stop_proxy(self.daemon)
157         self.assertFalse(self.node_setup.start.called)
158
159     def test_excess_counts_missing(self):
160         size = testutil.MockSize(1)
161         cloud_nodes = [testutil.cloud_node_mock(1), testutil.cloud_node_mock(2)]
162         self.make_daemon(cloud_nodes=cloud_nodes,
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])
166         self.assertEqual(2, self.alive_monitor_count())
167         for mon_ref in self.monitor_list():
168             self.daemon.node_can_shutdown(mon_ref.proxy()).get(self.TIMEOUT)
169         self.assertEqual(1, self.node_shutdown.start.call_count)
170
171     def test_missing_shutdown_not_excess(self):
172         size = testutil.MockSize(1)
173         cloud_nodes = [testutil.cloud_node_mock(1), testutil.cloud_node_mock(2)]
174         self.make_daemon(cloud_nodes=cloud_nodes,
175                          arvados_nodes=[testutil.arvados_node_mock(1),
176                                         testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
177                          want_sizes=[size])
178         self.daemon.shutdowns.get()[cloud_nodes[1].id] = mock.MagicMock(name='shutdown_proxy_mock')
179         self.assertEqual(2, self.alive_monitor_count())
180         for mon_ref in self.monitor_list():
181             self.daemon.node_can_shutdown(mon_ref.proxy()).get(self.TIMEOUT)
182         self.assertEqual(0, self.node_shutdown.start.call_count)
183
184     def test_booting_nodes_counted(self):
185         cloud_node = testutil.cloud_node_mock(1)
186         arv_node = testutil.arvados_node_mock(1)
187         server_wishlist = [testutil.MockSize(1)] * 2
188         self.make_daemon([cloud_node], [arv_node], server_wishlist)
189         self.daemon.max_nodes.get(self.TIMEOUT)
190         self.assertTrue(self.node_setup.start.called)
191         self.daemon.update_server_wishlist(server_wishlist).get(self.TIMEOUT)
192         self.stop_proxy(self.daemon)
193         self.assertEqual(1, self.node_setup.start.call_count)
194
195     def test_boot_new_node_when_all_nodes_busy(self):
196         arv_node = testutil.arvados_node_mock(2, job_uuid=True)
197         self.make_daemon([testutil.cloud_node_mock(2)], [arv_node],
198                          [testutil.MockSize(2)])
199         self.stop_proxy(self.daemon)
200         self.assertTrue(self.node_setup.start.called)
201
202     def test_boot_new_node_below_min_nodes(self):
203         min_size = testutil.MockSize(1)
204         wish_size = testutil.MockSize(3)
205         avail_sizes = [(min_size, {"cores": 1}),
206                        (wish_size, {"cores": 3})]
207         self.make_daemon([], [], None, avail_sizes=avail_sizes, min_nodes=2)
208         self.daemon.update_server_wishlist([wish_size]).get(self.TIMEOUT)
209         self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
210         self.daemon.update_server_wishlist([wish_size]).get(self.TIMEOUT)
211         self.stop_proxy(self.daemon)
212         self.assertEqual([wish_size, min_size],
213                          [call[1].get('cloud_size')
214                           for call in self.node_setup.start.call_args_list])
215
216     def test_no_new_node_when_ge_min_nodes_busy(self):
217         cloud_nodes = [testutil.cloud_node_mock(n) for n in range(1, 4)]
218         arv_nodes = [testutil.arvados_node_mock(n, job_uuid=True)
219                      for n in range(1, 4)]
220         self.make_daemon(cloud_nodes, arv_nodes, [], min_nodes=2)
221         self.stop_proxy(self.daemon)
222         self.assertEqual(0, self.node_setup.start.call_count)
223
224     def test_no_new_node_when_max_nodes_busy(self):
225         self.make_daemon([testutil.cloud_node_mock(3)],
226                          [testutil.arvados_node_mock(3, job_uuid=True)],
227                          [testutil.MockSize(3)],
228                          max_nodes=1)
229         self.stop_proxy(self.daemon)
230         self.assertFalse(self.node_setup.start.called)
231
232     def start_node_boot(self, cloud_node=None, arv_node=None, id_num=1):
233         if cloud_node is None:
234             cloud_node = testutil.cloud_node_mock(id_num)
235         if arv_node is None:
236             arv_node = testutil.arvados_node_mock(id_num)
237         self.make_daemon(want_sizes=[testutil.MockSize(id_num)])
238         self.daemon.max_nodes.get(self.TIMEOUT)
239         self.assertEqual(1, self.node_setup.start.call_count)
240         self.last_setup.cloud_node.get.return_value = cloud_node
241         self.last_setup.arvados_node.get.return_value = arv_node
242         return self.last_setup
243
244     def test_no_new_node_when_booted_node_not_usable(self):
245         cloud_node = testutil.cloud_node_mock(4)
246         arv_node = testutil.arvados_node_mock(4, crunch_worker_state='down')
247         setup = self.start_node_boot(cloud_node, arv_node)
248         self.daemon.node_up(setup).get(self.TIMEOUT)
249         self.assertEqual(1, self.alive_monitor_count())
250         self.daemon.update_cloud_nodes([cloud_node])
251         self.daemon.update_arvados_nodes([arv_node])
252         self.daemon.update_server_wishlist(
253             [testutil.MockSize(1)]).get(self.TIMEOUT)
254         self.stop_proxy(self.daemon)
255         self.assertEqual(1, self.node_setup.start.call_count)
256
257     def test_no_duplication_when_booting_node_listed_fast(self):
258         # Test that we don't start two ComputeNodeMonitorActors when
259         # we learn about a booting node through a listing before we
260         # get the "node up" message from CloudNodeSetupActor.
261         cloud_node = testutil.cloud_node_mock(1)
262         setup = self.start_node_boot(cloud_node)
263         self.daemon.update_cloud_nodes([cloud_node])
264         self.daemon.node_up(setup).get(self.TIMEOUT)
265         self.assertEqual(1, self.alive_monitor_count())
266
267     def test_no_duplication_when_booted_node_listed(self):
268         cloud_node = testutil.cloud_node_mock(2)
269         setup = self.start_node_boot(cloud_node, id_num=2)
270         self.daemon.node_up(setup)
271         self.daemon.update_cloud_nodes([cloud_node]).get(self.TIMEOUT)
272         self.assertEqual(1, self.alive_monitor_count())
273
274     def test_node_counted_after_boot_with_slow_listing(self):
275         # Test that, after we boot a compute node, we assume it exists
276         # even it doesn't appear in the listing (e.g., because of delays
277         # propagating tags).
278         setup = self.start_node_boot()
279         self.daemon.node_up(setup).get(self.TIMEOUT)
280         self.assertEqual(1, self.alive_monitor_count())
281         self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
282         self.assertEqual(1, self.alive_monitor_count())
283
284     def test_booted_unlisted_node_counted(self):
285         setup = self.start_node_boot(id_num=1)
286         self.daemon.node_up(setup)
287         self.daemon.update_server_wishlist(
288             [testutil.MockSize(1)]).get(self.TIMEOUT)
289         self.stop_proxy(self.daemon)
290         self.assertEqual(1, self.node_setup.start.call_count)
291
292     def test_booted_node_can_shutdown(self):
293         setup = self.start_node_boot()
294         self.daemon.node_up(setup).get(self.TIMEOUT)
295         self.assertEqual(1, self.alive_monitor_count())
296         monitor = self.monitor_list()[0].proxy()
297         self.daemon.update_server_wishlist([])
298         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
299         self.stop_proxy(self.daemon)
300         self.assertTrue(self.node_shutdown.start.called,
301                         "daemon did not shut down booted node on offer")
302
303     def test_booted_node_lifecycle(self):
304         cloud_node = testutil.cloud_node_mock(6)
305         setup = self.start_node_boot(cloud_node, id_num=6)
306         self.daemon.node_up(setup).get(self.TIMEOUT)
307         self.assertEqual(1, self.alive_monitor_count())
308         monitor = self.monitor_list()[0].proxy()
309         self.daemon.update_server_wishlist([])
310         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
311         self.assertShutdownCancellable(True)
312         shutdown = self.node_shutdown.start().proxy()
313         shutdown.cloud_node.get.return_value = cloud_node
314         self.daemon.node_finished_shutdown(shutdown).get(self.TIMEOUT)
315         self.assertTrue(shutdown.stop.called,
316                         "shutdown actor not stopped after finishing")
317         self.assertTrue(monitor.actor_ref.actor_stopped.wait(self.TIMEOUT),
318                         "monitor for booted node not stopped after shutdown")
319         self.daemon.update_server_wishlist(
320             [testutil.MockSize(2)]).get(self.TIMEOUT)
321         self.stop_proxy(self.daemon)
322         self.assertTrue(self.node_setup.start.called,
323                         "second node not started after booted node stopped")
324
325     def test_booted_node_shut_down_when_never_listed(self):
326         setup = self.start_node_boot()
327         self.daemon.node_up(setup).get(self.TIMEOUT)
328         self.assertEqual(1, self.alive_monitor_count())
329         self.assertFalse(self.node_shutdown.start.called)
330         self.timer.deliver()
331         self.stop_proxy(self.daemon)
332         self.assertShutdownCancellable(False)
333
334     def test_booted_node_shut_down_when_never_paired(self):
335         cloud_node = testutil.cloud_node_mock(2)
336         setup = self.start_node_boot(cloud_node)
337         self.daemon.node_up(setup).get(self.TIMEOUT)
338         self.assertEqual(1, self.alive_monitor_count())
339         self.daemon.update_cloud_nodes([cloud_node])
340         self.timer.deliver()
341         self.stop_proxy(self.daemon)
342         self.assertShutdownCancellable(False)
343
344     def test_booted_node_shut_down_when_never_working(self):
345         cloud_node = testutil.cloud_node_mock(4)
346         arv_node = testutil.arvados_node_mock(4, crunch_worker_state='down')
347         setup = self.start_node_boot(cloud_node, arv_node)
348         self.daemon.node_up(setup).get(self.TIMEOUT)
349         self.assertEqual(1, self.alive_monitor_count())
350         self.daemon.update_cloud_nodes([cloud_node])
351         self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
352         self.timer.deliver()
353         self.stop_proxy(self.daemon)
354         self.assertShutdownCancellable(False)
355
356     def test_node_that_pairs_not_considered_failed_boot(self):
357         cloud_node = testutil.cloud_node_mock(3)
358         arv_node = testutil.arvados_node_mock(3)
359         setup = self.start_node_boot(cloud_node, arv_node)
360         self.daemon.node_up(setup).get(self.TIMEOUT)
361         self.assertEqual(1, self.alive_monitor_count())
362         self.daemon.update_cloud_nodes([cloud_node])
363         self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
364         self.timer.deliver()
365         self.stop_proxy(self.daemon)
366         self.assertFalse(self.node_shutdown.start.called)
367
368     def test_node_that_pairs_busy_not_considered_failed_boot(self):
369         cloud_node = testutil.cloud_node_mock(5)
370         arv_node = testutil.arvados_node_mock(5, job_uuid=True)
371         setup = self.start_node_boot(cloud_node, arv_node)
372         self.daemon.node_up(setup).get(self.TIMEOUT)
373         self.assertEqual(1, self.alive_monitor_count())
374         self.daemon.update_cloud_nodes([cloud_node])
375         self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
376         self.timer.deliver()
377         self.stop_proxy(self.daemon)
378         self.assertFalse(self.node_shutdown.start.called)
379
380     def test_booting_nodes_shut_down(self):
381         self.make_daemon(want_sizes=[testutil.MockSize(1)])
382         self.daemon.update_server_wishlist([]).get(self.TIMEOUT)
383         self.stop_proxy(self.daemon)
384         self.assertTrue(self.last_setup.stop_if_no_cloud_node.called)
385
386     def test_all_booting_nodes_tried_to_shut_down(self):
387         size = testutil.MockSize(2)
388         self.make_daemon(want_sizes=[size])
389         self.daemon.max_nodes.get(self.TIMEOUT)
390         setup1 = self.last_setup
391         setup1.stop_if_no_cloud_node().get.return_value = False
392         setup1.stop_if_no_cloud_node.reset_mock()
393         self.daemon.update_server_wishlist([size, size]).get(self.TIMEOUT)
394         self.daemon.max_nodes.get(self.TIMEOUT)
395         self.assertIsNot(setup1, self.last_setup)
396         self.last_setup.stop_if_no_cloud_node().get.return_value = True
397         self.last_setup.stop_if_no_cloud_node.reset_mock()
398         self.daemon.update_server_wishlist([]).get(self.TIMEOUT)
399         self.daemon.max_nodes.get(self.TIMEOUT)
400         self.stop_proxy(self.daemon)
401         self.assertEqual(1, self.last_setup.stop_if_no_cloud_node.call_count)
402         self.assertTrue(setup1.stop_if_no_cloud_node.called)
403
404     def test_shutdown_declined_at_wishlist_capacity(self):
405         cloud_node = testutil.cloud_node_mock(1)
406         size = testutil.MockSize(1)
407         self.make_daemon(cloud_nodes=[cloud_node], want_sizes=[size])
408         self.assertEqual(1, self.alive_monitor_count())
409         monitor = self.monitor_list()[0].proxy()
410         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
411         self.stop_proxy(self.daemon)
412         self.assertFalse(self.node_shutdown.start.called)
413
414     def test_shutdown_declined_below_min_nodes(self):
415         cloud_node = testutil.cloud_node_mock(1)
416         self.make_daemon(cloud_nodes=[cloud_node], min_nodes=1)
417         self.assertEqual(1, self.alive_monitor_count())
418         monitor = self.monitor_list()[0].proxy()
419         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
420         self.stop_proxy(self.daemon)
421         self.assertFalse(self.node_shutdown.start.called)
422
423     def test_shutdown_accepted_below_capacity(self):
424         self.make_daemon(cloud_nodes=[testutil.cloud_node_mock()])
425         self.assertEqual(1, self.alive_monitor_count())
426         monitor = self.monitor_list()[0].proxy()
427         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
428         self.stop_proxy(self.daemon)
429         self.assertTrue(self.node_shutdown.start.called)
430
431     def test_shutdown_declined_when_idle_and_job_queued(self):
432         cloud_nodes = [testutil.cloud_node_mock(n) for n in [3, 4]]
433         arv_nodes = [testutil.arvados_node_mock(3, job_uuid=True),
434                      testutil.arvados_node_mock(4, job_uuid=None)]
435         self.make_daemon(cloud_nodes, arv_nodes, [testutil.MockSize(1)])
436         self.assertEqual(2, self.alive_monitor_count())
437         for mon_ref in self.monitor_list():
438             monitor = mon_ref.proxy()
439             if monitor.cloud_node.get(self.TIMEOUT) is cloud_nodes[-1]:
440                 break
441         else:
442             self.fail("monitor for idle node not found")
443         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
444         self.stop_proxy(self.daemon)
445         self.assertFalse(self.node_shutdown.start.called)
446
447     def test_node_shutdown_after_cancelled_shutdown(self):
448         cloud_node = testutil.cloud_node_mock(5)
449         self.make_daemon([cloud_node], [testutil.arvados_node_mock(5)])
450         self.assertEqual(1, self.alive_monitor_count())
451         monitor = self.monitor_list()[0].proxy()
452         shutdown_proxy = self.node_shutdown.start().proxy
453         shutdown_proxy().cloud_node.get.return_value = cloud_node
454         shutdown_proxy().success.get.return_value = False
455         shutdown_proxy.reset_mock()
456         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
457         self.assertTrue(shutdown_proxy.called)
458         self.daemon.node_finished_shutdown(shutdown_proxy()).get(self.TIMEOUT)
459         shutdown_proxy().success.get.return_value = True
460         shutdown_proxy.reset_mock()
461         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
462         self.assertTrue(shutdown_proxy.called)
463
464     def test_broken_node_blackholed_after_cancelled_shutdown(self):
465         cloud_node = testutil.cloud_node_mock(8)
466         wishlist = [testutil.MockSize(8)]
467         self.make_daemon([cloud_node], [testutil.arvados_node_mock(8)],
468                          wishlist)
469         self.assertEqual(1, self.alive_monitor_count())
470         self.assertFalse(self.node_setup.start.called)
471         monitor = self.monitor_list()[0].proxy()
472         shutdown_proxy = self.node_shutdown.start().proxy
473         shutdown_proxy().cloud_node.get.return_value = cloud_node
474         shutdown_proxy().success.get.return_value = False
475         shutdown_proxy().cancel_reason.get.return_value = self.node_shutdown.NODE_BROKEN
476         self.daemon.update_server_wishlist([]).get(self.TIMEOUT)
477         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
478         self.daemon.node_finished_shutdown(shutdown_proxy()).get(self.TIMEOUT)
479         self.daemon.update_cloud_nodes([cloud_node]).get(self.TIMEOUT)
480         self.daemon.update_server_wishlist(wishlist).get(self.TIMEOUT)
481         self.stop_proxy(self.daemon)
482         self.assertEqual(1, self.node_setup.start.call_count)
483
484     def test_nodes_shutting_down_replaced_below_max_nodes(self):
485         cloud_node = testutil.cloud_node_mock(6)
486         self.make_daemon([cloud_node], [testutil.arvados_node_mock(6)])
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.assertTrue(self.node_shutdown.start.called)
491         self.daemon.update_server_wishlist(
492             [testutil.MockSize(6)]).get(self.TIMEOUT)
493         self.stop_proxy(self.daemon)
494         self.assertTrue(self.node_setup.start.called)
495
496     def test_nodes_shutting_down_not_replaced_at_max_nodes(self):
497         cloud_node = testutil.cloud_node_mock(7)
498         self.make_daemon([cloud_node], [testutil.arvados_node_mock(7)],
499                          max_nodes=1)
500         self.assertEqual(1, self.alive_monitor_count())
501         monitor = self.monitor_list()[0].proxy()
502         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
503         self.assertTrue(self.node_shutdown.start.called)
504         self.daemon.update_server_wishlist(
505             [testutil.MockSize(7)]).get(self.TIMEOUT)
506         self.stop_proxy(self.daemon)
507         self.assertFalse(self.node_setup.start.called)
508
509     def test_nodes_shutting_down_count_against_excess(self):
510         cloud_nodes = [testutil.cloud_node_mock(n) for n in [8, 9]]
511         arv_nodes = [testutil.arvados_node_mock(n) for n in [8, 9]]
512         self.make_daemon(cloud_nodes, arv_nodes, [testutil.MockSize(8)])
513         self.assertEqual(2, self.alive_monitor_count())
514         for mon_ref in self.monitor_list():
515             self.daemon.node_can_shutdown(mon_ref.proxy()).get(self.TIMEOUT)
516         self.assertEqual(1, self.node_shutdown.start.call_count)
517
518     def test_clean_shutdown_waits_for_node_setup_finish(self):
519         new_node = self.start_node_boot()
520         new_node.stop_if_no_cloud_node().get.return_value = False
521         new_node.stop_if_no_cloud_node.reset_mock()
522         self.daemon.shutdown().get(self.TIMEOUT)
523         self.assertTrue(new_node.stop_if_no_cloud_node.called)
524         self.daemon.node_up(new_node).get(self.TIMEOUT)
525         self.assertTrue(new_node.stop.called)
526         self.timer.deliver()
527         self.assertTrue(
528             self.daemon.actor_ref.actor_stopped.wait(self.TIMEOUT))
529
530     def test_wishlist_ignored_after_shutdown(self):
531         new_node = self.start_node_boot()
532         new_node.stop_if_no_cloud_node().get.return_value = False
533         new_node.stop_if_no_cloud_node.reset_mock()
534         self.daemon.shutdown().get(self.TIMEOUT)
535         size = testutil.MockSize(2)
536         self.daemon.update_server_wishlist([size] * 2).get(self.TIMEOUT)
537         self.timer.deliver()
538         self.stop_proxy(self.daemon)
539         self.assertEqual(1, self.node_setup.start.call_count)
540
541     def test_shutdown_actor_stopped_when_cloud_node_delisted(self):
542         self.make_daemon(cloud_nodes=[testutil.cloud_node_mock()])
543         self.assertEqual(1, self.alive_monitor_count())
544         monitor = self.monitor_list()[0].proxy()
545         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
546         self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
547         self.stop_proxy(self.daemon)
548         self.assertEqual(
549             1, self.node_shutdown.start().proxy().stop().get.call_count)
550
551     def test_shutdown_actor_cleanup_copes_with_dead_actors(self):
552         self.make_daemon(cloud_nodes=[testutil.cloud_node_mock()])
553         self.assertEqual(1, self.alive_monitor_count())
554         monitor = self.monitor_list()[0].proxy()
555         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
556         # We're mainly testing that update_cloud_nodes catches and handles
557         # the ActorDeadError.
558         stop_method = self.node_shutdown.start().proxy().stop().get
559         stop_method.side_effect = pykka.ActorDeadError
560         self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
561         self.stop_proxy(self.daemon)
562         self.assertEqual(1, stop_method.call_count)