8561: Set ec2_instance_id and cloud_factory().node_id to mock cloud/arvados id
[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     maxDiff = None
20     def mock_node_start(self, **kwargs):
21         # Make sure that every time the daemon starts a setup actor,
22         # it gets a new mock object back.
23         get_cloud_size = mock.MagicMock()
24         get_cloud_size.get.return_value = kwargs["cloud_size"]
25         mock_actor = mock.MagicMock()
26         mock_proxy = mock.NonCallableMock(name='setup_mock_proxy',
27                                           cloud_size=get_cloud_size,
28                                           actor_ref=mock_actor)
29         mock_actor.proxy.return_value = mock_proxy
30         mock_actor.tell_proxy.return_value = mock_proxy
31
32         self.last_setup = mock_proxy
33         return mock_actor
34
35     def mock_node_shutdown(self, **kwargs):
36         # Make sure that every time the daemon starts a shutdown actor,
37         # it gets a new mock object back.
38         get_cloud_node = mock.MagicMock()
39         if "node_monitor" in kwargs:
40             get_cloud_node.get.return_value = kwargs["node_monitor"].proxy().cloud_node.get()
41         mock_actor = mock.MagicMock()
42         mock_proxy = mock.NonCallableMock(name='shutdown_mock_proxy',
43                                           cloud_node=get_cloud_node,
44                                           actor_ref=mock_actor)
45
46         mock_actor.proxy.return_value = mock_proxy
47         self.last_shutdown = mock_proxy
48
49         return mock_actor
50
51     def make_daemon(self, cloud_nodes=[], arvados_nodes=[], want_sizes=[],
52                     avail_sizes=[(testutil.MockSize(1), {"cores": 1})],
53                     min_nodes=0, max_nodes=8,
54                     shutdown_windows=[54, 5, 1],
55                     max_total_price=None):
56         for name in ['cloud_nodes', 'arvados_nodes', 'server_wishlist']:
57             setattr(self, name + '_poller', mock.MagicMock(name=name + '_mock'))
58         self.arv_factory = mock.MagicMock(name='arvados_mock')
59         self.cloud_factory = mock.MagicMock(name='cloud_mock')
60         self.cloud_factory().node_start_time.return_value = time.time()
61         self.cloud_updates = mock.MagicMock(name='updates_mock')
62         self.timer = testutil.MockTimer(deliver_immediately=False)
63         self.cloud_factory().node_id.side_effect = lambda node: node.id
64
65         self.node_setup = mock.MagicMock(name='setup_mock')
66         self.node_setup.start.side_effect = self.mock_node_start
67         self.node_setup.reset_mock()
68
69         self.node_shutdown = mock.MagicMock(name='shutdown_mock')
70         self.node_shutdown.start.side_effect = self.mock_node_shutdown
71
72         self.daemon = nmdaemon.NodeManagerDaemonActor.start(
73             self.server_wishlist_poller, self.arvados_nodes_poller,
74             self.cloud_nodes_poller, self.cloud_updates, self.timer,
75             self.arv_factory, self.cloud_factory,
76             shutdown_windows, ServerCalculator(avail_sizes),
77             min_nodes, max_nodes, 600, 1800, 3600,
78             self.node_setup, self.node_shutdown,
79             max_total_price=max_total_price).proxy()
80         if cloud_nodes is not None:
81             self.daemon.update_cloud_nodes(cloud_nodes).get(self.TIMEOUT)
82         if arvados_nodes is not None:
83             self.daemon.update_arvados_nodes(arvados_nodes).get(self.TIMEOUT)
84         if want_sizes is not None:
85             self.daemon.update_server_wishlist(want_sizes).get(self.TIMEOUT)
86
87     def monitor_list(self):
88         return pykka.ActorRegistry.get_by_class(ComputeNodeMonitorActor)
89
90     def monitored_arvados_nodes(self):
91         pairings = []
92         for future in [actor.proxy().arvados_node
93                        for actor in self.monitor_list()]:
94             try:
95                 pairings.append(future.get(self.TIMEOUT))
96             except pykka.ActorDeadError:
97                 pass
98         return pairings
99
100     def alive_monitor_count(self):
101         return len(self.monitored_arvados_nodes())
102
103     def assertShutdownCancellable(self, expected=True):
104         self.assertTrue(self.node_shutdown.start.called)
105         self.assertIs(expected,
106                       self.node_shutdown.start.call_args[1]['cancellable'],
107                       "ComputeNodeShutdownActor incorrectly cancellable")
108
109     def test_easy_node_creation(self):
110         size = testutil.MockSize(1)
111         self.make_daemon(want_sizes=[size])
112         self.stop_proxy(self.daemon)
113         self.assertTrue(self.node_setup.start.called)
114
115     def check_monitors_arvados_nodes(self, *arv_nodes):
116         self.assertItemsEqual(arv_nodes, self.monitored_arvados_nodes())
117
118     def test_node_pairing(self):
119         cloud_node = testutil.cloud_node_mock(1)
120         arv_node = testutil.arvados_node_mock(1)
121         self.make_daemon([cloud_node], [arv_node])
122         self.stop_proxy(self.daemon)
123         self.check_monitors_arvados_nodes(arv_node)
124
125     def test_node_pairing_after_arvados_update(self):
126         cloud_node = testutil.cloud_node_mock(2)
127         self.make_daemon([cloud_node],
128                          [testutil.arvados_node_mock(1, ip_address=None)])
129         arv_node = testutil.arvados_node_mock(2)
130         self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
131         self.stop_proxy(self.daemon)
132         self.check_monitors_arvados_nodes(arv_node)
133
134     def test_arvados_node_un_and_re_paired(self):
135         # We need to create the Arvados node mock after spinning up the daemon
136         # to make sure it's new enough to pair with the cloud node.
137         self.make_daemon([testutil.cloud_node_mock(3)], arvados_nodes=None)
138         arv_node = testutil.arvados_node_mock(3)
139         self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
140         self.check_monitors_arvados_nodes(arv_node)
141         self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
142         self.assertEqual(0, self.alive_monitor_count())
143         self.daemon.update_cloud_nodes([testutil.cloud_node_mock(3)])
144         self.stop_proxy(self.daemon)
145         self.check_monitors_arvados_nodes(arv_node)
146
147     def test_old_arvados_node_not_double_assigned(self):
148         arv_node = testutil.arvados_node_mock(3, age=9000)
149         size = testutil.MockSize(3)
150         self.make_daemon(arvados_nodes=[arv_node], avail_sizes=[(size, {"cores":1})])
151         self.daemon.update_server_wishlist([size]).get(self.TIMEOUT)
152         self.daemon.update_server_wishlist([size, size]).get(self.TIMEOUT)
153         self.stop_proxy(self.daemon)
154         used_nodes = [call[1].get('arvados_node')
155                       for call in self.node_setup.start.call_args_list]
156         self.assertEqual(2, len(used_nodes))
157         self.assertIn(arv_node, used_nodes)
158         self.assertIn(None, used_nodes)
159
160     def test_node_count_satisfied(self):
161         self.make_daemon([testutil.cloud_node_mock()],
162                          want_sizes=[testutil.MockSize(1)])
163         self.stop_proxy(self.daemon)
164         self.assertFalse(self.node_setup.start.called)
165
166     def test_dont_count_missing_as_busy(self):
167         size = testutil.MockSize(1)
168         self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(1),
169                                       testutil.cloud_node_mock(2)],
170                          arvados_nodes=[testutil.arvados_node_mock(1),
171                                       testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
172                          want_sizes=[size, size])
173         self.stop_proxy(self.daemon)
174         self.assertTrue(self.node_setup.start.called)
175
176     def test_missing_counts_towards_max(self):
177         size = testutil.MockSize(1)
178         self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(1),
179                                       testutil.cloud_node_mock(2)],
180                          arvados_nodes=[testutil.arvados_node_mock(1),
181                                         testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
182                          want_sizes=[size, size],
183                          max_nodes=2)
184         self.stop_proxy(self.daemon)
185         self.assertFalse(self.node_setup.start.called)
186
187     def test_excess_counts_missing(self):
188         size = testutil.MockSize(1)
189         cloud_nodes = [testutil.cloud_node_mock(1), testutil.cloud_node_mock(2)]
190         self.make_daemon(cloud_nodes=cloud_nodes,
191                          arvados_nodes=[testutil.arvados_node_mock(1),
192                                         testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
193                          want_sizes=[size])
194         self.assertEqual(2, self.alive_monitor_count())
195         for mon_ref in self.monitor_list():
196             self.daemon.node_can_shutdown(mon_ref.proxy()).get(self.TIMEOUT)
197         self.assertEqual(1, self.node_shutdown.start.call_count)
198
199     def test_missing_shutdown_not_excess(self):
200         size = testutil.MockSize(1)
201         cloud_nodes = [testutil.cloud_node_mock(1), testutil.cloud_node_mock(2)]
202         self.make_daemon(cloud_nodes=cloud_nodes,
203                          arvados_nodes=[testutil.arvados_node_mock(1),
204                                         testutil.arvados_node_mock(2, last_ping_at='1970-01-01T01:02:03.04050607Z')],
205                          want_sizes=[size])
206
207         get_cloud_node = mock.MagicMock(name="get_cloud_node")
208         get_cloud_node.get.return_value = cloud_nodes[1]
209         mock_node_monitor = mock.MagicMock()
210         mock_node_monitor.proxy.return_value = mock.NonCallableMock(cloud_node=get_cloud_node)
211         mock_shutdown = self.node_shutdown.start(node_monitor=mock_node_monitor)
212
213         self.daemon.shutdowns.get()[cloud_nodes[1].id] = mock_shutdown.proxy()
214         self.daemon.sizes_booting_shutdown.get()[cloud_nodes[1].id] = size
215
216         self.assertEqual(2, self.alive_monitor_count())
217         for mon_ref in self.monitor_list():
218             self.daemon.node_can_shutdown(mon_ref.proxy()).get(self.TIMEOUT)
219         self.assertEqual(1, self.node_shutdown.start.call_count)
220
221     def test_booting_nodes_counted(self):
222         cloud_node = testutil.cloud_node_mock(1)
223         arv_node = testutil.arvados_node_mock(1)
224         server_wishlist = [testutil.MockSize(1)] * 2
225         self.make_daemon([cloud_node], [arv_node], server_wishlist)
226         self.daemon.max_nodes.get(self.TIMEOUT)
227         self.assertTrue(self.node_setup.start.called)
228         self.daemon.update_server_wishlist(server_wishlist).get(self.TIMEOUT)
229         self.stop_proxy(self.daemon)
230         self.assertEqual(1, self.node_setup.start.call_count)
231
232     def test_boot_new_node_when_all_nodes_busy(self):
233         size = testutil.MockSize(2)
234         arv_node = testutil.arvados_node_mock(2, job_uuid=True)
235         self.make_daemon([testutil.cloud_node_mock(2, size=size)], [arv_node],
236                          [size], avail_sizes=[(size, {"cores":1})])
237         self.stop_proxy(self.daemon)
238         self.assertTrue(self.node_setup.start.called)
239
240     def test_boot_new_node_below_min_nodes(self):
241         min_size = testutil.MockSize(1)
242         wish_size = testutil.MockSize(3)
243         avail_sizes = [(min_size, {"cores": 1}),
244                        (wish_size, {"cores": 3})]
245         self.make_daemon([], [], None, avail_sizes=avail_sizes, min_nodes=2)
246         self.daemon.update_server_wishlist([wish_size]).get(self.TIMEOUT)
247         self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
248         self.daemon.update_server_wishlist([wish_size]).get(self.TIMEOUT)
249         self.stop_proxy(self.daemon)
250         self.assertEqual([wish_size, min_size],
251                          [call[1].get('cloud_size')
252                           for call in self.node_setup.start.call_args_list])
253
254     def test_no_new_node_when_ge_min_nodes_busy(self):
255         cloud_nodes = [testutil.cloud_node_mock(n) for n in range(1, 4)]
256         arv_nodes = [testutil.arvados_node_mock(n, job_uuid=True)
257                      for n in range(1, 4)]
258         self.make_daemon(cloud_nodes, arv_nodes, [], min_nodes=2)
259         self.stop_proxy(self.daemon)
260         self.assertEqual(0, self.node_setup.start.call_count)
261
262     def test_no_new_node_when_max_nodes_busy(self):
263         self.make_daemon([testutil.cloud_node_mock(3)],
264                          [testutil.arvados_node_mock(3, job_uuid=True)],
265                          [testutil.MockSize(3)],
266                          max_nodes=1)
267         self.stop_proxy(self.daemon)
268         self.assertFalse(self.node_setup.start.called)
269
270     def start_node_boot(self, cloud_node=None, arv_node=None, id_num=1):
271         if cloud_node is None:
272             cloud_node = testutil.cloud_node_mock(id_num)
273         if arv_node is None:
274             arv_node = testutil.arvados_node_mock(id_num)
275         self.make_daemon(want_sizes=[testutil.MockSize(id_num)],
276                          avail_sizes=[(testutil.MockSize(id_num), {"cores":1})])
277         self.daemon.max_nodes.get(self.TIMEOUT)
278         self.assertEqual(1, self.node_setup.start.call_count)
279         self.last_setup.cloud_node.get.return_value = cloud_node
280         self.last_setup.arvados_node.get.return_value = arv_node
281         return self.last_setup
282
283     def test_no_new_node_when_booted_node_not_usable(self):
284         cloud_node = testutil.cloud_node_mock(4)
285         arv_node = testutil.arvados_node_mock(4, crunch_worker_state='down')
286         setup = self.start_node_boot(cloud_node, arv_node)
287         self.daemon.node_up(setup).get(self.TIMEOUT)
288         self.assertEqual(1, self.alive_monitor_count())
289         self.daemon.update_cloud_nodes([cloud_node])
290         self.daemon.update_arvados_nodes([arv_node])
291         self.daemon.update_server_wishlist(
292             [testutil.MockSize(1)]).get(self.TIMEOUT)
293         self.stop_proxy(self.daemon)
294         self.assertEqual(1, self.node_setup.start.call_count)
295
296     def test_no_duplication_when_booting_node_listed_fast(self):
297         # Test that we don't start two ComputeNodeMonitorActors when
298         # we learn about a booting node through a listing before we
299         # get the "node up" message from CloudNodeSetupActor.
300         cloud_node = testutil.cloud_node_mock(1)
301         setup = self.start_node_boot(cloud_node)
302         self.daemon.update_cloud_nodes([cloud_node])
303         self.daemon.node_up(setup).get(self.TIMEOUT)
304         self.assertEqual(1, self.alive_monitor_count())
305
306     def test_no_duplication_when_booted_node_listed(self):
307         cloud_node = testutil.cloud_node_mock(2)
308         setup = self.start_node_boot(cloud_node, id_num=2)
309         self.daemon.node_up(setup)
310         self.daemon.update_cloud_nodes([cloud_node]).get(self.TIMEOUT)
311         self.assertEqual(1, self.alive_monitor_count())
312
313     def test_node_counted_after_boot_with_slow_listing(self):
314         # Test that, after we boot a compute node, we assume it exists
315         # even it doesn't appear in the listing (e.g., because of delays
316         # propagating tags).
317         setup = self.start_node_boot()
318         self.daemon.node_up(setup).get(self.TIMEOUT)
319         self.assertEqual(1, self.alive_monitor_count())
320         self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
321         self.assertEqual(1, self.alive_monitor_count())
322
323     def test_booted_unlisted_node_counted(self):
324         setup = self.start_node_boot(id_num=1)
325         self.daemon.node_up(setup)
326         self.daemon.update_server_wishlist(
327             [testutil.MockSize(1)]).get(self.TIMEOUT)
328         self.stop_proxy(self.daemon)
329         self.assertEqual(1, self.node_setup.start.call_count)
330
331     def test_booted_node_can_shutdown(self):
332         setup = self.start_node_boot()
333         self.daemon.node_up(setup).get(self.TIMEOUT)
334         self.assertEqual(1, self.alive_monitor_count())
335         monitor = self.monitor_list()[0].proxy()
336         self.daemon.update_server_wishlist([])
337         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
338         self.stop_proxy(self.daemon)
339         self.assertTrue(self.node_shutdown.start.called,
340                         "daemon did not shut down booted node on offer")
341
342     def test_booted_node_lifecycle(self):
343         cloud_node = testutil.cloud_node_mock(6)
344         setup = self.start_node_boot(cloud_node, id_num=6)
345         self.daemon.node_up(setup).get(self.TIMEOUT)
346         self.assertEqual(1, self.alive_monitor_count())
347         monitor = self.monitor_list()[0].proxy()
348         self.daemon.update_server_wishlist([])
349         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
350         self.assertShutdownCancellable(True)
351         shutdown = self.node_shutdown.start().proxy()
352         shutdown.cloud_node.get.return_value = cloud_node
353         self.daemon.node_finished_shutdown(shutdown).get(self.TIMEOUT)
354         self.assertTrue(shutdown.stop.called,
355                         "shutdown actor not stopped after finishing")
356         self.assertTrue(monitor.actor_ref.actor_stopped.wait(self.TIMEOUT),
357                         "monitor for booted node not stopped after shutdown")
358         self.daemon.update_server_wishlist(
359             [testutil.MockSize(2)]).get(self.TIMEOUT)
360         self.stop_proxy(self.daemon)
361         self.assertTrue(self.node_setup.start.called,
362                         "second node not started after booted node stopped")
363
364     def test_booted_node_shut_down_when_never_listed(self):
365         setup = self.start_node_boot()
366         self.daemon.node_up(setup).get(self.TIMEOUT)
367         self.assertEqual(1, self.alive_monitor_count())
368         self.assertFalse(self.node_shutdown.start.called)
369         self.timer.deliver()
370         self.stop_proxy(self.daemon)
371         self.assertShutdownCancellable(False)
372
373     def test_booted_node_shut_down_when_never_paired(self):
374         cloud_node = testutil.cloud_node_mock(2)
375         setup = self.start_node_boot(cloud_node)
376         self.daemon.node_up(setup).get(self.TIMEOUT)
377         self.assertEqual(1, self.alive_monitor_count())
378         self.daemon.update_cloud_nodes([cloud_node])
379         self.timer.deliver()
380         self.stop_proxy(self.daemon)
381         self.assertShutdownCancellable(False)
382
383     def test_booted_node_shut_down_when_never_working(self):
384         cloud_node = testutil.cloud_node_mock(4)
385         arv_node = testutil.arvados_node_mock(4, crunch_worker_state='down')
386         setup = self.start_node_boot(cloud_node, arv_node)
387         self.daemon.node_up(setup).get(self.TIMEOUT)
388         self.assertEqual(1, self.alive_monitor_count())
389         self.daemon.update_cloud_nodes([cloud_node])
390         self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
391         self.timer.deliver()
392         self.stop_proxy(self.daemon)
393         self.assertShutdownCancellable(False)
394
395     def test_node_that_pairs_not_considered_failed_boot(self):
396         cloud_node = testutil.cloud_node_mock(3)
397         arv_node = testutil.arvados_node_mock(3)
398         setup = self.start_node_boot(cloud_node, arv_node)
399         self.daemon.node_up(setup).get(self.TIMEOUT)
400         self.assertEqual(1, self.alive_monitor_count())
401         self.daemon.update_cloud_nodes([cloud_node])
402         self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
403         self.timer.deliver()
404         self.stop_proxy(self.daemon)
405         self.assertFalse(self.node_shutdown.start.called)
406
407     def test_node_that_pairs_busy_not_considered_failed_boot(self):
408         cloud_node = testutil.cloud_node_mock(5)
409         arv_node = testutil.arvados_node_mock(5, job_uuid=True)
410         setup = self.start_node_boot(cloud_node, arv_node)
411         self.daemon.node_up(setup).get(self.TIMEOUT)
412         self.assertEqual(1, self.alive_monitor_count())
413         self.daemon.update_cloud_nodes([cloud_node])
414         self.daemon.update_arvados_nodes([arv_node]).get(self.TIMEOUT)
415         self.timer.deliver()
416         self.stop_proxy(self.daemon)
417         self.assertFalse(self.node_shutdown.start.called)
418
419     def test_booting_nodes_shut_down(self):
420         self.make_daemon(want_sizes=[testutil.MockSize(1)])
421         self.daemon.update_server_wishlist([]).get(self.TIMEOUT)
422         self.stop_proxy(self.daemon)
423         self.assertTrue(self.last_setup.stop_if_no_cloud_node.called)
424
425     def test_all_booting_nodes_tried_to_shut_down(self):
426         size = testutil.MockSize(2)
427         self.make_daemon(want_sizes=[size], avail_sizes=[(size, {"cores":1})])
428         self.daemon.max_nodes.get(self.TIMEOUT)
429         setup1 = self.last_setup
430         setup1.stop_if_no_cloud_node().get.return_value = False
431         setup1.stop_if_no_cloud_node.reset_mock()
432         self.daemon.update_server_wishlist([size, size]).get(self.TIMEOUT)
433         self.daemon.max_nodes.get(self.TIMEOUT)
434         self.assertIsNot(setup1, self.last_setup)
435         self.last_setup.stop_if_no_cloud_node().get.return_value = True
436         self.last_setup.stop_if_no_cloud_node.reset_mock()
437         self.daemon.update_server_wishlist([]).get(self.TIMEOUT)
438         self.daemon.max_nodes.get(self.TIMEOUT)
439         self.stop_proxy(self.daemon)
440         self.assertEqual(1, self.last_setup.stop_if_no_cloud_node.call_count)
441         self.assertTrue(setup1.stop_if_no_cloud_node.called)
442
443     def test_shutdown_declined_at_wishlist_capacity(self):
444         cloud_node = testutil.cloud_node_mock(1)
445         size = testutil.MockSize(1)
446         self.make_daemon(cloud_nodes=[cloud_node], want_sizes=[size])
447         self.assertEqual(1, self.alive_monitor_count())
448         monitor = self.monitor_list()[0].proxy()
449         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
450         self.stop_proxy(self.daemon)
451         self.assertFalse(self.node_shutdown.start.called)
452
453     def test_shutdown_declined_below_min_nodes(self):
454         cloud_node = testutil.cloud_node_mock(1)
455         self.make_daemon(cloud_nodes=[cloud_node], min_nodes=1)
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.assertFalse(self.node_shutdown.start.called)
461
462     def test_shutdown_accepted_below_capacity(self):
463         self.make_daemon(cloud_nodes=[testutil.cloud_node_mock()])
464         self.assertEqual(1, self.alive_monitor_count())
465         monitor = self.monitor_list()[0].proxy()
466         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
467         self.stop_proxy(self.daemon)
468         self.assertTrue(self.node_shutdown.start.called)
469
470     def test_shutdown_declined_when_idle_and_job_queued(self):
471         cloud_nodes = [testutil.cloud_node_mock(n) for n in [3, 4]]
472         arv_nodes = [testutil.arvados_node_mock(3, job_uuid=True),
473                      testutil.arvados_node_mock(4, job_uuid=None)]
474         self.make_daemon(cloud_nodes, arv_nodes, [testutil.MockSize(1)])
475         self.assertEqual(2, self.alive_monitor_count())
476         for mon_ref in self.monitor_list():
477             monitor = mon_ref.proxy()
478             if monitor.cloud_node.get(self.TIMEOUT) is cloud_nodes[-1]:
479                 break
480         else:
481             self.fail("monitor for idle node not found")
482         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
483         self.stop_proxy(self.daemon)
484         self.assertFalse(self.node_shutdown.start.called)
485
486     def test_node_shutdown_after_cancelled_shutdown(self):
487         cloud_node = testutil.cloud_node_mock(5)
488         self.make_daemon([cloud_node], [testutil.arvados_node_mock(5)])
489         self.assertEqual(1, self.alive_monitor_count())
490         monitor = self.monitor_list()[0].proxy()
491         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
492         self.last_shutdown.success.get.return_value = False
493         self.daemon.node_finished_shutdown(self.last_shutdown).get(self.TIMEOUT)
494         self.assertEqual(1, self.alive_monitor_count())
495
496         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
497         self.last_shutdown.success.get.return_value = True
498         self.last_shutdown.stop.side_effect = lambda: monitor.stop()
499         self.daemon.node_finished_shutdown(self.last_shutdown).get(self.TIMEOUT)
500         self.assertEqual(0, self.alive_monitor_count())
501
502     def test_broken_node_blackholed_after_cancelled_shutdown(self):
503         size = testutil.MockSize(8)
504         cloud_node = testutil.cloud_node_mock(8, size=size)
505         wishlist = [size]
506         self.make_daemon([cloud_node], [testutil.arvados_node_mock(8)],
507                          wishlist, avail_sizes=[(size, {"cores":1})])
508         self.assertEqual(1, self.alive_monitor_count())
509         self.assertFalse(self.node_setup.start.called)
510         monitor = self.monitor_list()[0].proxy()
511         shutdown_proxy = self.node_shutdown.start().proxy
512         shutdown_proxy().cloud_node.get.return_value = cloud_node
513         shutdown_proxy().success.get.return_value = False
514         shutdown_proxy().cancel_reason.get.return_value = self.node_shutdown.NODE_BROKEN
515         self.daemon.update_server_wishlist([]).get(self.TIMEOUT)
516         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
517         self.daemon.node_finished_shutdown(shutdown_proxy()).get(self.TIMEOUT)
518         self.daemon.update_cloud_nodes([cloud_node]).get(self.TIMEOUT)
519         self.daemon.update_server_wishlist(wishlist).get(self.TIMEOUT)
520         self.stop_proxy(self.daemon)
521         self.assertEqual(1, self.node_setup.start.call_count)
522
523     def test_nodes_shutting_down_replaced_below_max_nodes(self):
524         size = testutil.MockSize(6)
525         cloud_node = testutil.cloud_node_mock(6, size=size)
526         self.make_daemon([cloud_node], [testutil.arvados_node_mock(6)],
527                          avail_sizes=[(size, {"cores":1})])
528         self.assertEqual(1, self.alive_monitor_count())
529         monitor = self.monitor_list()[0].proxy()
530         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
531         self.assertTrue(self.node_shutdown.start.called)
532         self.daemon.update_server_wishlist(
533             [testutil.MockSize(6)]).get(self.TIMEOUT)
534         self.stop_proxy(self.daemon)
535         self.assertTrue(self.node_setup.start.called)
536
537     def test_nodes_shutting_down_not_replaced_at_max_nodes(self):
538         cloud_node = testutil.cloud_node_mock(7)
539         self.make_daemon([cloud_node], [testutil.arvados_node_mock(7)],
540                          max_nodes=1)
541         self.assertEqual(1, self.alive_monitor_count())
542         monitor = self.monitor_list()[0].proxy()
543         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
544         self.assertTrue(self.node_shutdown.start.called)
545         self.daemon.update_server_wishlist(
546             [testutil.MockSize(7)]).get(self.TIMEOUT)
547         self.stop_proxy(self.daemon)
548         self.assertFalse(self.node_setup.start.called)
549
550     def test_nodes_shutting_down_count_against_excess(self):
551         size = testutil.MockSize(8)
552         cloud_nodes = [testutil.cloud_node_mock(n, size=size) for n in [8, 9]]
553         arv_nodes = [testutil.arvados_node_mock(n, size=size) for n in [8, 9]]
554         self.make_daemon(cloud_nodes, arv_nodes, [size],
555                          avail_sizes=[(size, {"cores":1})])
556         self.assertEqual(2, self.alive_monitor_count())
557         for mon_ref in self.monitor_list():
558             self.daemon.node_can_shutdown(mon_ref.proxy()).get(self.TIMEOUT)
559         self.assertEqual(1, self.node_shutdown.start.call_count)
560
561     def test_clean_shutdown_waits_for_node_setup_finish(self):
562         new_node = self.start_node_boot()
563         new_node.stop_if_no_cloud_node().get.return_value = False
564         new_node.stop_if_no_cloud_node.reset_mock()
565         self.daemon.shutdown().get(self.TIMEOUT)
566         self.assertTrue(new_node.stop_if_no_cloud_node.called)
567         self.daemon.node_up(new_node).get(self.TIMEOUT)
568         self.assertTrue(new_node.stop.called)
569         self.timer.deliver()
570         self.assertTrue(
571             self.daemon.actor_ref.actor_stopped.wait(self.TIMEOUT))
572
573     def test_wishlist_ignored_after_shutdown(self):
574         new_node = self.start_node_boot()
575         new_node.stop_if_no_cloud_node().get.return_value = False
576         new_node.stop_if_no_cloud_node.reset_mock()
577         self.daemon.shutdown().get(self.TIMEOUT)
578         size = testutil.MockSize(2)
579         self.daemon.update_server_wishlist([size] * 2).get(self.TIMEOUT)
580         self.timer.deliver()
581         self.stop_proxy(self.daemon)
582         self.assertEqual(1, self.node_setup.start.call_count)
583
584     def test_shutdown_actor_stopped_when_cloud_node_delisted(self):
585         self.make_daemon(cloud_nodes=[testutil.cloud_node_mock()])
586         self.assertEqual(1, self.alive_monitor_count())
587         monitor = self.monitor_list()[0].proxy()
588         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
589         self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
590         self.stop_proxy(self.daemon)
591         self.assertEqual(
592             1, self.last_shutdown.stop.call_count)
593
594     def test_shutdown_actor_cleanup_copes_with_dead_actors(self):
595         self.make_daemon(cloud_nodes=[testutil.cloud_node_mock()])
596         self.assertEqual(1, self.alive_monitor_count())
597         monitor = self.monitor_list()[0].proxy()
598         self.daemon.node_can_shutdown(monitor).get(self.TIMEOUT)
599         # We're mainly testing that update_cloud_nodes catches and handles
600         # the ActorDeadError.
601         self.last_shutdown.stop.side_effect = pykka.ActorDeadError
602         self.daemon.update_cloud_nodes([]).get(self.TIMEOUT)
603         self.stop_proxy(self.daemon)
604         self.assertEqual(1, self.last_shutdown.stop.call_count)
605
606     def busywait(self, f):
607         n = 0
608         while not f() and n < 10:
609             time.sleep(.1)
610             n += 1
611         self.assertTrue(f())
612
613     def test_node_create_two_sizes(self):
614         small = testutil.MockSize(1)
615         big = testutil.MockSize(2)
616         avail_sizes = [(testutil.MockSize(1), {"cores":1}),
617                         (testutil.MockSize(2), {"cores":2})]
618         self.make_daemon(want_sizes=[small, small, small, big],
619                          avail_sizes=avail_sizes, max_nodes=4)
620
621         # the daemon runs in another thread, so we need to wait and see
622         # if it does all the work we're expecting it to do before stopping it.
623         self.busywait(lambda: self.node_setup.start.call_count == 4)
624         booting = self.daemon.booting.get(self.TIMEOUT)
625         self.stop_proxy(self.daemon)
626         sizecounts = {a[0].id: 0 for a in avail_sizes}
627         for b in booting.itervalues():
628             sizecounts[b.cloud_size.get().id] += 1
629         logging.info(sizecounts)
630         self.assertEqual(3, sizecounts[small.id])
631         self.assertEqual(1, sizecounts[big.id])
632
633     def test_node_max_nodes_two_sizes(self):
634         small = testutil.MockSize(1)
635         big = testutil.MockSize(2)
636         avail_sizes = [(testutil.MockSize(1), {"cores":1}),
637                         (testutil.MockSize(2), {"cores":2})]
638         self.make_daemon(want_sizes=[small, small, small, big],
639                          avail_sizes=avail_sizes, max_nodes=3)
640
641         # the daemon runs in another thread, so we need to wait and see
642         # if it does all the work we're expecting it to do before stopping it.
643         self.busywait(lambda: self.node_setup.start.call_count == 3)
644         booting = self.daemon.booting.get(self.TIMEOUT)
645         self.stop_proxy(self.daemon)
646         sizecounts = {a[0].id: 0 for a in avail_sizes}
647         for b in booting.itervalues():
648             sizecounts[b.cloud_size.get().id] += 1
649         self.assertEqual(2, sizecounts[small.id])
650         self.assertEqual(1, sizecounts[big.id])
651
652     def test_wishlist_reconfigure(self):
653         small = testutil.MockSize(1)
654         big = testutil.MockSize(2)
655         avail_sizes = [(small, {"cores":1}), (big, {"cores":2})]
656
657         self.make_daemon(cloud_nodes=[testutil.cloud_node_mock(1, small),
658                                       testutil.cloud_node_mock(2, small),
659                                       testutil.cloud_node_mock(3, big)],
660                          arvados_nodes=[testutil.arvados_node_mock(1),
661                                         testutil.arvados_node_mock(2),
662                                         testutil.arvados_node_mock(3)],
663                          want_sizes=[small, small, big],
664                          avail_sizes=avail_sizes)
665
666         self.daemon.update_server_wishlist([small, big, big]).get(self.TIMEOUT)
667
668         self.assertEqual(0, self.node_shutdown.start.call_count)
669
670         for c in self.daemon.cloud_nodes.get().nodes.itervalues():
671             self.daemon.node_can_shutdown(c.actor)
672
673         booting = self.daemon.booting.get()
674         shutdowns = self.daemon.shutdowns.get()
675
676         self.stop_proxy(self.daemon)
677
678         self.assertEqual(1, self.node_setup.start.call_count)
679         self.assertEqual(1, self.node_shutdown.start.call_count)
680
681         # booting a new big node
682         sizecounts = {a[0].id: 0 for a in avail_sizes}
683         for b in booting.itervalues():
684             sizecounts[b.cloud_size.get().id] += 1
685         self.assertEqual(0, sizecounts[small.id])
686         self.assertEqual(1, sizecounts[big.id])
687
688         # shutting down a small node
689         sizecounts = {a[0].id: 0 for a in avail_sizes}
690         for b in shutdowns.itervalues():
691             sizecounts[b.cloud_node.get().size.id] += 1
692         self.assertEqual(1, sizecounts[small.id])
693         self.assertEqual(0, sizecounts[big.id])
694
695     def test_node_max_price(self):
696         small = testutil.MockSize(1)
697         big = testutil.MockSize(2)
698         avail_sizes = [(testutil.MockSize(1), {"cores":1, "price":1}),
699                         (testutil.MockSize(2), {"cores":2, "price":2})]
700         self.make_daemon(want_sizes=[small, small, small, big],
701                          avail_sizes=avail_sizes,
702                          max_nodes=4,
703                          max_total_price=4)
704         # the daemon runs in another thread, so we need to wait and see
705         # if it does all the work we're expecting it to do before stopping it.
706         self.busywait(lambda: self.node_setup.start.call_count == 3)
707         booting = self.daemon.booting.get()
708         self.stop_proxy(self.daemon)
709
710         sizecounts = {a[0].id: 0 for a in avail_sizes}
711         for b in booting.itervalues():
712             sizecounts[b.cloud_size.get().id] += 1
713         logging.info(sizecounts)
714
715         # Booting 3 small nodes and not booting a big node would also partially
716         # satisfy the wishlist and come in under the price cap, however the way
717         # the update_server_wishlist() currently works effectively results in a
718         # round-robin creation of one node of each size in the wishlist, so
719         # test for that.
720         self.assertEqual(2, sizecounts[small.id])
721         self.assertEqual(1, sizecounts[big.id])