Merge branch 'master' into 3198-writable-fuse
[arvados.git] / services / nodemanager / tests / test_computenode_dispatch.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 arvados.errors as arverror
9 import httplib2
10 import mock
11 import pykka
12
13 import arvnodeman.computenode.dispatch as dispatch
14 from . import testutil
15
16 class ComputeNodeSetupActorTestCase(testutil.ActorTestMixin, unittest.TestCase):
17     def make_mocks(self, arvados_effect=None):
18         if arvados_effect is None:
19             arvados_effect = [testutil.arvados_node_mock()]
20         self.arvados_effect = arvados_effect
21         self.timer = testutil.MockTimer()
22         self.api_client = mock.MagicMock(name='api_client')
23         self.api_client.nodes().create().execute.side_effect = arvados_effect
24         self.api_client.nodes().update().execute.side_effect = arvados_effect
25         self.cloud_client = mock.MagicMock(name='cloud_client')
26         self.cloud_client.create_node.return_value = testutil.cloud_node_mock(1)
27
28     def make_actor(self, arv_node=None):
29         if not hasattr(self, 'timer'):
30             self.make_mocks(arvados_effect=[arv_node])
31         self.setup_actor = dispatch.ComputeNodeSetupActor.start(
32             self.timer, self.api_client, self.cloud_client,
33             testutil.MockSize(1), arv_node).proxy()
34
35     def test_creation_without_arvados_node(self):
36         self.make_actor()
37         self.assertEqual(self.arvados_effect[-1],
38                          self.setup_actor.arvados_node.get(self.TIMEOUT))
39         self.assertTrue(self.api_client.nodes().create().execute.called)
40         self.assertEqual(self.cloud_client.create_node(),
41                          self.setup_actor.cloud_node.get(self.TIMEOUT))
42
43     def test_creation_with_arvados_node(self):
44         self.make_actor(testutil.arvados_node_mock())
45         self.assertEqual(self.arvados_effect[-1],
46                          self.setup_actor.arvados_node.get(self.TIMEOUT))
47         self.assertTrue(self.api_client.nodes().update().execute.called)
48         self.assertEqual(self.cloud_client.create_node(),
49                          self.setup_actor.cloud_node.get(self.TIMEOUT))
50
51     def test_failed_arvados_calls_retried(self):
52         self.make_mocks([
53                 arverror.ApiError(httplib2.Response({'status': '500'}), ""),
54                 testutil.arvados_node_mock(),
55                 ])
56         self.make_actor()
57         self.wait_for_assignment(self.setup_actor, 'arvados_node')
58
59     def test_failed_cloud_calls_retried(self):
60         self.make_mocks()
61         self.cloud_client.create_node.side_effect = [
62             Exception("test cloud creation error"),
63             self.cloud_client.create_node.return_value,
64             ]
65         self.make_actor()
66         self.wait_for_assignment(self.setup_actor, 'cloud_node')
67
68     def test_failed_post_create_retried(self):
69         self.make_mocks()
70         self.cloud_client.post_create_node.side_effect = [
71             Exception("test cloud post-create error"), None]
72         self.make_actor()
73         done = self.FUTURE_CLASS()
74         self.setup_actor.subscribe(done.set)
75         done.get(self.TIMEOUT)
76         self.assertEqual(2, self.cloud_client.post_create_node.call_count)
77
78     def test_stop_when_no_cloud_node(self):
79         self.make_mocks(
80             arverror.ApiError(httplib2.Response({'status': '500'}), ""))
81         self.make_actor()
82         self.assertTrue(
83             self.setup_actor.stop_if_no_cloud_node().get(self.TIMEOUT))
84         self.assertTrue(
85             self.setup_actor.actor_ref.actor_stopped.wait(self.TIMEOUT))
86
87     def test_no_stop_when_cloud_node(self):
88         self.make_actor()
89         self.wait_for_assignment(self.setup_actor, 'cloud_node')
90         self.assertFalse(
91             self.setup_actor.stop_if_no_cloud_node().get(self.TIMEOUT))
92         self.assertTrue(self.stop_proxy(self.setup_actor),
93                         "actor was stopped by stop_if_no_cloud_node")
94
95     def test_subscribe(self):
96         self.make_mocks(
97             arverror.ApiError(httplib2.Response({'status': '500'}), ""))
98         self.make_actor()
99         subscriber = mock.Mock(name='subscriber_mock')
100         self.setup_actor.subscribe(subscriber)
101         self.api_client.nodes().create().execute.side_effect = [
102             testutil.arvados_node_mock()]
103         self.wait_for_assignment(self.setup_actor, 'cloud_node')
104         self.assertEqual(self.setup_actor.actor_ref.actor_urn,
105                          subscriber.call_args[0][0].actor_ref.actor_urn)
106
107     def test_late_subscribe(self):
108         self.make_actor()
109         subscriber = mock.Mock(name='subscriber_mock')
110         self.wait_for_assignment(self.setup_actor, 'cloud_node')
111         self.setup_actor.subscribe(subscriber).get(self.TIMEOUT)
112         self.stop_proxy(self.setup_actor)
113         self.assertEqual(self.setup_actor.actor_ref.actor_urn,
114                          subscriber.call_args[0][0].actor_ref.actor_urn)
115
116
117 class ComputeNodeShutdownActorMixin(testutil.ActorTestMixin):
118     def make_mocks(self, cloud_node=None, arvados_node=None,
119                    shutdown_open=True):
120         self.timer = testutil.MockTimer()
121         self.shutdowns = testutil.MockShutdownTimer()
122         self.shutdowns._set_state(shutdown_open, 300)
123         self.cloud_client = mock.MagicMock(name='cloud_client')
124         self.arvados_client = mock.MagicMock(name='arvados_client')
125         self.updates = mock.MagicMock(name='update_mock')
126         if cloud_node is None:
127             cloud_node = testutil.cloud_node_mock()
128         self.cloud_node = cloud_node
129         self.arvados_node = arvados_node
130
131     def make_actor(self, cancellable=True):
132         if not hasattr(self, 'timer'):
133             self.make_mocks()
134         monitor_actor = dispatch.ComputeNodeMonitorActor.start(
135             self.cloud_node, time.time(), self.shutdowns,
136             testutil.cloud_node_fqdn, self.timer, self.updates,
137             self.arvados_node)
138         self.shutdown_actor = self.ACTOR_CLASS.start(
139             self.timer, self.cloud_client, self.arvados_client, monitor_actor,
140             cancellable).proxy()
141         self.monitor_actor = monitor_actor.proxy()
142
143     def check_success_flag(self, expected, allow_msg_count=1):
144         # allow_msg_count is the number of internal messages that may
145         # need to be handled for shutdown to finish.
146         for try_num in range(1 + allow_msg_count):
147             last_flag = self.shutdown_actor.success.get(self.TIMEOUT)
148             if last_flag is expected:
149                 break
150         else:
151             self.fail("success flag {} is not {}".format(last_flag, expected))
152
153     def test_uncancellable_shutdown(self, *mocks):
154         self.make_mocks(shutdown_open=False)
155         self.cloud_client.destroy_node.return_value = False
156         self.make_actor(cancellable=False)
157         self.check_success_flag(None, 0)
158         self.shutdowns._set_state(True, 600)
159         self.cloud_client.destroy_node.return_value = True
160         self.check_success_flag(True)
161
162     def test_arvados_node_cleaned_after_shutdown(self, *mocks):
163         cloud_node = testutil.cloud_node_mock(62)
164         arv_node = testutil.arvados_node_mock(62)
165         self.make_mocks(cloud_node, arv_node)
166         self.make_actor()
167         self.check_success_flag(True, 3)
168         update_mock = self.arvados_client.nodes().update
169         self.assertTrue(update_mock.called)
170         update_kwargs = update_mock.call_args_list[0][1]
171         self.assertEqual(arv_node['uuid'], update_kwargs.get('uuid'))
172         self.assertIn('body', update_kwargs)
173         for clear_key in ['slot_number', 'hostname', 'ip_address',
174                           'first_ping_at', 'last_ping_at']:
175             self.assertIn(clear_key, update_kwargs['body'])
176             self.assertIsNone(update_kwargs['body'][clear_key])
177         self.assertTrue(update_mock().execute.called)
178
179     def test_arvados_node_not_cleaned_after_shutdown_cancelled(self, *mocks):
180         cloud_node = testutil.cloud_node_mock(61)
181         arv_node = testutil.arvados_node_mock(61)
182         self.make_mocks(cloud_node, arv_node, shutdown_open=False)
183         self.make_actor(cancellable=True)
184         self.check_success_flag(False, 2)
185         self.assertFalse(self.arvados_client.nodes().update.called)
186
187
188 class ComputeNodeShutdownActorTestCase(ComputeNodeShutdownActorMixin,
189                                        unittest.TestCase):
190     ACTOR_CLASS = dispatch.ComputeNodeShutdownActor
191
192     def test_easy_shutdown(self):
193         self.make_actor()
194         self.check_success_flag(True)
195         self.assertTrue(self.cloud_client.destroy_node.called)
196
197     def test_shutdown_cancelled_when_window_closes(self):
198         self.make_mocks(shutdown_open=False)
199         self.make_actor()
200         self.check_success_flag(False, 2)
201         self.assertFalse(self.cloud_client.destroy_node.called)
202
203     def test_shutdown_retries_when_cloud_fails(self):
204         self.make_mocks()
205         self.cloud_client.destroy_node.return_value = False
206         self.make_actor()
207         self.assertIsNone(self.shutdown_actor.success.get(self.TIMEOUT))
208         self.cloud_client.destroy_node.return_value = True
209         self.check_success_flag(True)
210
211     def test_late_subscribe(self):
212         self.make_actor()
213         subscriber = mock.Mock(name='subscriber_mock')
214         self.shutdown_actor.subscribe(subscriber).get(self.TIMEOUT)
215         self.stop_proxy(self.shutdown_actor)
216         self.assertTrue(subscriber.called)
217         self.assertEqual(self.shutdown_actor.actor_ref.actor_urn,
218                          subscriber.call_args[0][0].actor_ref.actor_urn)
219
220
221 class ComputeNodeUpdateActorTestCase(testutil.ActorTestMixin,
222                                      unittest.TestCase):
223     def make_actor(self):
224         self.driver = mock.MagicMock(name='driver_mock')
225         self.updater = dispatch.ComputeNodeUpdateActor.start(self.driver).proxy()
226
227     def test_node_sync(self):
228         self.make_actor()
229         cloud_node = testutil.cloud_node_mock()
230         arv_node = testutil.arvados_node_mock()
231         self.updater.sync_node(cloud_node, arv_node).get(self.TIMEOUT)
232         self.driver().sync_node.assert_called_with(cloud_node, arv_node)
233
234
235 class ComputeNodeMonitorActorTestCase(testutil.ActorTestMixin,
236                                       unittest.TestCase):
237     def make_mocks(self, node_num):
238         self.shutdowns = testutil.MockShutdownTimer()
239         self.shutdowns._set_state(False, 300)
240         self.timer = mock.MagicMock(name='timer_mock')
241         self.updates = mock.MagicMock(name='update_mock')
242         self.cloud_mock = testutil.cloud_node_mock(node_num)
243         self.subscriber = mock.Mock(name='subscriber_mock')
244
245     def make_actor(self, node_num=1, arv_node=None, start_time=None):
246         if not hasattr(self, 'cloud_mock'):
247             self.make_mocks(node_num)
248         if start_time is None:
249             start_time = time.time()
250         self.node_actor = dispatch.ComputeNodeMonitorActor.start(
251             self.cloud_mock, start_time, self.shutdowns,
252             testutil.cloud_node_fqdn, self.timer, self.updates,
253             arv_node).proxy()
254         self.node_actor.subscribe(self.subscriber).get(self.TIMEOUT)
255
256     def node_state(self, *states):
257         return self.node_actor.in_state(*states).get(self.TIMEOUT)
258
259     def test_in_state_when_unpaired(self):
260         self.make_actor()
261         self.assertIsNone(self.node_state('idle', 'busy'))
262
263     def test_in_state_when_pairing_stale(self):
264         self.make_actor(arv_node=testutil.arvados_node_mock(
265                 job_uuid=None, age=90000))
266         self.assertIsNone(self.node_state('idle', 'busy'))
267
268     def test_in_state_when_no_state_available(self):
269         self.make_actor(arv_node=testutil.arvados_node_mock(
270                 crunch_worker_state=None))
271         self.assertIsNone(self.node_state('idle', 'busy'))
272
273     def test_in_idle_state(self):
274         self.make_actor(2, arv_node=testutil.arvados_node_mock(job_uuid=None))
275         self.assertTrue(self.node_state('idle'))
276         self.assertFalse(self.node_state('busy'))
277         self.assertTrue(self.node_state('idle', 'busy'))
278
279     def test_in_busy_state(self):
280         self.make_actor(3, arv_node=testutil.arvados_node_mock(job_uuid=True))
281         self.assertFalse(self.node_state('idle'))
282         self.assertTrue(self.node_state('busy'))
283         self.assertTrue(self.node_state('idle', 'busy'))
284
285     def test_init_shutdown_scheduling(self):
286         self.make_actor()
287         self.assertTrue(self.timer.schedule.called)
288         self.assertEqual(300, self.timer.schedule.call_args[0][0])
289
290     def test_shutdown_window_close_scheduling(self):
291         self.make_actor()
292         self.shutdowns._set_state(False, 600)
293         self.timer.schedule.reset_mock()
294         self.node_actor.consider_shutdown().get(self.TIMEOUT)
295         self.stop_proxy(self.node_actor)
296         self.assertTrue(self.timer.schedule.called)
297         self.assertEqual(600, self.timer.schedule.call_args[0][0])
298         self.assertFalse(self.subscriber.called)
299
300     def test_shutdown_subscription(self):
301         self.make_actor()
302         self.shutdowns._set_state(True, 600)
303         self.node_actor.consider_shutdown().get(self.TIMEOUT)
304         self.assertTrue(self.subscriber.called)
305         self.assertEqual(self.node_actor.actor_ref.actor_urn,
306                          self.subscriber.call_args[0][0].actor_ref.actor_urn)
307
308     def test_shutdown_without_arvados_node(self):
309         self.make_actor()
310         self.shutdowns._set_state(True, 600)
311         self.assertTrue(self.node_actor.shutdown_eligible().get(self.TIMEOUT))
312
313     def test_no_shutdown_without_arvados_node_and_old_cloud_node(self):
314         self.make_actor(start_time=0)
315         self.shutdowns._set_state(True, 600)
316         self.assertFalse(self.node_actor.shutdown_eligible().get(self.TIMEOUT))
317
318     def test_no_shutdown_when_window_closed(self):
319         self.make_actor(3, testutil.arvados_node_mock(3, job_uuid=None))
320         self.assertFalse(self.node_actor.shutdown_eligible().get(self.TIMEOUT))
321
322     def test_no_shutdown_when_node_running_job(self):
323         self.make_actor(4, testutil.arvados_node_mock(4, job_uuid=True))
324         self.shutdowns._set_state(True, 600)
325         self.assertFalse(self.node_actor.shutdown_eligible().get(self.TIMEOUT))
326
327     def test_no_shutdown_when_node_state_unknown(self):
328         self.make_actor(5, testutil.arvados_node_mock(
329             5, crunch_worker_state=None))
330         self.shutdowns._set_state(True, 600)
331         self.assertFalse(self.node_actor.shutdown_eligible().get(self.TIMEOUT))
332
333     def test_no_shutdown_when_node_state_stale(self):
334         self.make_actor(6, testutil.arvados_node_mock(6, age=90000))
335         self.shutdowns._set_state(True, 600)
336         self.assertFalse(self.node_actor.shutdown_eligible().get(self.TIMEOUT))
337
338     def test_arvados_node_match(self):
339         self.make_actor(2)
340         arv_node = testutil.arvados_node_mock(
341             2, hostname='compute-two.zzzzz.arvadosapi.com')
342         pair_id = self.node_actor.offer_arvados_pair(arv_node).get(self.TIMEOUT)
343         self.assertEqual(self.cloud_mock.id, pair_id)
344         self.stop_proxy(self.node_actor)
345         self.updates.sync_node.assert_called_with(self.cloud_mock, arv_node)
346
347     def test_arvados_node_mismatch(self):
348         self.make_actor(3)
349         arv_node = testutil.arvados_node_mock(1)
350         self.assertIsNone(
351             self.node_actor.offer_arvados_pair(arv_node).get(self.TIMEOUT))
352
353     def test_arvados_node_mismatch_first_ping_too_early(self):
354         self.make_actor(4)
355         arv_node = testutil.arvados_node_mock(
356             4, first_ping_at='1971-03-02T14:15:16.1717282Z')
357         self.assertIsNone(
358             self.node_actor.offer_arvados_pair(arv_node).get(self.TIMEOUT))
359
360     def test_update_cloud_node(self):
361         self.make_actor(1)
362         self.make_mocks(2)
363         self.cloud_mock.id = '1'
364         self.node_actor.update_cloud_node(self.cloud_mock)
365         current_cloud = self.node_actor.cloud_node.get(self.TIMEOUT)
366         self.assertEqual([testutil.ip_address_mock(2)],
367                          current_cloud.private_ips)
368
369     def test_missing_cloud_node_update(self):
370         self.make_actor(1)
371         self.node_actor.update_cloud_node(None)
372         current_cloud = self.node_actor.cloud_node.get(self.TIMEOUT)
373         self.assertEqual([testutil.ip_address_mock(1)],
374                          current_cloud.private_ips)
375
376     def test_update_arvados_node(self):
377         self.make_actor(3)
378         job_uuid = 'zzzzz-jjjjj-updatejobnode00'
379         new_arvados = testutil.arvados_node_mock(3, job_uuid)
380         self.node_actor.update_arvados_node(new_arvados)
381         current_arvados = self.node_actor.arvados_node.get(self.TIMEOUT)
382         self.assertEqual(job_uuid, current_arvados['job_uuid'])
383
384     def test_missing_arvados_node_update(self):
385         self.make_actor(4, testutil.arvados_node_mock(4))
386         self.node_actor.update_arvados_node(None)
387         current_arvados = self.node_actor.arvados_node.get(self.TIMEOUT)
388         self.assertEqual(testutil.ip_address_mock(4),
389                          current_arvados['ip_address'])
390
391     def test_update_arvados_node_syncs_when_fqdn_mismatch(self):
392         self.make_mocks(5)
393         self.cloud_mock.extra['testname'] = 'cloudfqdn.zzzzz.arvadosapi.com'
394         self.make_actor()
395         arv_node = testutil.arvados_node_mock(5)
396         self.node_actor.update_arvados_node(arv_node).get(self.TIMEOUT)
397         self.assertEqual(1, self.updates.sync_node.call_count)
398
399     def test_update_arvados_node_skips_sync_when_fqdn_match(self):
400         self.make_mocks(6)
401         arv_node = testutil.arvados_node_mock(6)
402         self.cloud_mock.extra['testname'] ='{n[hostname]}.{n[domain]}'.format(
403             n=arv_node)
404         self.make_actor()
405         self.node_actor.update_arvados_node(arv_node).get(self.TIMEOUT)
406         self.assertEqual(0, self.updates.sync_node.call_count)