Merge branch 'master' into 4486-document-arv
[arvados.git] / services / nodemanager / tests / test_computenode.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 as cnode
14 from . import testutil
15
16 class ComputeNodeSetupActorTestCase(testutil.ActorTestMixin, unittest.TestCase):
17     def make_mocks(self, arvados_effect=None, cloud_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 = cnode.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_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, 'cloud_node')
58
59     def test_stop_when_no_cloud_node(self):
60         self.make_mocks(
61             arverror.ApiError(httplib2.Response({'status': '500'}), ""))
62         self.make_actor()
63         self.setup_actor.stop_if_no_cloud_node()
64         self.assertTrue(
65             self.setup_actor.actor_ref.actor_stopped.wait(self.TIMEOUT))
66
67     def test_no_stop_when_cloud_node(self):
68         self.make_actor()
69         self.wait_for_assignment(self.setup_actor, 'cloud_node')
70         self.setup_actor.stop_if_no_cloud_node().get(self.TIMEOUT)
71         self.assertTrue(self.stop_proxy(self.setup_actor),
72                         "actor was stopped by stop_if_no_cloud_node")
73
74     def test_subscribe(self):
75         self.make_mocks(
76             arverror.ApiError(httplib2.Response({'status': '500'}), ""))
77         self.make_actor()
78         subscriber = mock.Mock(name='subscriber_mock')
79         self.setup_actor.subscribe(subscriber)
80         self.api_client.nodes().create().execute.side_effect = [
81             testutil.arvados_node_mock()]
82         self.wait_for_assignment(self.setup_actor, 'cloud_node')
83         self.assertEqual(self.setup_actor.actor_ref.actor_urn,
84                          subscriber.call_args[0][0].actor_ref.actor_urn)
85
86     def test_late_subscribe(self):
87         self.make_actor()
88         subscriber = mock.Mock(name='subscriber_mock')
89         self.wait_for_assignment(self.setup_actor, 'cloud_node')
90         self.setup_actor.subscribe(subscriber).get(self.TIMEOUT)
91         self.stop_proxy(self.setup_actor)
92         self.assertEqual(self.setup_actor.actor_ref.actor_urn,
93                          subscriber.call_args[0][0].actor_ref.actor_urn)
94
95
96 class ComputeNodeShutdownActorTestCase(testutil.ActorTestMixin,
97                                        unittest.TestCase):
98     def make_mocks(self, cloud_node=None):
99         self.timer = testutil.MockTimer()
100         self.cloud_client = mock.MagicMock(name='cloud_client')
101         if cloud_node is None:
102             cloud_node = testutil.cloud_node_mock()
103         self.cloud_node = cloud_node
104
105     def make_actor(self, arv_node=None):
106         if not hasattr(self, 'timer'):
107             self.make_mocks()
108         self.shutdown_actor = cnode.ComputeNodeShutdownActor.start(
109             self.timer, self.cloud_client, self.cloud_node).proxy()
110
111     def test_easy_shutdown(self):
112         self.make_actor()
113         self.shutdown_actor.cloud_node.get(self.TIMEOUT)
114         self.stop_proxy(self.shutdown_actor)
115         self.assertTrue(self.cloud_client.destroy_node.called)
116
117     def test_late_subscribe(self):
118         self.make_actor()
119         subscriber = mock.Mock(name='subscriber_mock')
120         self.shutdown_actor.subscribe(subscriber).get(self.TIMEOUT)
121         self.stop_proxy(self.shutdown_actor)
122         self.assertEqual(self.shutdown_actor.actor_ref.actor_urn,
123                          subscriber.call_args[0][0].actor_ref.actor_urn)
124
125
126 class ComputeNodeUpdateActorTestCase(testutil.ActorTestMixin,
127                                      unittest.TestCase):
128     def make_actor(self):
129         self.driver = mock.MagicMock(name='driver_mock')
130         self.updater = cnode.ComputeNodeUpdateActor.start(self.driver).proxy()
131
132     def test_node_sync(self):
133         self.make_actor()
134         cloud_node = testutil.cloud_node_mock()
135         arv_node = testutil.arvados_node_mock()
136         self.updater.sync_node(cloud_node, arv_node).get(self.TIMEOUT)
137         self.driver().sync_node.assert_called_with(cloud_node, arv_node)
138
139
140 @mock.patch('time.time', return_value=1)
141 class ShutdownTimerTestCase(unittest.TestCase):
142     def test_two_length_window(self, time_mock):
143         timer = cnode.ShutdownTimer(time_mock.return_value, [8, 2])
144         self.assertEqual(481, timer.next_opening())
145         self.assertFalse(timer.window_open())
146         time_mock.return_value += 500
147         self.assertEqual(1081, timer.next_opening())
148         self.assertTrue(timer.window_open())
149         time_mock.return_value += 200
150         self.assertEqual(1081, timer.next_opening())
151         self.assertFalse(timer.window_open())
152
153     def test_three_length_window(self, time_mock):
154         timer = cnode.ShutdownTimer(time_mock.return_value, [6, 3, 1])
155         self.assertEqual(361, timer.next_opening())
156         self.assertFalse(timer.window_open())
157         time_mock.return_value += 400
158         self.assertEqual(961, timer.next_opening())
159         self.assertTrue(timer.window_open())
160         time_mock.return_value += 200
161         self.assertEqual(961, timer.next_opening())
162         self.assertFalse(timer.window_open())
163
164
165 class ComputeNodeMonitorActorTestCase(testutil.ActorTestMixin,
166                                       unittest.TestCase):
167     class MockShutdownTimer(object):
168         def _set_state(self, is_open, next_opening):
169             self.window_open = lambda: is_open
170             self.next_opening = lambda: next_opening
171
172
173     def make_mocks(self, node_num):
174         self.shutdowns = self.MockShutdownTimer()
175         self.shutdowns._set_state(False, 300)
176         self.timer = mock.MagicMock(name='timer_mock')
177         self.updates = mock.MagicMock(name='update_mock')
178         self.cloud_mock = testutil.cloud_node_mock(node_num)
179         self.subscriber = mock.Mock(name='subscriber_mock')
180
181     def make_actor(self, node_num=1, arv_node=None, start_time=None):
182         if not hasattr(self, 'cloud_mock'):
183             self.make_mocks(node_num)
184         if start_time is None:
185             start_time = time.time()
186         self.node_actor = cnode.ComputeNodeMonitorActor.start(
187             self.cloud_mock, start_time, self.shutdowns, self.timer,
188             self.updates, arv_node).proxy()
189         self.node_actor.subscribe(self.subscriber).get(self.TIMEOUT)
190
191     def node_state(self, *states):
192         return self.node_actor.in_state(*states).get(self.TIMEOUT)
193
194     def test_in_state_when_unpaired(self):
195         self.make_actor()
196         self.assertIsNone(self.node_state('idle', 'alloc'))
197
198     def test_in_state_when_pairing_stale(self):
199         self.make_actor(arv_node=testutil.arvados_node_mock(
200                 job_uuid=None, age=90000))
201         self.assertIsNone(self.node_state('idle', 'alloc'))
202
203     def test_in_state_when_no_state_available(self):
204         self.make_actor(arv_node=testutil.arvados_node_mock(info={}))
205         self.assertIsNone(self.node_state('idle', 'alloc'))
206
207     def test_in_idle_state(self):
208         self.make_actor(2, arv_node=testutil.arvados_node_mock(job_uuid=None))
209         self.assertTrue(self.node_state('idle'))
210         self.assertFalse(self.node_state('alloc'))
211         self.assertTrue(self.node_state('idle', 'alloc'))
212
213     def test_in_alloc_state(self):
214         self.make_actor(3, arv_node=testutil.arvados_node_mock(job_uuid=True))
215         self.assertFalse(self.node_state('idle'))
216         self.assertTrue(self.node_state('alloc'))
217         self.assertTrue(self.node_state('idle', 'alloc'))
218
219     def test_init_shutdown_scheduling(self):
220         self.make_actor()
221         self.assertTrue(self.timer.schedule.called)
222         self.assertEqual(300, self.timer.schedule.call_args[0][0])
223
224     def test_shutdown_subscription(self):
225         self.make_actor()
226         self.shutdowns._set_state(True, 600)
227         self.node_actor.consider_shutdown().get(self.TIMEOUT)
228         self.assertTrue(self.subscriber.called)
229         self.assertEqual(self.node_actor.actor_ref.actor_urn,
230                          self.subscriber.call_args[0][0].actor_ref.actor_urn)
231
232     def test_shutdown_without_arvados_node(self):
233         self.make_actor()
234         self.shutdowns._set_state(True, 600)
235         self.node_actor.consider_shutdown().get(self.TIMEOUT)
236         self.assertTrue(self.subscriber.called)
237
238     def test_no_shutdown_without_arvados_node_and_old_cloud_node(self):
239         self.make_actor(start_time=0)
240         self.shutdowns._set_state(True, 600)
241         self.node_actor.consider_shutdown().get(self.TIMEOUT)
242         self.assertFalse(self.subscriber.called)
243
244     def check_shutdown_rescheduled(self, window_open, next_window,
245                                    schedule_time=None):
246         self.shutdowns._set_state(window_open, next_window)
247         self.timer.schedule.reset_mock()
248         self.node_actor.consider_shutdown().get(self.TIMEOUT)
249         self.stop_proxy(self.node_actor)
250         self.assertTrue(self.timer.schedule.called)
251         if schedule_time is not None:
252             self.assertEqual(schedule_time, self.timer.schedule.call_args[0][0])
253         self.assertFalse(self.subscriber.called)
254
255     def test_shutdown_window_close_scheduling(self):
256         self.make_actor()
257         self.check_shutdown_rescheduled(False, 600, 600)
258
259     def test_no_shutdown_when_node_running_job(self):
260         self.make_actor(4, testutil.arvados_node_mock(4, job_uuid=True))
261         self.check_shutdown_rescheduled(True, 600)
262
263     def test_no_shutdown_when_node_state_unknown(self):
264         self.make_actor(5, testutil.arvados_node_mock(5, info={}))
265         self.check_shutdown_rescheduled(True, 600)
266
267     def test_no_shutdown_when_node_state_stale(self):
268         self.make_actor(6, testutil.arvados_node_mock(6, age=90000))
269         self.check_shutdown_rescheduled(True, 600)
270
271     def test_arvados_node_match(self):
272         self.make_actor(2)
273         arv_node = testutil.arvados_node_mock(
274             2, hostname='compute-two.zzzzz.arvadosapi.com')
275         pair_id = self.node_actor.offer_arvados_pair(arv_node).get(self.TIMEOUT)
276         self.assertEqual(self.cloud_mock.id, pair_id)
277         self.stop_proxy(self.node_actor)
278         self.updates.sync_node.assert_called_with(self.cloud_mock, arv_node)
279
280     def test_arvados_node_mismatch(self):
281         self.make_actor(3)
282         arv_node = testutil.arvados_node_mock(1)
283         self.assertIsNone(
284             self.node_actor.offer_arvados_pair(arv_node).get(self.TIMEOUT))
285
286     def test_update_cloud_node(self):
287         self.make_actor(1)
288         self.make_mocks(2)
289         self.cloud_mock.id = '1'
290         self.node_actor.update_cloud_node(self.cloud_mock)
291         current_cloud = self.node_actor.cloud_node.get(self.TIMEOUT)
292         self.assertEqual([testutil.ip_address_mock(2)],
293                          current_cloud.private_ips)
294
295     def test_missing_cloud_node_update(self):
296         self.make_actor(1)
297         self.node_actor.update_cloud_node(None)
298         current_cloud = self.node_actor.cloud_node.get(self.TIMEOUT)
299         self.assertEqual([testutil.ip_address_mock(1)],
300                          current_cloud.private_ips)
301
302     def test_update_arvados_node(self):
303         self.make_actor(3)
304         job_uuid = 'zzzzz-jjjjj-updatejobnode00'
305         new_arvados = testutil.arvados_node_mock(3, job_uuid)
306         self.node_actor.update_arvados_node(new_arvados)
307         current_arvados = self.node_actor.arvados_node.get(self.TIMEOUT)
308         self.assertEqual(job_uuid, current_arvados['job_uuid'])
309
310     def test_missing_arvados_node_update(self):
311         self.make_actor(4, testutil.arvados_node_mock(4))
312         self.node_actor.update_arvados_node(None)
313         current_arvados = self.node_actor.arvados_node.get(self.TIMEOUT)
314         self.assertEqual(testutil.ip_address_mock(4),
315                          current_arvados['ip_address'])