services/nodemanager/tests/test_computenode_dispatch_slurm.py

   1 #!/usr/bin/env python
   2 # Copyright (C) The Arvados Authors. All rights reserved.
   3 #
   4 # SPDX-License-Identifier: AGPL-3.0
   5
   6 from __future__ import absolute_import, print_function
   7
   8 import subprocess
   9 import time
  10 import unittest
  11
  12 import mock
  13
  14 import arvnodeman.computenode.dispatch.slurm as slurm_dispatch
  15 from . import testutil
  16 from .test_computenode_dispatch import \
  17     ComputeNodeShutdownActorMixin, \
  18     ComputeNodeSetupActorTestCase, \
  19     ComputeNodeUpdateActorTestCase
  20
  21 @mock.patch('subprocess.check_output')
  22 class SLURMComputeNodeShutdownActorTestCase(ComputeNodeShutdownActorMixin,
  23                                             unittest.TestCase):
  24     ACTOR_CLASS = slurm_dispatch.ComputeNodeShutdownActor
  25
  26     def check_slurm_got_args(self, proc_mock, *args):
  27         self.assertTrue(proc_mock.called)
  28         slurm_cmd = proc_mock.call_args[0][0]
  29         for s in args:
  30             self.assertIn(s, slurm_cmd)
  31
  32     def check_success_after_reset(self, proc_mock, end_state='drain\n', timer=False):
  33         self.make_mocks(arvados_node=testutil.arvados_node_mock(63))
  34         if not timer:
  35             self.timer = testutil.MockTimer(False)
  36         self.make_actor()
  37         self.check_success_flag(None, 0)
  38         # At this point, 1st try should have happened.
  39
  40         self.timer.deliver()
  41         self.check_success_flag(None, 0)
  42         # At this point, 2nd try should have happened.
  43
  44         # Order is critical here: if the mock gets called when no return value
  45         # or side effect is set, we may invoke a real subprocess.
  46         proc_mock.return_value = end_state
  47         proc_mock.side_effect = None
  48
  49         # 3rd try
  50         self.timer.deliver()
  51
  52         self.check_success_flag(True, 3)
  53         self.check_slurm_got_args(proc_mock, 'NodeName=compute63')
  54
  55     def make_wait_state_test(start_state='drng\n', end_state='drain\n'):
  56         def test(self, proc_mock):
  57             proc_mock.return_value = start_state
  58             self.check_success_after_reset(proc_mock, end_state)
  59         return test
  60
  61     for wait_state in ['alloc\n', 'drng\n']:
  62         locals()['test_wait_while_' + wait_state.strip()
  63                  ] = make_wait_state_test(start_state=wait_state)
  64
  65     for end_state in ['idle*\n', 'down\n', 'down*\n', 'drain\n', 'fail\n']:
  66         locals()['test_wait_until_' + end_state.strip()
  67                  ] = make_wait_state_test(end_state=end_state)
  68
  69     def test_retry_failed_slurm_calls(self, proc_mock):
  70         proc_mock.side_effect = subprocess.CalledProcessError(1, ["mock"])
  71         self.check_success_after_reset(proc_mock)
  72
  73     def test_slurm_bypassed_when_no_arvados_node(self, proc_mock):
  74         # Test we correctly handle a node that failed to bootstrap.
  75         proc_mock.return_value = 'down\n'
  76         self.make_actor(start_time=0)
  77         self.check_success_flag(True)
  78         self.assertFalse(proc_mock.called)
  79
  80     def test_node_resumed_when_shutdown_cancelled(self, proc_mock):
  81         try:
  82             proc_mock.side_effect = iter(['', 'drng\n', 'drng\n', ''])
  83             self.make_mocks(arvados_node=testutil.arvados_node_mock(job_uuid=True))
  84             self.timer = testutil.MockTimer(False)
  85             self.make_actor()
  86             self.busywait(lambda: proc_mock.call_args is not None)
  87             self.shutdown_actor.cancel_shutdown("test")
  88             self.check_success_flag(False, 2)
  89             self.assertEqual(proc_mock.call_args_list[0], mock.call(['scontrol', 'update', 'NodeName=compute99', 'State=DRAIN', 'Reason=Node Manager shutdown']))
  90             self.assertEqual(proc_mock.call_args_list[-1], mock.call(['scontrol', 'update', 'NodeName=compute99', 'State=RESUME']))
  91
  92         finally:
  93             self.shutdown_actor.actor_ref.stop()
  94
  95     def test_cancel_shutdown_retry(self, proc_mock):
  96         proc_mock.side_effect = iter([OSError, 'drain\n', OSError, 'idle\n', 'idle\n'])
  97         self.make_mocks(arvados_node=testutil.arvados_node_mock(job_uuid=True))
  98         self.make_actor()
  99         self.check_success_flag(False, 5)
 100
 101     def test_issue_slurm_drain_retry(self, proc_mock):
 102         proc_mock.side_effect = iter([OSError, OSError, 'drng\n', 'drain\n'])
 103         self.check_success_after_reset(proc_mock, timer=False)
 104
 105     def test_arvados_node_cleaned_after_shutdown(self, proc_mock):
 106         proc_mock.return_value = 'drain\n'
 107         super(SLURMComputeNodeShutdownActorTestCase,
 108               self).test_arvados_node_cleaned_after_shutdown()
 109
 110     def test_cancellable_shutdown(self, proc_mock):
 111         proc_mock.return_value = 'other\n'
 112         super(SLURMComputeNodeShutdownActorTestCase,
 113               self).test_cancellable_shutdown()
 114
 115     def test_uncancellable_shutdown(self, proc_mock):
 116         proc_mock.return_value = 'other\n'
 117         super(SLURMComputeNodeShutdownActorTestCase,
 118               self).test_uncancellable_shutdown()
 119
 120 @mock.patch('subprocess.check_output')
 121 class SLURMComputeNodeUpdateActorTestCase(ComputeNodeUpdateActorTestCase):
 122     ACTOR_CLASS = slurm_dispatch.ComputeNodeUpdateActor
 123
 124     def test_update_node_weight(self, check_output):
 125         self.make_actor()
 126         cloud_node = testutil.cloud_node_mock()
 127         arv_node = testutil.arvados_node_mock()
 128         self.updater.sync_node(cloud_node, arv_node).get(self.TIMEOUT)
 129         check_output.assert_called_with(['scontrol', 'update', 'NodeName=compute99', 'Weight=99000', 'Features=instancetype=z99.test'])
 130
 131 class SLURMComputeNodeSetupActorTestCase(ComputeNodeSetupActorTestCase):
 132     ACTOR_CLASS = slurm_dispatch.ComputeNodeSetupActor
 133
 134     @mock.patch('subprocess.check_output')
 135     def test_update_node_features(self, check_output):
 136         # `scontrol update` happens only if the Arvados node record
 137         # has a hostname. ComputeNodeSetupActorTestCase.make_mocks
 138         # uses mocks with scrubbed hostnames, so we override with the
 139         # default testutil.arvados_node_mock.
 140         self.make_mocks(arvados_effect=[testutil.arvados_node_mock()])
 141         self.make_actor()
 142         self.wait_for_assignment(self.setup_actor, 'cloud_node')
 143         check_output.assert_called_with(['scontrol', 'update', 'NodeName=compute99', 'Weight=1000', 'Features=instancetype=z1.test'])
 144
 145     @mock.patch('subprocess.check_output')
 146     def test_failed_arvados_calls_retried(self, check_output):
 147         super(SLURMComputeNodeSetupActorTestCase, self).test_failed_arvados_calls_retried()
 148
 149     @mock.patch('subprocess.check_output')
 150     def test_subscribe(self, check_output):
 151         super(SLURMComputeNodeSetupActorTestCase, self).test_subscribe()
 152
 153     @mock.patch('subprocess.check_output')
 154     def test_creation_with_arvados_node(self, check_output):
 155         super(SLURMComputeNodeSetupActorTestCase, self).test_creation_with_arvados_node()