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 ComputeNodeShutdownActorMixin, ComputeNodeUpdateActorTestCase
  17
  18 @mock.patch('subprocess.check_output')
  19 class SLURMComputeNodeShutdownActorTestCase(ComputeNodeShutdownActorMixin,
  20                                             unittest.TestCase):
  21     ACTOR_CLASS = slurm_dispatch.ComputeNodeShutdownActor
  22
  23     def check_slurm_got_args(self, proc_mock, *args):
  24         self.assertTrue(proc_mock.called)
  25         slurm_cmd = proc_mock.call_args[0][0]
  26         for s in args:
  27             self.assertIn(s, slurm_cmd)
  28
  29     def check_success_after_reset(self, proc_mock, end_state='drain\n', timer=False):
  30         self.make_mocks(arvados_node=testutil.arvados_node_mock(63))
  31         if not timer:
  32             self.timer = testutil.MockTimer(False)
  33         self.make_actor()
  34         self.check_success_flag(None, 0)
  35         self.timer.deliver()
  36         self.check_success_flag(None, 0)
  37         self.timer.deliver()
  38         # Order is critical here: if the mock gets called when no return value
  39         # or side effect is set, we may invoke a real subprocess.
  40         proc_mock.return_value = end_state
  41         proc_mock.side_effect = None
  42         self.check_success_flag(True, 3)
  43         self.check_slurm_got_args(proc_mock, 'NodeName=compute63')
  44
  45     def make_wait_state_test(start_state='drng\n', end_state='drain\n'):
  46         def test(self, proc_mock):
  47             proc_mock.return_value = start_state
  48             self.check_success_after_reset(proc_mock, end_state)
  49         return test
  50
  51     for wait_state in ['alloc\n', 'drng\n']:
  52         locals()['test_wait_while_' + wait_state.strip()
  53                  ] = make_wait_state_test(start_state=wait_state)
  54
  55     for end_state in ['idle*\n', 'down\n', 'down*\n', 'drain\n', 'fail\n']:
  56         locals()['test_wait_until_' + end_state.strip()
  57                  ] = make_wait_state_test(end_state=end_state)
  58
  59     def test_retry_failed_slurm_calls(self, proc_mock):
  60         proc_mock.side_effect = subprocess.CalledProcessError(1, ["mock"])
  61         self.check_success_after_reset(proc_mock)
  62
  63     def test_slurm_bypassed_when_no_arvados_node(self, proc_mock):
  64         # Test we correctly handle a node that failed to bootstrap.
  65         proc_mock.return_value = 'down\n'
  66         self.make_actor(start_time=0)
  67         self.check_success_flag(True)
  68         self.assertFalse(proc_mock.called)
  69
  70     def test_node_undrained_when_shutdown_cancelled(self, proc_mock):
  71         try:
  72             proc_mock.side_effect = iter(['', 'drng\n', 'drng\n', ''])
  73             self.make_mocks(arvados_node=testutil.arvados_node_mock(job_uuid=True))
  74             self.timer = testutil.MockTimer(False)
  75             self.make_actor()
  76             self.busywait(lambda: proc_mock.call_args is not None)
  77             self.shutdown_actor.cancel_shutdown("test").get(self.TIMEOUT)
  78             self.check_success_flag(False, 2)
  79             self.assertEqual(proc_mock.call_args_list,
  80                              [mock.call(['scontrol', 'update', 'NodeName=compute99', 'State=DRAIN', 'Reason=Node Manager shutdown']),
  81                               mock.call(['sinfo', '--noheader', '-o', '%t', '-n', 'compute99']),
  82                               mock.call(['sinfo', '--noheader', '-o', '%t', '-n', 'compute99']),
  83                               mock.call(['scontrol', 'update', 'NodeName=compute99', 'State=RESUME'])])
  84         finally:
  85             self.shutdown_actor.actor_ref.stop()
  86
  87     def test_cancel_shutdown_retry(self, proc_mock):
  88         proc_mock.side_effect = iter([OSError, 'drain\n', OSError, 'idle\n', 'idle\n'])
  89         self.make_mocks(arvados_node=testutil.arvados_node_mock(job_uuid=True))
  90         self.make_actor()
  91         self.check_success_flag(False, 2)
  92
  93     def test_issue_slurm_drain_retry(self, proc_mock):
  94         proc_mock.side_effect = iter([OSError, '', OSError, 'drng\n'])
  95         self.check_success_after_reset(proc_mock, timer=False)
  96
  97     def test_arvados_node_cleaned_after_shutdown(self, proc_mock):
  98         proc_mock.return_value = 'drain\n'
  99         super(SLURMComputeNodeShutdownActorTestCase,
 100               self).test_arvados_node_cleaned_after_shutdown()
 101
 102     def test_cancellable_shutdown(self, proc_mock):
 103         proc_mock.return_value = 'other\n'
 104         super(SLURMComputeNodeShutdownActorTestCase,
 105               self).test_cancellable_shutdown()
 106
 107     def test_uncancellable_shutdown(self, proc_mock):
 108         proc_mock.return_value = 'other\n'
 109         super(SLURMComputeNodeShutdownActorTestCase,
 110               self).test_uncancellable_shutdown()
 111
 112 @mock.patch('subprocess.check_output')
 113 class SLURMComputeNodeUpdateActorTestCase(ComputeNodeUpdateActorTestCase):
 114     ACTOR_CLASS = slurm_dispatch.ComputeNodeUpdateActor
 115
 116     def test_update_node_weight(self, check_output):
 117         self.make_actor()
 118         cloud_node = testutil.cloud_node_mock()
 119         arv_node = testutil.arvados_node_mock()
 120         self.updater.sync_node(cloud_node, arv_node).get(self.TIMEOUT)
 121         check_output.assert_called_with(['scontrol', 'update', 'NodeName=compute99', 'Weight=99000'])