services/nodemanager/tests/test_computenode_dispatch_slurm.py

   1 #!/usr/bin/env python
   2
   3 from __future__ import absolute_import, print_function
   4
   5 import subprocess
   6 import time
   7 import unittest
   8
   9 import mock
  10
  11 import arvnodeman.computenode.dispatch.slurm as slurm_dispatch
  12 from . import testutil
  13 from .test_computenode_dispatch import ComputeNodeShutdownActorMixin, ComputeNodeUpdateActorTestCase
  14
  15 @mock.patch('subprocess.check_output')
  16 class SLURMComputeNodeShutdownActorTestCase(ComputeNodeShutdownActorMixin,
  17                                             unittest.TestCase):
  18     ACTOR_CLASS = slurm_dispatch.ComputeNodeShutdownActor
  19
  20     def check_slurm_got_args(self, proc_mock, *args):
  21         self.assertTrue(proc_mock.called)
  22         slurm_cmd = proc_mock.call_args[0][0]
  23         for s in args:
  24             self.assertIn(s, slurm_cmd)
  25
  26     def check_success_after_reset(self, proc_mock, end_state='drain\n', timer=False):
  27         self.make_mocks(arvados_node=testutil.arvados_node_mock(63))
  28         if not timer:
  29             self.timer = testutil.MockTimer(False)
  30         self.make_actor()
  31         self.check_success_flag(None, 0)
  32         self.timer.deliver()
  33         self.check_success_flag(None, 0)
  34         self.timer.deliver()
  35         # Order is critical here: if the mock gets called when no return value
  36         # or side effect is set, we may invoke a real subprocess.
  37         proc_mock.return_value = end_state
  38         proc_mock.side_effect = None
  39         self.check_success_flag(True, 3)
  40         self.check_slurm_got_args(proc_mock, 'NodeName=compute63')
  41
  42     def make_wait_state_test(start_state='drng\n', end_state='drain\n'):
  43         def test(self, proc_mock):
  44             proc_mock.return_value = start_state
  45             self.check_success_after_reset(proc_mock, end_state)
  46         return test
  47
  48     for wait_state in ['alloc\n', 'drng\n']:
  49         locals()['test_wait_while_' + wait_state.strip()
  50                  ] = make_wait_state_test(start_state=wait_state)
  51
  52     for end_state in ['idle*\n', 'down\n', 'down*\n', 'drain\n', 'fail\n']:
  53         locals()['test_wait_until_' + end_state.strip()
  54                  ] = make_wait_state_test(end_state=end_state)
  55
  56     def test_retry_failed_slurm_calls(self, proc_mock):
  57         proc_mock.side_effect = subprocess.CalledProcessError(1, ["mock"])
  58         self.check_success_after_reset(proc_mock)
  59
  60     def test_slurm_bypassed_when_no_arvados_node(self, proc_mock):
  61         # Test we correctly handle a node that failed to bootstrap.
  62         proc_mock.return_value = 'down\n'
  63         self.make_actor(start_time=0)
  64         self.check_success_flag(True)
  65         self.assertFalse(proc_mock.called)
  66
  67     def test_node_undrained_when_shutdown_cancelled(self, proc_mock):
  68         try:
  69             proc_mock.side_effect = iter(['', 'drng\n', 'drng\n', ''])
  70             self.make_mocks(arvados_node=testutil.arvados_node_mock(job_uuid=True))
  71             self.timer = testutil.MockTimer(False)
  72             self.make_actor()
  73             self.busywait(lambda: proc_mock.call_args is not None)
  74             self.shutdown_actor.cancel_shutdown("test").get(self.TIMEOUT)
  75             self.check_success_flag(False, 2)
  76             self.assertEqual(proc_mock.call_args_list,
  77                              [mock.call(['scontrol', 'update', 'NodeName=compute99', 'State=DRAIN', 'Reason=Node Manager shutdown']),
  78                               mock.call(['sinfo', '--noheader', '-o', '%t', '-n', 'compute99']),
  79                               mock.call(['sinfo', '--noheader', '-o', '%t', '-n', 'compute99']),
  80                               mock.call(['scontrol', 'update', 'NodeName=compute99', 'State=RESUME'])])
  81         finally:
  82             self.shutdown_actor.actor_ref.stop()
  83
  84     def test_cancel_shutdown_retry(self, proc_mock):
  85         proc_mock.side_effect = iter([OSError, 'drain\n', OSError, 'idle\n', 'idle\n'])
  86         self.make_mocks(arvados_node=testutil.arvados_node_mock(job_uuid=True))
  87         self.make_actor()
  88         self.check_success_flag(False, 2)
  89
  90     def test_issue_slurm_drain_retry(self, proc_mock):
  91         proc_mock.side_effect = iter([OSError, '', OSError, 'drng\n'])
  92         self.check_success_after_reset(proc_mock, timer=False)
  93
  94     def test_arvados_node_cleaned_after_shutdown(self, proc_mock):
  95         proc_mock.return_value = 'drain\n'
  96         super(SLURMComputeNodeShutdownActorTestCase,
  97               self).test_arvados_node_cleaned_after_shutdown()
  98
  99     def test_cancellable_shutdown(self, proc_mock):
 100         proc_mock.return_value = 'other\n'
 101         super(SLURMComputeNodeShutdownActorTestCase,
 102               self).test_cancellable_shutdown()
 103
 104     def test_uncancellable_shutdown(self, proc_mock):
 105         proc_mock.return_value = 'other\n'
 106         super(SLURMComputeNodeShutdownActorTestCase,
 107               self).test_uncancellable_shutdown()
 108
 109 @mock.patch('subprocess.check_output')
 110 class SLURMComputeNodeUpdateActorTestCase(ComputeNodeUpdateActorTestCase):
 111     ACTOR_CLASS = slurm_dispatch.ComputeNodeUpdateActor
 112
 113     def test_update_node_weight(self, check_output):
 114         self.make_actor()
 115         cloud_node = testutil.cloud_node_mock()
 116         arv_node = testutil.arvados_node_mock()
 117         self.updater.sync_node(cloud_node, arv_node).get(self.TIMEOUT)
 118         check_output.assert_called_with(['scontrol', 'update', 'NodeName=compute99', 'Weight=99000'])