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