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 unittest
   7
   8 import mock
   9
  10 import arvnodeman.computenode.dispatch.slurm as slurm_dispatch
  11 from . import testutil
  12 from .test_computenode_dispatch import ComputeNodeShutdownActorMixin
  13
  14 @mock.patch('subprocess.check_output')
  15 class SLURMComputeNodeShutdownActorTestCase(ComputeNodeShutdownActorMixin,
  16                                             unittest.TestCase):
  17     ACTOR_CLASS = slurm_dispatch.ComputeNodeShutdownActor
  18
  19     def check_slurm_got_args(self, proc_mock, *args):
  20         self.assertTrue(proc_mock.called)
  21         slurm_cmd = proc_mock.call_args[0][0]
  22         for s in args:
  23             self.assertIn(s, slurm_cmd)
  24
  25     def check_success_after_reset(self, proc_mock, end_state='drain\n'):
  26         self.make_mocks(arvados_node=testutil.arvados_node_mock(63))
  27         self.make_actor()
  28         self.check_success_flag(None, 0)
  29         self.check_success_flag(None, 0)
  30         # Order is critical here: if the mock gets called when no return value
  31         # or side effect is set, we may invoke a real subprocess.
  32         proc_mock.return_value = end_state
  33         proc_mock.side_effect = None
  34         self.check_success_flag(True, 3)
  35         self.check_slurm_got_args(proc_mock, 'compute63')
  36
  37     def make_wait_state_test(start_state='drng\n', end_state='drain\n'):
  38         def test(self, proc_mock):
  39             proc_mock.return_value = start_state
  40             self.check_success_after_reset(proc_mock, end_state)
  41         return test
  42
  43     for wait_state in ['alloc\n', 'drng\n', 'idle*\n']:
  44         locals()['test_wait_while_' + wait_state.strip()
  45                  ] = make_wait_state_test(start_state=wait_state)
  46
  47     for end_state in ['down\n', 'down*\n', 'drain\n', 'fail\n']:
  48         locals()['test_wait_until_' + end_state.strip()
  49                  ] = make_wait_state_test(end_state=end_state)
  50
  51     def test_retry_failed_slurm_calls(self, proc_mock):
  52         proc_mock.side_effect = subprocess.CalledProcessError(1, ["mock"])
  53         self.check_success_after_reset(proc_mock)
  54
  55     def test_slurm_bypassed_when_no_arvados_node(self, proc_mock):
  56         # Test we correctly handle a node that failed to bootstrap.
  57         proc_mock.return_value = 'idle\n'
  58         self.make_actor(start_time=0)
  59         self.check_success_flag(True)
  60         self.assertFalse(proc_mock.called)
  61
  62     def test_node_undrained_when_shutdown_window_closes(self, proc_mock):
  63         proc_mock.side_effect = iter(['drng\n', 'idle\n'])
  64         self.make_mocks(arvados_node=testutil.arvados_node_mock(job_uuid=True))
  65         self.make_actor()
  66         self.check_success_flag(False, 2)
  67         self.check_slurm_got_args(proc_mock, 'NodeName=compute99', 'State=RESUME')
  68
  69     def test_alloc_node_undrained_when_shutdown_window_closes(self, proc_mock):
  70         proc_mock.side_effect = iter(['alloc\n'])
  71         self.make_mocks(arvados_node=testutil.arvados_node_mock(job_uuid=True))
  72         self.make_actor()
  73         self.check_success_flag(False, 2)
  74         self.check_slurm_got_args(proc_mock, 'sinfo', '--noheader', '-o', '%t', '-n', 'compute99')
  75
  76     def test_cancel_shutdown_retry(self, proc_mock):
  77         proc_mock.side_effect = iter([OSError, 'drain\n', OSError, 'idle\n'])
  78         self.make_mocks(arvados_node=testutil.arvados_node_mock(job_uuid=True))
  79         self.make_actor()
  80         self.check_success_flag(False, 2)
  81
  82     def test_issue_slurm_drain_retry(self, proc_mock):
  83         proc_mock.side_effect = iter([OSError, '', OSError, 'drng\n'])
  84         self.check_success_after_reset(proc_mock)
  85
  86     def test_arvados_node_cleaned_after_shutdown(self, proc_mock):
  87         proc_mock.return_value = 'drain\n'
  88         super(SLURMComputeNodeShutdownActorTestCase,
  89               self).test_arvados_node_cleaned_after_shutdown()