|
| 1 | +import unittest |
| 2 | + |
| 3 | +from deepmd.cluster import local, slurm |
| 4 | +from unittest import mock |
| 5 | + |
| 6 | + |
| 7 | +kHostName = 'compute-b24-1' |
| 8 | + |
| 9 | + |
| 10 | +class FakePopen(object): |
| 11 | + def __init__(self, stdout=b'', stderr=b'', returncode=0): |
| 12 | + self._stdout = stdout |
| 13 | + self._stderr = stderr |
| 14 | + self._returncode = returncode |
| 15 | + |
| 16 | + def communicate(self): |
| 17 | + return self._stdout, self._stderr |
| 18 | + |
| 19 | + @property |
| 20 | + def returncode(self): |
| 21 | + return self._returncode |
| 22 | + |
| 23 | + |
| 24 | +class TestGPU(unittest.TestCase): |
| 25 | + @mock.patch('subprocess.Popen') |
| 26 | + def test_none(self, mock_Popen): |
| 27 | + mock_Popen.return_value.__enter__.return_value = FakePopen(b'0', b'') |
| 28 | + gpus = local.get_gpus() |
| 29 | + self.assertIsNone(gpus) |
| 30 | + |
| 31 | + @mock.patch('subprocess.Popen') |
| 32 | + def test_valid(self, mock_Popen): |
| 33 | + mock_Popen.return_value.__enter__.return_value = FakePopen(b'2', b'') |
| 34 | + gpus = local.get_gpus() |
| 35 | + self.assertEqual(gpus, [0, 1]) |
| 36 | + |
| 37 | + @mock.patch('subprocess.Popen') |
| 38 | + def test_error(self, mock_Popen): |
| 39 | + mock_Popen.return_value.__enter__.return_value = \ |
| 40 | + FakePopen(stderr=b'!', returncode=1) |
| 41 | + with self.assertRaises(RuntimeError) as cm: |
| 42 | + _ = local.get_gpus() |
| 43 | + self.assertIn('Failed to detect', str(cm.exception)) |
| 44 | + |
| 45 | + |
| 46 | +class TestLocal(unittest.TestCase): |
| 47 | + @mock.patch('socket.gethostname') |
| 48 | + def test_resource(self, mock_gethostname): |
| 49 | + mock_gethostname.return_value = kHostName |
| 50 | + nodename, nodelist, _ = local.get_resource() |
| 51 | + self.assertEqual(nodename, kHostName) |
| 52 | + self.assertEqual(nodelist, [kHostName]) |
| 53 | + |
| 54 | + |
| 55 | +class TestSlurm(unittest.TestCase): |
| 56 | + @mock.patch.dict('os.environ', values={ |
| 57 | + 'SLURM_JOB_NODELIST': kHostName, |
| 58 | + 'SLURMD_NODENAME': kHostName, |
| 59 | + 'SLURM_JOB_NUM_NODES': '1' |
| 60 | + }) |
| 61 | + def test_single(self): |
| 62 | + nodename, nodelist, _ = slurm.get_resource() |
| 63 | + self.assertEqual(nodename, kHostName) |
| 64 | + self.assertEqual(nodelist, [kHostName]) |
| 65 | + |
| 66 | + @mock.patch.dict('os.environ', values={ |
| 67 | + 'SLURM_JOB_NODELIST': 'compute-b24-[1-3,5-9],compute-b25-[4,8]', |
| 68 | + 'SLURMD_NODENAME': 'compute-b24-2', |
| 69 | + 'SLURM_JOB_NUM_NODES': '10' |
| 70 | + }) |
| 71 | + def test_multiple(self): |
| 72 | + nodename, nodelist, _ = slurm.get_resource() |
| 73 | + self.assertEqual(nodename, 'compute-b24-2') |
| 74 | + self.assertEqual(nodelist, [ |
| 75 | + 'compute-b24-1', |
| 76 | + 'compute-b24-2', |
| 77 | + 'compute-b24-3', |
| 78 | + 'compute-b24-5', |
| 79 | + 'compute-b24-6', |
| 80 | + 'compute-b24-7', |
| 81 | + 'compute-b24-8', |
| 82 | + 'compute-b24-9', |
| 83 | + 'compute-b25-4', |
| 84 | + 'compute-b25-8' |
| 85 | + ]) |
| 86 | + |
| 87 | + def test_illegal(self): |
| 88 | + environ = { |
| 89 | + 'SLURM_JOB_NODELIST': 'compute-b24-[3-5]', |
| 90 | + 'SLURMD_NODENAME': 'compute-b24-4' |
| 91 | + } |
| 92 | + with mock.patch.dict('os.environ', environ): |
| 93 | + with self.assertRaises(RuntimeError) as cm: |
| 94 | + _ = slurm.get_resource() |
| 95 | + self.assertIn('Could not get SLURM number', str(cm.exception)) |
| 96 | + |
| 97 | + environ = { |
| 98 | + 'SLURM_JOB_NODELIST': 'compute-b24-1,compute-b25-2', |
| 99 | + 'SLURMD_NODENAME': 'compute-b25-2', |
| 100 | + 'SLURM_JOB_NUM_NODES': '4' |
| 101 | + } |
| 102 | + with mock.patch.dict('os.environ', environ): |
| 103 | + with self.assertRaises(ValueError) as cm: |
| 104 | + _ = slurm.get_resource() |
| 105 | + self.assertIn('Number of slurm nodes 2', str(cm.exception)) |
| 106 | + |
| 107 | + environ = { |
| 108 | + 'SLURM_JOB_NODELIST': 'compute-b24-1,compute-b25-3', |
| 109 | + 'SLURMD_NODENAME': 'compute-b25-2', |
| 110 | + 'SLURM_JOB_NUM_NODES': '2' |
| 111 | + } |
| 112 | + with mock.patch.dict('os.environ', environ): |
| 113 | + with self.assertRaises(ValueError) as cm: |
| 114 | + _ = slurm.get_resource() |
| 115 | + self.assertIn('Nodename(compute-b25-2', str(cm.exception)) |
0 commit comments