1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
|
# Owner(s): ["oncall: r2p"]
import sys
import tempfile
import time
import unittest
from datetime import datetime, timedelta
from torch.monitor import (
_WaitCounter,
Aggregation,
Event,
log_event,
register_event_handler,
Stat,
TensorboardEventHandler,
unregister_event_handler,
)
from torch.testing._internal.common_utils import run_tests, skipIfTorchDynamo, TestCase
class TestMonitor(TestCase):
def test_interval_stat(self) -> None:
events = []
def handler(event):
events.append(event)
handle = register_event_handler(handler)
s = Stat(
"asdf",
(Aggregation.SUM, Aggregation.COUNT),
timedelta(milliseconds=1),
)
self.assertEqual(s.name, "asdf")
s.add(2)
for _ in range(100):
# NOTE: different platforms sleep may be inaccurate so we loop
# instead (i.e. win)
time.sleep(1 / 1000) # ms
s.add(3)
if len(events) >= 1:
break
self.assertGreaterEqual(len(events), 1)
unregister_event_handler(handle)
def test_fixed_count_stat(self) -> None:
s = Stat(
"asdf",
(Aggregation.SUM, Aggregation.COUNT),
timedelta(hours=100),
3,
)
s.add(1)
s.add(2)
name = s.name
self.assertEqual(name, "asdf")
self.assertEqual(s.count, 2)
s.add(3)
self.assertEqual(s.count, 0)
self.assertEqual(s.get(), {Aggregation.SUM: 6.0, Aggregation.COUNT: 3})
def test_log_event(self) -> None:
e = Event(
name="torch.monitor.TestEvent",
timestamp=datetime.now(),
data={
"str": "a string",
"float": 1234.0,
"int": 1234,
},
)
self.assertEqual(e.name, "torch.monitor.TestEvent")
self.assertIsNotNone(e.timestamp)
self.assertIsNotNone(e.data)
log_event(e)
@skipIfTorchDynamo("Really weird error")
def test_event_handler(self) -> None:
events = []
def handler(event: Event) -> None:
events.append(event)
handle = register_event_handler(handler)
e = Event(
name="torch.monitor.TestEvent",
timestamp=datetime.now(),
data={},
)
log_event(e)
self.assertEqual(len(events), 1)
self.assertEqual(events[0], e)
log_event(e)
self.assertEqual(len(events), 2)
unregister_event_handler(handle)
log_event(e)
self.assertEqual(len(events), 2)
def test_wait_counter(self) -> None:
wait_counter = _WaitCounter(
"test_wait_counter",
)
with wait_counter.guard() as wcg:
pass
@skipIfTorchDynamo("Really weird error")
class TestMonitorTensorboard(TestCase):
def setUp(self):
global SummaryWriter, event_multiplexer
try:
from tensorboard.backend.event_processing import (
plugin_event_multiplexer as event_multiplexer,
)
from torch.utils.tensorboard import SummaryWriter
except ImportError:
return self.skipTest("Skip the test since TensorBoard is not installed")
self.temp_dirs = []
def create_summary_writer(self):
temp_dir = tempfile.TemporaryDirectory() # noqa: P201
self.temp_dirs.append(temp_dir)
return SummaryWriter(temp_dir.name)
def tearDown(self):
# Remove directories created by SummaryWriter
for temp_dir in self.temp_dirs:
temp_dir.cleanup()
@unittest.skipIf(
sys.version_info >= (3, 13),
"numpy failure, likely caused by old tensorboard version",
)
def test_event_handler(self):
with self.create_summary_writer() as w:
handle = register_event_handler(TensorboardEventHandler(w))
s = Stat(
"asdf",
(Aggregation.SUM, Aggregation.COUNT),
timedelta(hours=1),
5,
)
for i in range(10):
s.add(i)
self.assertEqual(s.count, 0)
unregister_event_handler(handle)
mul = event_multiplexer.EventMultiplexer()
mul.AddRunsFromDirectory(self.temp_dirs[-1].name)
mul.Reload()
scalar_dict = mul.PluginRunToTagToContent("scalars")
raw_result = {
tag: mul.Tensors(run, tag)
for run, run_dict in scalar_dict.items()
for tag in run_dict
}
scalars = {
tag: [e.tensor_proto.float_val[0] for e in events]
for tag, events in raw_result.items()
}
self.assertEqual(
scalars,
{
"asdf.sum": [10],
"asdf.count": [5],
},
)
if __name__ == "__main__":
run_tests()
|
