{{ message }}
Repository navigation
Expand file tree
/
Copy pathoracle_external_inspection.py
More file actions
436 lines (383 loc) · 13 KB
/
Copy pathoracle_external_inspection.py
File metadata and controls
436 lines (383 loc) · 13 KB
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
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
"""Compare external inspection modes against a reference mode ("Oracle").
This script reports the following validation metrics:
- impossible: Number of stacks matching a known impossible pattern. The
classifiers are not exhaustive.
- raw_tvd: Total variation distance between this mode's stack distribution
and the reference's distribution. Stacks classified as impossible are
excluded.
- tvd_excess: raw_tvd minus tvd_floor.
- tvd_floor: IID heuristic for the TVD expected from finite sampling of both
distributions. This is not a lower bound.
"""
import argparse
import contextlib
import math
import os
import random
import subprocess
import statistics
import sys
import tempfile
import time
from collections import Counter
import _remote_debugging
from snippets import CASES, _get_lineno
TRANSIENT_ERRORS = (OSError, RuntimeError, UnicodeDecodeError)
MODES = {
"live-cache": (False, True),
"live-nocache": (False, False),
"blocking-cache": (True, True),
"blocking-nocache": (True, False),
}
def collapse_cache(mode):
blocking, _ = MODES[mode]
return "blocking-nocache" if blocking else "live-nocache"
def tvd(left, right):
lt, rt = sum(left.values()), sum(right.values())
if not lt or not rt:
return None
return 0.5 * sum(
abs(left[k] / lt - right[k] / rt) for k in set(left) | set(right)
)
def tvd_floor(reference_obs, n_live):
n_ref = sum(reference_obs.values())
if not n_ref or not n_live:
return None
spread = sum(
math.sqrt(p * (1 - p))
for p in (c / n_ref for c in reference_obs.values())
)
return (
0.5
* math.sqrt(2 / math.pi)
* math.sqrt(1 / n_live + 1 / n_ref)
* spread
)
def print_run_info(args, cases):
print(sys.version.replace("\n", " "))
print(
f"cases={','.join(cases)} runs={args.runs} "
f"duration={args.duration} "
f"rate_khz={args.rate_khz} warmup={args.warmup} "
f"poisson_sampling={args.poisson_sampling}"
)
def terminate_process(proc):
if proc.poll() is not None:
return
proc.terminate()
try:
proc.wait(timeout=5)
except subprocess.TimeoutExpired:
proc.kill()
proc.wait()
@contextlib.contextmanager
def target_process(code, warmup):
with tempfile.NamedTemporaryFile("w", suffix=".py", delete=False) as tmp:
tmp.write(code)
tmp.flush()
tmp_name = tmp.name
proc = None
try:
proc = subprocess.Popen(
[sys.executable, tmp_name],
stdout=subprocess.DEVNULL,
)
time.sleep(warmup)
if proc.poll() is not None:
raise RuntimeError(
f"target exited unexpectedly with code {proc.returncode}"
)
yield proc
finally:
with contextlib.suppress(Exception):
if proc is not None:
terminate_process(proc)
with contextlib.suppress(OSError):
os.unlink(tmp_name)
def get_trace(unwinder, blocking, op="get_stack_trace"):
call = getattr(unwinder, op)
if not blocking:
return call()
unwinder.pause_threads()
try:
return call()
finally:
unwinder.resume_threads()
def iter_units(raw, op):
if op == "get_stack_trace":
for interp in raw:
for thread in interp.threads:
yield (
thread.thread_id,
None,
thread.status,
thread.frame_info,
)
else:
for awaited_info in raw:
for task in awaited_info.awaited_by:
frames = [
frame
for coro in task.coroutine_stack
for frame in coro.call_stack
]
if frames:
yield (task.task_id, task.task_name, None, frames)
def run_mode(case, mode_name, args):
code, classify, *rest = case
op = rest[0] if rest else "get_stack_trace"
classify_units = op != "get_stack_trace"
blocking, cache_frames = MODES[mode_name]
result = {
"attempts": 0,
"samples": 0,
"stacks": 0,
"errors": 0,
"observations": Counter(),
"impossible": 0,
"work_time": 0.0,
}
with target_process(code, args.warmup) as proc:
unwinder = _remote_debugging.RemoteUnwinder(
proc.pid,
all_threads=True,
cache_frames=cache_frames,
)
rate_hz = args.rate_khz * 1000
period = 1.0 / rate_hz if rate_hz else 0
next_sample = time.perf_counter()
deadline = next_sample + args.duration
while time.perf_counter() < deadline:
if period:
if args.poisson_sampling:
next_sample += random.expovariate(rate_hz)
if next_sample >= deadline:
break
now = time.perf_counter()
if next_sample > now:
time.sleep(next_sample - now)
if not args.poisson_sampling:
next_sample += period
if time.perf_counter() >= deadline:
break
result["attempts"] += 1
work_start = time.perf_counter()
try:
trace = get_trace(unwinder, blocking, op)
except TRANSIENT_ERRORS:
trace = None
result["errors"] += 1
result["work_time"] += time.perf_counter() - work_start
if not trace:
continue
result["samples"] += 1
for unit in iter_units(trace, op):
frames = unit[3]
impossible = classify is not None and (
classify(unit) if classify_units else classify(frames)
)
result["stacks"] += 1
if impossible:
result["impossible"] += 1
else:
result["observations"][
";".join(
f"{frame.funcname}:{_get_lineno(frame)}"
for frame in frames
)
] += 1
return result
def result_metrics(result, reference_obs, is_reference, op):
skip_tvd = is_reference or op != "get_stack_trace"
n_live = sum(result["observations"].values())
raw_tvd = None if skip_tvd else tvd(result["observations"], reference_obs)
# IID heuristic, not a calibrated noise bound.
floor = None if skip_tvd else tvd_floor(reference_obs, n_live)
return {
"samples": result["samples"],
"stacks": result["stacks"],
"empty": result["attempts"] - result["samples"] - result["errors"],
"impossible": result["impossible"],
"error_percent": (
100.0 * result["errors"] / result["attempts"]
if result["attempts"]
else 0.0
),
"impossible_percent": (
100.0 * result["impossible"] / result["stacks"]
if result["stacks"]
else 0.0
),
"avg_us": (
1e6 * result["work_time"] / result["attempts"]
if result["attempts"]
else 0.0
),
"raw_tvd": raw_tvd,
"tvd_floor": floor,
"tvd_excess": (
None if (raw_tvd is None or floor is None) else raw_tvd - floor
),
}
def fmt_stat(values, precision):
vals = [value for value in values if value is not None]
if not vals:
return "n/a"
mean = statistics.mean(vals)
if len(vals) > 1:
return f"{mean:.{precision}f}±{statistics.stdev(vals):.{precision}f}"
return f"{mean:.{precision}f}"
def fmt_floor(values):
vals = [value for value in values if value is not None]
return "n/a" if not vals else f"{statistics.median(vals):.3f}"
def print_results(
case_name, run_results, modes, reference_mode, op, has_classify
):
is_sync = op == "get_stack_trace"
ref_keys = len(run_results[0][reference_mode]["observations"])
rows = {}
for mode in modes:
metrics = [
result_metrics(
results[mode],
results[reference_mode]["observations"],
mode == reference_mode,
op,
)
for results in run_results
]
rows[mode] = {
"mode": mode,
"samples": sum(item["samples"] for item in metrics),
"stacks": sum(item["stacks"] for item in metrics),
"empty": sum(item["empty"] for item in metrics),
"impossible": sum(item["impossible"] for item in metrics),
"errors": fmt_stat([item["error_percent"] for item in metrics], 2),
"us": fmt_stat([item["avg_us"] for item in metrics], 2),
"impossible_pct": fmt_stat(
[item["impossible_percent"] for item in metrics], 2
),
"raw_tvd": "ref"
if mode == reference_mode
else fmt_stat([item["raw_tvd"] for item in metrics], 3),
"tvd_excess": "ref"
if mode == reference_mode
else fmt_stat([item["tvd_excess"] for item in metrics], 3),
"tvd_floor": "-"
if mode == reference_mode
else fmt_floor([item["tvd_floor"] for item in metrics]),
}
show_stacks = any(row["stacks"] != row["samples"] for row in rows.values())
ordered = [reference_mode] + [m for m in modes if m != reference_mode]
columns = [("mode", "<18", "mode"), ("samples", ">9", "samples")]
if show_stacks:
columns.append(("stacks", ">9", "stacks"))
if not is_sync:
columns.append(("empty", ">9", "empty"))
columns += [("µs", ">12", "us"), ("errors%", ">12", "errors")]
if has_classify:
columns += [
("impossible", ">10", "impossible"),
("impossible%", ">12", "impossible_pct"),
]
if is_sync:
columns += [
("raw_tvd", ">12", "raw_tvd"),
("tvd_excess", ">12", "tvd_excess"),
("tvd_floor", ">10", "tvd_floor"),
]
print(f"\n{case_name} ({op}) ref_keys={ref_keys}")
print(" ".join(f"{label:{spec}}" for label, spec, _ in columns))
for mode in ordered:
row = rows[mode]
print(" ".join(f"{row[key]:{spec}}" for _, spec, key in columns))
def parse_args():
parser = argparse.ArgumentParser(
formatter_class=argparse.ArgumentDefaultsHelpFormatter
)
parser.add_argument(
"--snippet",
action="append",
help="snippet name or Python file; may be passed more than once",
)
parser.add_argument(
"--mode",
choices=sorted(MODES),
action="append",
help="mode to run; may be passed more than once; omit to run all modes",
)
parser.add_argument(
"--reference-mode",
choices=sorted(MODES),
default="blocking-nocache",
help="mode used as the distribution reference",
)
parser.add_argument(
"--duration",
type=float,
default=3.0,
help="seconds to sample each mode",
)
parser.add_argument(
"--runs",
type=int,
default=1,
help="number of independent runs per case",
)
parser.add_argument(
"--rate-khz",
type=float,
default=100.0,
help="target sampling rate in kHz; 0 samples as fast as possible",
)
parser.add_argument(
"--warmup",
type=float,
default=0.7,
help="seconds to let the target run before sampling",
)
parser.add_argument(
"--poisson-sampling",
action="store_true",
help=(
"sample with exponential inter-arrival times instead of a fixed "
"period"
),
)
args = parser.parse_args()
if args.runs < 1:
parser.error("--runs must be greater than zero")
return args
def main():
args = parse_args()
cases = list(args.snippet) if args.snippet else sorted(CASES)
modes = list(args.mode) if args.mode else sorted(MODES)
if args.reference_mode not in modes:
modes.append(args.reference_mode)
print_run_info(args, cases)
for name in cases:
if name in CASES:
case = CASES[name]
else:
with open(name, encoding="utf-8") as file:
case = (file.read(), None)
op = case[2] if len(case) > 2 else "get_stack_trace"
case_ref = args.reference_mode
case_modes = list(modes)
if op != "get_stack_trace":
case_ref = collapse_cache(case_ref)
case_modes = list(
dict.fromkeys(collapse_cache(m) for m in case_modes)
)
if case_ref not in case_modes:
case_modes.append(case_ref)
run_results = [
{mode: run_mode(case, mode, args) for mode in case_modes}
for _ in range(args.runs)
]
print_results(
name, run_results, case_modes, case_ref, op, case[1] is not None
)
return 0
if __name__ == "__main__":
sys.exit(main())
You can’t perform that action at this time.