Show More
@@ -1,797 +1,796 b'' | |||
|
1 | 1 | #!/usr/bin/env python |
|
2 | 2 | ## statprof.py |
|
3 | 3 | ## Copyright (C) 2012 Bryan O'Sullivan <bos@serpentine.com> |
|
4 | 4 | ## Copyright (C) 2011 Alex Fraser <alex at phatcore dot com> |
|
5 | 5 | ## Copyright (C) 2004,2005 Andy Wingo <wingo at pobox dot com> |
|
6 | 6 | ## Copyright (C) 2001 Rob Browning <rlb at defaultvalue dot org> |
|
7 | 7 | |
|
8 | 8 | ## This library is free software; you can redistribute it and/or |
|
9 | 9 | ## modify it under the terms of the GNU Lesser General Public |
|
10 | 10 | ## License as published by the Free Software Foundation; either |
|
11 | 11 | ## version 2.1 of the License, or (at your option) any later version. |
|
12 | 12 | ## |
|
13 | 13 | ## This library is distributed in the hope that it will be useful, |
|
14 | 14 | ## but WITHOUT ANY WARRANTY; without even the implied warranty of |
|
15 | 15 | ## MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
|
16 | 16 | ## Lesser General Public License for more details. |
|
17 | 17 | ## |
|
18 | 18 | ## You should have received a copy of the GNU Lesser General Public |
|
19 | 19 | ## License along with this program; if not, contact: |
|
20 | 20 | ## |
|
21 | 21 | ## Free Software Foundation Voice: +1-617-542-5942 |
|
22 | 22 | ## 59 Temple Place - Suite 330 Fax: +1-617-542-2652 |
|
23 | 23 | ## Boston, MA 02111-1307, USA gnu@gnu.org |
|
24 | 24 | |
|
25 | 25 | """ |
|
26 | 26 | statprof is intended to be a fairly simple statistical profiler for |
|
27 | 27 | python. It was ported directly from a statistical profiler for guile, |
|
28 | 28 | also named statprof, available from guile-lib [0]. |
|
29 | 29 | |
|
30 | 30 | [0] http://wingolog.org/software/guile-lib/statprof/ |
|
31 | 31 | |
|
32 | 32 | To start profiling, call statprof.start(): |
|
33 | 33 | >>> start() |
|
34 | 34 | |
|
35 | 35 | Then run whatever it is that you want to profile, for example: |
|
36 | 36 | >>> import test.pystone; test.pystone.pystones() |
|
37 | 37 | |
|
38 | 38 | Then stop the profiling and print out the results: |
|
39 | 39 | >>> stop() |
|
40 | 40 | >>> display() |
|
41 | 41 | % cumulative self |
|
42 | 42 | time seconds seconds name |
|
43 | 43 | 26.72 1.40 0.37 pystone.py:79:Proc0 |
|
44 | 44 | 13.79 0.56 0.19 pystone.py:133:Proc1 |
|
45 | 45 | 13.79 0.19 0.19 pystone.py:208:Proc8 |
|
46 | 46 | 10.34 0.16 0.14 pystone.py:229:Func2 |
|
47 | 47 | 6.90 0.10 0.10 pystone.py:45:__init__ |
|
48 | 48 | 4.31 0.16 0.06 pystone.py:53:copy |
|
49 | 49 | ... |
|
50 | 50 | |
|
51 | 51 | All of the numerical data is statistically approximate. In the |
|
52 | 52 | following column descriptions, and in all of statprof, "time" refers |
|
53 | 53 | to execution time (both user and system), not wall clock time. |
|
54 | 54 | |
|
55 | 55 | % time |
|
56 | 56 | The percent of the time spent inside the procedure itself (not |
|
57 | 57 | counting children). |
|
58 | 58 | |
|
59 | 59 | cumulative seconds |
|
60 | 60 | The total number of seconds spent in the procedure, including |
|
61 | 61 | children. |
|
62 | 62 | |
|
63 | 63 | self seconds |
|
64 | 64 | The total number of seconds spent in the procedure itself (not |
|
65 | 65 | counting children). |
|
66 | 66 | |
|
67 | 67 | name |
|
68 | 68 | The name of the procedure. |
|
69 | 69 | |
|
70 | 70 | By default statprof keeps the data collected from previous runs. If you |
|
71 | 71 | want to clear the collected data, call reset(): |
|
72 | 72 | >>> reset() |
|
73 | 73 | |
|
74 | 74 | reset() can also be used to change the sampling frequency from the |
|
75 | 75 | default of 1000 Hz. For example, to tell statprof to sample 50 times a |
|
76 | 76 | second: |
|
77 | 77 | >>> reset(50) |
|
78 | 78 | |
|
79 | 79 | This means that statprof will sample the call stack after every 1/50 of |
|
80 | 80 | a second of user + system time spent running on behalf of the python |
|
81 | 81 | process. When your process is idle (for example, blocking in a read(), |
|
82 | 82 | as is the case at the listener), the clock does not advance. For this |
|
83 | 83 | reason statprof is not currently not suitable for profiling io-bound |
|
84 | 84 | operations. |
|
85 | 85 | |
|
86 | 86 | The profiler uses the hash of the code object itself to identify the |
|
87 | 87 | procedures, so it won't confuse different procedures with the same name. |
|
88 | 88 | They will show up as two different rows in the output. |
|
89 | 89 | |
|
90 | 90 | Right now the profiler is quite simplistic. I cannot provide |
|
91 | 91 | call-graphs or other higher level information. What you see in the |
|
92 | 92 | table is pretty much all there is. Patches are welcome :-) |
|
93 | 93 | |
|
94 | 94 | |
|
95 | 95 | Threading |
|
96 | 96 | --------- |
|
97 | 97 | |
|
98 | 98 | Because signals only get delivered to the main thread in Python, |
|
99 | 99 | statprof only profiles the main thread. However because the time |
|
100 | 100 | reporting function uses per-process timers, the results can be |
|
101 | 101 | significantly off if other threads' work patterns are not similar to the |
|
102 | 102 | main thread's work patterns. |
|
103 | 103 | """ |
|
104 | 104 | # no-check-code |
|
105 | 105 | from __future__ import division |
|
106 | 106 | |
|
107 |
import inspect, json, os, signal, tempfile, sys, getopt, threading |
|
|
107 | import inspect, json, os, signal, tempfile, sys, getopt, threading | |
|
108 | 108 | import time |
|
109 | 109 | from collections import defaultdict |
|
110 | 110 | from contextlib import contextmanager |
|
111 | from subprocess import call | |
|
112 | 111 | |
|
113 | 112 | __all__ = ['start', 'stop', 'reset', 'display', 'profile'] |
|
114 | 113 | |
|
115 | 114 | skips = set(["util.py:check", "extensions.py:closure", |
|
116 | 115 | "color.py:colorcmd", "dispatch.py:checkargs", |
|
117 | 116 | "dispatch.py:<lambda>", "dispatch.py:_runcatch", |
|
118 | 117 | "dispatch.py:_dispatch", "dispatch.py:_runcommand", |
|
119 | 118 | "pager.py:pagecmd", "dispatch.py:run", |
|
120 | 119 | "dispatch.py:dispatch", "dispatch.py:runcommand", |
|
121 | 120 | "hg.py:<module>", "evolve.py:warnobserrors", |
|
122 | 121 | ]) |
|
123 | 122 | |
|
124 | 123 | ########################################################################### |
|
125 | 124 | ## Utils |
|
126 | 125 | |
|
127 | 126 | def clock(): |
|
128 | 127 | times = os.times() |
|
129 | 128 | return times[0] + times[1] |
|
130 | 129 | |
|
131 | 130 | |
|
132 | 131 | ########################################################################### |
|
133 | 132 | ## Collection data structures |
|
134 | 133 | |
|
135 | 134 | class ProfileState(object): |
|
136 | 135 | def __init__(self, frequency=None): |
|
137 | 136 | self.reset(frequency) |
|
138 | 137 | |
|
139 | 138 | def reset(self, frequency=None): |
|
140 | 139 | # total so far |
|
141 | 140 | self.accumulated_time = 0.0 |
|
142 | 141 | # start_time when timer is active |
|
143 | 142 | self.last_start_time = None |
|
144 | 143 | # a float |
|
145 | 144 | if frequency: |
|
146 | 145 | self.sample_interval = 1.0 / frequency |
|
147 | 146 | elif not hasattr(self, 'sample_interval'): |
|
148 | 147 | # default to 1000 Hz |
|
149 | 148 | self.sample_interval = 1.0 / 1000.0 |
|
150 | 149 | else: |
|
151 | 150 | # leave the frequency as it was |
|
152 | 151 | pass |
|
153 | 152 | self.remaining_prof_time = None |
|
154 | 153 | # for user start/stop nesting |
|
155 | 154 | self.profile_level = 0 |
|
156 | 155 | |
|
157 | 156 | self.samples = [] |
|
158 | 157 | |
|
159 | 158 | def accumulate_time(self, stop_time): |
|
160 | 159 | self.accumulated_time += stop_time - self.last_start_time |
|
161 | 160 | |
|
162 | 161 | def seconds_per_sample(self): |
|
163 | 162 | return self.accumulated_time / len(self.samples) |
|
164 | 163 | |
|
165 | 164 | state = ProfileState() |
|
166 | 165 | |
|
167 | 166 | |
|
168 | 167 | class CodeSite(object): |
|
169 | 168 | cache = {} |
|
170 | 169 | |
|
171 | 170 | __slots__ = ('path', 'lineno', 'function', 'source') |
|
172 | 171 | |
|
173 | 172 | def __init__(self, path, lineno, function): |
|
174 | 173 | self.path = path |
|
175 | 174 | self.lineno = lineno |
|
176 | 175 | self.function = function |
|
177 | 176 | self.source = None |
|
178 | 177 | |
|
179 | 178 | def __eq__(self, other): |
|
180 | 179 | try: |
|
181 | 180 | return (self.lineno == other.lineno and |
|
182 | 181 | self.path == other.path) |
|
183 | 182 | except: |
|
184 | 183 | return False |
|
185 | 184 | |
|
186 | 185 | def __hash__(self): |
|
187 | 186 | return hash((self.lineno, self.path)) |
|
188 | 187 | |
|
189 | 188 | @classmethod |
|
190 | 189 | def get(cls, path, lineno, function): |
|
191 | 190 | k = (path, lineno) |
|
192 | 191 | try: |
|
193 | 192 | return cls.cache[k] |
|
194 | 193 | except KeyError: |
|
195 | 194 | v = cls(path, lineno, function) |
|
196 | 195 | cls.cache[k] = v |
|
197 | 196 | return v |
|
198 | 197 | |
|
199 | 198 | def getsource(self, length): |
|
200 | 199 | if self.source is None: |
|
201 | 200 | lineno = self.lineno - 1 |
|
202 | 201 | fp = None |
|
203 | 202 | try: |
|
204 | 203 | fp = open(self.path) |
|
205 | 204 | for i, line in enumerate(fp): |
|
206 | 205 | if i == lineno: |
|
207 | 206 | self.source = line.strip() |
|
208 | 207 | break |
|
209 | 208 | except: |
|
210 | 209 | pass |
|
211 | 210 | finally: |
|
212 | 211 | if fp: |
|
213 | 212 | fp.close() |
|
214 | 213 | if self.source is None: |
|
215 | 214 | self.source = '' |
|
216 | 215 | |
|
217 | 216 | source = self.source |
|
218 | 217 | if len(source) > length: |
|
219 | 218 | source = source[:(length - 3)] + "..." |
|
220 | 219 | return source |
|
221 | 220 | |
|
222 | 221 | def filename(self): |
|
223 | 222 | return os.path.basename(self.path) |
|
224 | 223 | |
|
225 | 224 | class Sample(object): |
|
226 | 225 | __slots__ = ('stack', 'time') |
|
227 | 226 | |
|
228 | 227 | def __init__(self, stack, time): |
|
229 | 228 | self.stack = stack |
|
230 | 229 | self.time = time |
|
231 | 230 | |
|
232 | 231 | @classmethod |
|
233 | 232 | def from_frame(cls, frame, time): |
|
234 | 233 | stack = [] |
|
235 | 234 | |
|
236 | 235 | while frame: |
|
237 | 236 | stack.append(CodeSite.get(frame.f_code.co_filename, frame.f_lineno, |
|
238 | 237 | frame.f_code.co_name)) |
|
239 | 238 | frame = frame.f_back |
|
240 | 239 | |
|
241 | 240 | return Sample(stack, time) |
|
242 | 241 | |
|
243 | 242 | ########################################################################### |
|
244 | 243 | ## SIGPROF handler |
|
245 | 244 | |
|
246 | 245 | def profile_signal_handler(signum, frame): |
|
247 | 246 | if state.profile_level > 0: |
|
248 | 247 | now = clock() |
|
249 | 248 | state.accumulate_time(now) |
|
250 | 249 | |
|
251 | 250 | state.samples.append(Sample.from_frame(frame, state.accumulated_time)) |
|
252 | 251 | |
|
253 | 252 | signal.setitimer(signal.ITIMER_PROF, |
|
254 | 253 | state.sample_interval, 0.0) |
|
255 | 254 | state.last_start_time = now |
|
256 | 255 | |
|
257 | 256 | stopthread = threading.Event() |
|
258 | 257 | def samplerthread(tid): |
|
259 | 258 | while not stopthread.is_set(): |
|
260 | 259 | now = clock() |
|
261 | 260 | state.accumulate_time(now) |
|
262 | 261 | |
|
263 | 262 | frame = sys._current_frames()[tid] |
|
264 | 263 | state.samples.append(Sample.from_frame(frame, state.accumulated_time)) |
|
265 | 264 | |
|
266 | 265 | state.last_start_time = now |
|
267 | 266 | time.sleep(state.sample_interval) |
|
268 | 267 | |
|
269 | 268 | stopthread.clear() |
|
270 | 269 | |
|
271 | 270 | ########################################################################### |
|
272 | 271 | ## Profiling API |
|
273 | 272 | |
|
274 | 273 | def is_active(): |
|
275 | 274 | return state.profile_level > 0 |
|
276 | 275 | |
|
277 | 276 | lastmechanism = None |
|
278 | 277 | def start(mechanism='thread'): |
|
279 | 278 | '''Install the profiling signal handler, and start profiling.''' |
|
280 | 279 | state.profile_level += 1 |
|
281 | 280 | if state.profile_level == 1: |
|
282 | 281 | state.last_start_time = clock() |
|
283 | 282 | rpt = state.remaining_prof_time |
|
284 | 283 | state.remaining_prof_time = None |
|
285 | 284 | |
|
286 | 285 | global lastmechanism |
|
287 | 286 | lastmechanism = mechanism |
|
288 | 287 | |
|
289 | 288 | if mechanism == 'signal': |
|
290 | 289 | signal.signal(signal.SIGPROF, profile_signal_handler) |
|
291 | 290 | signal.setitimer(signal.ITIMER_PROF, |
|
292 | 291 | rpt or state.sample_interval, 0.0) |
|
293 | 292 | elif mechanism == 'thread': |
|
294 | 293 | frame = inspect.currentframe() |
|
295 | 294 | tid = [k for k, f in sys._current_frames().items() if f == frame][0] |
|
296 | 295 | state.thread = threading.Thread(target=samplerthread, |
|
297 | 296 | args=(tid,), name="samplerthread") |
|
298 | 297 | state.thread.start() |
|
299 | 298 | |
|
300 | 299 | def stop(): |
|
301 | 300 | '''Stop profiling, and uninstall the profiling signal handler.''' |
|
302 | 301 | state.profile_level -= 1 |
|
303 | 302 | if state.profile_level == 0: |
|
304 | 303 | if lastmechanism == 'signal': |
|
305 | 304 | rpt = signal.setitimer(signal.ITIMER_PROF, 0.0, 0.0) |
|
306 | 305 | signal.signal(signal.SIGPROF, signal.SIG_IGN) |
|
307 | 306 | state.remaining_prof_time = rpt[0] |
|
308 | 307 | elif lastmechanism == 'thread': |
|
309 | 308 | stopthread.set() |
|
310 | 309 | state.thread.join() |
|
311 | 310 | |
|
312 | 311 | state.accumulate_time(clock()) |
|
313 | 312 | state.last_start_time = None |
|
314 | 313 | statprofpath = os.environ.get('STATPROF_DEST') |
|
315 | 314 | save_data(statprofpath) |
|
316 | 315 | |
|
317 | 316 | def save_data(path=None): |
|
318 | 317 | try: |
|
319 | 318 | path = path or (os.environ['HOME'] + '/statprof.data') |
|
320 | 319 | file = open(path, "w+") |
|
321 | 320 | |
|
322 | 321 | file.write(str(state.accumulated_time) + '\n') |
|
323 | 322 | for sample in state.samples: |
|
324 | 323 | time = str(sample.time) |
|
325 | 324 | stack = sample.stack |
|
326 | 325 | sites = ['\1'.join([s.path, str(s.lineno), s.function]) |
|
327 | 326 | for s in stack] |
|
328 | 327 | file.write(time + '\0' + '\0'.join(sites) + '\n') |
|
329 | 328 | |
|
330 | 329 | file.close() |
|
331 |
except (IOError, OSError) |
|
|
330 | except (IOError, OSError): | |
|
332 | 331 | # The home directory probably didn't exist, or wasn't writable. Oh well. |
|
333 | 332 | pass |
|
334 | 333 | |
|
335 | 334 | def load_data(path=None): |
|
336 | 335 | path = path or (os.environ['HOME'] + '/statprof.data') |
|
337 | 336 | lines = open(path, 'r').read().splitlines() |
|
338 | 337 | |
|
339 | 338 | state.accumulated_time = float(lines[0]) |
|
340 | 339 | state.samples = [] |
|
341 | 340 | for line in lines[1:]: |
|
342 | 341 | parts = line.split('\0') |
|
343 | 342 | time = float(parts[0]) |
|
344 | 343 | rawsites = parts[1:] |
|
345 | 344 | sites = [] |
|
346 | 345 | for rawsite in rawsites: |
|
347 | 346 | siteparts = rawsite.split('\1') |
|
348 | 347 | sites.append(CodeSite.get(siteparts[0], int(siteparts[1]), |
|
349 | 348 | siteparts[2])) |
|
350 | 349 | |
|
351 | 350 | state.samples.append(Sample(sites, time)) |
|
352 | 351 | |
|
353 | 352 | |
|
354 | 353 | |
|
355 | 354 | def reset(frequency=None): |
|
356 | 355 | '''Clear out the state of the profiler. Do not call while the |
|
357 | 356 | profiler is running. |
|
358 | 357 | |
|
359 | 358 | The optional frequency argument specifies the number of samples to |
|
360 | 359 | collect per second.''' |
|
361 | 360 | assert state.profile_level == 0, "Can't reset() while statprof is running" |
|
362 | 361 | CodeSite.cache.clear() |
|
363 | 362 | state.reset(frequency) |
|
364 | 363 | |
|
365 | 364 | |
|
366 | 365 | @contextmanager |
|
367 | 366 | def profile(): |
|
368 | 367 | start() |
|
369 | 368 | try: |
|
370 | 369 | yield |
|
371 | 370 | finally: |
|
372 | 371 | stop() |
|
373 | 372 | display() |
|
374 | 373 | |
|
375 | 374 | |
|
376 | 375 | ########################################################################### |
|
377 | 376 | ## Reporting API |
|
378 | 377 | |
|
379 | 378 | class SiteStats(object): |
|
380 | 379 | def __init__(self, site): |
|
381 | 380 | self.site = site |
|
382 | 381 | self.selfcount = 0 |
|
383 | 382 | self.totalcount = 0 |
|
384 | 383 | |
|
385 | 384 | def addself(self): |
|
386 | 385 | self.selfcount += 1 |
|
387 | 386 | |
|
388 | 387 | def addtotal(self): |
|
389 | 388 | self.totalcount += 1 |
|
390 | 389 | |
|
391 | 390 | def selfpercent(self): |
|
392 | 391 | return self.selfcount / len(state.samples) * 100 |
|
393 | 392 | |
|
394 | 393 | def totalpercent(self): |
|
395 | 394 | return self.totalcount / len(state.samples) * 100 |
|
396 | 395 | |
|
397 | 396 | def selfseconds(self): |
|
398 | 397 | return self.selfcount * state.seconds_per_sample() |
|
399 | 398 | |
|
400 | 399 | def totalseconds(self): |
|
401 | 400 | return self.totalcount * state.seconds_per_sample() |
|
402 | 401 | |
|
403 | 402 | @classmethod |
|
404 | 403 | def buildstats(cls, samples): |
|
405 | 404 | stats = {} |
|
406 | 405 | |
|
407 | 406 | for sample in samples: |
|
408 | 407 | for i, site in enumerate(sample.stack): |
|
409 | 408 | sitestat = stats.get(site) |
|
410 | 409 | if not sitestat: |
|
411 | 410 | sitestat = SiteStats(site) |
|
412 | 411 | stats[site] = sitestat |
|
413 | 412 | |
|
414 | 413 | sitestat.addtotal() |
|
415 | 414 | |
|
416 | 415 | if i == 0: |
|
417 | 416 | sitestat.addself() |
|
418 | 417 | |
|
419 | 418 | return [s for s in stats.itervalues()] |
|
420 | 419 | |
|
421 | 420 | class DisplayFormats: |
|
422 | 421 | ByLine = 0 |
|
423 | 422 | ByMethod = 1 |
|
424 | 423 | AboutMethod = 2 |
|
425 | 424 | Hotpath = 3 |
|
426 | 425 | FlameGraph = 4 |
|
427 | 426 | Json = 5 |
|
428 | 427 | |
|
429 | 428 | def display(fp=None, format=3, **kwargs): |
|
430 | 429 | '''Print statistics, either to stdout or the given file object.''' |
|
431 | 430 | |
|
432 | 431 | if fp is None: |
|
433 | 432 | import sys |
|
434 | 433 | fp = sys.stdout |
|
435 | 434 | if len(state.samples) == 0: |
|
436 | 435 | print >> fp, ('No samples recorded.') |
|
437 | 436 | return |
|
438 | 437 | |
|
439 | 438 | if format == DisplayFormats.ByLine: |
|
440 | 439 | display_by_line(fp) |
|
441 | 440 | elif format == DisplayFormats.ByMethod: |
|
442 | 441 | display_by_method(fp) |
|
443 | 442 | elif format == DisplayFormats.AboutMethod: |
|
444 | 443 | display_about_method(fp, **kwargs) |
|
445 | 444 | elif format == DisplayFormats.Hotpath: |
|
446 | 445 | display_hotpath(fp, **kwargs) |
|
447 | 446 | elif format == DisplayFormats.FlameGraph: |
|
448 | 447 | write_to_flame(fp, **kwargs) |
|
449 | 448 | elif format == DisplayFormats.Json: |
|
450 | 449 | write_to_json(fp) |
|
451 | 450 | else: |
|
452 | 451 | raise Exception("Invalid display format") |
|
453 | 452 | |
|
454 | 453 | if format != DisplayFormats.Json: |
|
455 | 454 | print >> fp, ('---') |
|
456 | 455 | print >> fp, ('Sample count: %d' % len(state.samples)) |
|
457 | 456 | print >> fp, ('Total time: %f seconds' % state.accumulated_time) |
|
458 | 457 | |
|
459 | 458 | def display_by_line(fp): |
|
460 | 459 | '''Print the profiler data with each sample line represented |
|
461 | 460 | as one row in a table. Sorted by self-time per line.''' |
|
462 | 461 | stats = SiteStats.buildstats(state.samples) |
|
463 | 462 | stats.sort(reverse=True, key=lambda x: x.selfseconds()) |
|
464 | 463 | |
|
465 | 464 | print >> fp, ('%5.5s %10.10s %7.7s %-8.8s' % |
|
466 | 465 | ('% ', 'cumulative', 'self', '')) |
|
467 | 466 | print >> fp, ('%5.5s %9.9s %8.8s %-8.8s' % |
|
468 | 467 | ("time", "seconds", "seconds", "name")) |
|
469 | 468 | |
|
470 | 469 | for stat in stats: |
|
471 | 470 | site = stat.site |
|
472 | 471 | sitelabel = '%s:%d:%s' % (site.filename(), site.lineno, site.function) |
|
473 | 472 | print >> fp, ('%6.2f %9.2f %9.2f %s' % (stat.selfpercent(), |
|
474 | 473 | stat.totalseconds(), |
|
475 | 474 | stat.selfseconds(), |
|
476 | 475 | sitelabel)) |
|
477 | 476 | |
|
478 | 477 | def display_by_method(fp): |
|
479 | 478 | '''Print the profiler data with each sample function represented |
|
480 | 479 | as one row in a table. Important lines within that function are |
|
481 | 480 | output as nested rows. Sorted by self-time per line.''' |
|
482 | 481 | print >> fp, ('%5.5s %10.10s %7.7s %-8.8s' % |
|
483 | 482 | ('% ', 'cumulative', 'self', '')) |
|
484 | 483 | print >> fp, ('%5.5s %9.9s %8.8s %-8.8s' % |
|
485 | 484 | ("time", "seconds", "seconds", "name")) |
|
486 | 485 | |
|
487 | 486 | stats = SiteStats.buildstats(state.samples) |
|
488 | 487 | |
|
489 | 488 | grouped = defaultdict(list) |
|
490 | 489 | for stat in stats: |
|
491 | 490 | grouped[stat.site.filename() + ":" + stat.site.function].append(stat) |
|
492 | 491 | |
|
493 | 492 | # compute sums for each function |
|
494 | 493 | functiondata = [] |
|
495 | 494 | for fname, sitestats in grouped.iteritems(): |
|
496 | 495 | total_cum_sec = 0 |
|
497 | 496 | total_self_sec = 0 |
|
498 | 497 | total_percent = 0 |
|
499 | 498 | for stat in sitestats: |
|
500 | 499 | total_cum_sec += stat.totalseconds() |
|
501 | 500 | total_self_sec += stat.selfseconds() |
|
502 | 501 | total_percent += stat.selfpercent() |
|
503 | 502 | |
|
504 | 503 | functiondata.append((fname, |
|
505 | 504 | total_cum_sec, |
|
506 | 505 | total_self_sec, |
|
507 | 506 | total_percent, |
|
508 | 507 | sitestats)) |
|
509 | 508 | |
|
510 | 509 | # sort by total self sec |
|
511 | 510 | functiondata.sort(reverse=True, key=lambda x: x[2]) |
|
512 | 511 | |
|
513 | 512 | for function in functiondata: |
|
514 | 513 | if function[3] < 0.05: |
|
515 | 514 | continue |
|
516 | 515 | print >> fp, ('%6.2f %9.2f %9.2f %s' % (function[3], # total percent |
|
517 | 516 | function[1], # total cum sec |
|
518 | 517 | function[2], # total self sec |
|
519 | 518 | function[0])) # file:function |
|
520 | 519 | function[4].sort(reverse=True, key=lambda i: i.selfseconds()) |
|
521 | 520 | for stat in function[4]: |
|
522 | 521 | # only show line numbers for significant locations (>1% time spent) |
|
523 | 522 | if stat.selfpercent() > 1: |
|
524 | 523 | source = stat.site.getsource(25) |
|
525 | 524 | stattuple = (stat.selfpercent(), stat.selfseconds(), |
|
526 | 525 | stat.site.lineno, source) |
|
527 | 526 | |
|
528 | 527 | print >> fp, ('%33.0f%% %6.2f line %s: %s' % (stattuple)) |
|
529 | 528 | |
|
530 | 529 | def display_about_method(fp, function=None, **kwargs): |
|
531 | 530 | if function is None: |
|
532 | 531 | raise Exception("Invalid function") |
|
533 | 532 | |
|
534 | 533 | filename = None |
|
535 | 534 | if ':' in function: |
|
536 | 535 | filename, function = function.split(':') |
|
537 | 536 | |
|
538 | 537 | relevant_samples = 0 |
|
539 | 538 | parents = {} |
|
540 | 539 | children = {} |
|
541 | 540 | |
|
542 | 541 | for sample in state.samples: |
|
543 | 542 | for i, site in enumerate(sample.stack): |
|
544 | 543 | if site.function == function and (not filename |
|
545 | 544 | or site.filename() == filename): |
|
546 | 545 | relevant_samples += 1 |
|
547 | 546 | if i != len(sample.stack) - 1: |
|
548 | 547 | parent = sample.stack[i + 1] |
|
549 | 548 | if parent in parents: |
|
550 | 549 | parents[parent] = parents[parent] + 1 |
|
551 | 550 | else: |
|
552 | 551 | parents[parent] = 1 |
|
553 | 552 | |
|
554 | 553 | if site in children: |
|
555 | 554 | children[site] = children[site] + 1 |
|
556 | 555 | else: |
|
557 | 556 | children[site] = 1 |
|
558 | 557 | |
|
559 | 558 | parents = [(parent, count) for parent, count in parents.iteritems()] |
|
560 | 559 | parents.sort(reverse=True, key=lambda x: x[1]) |
|
561 | 560 | for parent, count in parents: |
|
562 | 561 | print >> fp, ('%6.2f%% %s:%s line %s: %s' % |
|
563 | 562 | (count / relevant_samples * 100, parent.filename(), |
|
564 | 563 | parent.function, parent.lineno, parent.getsource(50))) |
|
565 | 564 | |
|
566 | 565 | stats = SiteStats.buildstats(state.samples) |
|
567 | 566 | stats = [s for s in stats |
|
568 | 567 | if s.site.function == function and |
|
569 | 568 | (not filename or s.site.filename() == filename)] |
|
570 | 569 | |
|
571 | 570 | total_cum_sec = 0 |
|
572 | 571 | total_self_sec = 0 |
|
573 | 572 | total_self_percent = 0 |
|
574 | 573 | total_cum_percent = 0 |
|
575 | 574 | for stat in stats: |
|
576 | 575 | total_cum_sec += stat.totalseconds() |
|
577 | 576 | total_self_sec += stat.selfseconds() |
|
578 | 577 | total_self_percent += stat.selfpercent() |
|
579 | 578 | total_cum_percent += stat.totalpercent() |
|
580 | 579 | |
|
581 | 580 | print >> fp, ( |
|
582 | 581 | '\n %s:%s Total: %0.2fs (%0.2f%%) Self: %0.2fs (%0.2f%%)\n' % |
|
583 | 582 | ( |
|
584 | 583 | filename or '___', |
|
585 | 584 | function, |
|
586 | 585 | total_cum_sec, |
|
587 | 586 | total_cum_percent, |
|
588 | 587 | total_self_sec, |
|
589 | 588 | total_self_percent |
|
590 | 589 | )) |
|
591 | 590 | |
|
592 | 591 | children = [(child, count) for child, count in children.iteritems()] |
|
593 | 592 | children.sort(reverse=True, key=lambda x: x[1]) |
|
594 | 593 | for child, count in children: |
|
595 | 594 | print >> fp, (' %6.2f%% line %s: %s' % |
|
596 | 595 | (count / relevant_samples * 100, child.lineno, child.getsource(50))) |
|
597 | 596 | |
|
598 | 597 | def display_hotpath(fp, limit=0.05, **kwargs): |
|
599 | 598 | class HotNode(object): |
|
600 | 599 | def __init__(self, site): |
|
601 | 600 | self.site = site |
|
602 | 601 | self.count = 0 |
|
603 | 602 | self.children = {} |
|
604 | 603 | |
|
605 | 604 | def add(self, stack, time): |
|
606 | 605 | self.count += time |
|
607 | 606 | site = stack[0] |
|
608 | 607 | child = self.children.get(site) |
|
609 | 608 | if not child: |
|
610 | 609 | child = HotNode(site) |
|
611 | 610 | self.children[site] = child |
|
612 | 611 | |
|
613 | 612 | if len(stack) > 1: |
|
614 | 613 | i = 1 |
|
615 | 614 | # Skip boiler plate parts of the stack |
|
616 | 615 | while i < len(stack) and '%s:%s' % (stack[i].filename(), stack[i].function) in skips: |
|
617 | 616 | i += 1 |
|
618 | 617 | if i < len(stack): |
|
619 | 618 | child.add(stack[i:], time) |
|
620 | 619 | |
|
621 | 620 | root = HotNode(None) |
|
622 | 621 | lasttime = state.samples[0].time |
|
623 | 622 | for sample in state.samples: |
|
624 | 623 | root.add(sample.stack[::-1], sample.time - lasttime) |
|
625 | 624 | lasttime = sample.time |
|
626 | 625 | |
|
627 | 626 | def _write(node, depth, multiple_siblings): |
|
628 | 627 | site = node.site |
|
629 | 628 | visiblechildren = [c for c in node.children.itervalues() |
|
630 | 629 | if c.count >= (limit * root.count)] |
|
631 | 630 | if site: |
|
632 | 631 | indent = depth * 2 - 1 |
|
633 | 632 | filename = '' |
|
634 | 633 | function = '' |
|
635 | 634 | if len(node.children) > 0: |
|
636 | 635 | childsite = list(node.children.itervalues())[0].site |
|
637 | 636 | filename = (childsite.filename() + ':').ljust(15) |
|
638 | 637 | function = childsite.function |
|
639 | 638 | |
|
640 | 639 | # lots of string formatting |
|
641 | 640 | listpattern = ''.ljust(indent) +\ |
|
642 | 641 | ('\\' if multiple_siblings else '|') +\ |
|
643 | 642 | ' %4.1f%% %s %s' |
|
644 | 643 | liststring = listpattern % (node.count / root.count * 100, |
|
645 | 644 | filename, function) |
|
646 | 645 | codepattern = '%' + str(55 - len(liststring)) + 's %s: %s' |
|
647 | 646 | codestring = codepattern % ('line', site.lineno, site.getsource(30)) |
|
648 | 647 | |
|
649 | 648 | finalstring = liststring + codestring |
|
650 | 649 | childrensamples = sum([c.count for c in node.children.itervalues()]) |
|
651 | 650 | # Make frames that performed more than 10% of the operation red |
|
652 | 651 | if node.count - childrensamples > (0.1 * root.count): |
|
653 | 652 | finalstring = '\033[91m' + finalstring + '\033[0m' |
|
654 | 653 | # Make frames that didn't actually perform work dark grey |
|
655 | 654 | elif node.count - childrensamples == 0: |
|
656 | 655 | finalstring = '\033[90m' + finalstring + '\033[0m' |
|
657 | 656 | print >> fp, finalstring |
|
658 | 657 | |
|
659 | 658 | newdepth = depth |
|
660 | 659 | if len(visiblechildren) > 1 or multiple_siblings: |
|
661 | 660 | newdepth += 1 |
|
662 | 661 | |
|
663 | 662 | visiblechildren.sort(reverse=True, key=lambda x: x.count) |
|
664 | 663 | for child in visiblechildren: |
|
665 | 664 | _write(child, newdepth, len(visiblechildren) > 1) |
|
666 | 665 | |
|
667 | 666 | if root.count > 0: |
|
668 | 667 | _write(root, 0, False) |
|
669 | 668 | |
|
670 | 669 | def write_to_flame(fp, scriptpath=None, outputfile=None, **kwargs): |
|
671 | 670 | if scriptpath is None: |
|
672 | 671 | scriptpath = os.environ['HOME'] + '/flamegraph.pl' |
|
673 | 672 | if not os.path.exists(scriptpath): |
|
674 | 673 | print >> fp, "error: missing %s" % scriptpath |
|
675 | 674 | print >> fp, "get it here: https://github.com/brendangregg/FlameGraph" |
|
676 | 675 | return |
|
677 | 676 | |
|
678 | 677 | fd, path = tempfile.mkstemp() |
|
679 | 678 | |
|
680 | 679 | file = open(path, "w+") |
|
681 | 680 | |
|
682 | 681 | lines = {} |
|
683 | 682 | for sample in state.samples: |
|
684 | 683 | sites = [s.function for s in sample.stack] |
|
685 | 684 | sites.reverse() |
|
686 | 685 | line = ';'.join(sites) |
|
687 | 686 | if line in lines: |
|
688 | 687 | lines[line] = lines[line] + 1 |
|
689 | 688 | else: |
|
690 | 689 | lines[line] = 1 |
|
691 | 690 | |
|
692 | 691 | for line, count in lines.iteritems(): |
|
693 | 692 | file.write("%s %s\n" % (line, count)) |
|
694 | 693 | |
|
695 | 694 | file.close() |
|
696 | 695 | |
|
697 | 696 | if outputfile is None: |
|
698 | 697 | outputfile = '~/flamegraph.svg' |
|
699 | 698 | |
|
700 | 699 | os.system("perl ~/flamegraph.pl %s > %s" % (path, outputfile)) |
|
701 | 700 | print "Written to %s" % outputfile |
|
702 | 701 | |
|
703 | 702 | def write_to_json(fp): |
|
704 | 703 | samples = [] |
|
705 | 704 | |
|
706 | 705 | for sample in state.samples: |
|
707 | 706 | stack = [] |
|
708 | 707 | |
|
709 | 708 | for frame in sample.stack: |
|
710 | 709 | stack.append((frame.path, frame.lineno, frame.function)) |
|
711 | 710 | |
|
712 | 711 | samples.append((sample.time, stack)) |
|
713 | 712 | |
|
714 | 713 | print >> fp, json.dumps(samples) |
|
715 | 714 | |
|
716 | 715 | def printusage(): |
|
717 | 716 | print """ |
|
718 | 717 | The statprof command line allows you to inspect the last profile's results in |
|
719 | 718 | the following forms: |
|
720 | 719 | |
|
721 | 720 | usage: |
|
722 | 721 | hotpath [-l --limit percent] |
|
723 | 722 | Shows a graph of calls with the percent of time each takes. |
|
724 | 723 | Red calls take over 10%% of the total time themselves. |
|
725 | 724 | lines |
|
726 | 725 | Shows the actual sampled lines. |
|
727 | 726 | functions |
|
728 | 727 | Shows the samples grouped by function. |
|
729 | 728 | function [filename:]functionname |
|
730 | 729 | Shows the callers and callees of a particular function. |
|
731 | 730 | flame [-s --script-path] [-o --output-file path] |
|
732 | 731 | Writes out a flamegraph to output-file (defaults to ~/flamegraph.svg) |
|
733 | 732 | Requires that ~/flamegraph.pl exist. |
|
734 | 733 | (Specify alternate script path with --script-path.)""" |
|
735 | 734 | |
|
736 | 735 | def main(argv=None): |
|
737 | 736 | if argv is None: |
|
738 | 737 | argv = sys.argv |
|
739 | 738 | |
|
740 | 739 | if len(argv) == 1: |
|
741 | 740 | printusage() |
|
742 | 741 | return 0 |
|
743 | 742 | |
|
744 | 743 | displayargs = {} |
|
745 | 744 | |
|
746 | 745 | optstart = 2 |
|
747 | 746 | displayargs['function'] = None |
|
748 | 747 | if argv[1] == 'hotpath': |
|
749 | 748 | displayargs['format'] = DisplayFormats.Hotpath |
|
750 | 749 | elif argv[1] == 'lines': |
|
751 | 750 | displayargs['format'] = DisplayFormats.ByLine |
|
752 | 751 | elif argv[1] == 'functions': |
|
753 | 752 | displayargs['format'] = DisplayFormats.ByMethod |
|
754 | 753 | elif argv[1] == 'function': |
|
755 | 754 | displayargs['format'] = DisplayFormats.AboutMethod |
|
756 | 755 | displayargs['function'] = argv[2] |
|
757 | 756 | optstart = 3 |
|
758 | 757 | elif argv[1] == 'flame': |
|
759 | 758 | displayargs['format'] = DisplayFormats.FlameGraph |
|
760 | 759 | else: |
|
761 | 760 | printusage() |
|
762 | 761 | return 0 |
|
763 | 762 | |
|
764 | 763 | # process options |
|
765 | 764 | try: |
|
766 | 765 | opts, args = getopt.getopt(sys.argv[optstart:], "hl:f:o:p:", |
|
767 | 766 | ["help", "limit=", "file=", "output-file=", "script-path="]) |
|
768 | 767 | except getopt.error as msg: |
|
769 | 768 | print msg |
|
770 | 769 | printusage() |
|
771 | 770 | return 2 |
|
772 | 771 | |
|
773 | 772 | displayargs['limit'] = 0.05 |
|
774 | 773 | path = None |
|
775 | 774 | for o, value in opts: |
|
776 | 775 | if o in ("-l", "--limit"): |
|
777 | 776 | displayargs['limit'] = float(value) |
|
778 | 777 | elif o in ("-f", "--file"): |
|
779 | 778 | path = value |
|
780 | 779 | elif o in ("-o", "--output-file"): |
|
781 | 780 | displayargs['outputfile'] = value |
|
782 | 781 | elif o in ("-p", "--script-path"): |
|
783 | 782 | displayargs['scriptpath'] = value |
|
784 | 783 | elif o in ("-h", "help"): |
|
785 | 784 | printusage() |
|
786 | 785 | return 0 |
|
787 | 786 | else: |
|
788 | 787 | assert False, "unhandled option %s" % o |
|
789 | 788 | |
|
790 | 789 | load_data(path=path) |
|
791 | 790 | |
|
792 | 791 | display(**displayargs) |
|
793 | 792 | |
|
794 | 793 | return 0 |
|
795 | 794 | |
|
796 | 795 | if __name__ == "__main__": |
|
797 | 796 | sys.exit(main()) |
General Comments 0
You need to be logged in to leave comments.
Login now