forked from zooko/pyutil
-
Notifications
You must be signed in to change notification settings - Fork 3
/
memutil.py
589 lines (522 loc) · 20.3 KB
/
memutil.py
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
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
# -*- coding: utf-8; fill-column: 77 -*-
# -*- indent-tabs-mode: nil -*-
# This file is part of pyutil; see README.rst for licensing terms.
from __future__ import print_function
# from the Python Standard Library
import gc, math, operator, os, sys, types
try:
StandardError # PY2
except NameError:
StandardError = Exception # PY3
# from the pyutil library
from .assertutil import precondition
from . import mathutil
from . import dictutil
class Canary:
"""
Want to get a printout when your object is garbage collected? Then put "self.canary = Canary(self)" in your object's constructor.
"""
def __init__(self, owner):
self.ownerdesc = repr(owner)
def __del__(self):
print("Canary says that %s is gone." % self.ownerdesc)
def estimate_mem_of_obj(o):
# assumes 32-bit CPUs...
PY_STRUCT_HEAD_LEN=4
if hasattr(o, '__len__'):
if isinstance(o, str):
return PY_STRUCT_HEAD_LEN + o.__len__() * 1
if isinstance(o, unicode):
return PY_STRUCT_HEAD_LEN + o.__len__() * 4 # 4 depends on implementation and is approximate
if isinstance(o, (tuple, list,)):
return PY_STRUCT_HEAD_LEN + o.__len__() * 4
if isinstance(o, (dict, set,)):
return PY_STRUCT_HEAD_LEN + o.__len__() * 4 * 2 * 2 # approximate
if isinstance(o, int):
return PY_STRUCT_HEAD_LEN + 4
if isinstance(o, long):
return PY_STRUCT_HEAD_LEN + 4
if o < 1:
return PY_STRUCT_HEAD_LEN
else:
return PY_STRUCT_HEAD_LEN + math.log(o) / 5 # the 5 was empirically determined (it is approximate)
if isinstance(o, float):
return PY_STRUCT_HEAD_LEN + 8
# Uh-oh... I wonder what we are missing here...
return PY_STRUCT_HEAD_LEN
def check_for_obj_leakage(f, *args, **kwargs):
"""
The idea is that I am going to invoke f(), then run gc.collect(), then run
gc.get_objects() to get a complete list of all objects in the system, then
invoke f() a second time, then run gc.collect(), then run gc.get_objects()
to get a list of all the objects *now* in the system.
Then I return a tuple two things: the first element of the tuple is the
difference between the number of objects in the second list and the number
of objects in the first list.
I.e., if this number is zero then you can be pretty sure there is no memory
leak, unless f is deleting some objects and replacing them by exactly the
same number of objects but the new objects take up more memory. If this
number is greater than zero then you can pretty sure there is a memory
leak, unless f is doing some memoization/caching behavior and it will
eventually stabilize, which you can detect by running
check_for_obj_leakage() more times and seeing if it stabilizes.
(Actually we run f() followed by gc.collect() one time before we start in
order to account for any static objects which are created the first time
you run f() and then re-used after that.)
The second element in the return value is the set of all objects which were
present in the second list and not in the first. Some of these objects
might be memory-leaked objects, or perhaps f deleted some objects and
replaced them with equivalent objects, in which case these objects are not
leaked.
(We actually invoke gc.collect() three times in a row in case there are
objects which get collected in the first pass that have finalizers which
create new reference-cycled objects... "3" is a superstitious number -- we
figure most of the time the finalizers of the things produced by the first
round of finalizers won't themselves product another round of
reference-cycled objects.)
"""
f()
gc.collect();gc.collect();gc.collect()
f()
gc.collect();gc.collect();gc.collect()
r1 = gc.get_objects()
f()
gc.collect();gc.collect();gc.collect()
r2 = gc.get_objects()
d2 = dict([(id(x), x) for x in r2])
# Now remove everything from r1, and r1 itself, from d2.
del d2[id(r1)]
for o in r1:
if id(o) in d2:
del d2[id(o)]
return (len(r2) - len(r1) - 1, d2)
def measure_obj_leakage(f, numsamples=2**7, iterspersample=2**4, *args, **kwargs):
"""
The idea is we are going to use count_all_objects() to see how many
objects are in use, and keep track of that number with respect to how
many times we've invoked f(), and return the slope of the best linear
fit.
@param numsamples: recommended: 2**7
@param iterspersample: how many times f() should be invoked per sample;
Basically, choose iterspersample such that
iterspersample * numsamples *
how-long-it-takes-to-compute-f() is slightly less
than how long you are willing to wait for this
leak test.
@return: the slope of the best linear fit, which can be interpreted as 'the
approximate number of Python objects created and not destroyed
per invocation of f()'
"""
precondition(numsamples > 0, "numsamples is required to be positive.", numsamples)
precondition(iterspersample > 0, "iterspersample is required to be positive.", iterspersample)
resiters = [None]*numsamples # values: iters
resnumobjs = [None]*numsamples # values: numobjs
totaliters = 0
for i in range(numsamples):
for j in range(iterspersample):
f(*args, **kwargs)
totaliters = totaliters + iterspersample
resiters[i] = totaliters
gc.collect()
resnumobjs[i] = count_all_objects()
# print "totaliters: %s, numobjs: %s" % (resiters[-1], resnumobjs[-1],)
avex = float(reduce(operator.__add__, resiters)) / len(resiters)
avey = float(reduce(operator.__add__, resnumobjs)) / len(resnumobjs)
sxy = reduce(operator.__add__, map(lambda a, avex=avex, avey=avey: (a[0] - avex) * (a[1] - avey), zip(resiters, resnumobjs)))
sxx = reduce(operator.__add__, map(lambda a, avex=avex: (a - avex) ** 2, resiters))
return sxy / sxx
def linear_fit_slope(xs, ys):
avex = float(reduce(operator.__add__, xs)) / len(xs)
avey = float(reduce(operator.__add__, ys)) / len(ys)
sxy = reduce(operator.__add__, map(lambda a, avex=avex, avey=avey: (a[0] - avex) * (a[1] - avey), zip(xs, ys)))
sxx = reduce(operator.__add__, map(lambda a, avex=avex: (a - avex) ** 2, xs))
return sxy / sxx
def measure_ref_leakage(f, numsamples=2**7, iterspersample=2**4, *args, **kwargs):
"""
The idea is we are going to use sys.gettotalrefcount() to see how many
references are extant, and keep track of that number with respect to how
many times we've invoked f(), and return the slope of the best linear
fit.
@param numsamples: recommended: 2**7
@param iterspersample: how many times f() should be invoked per sample;
Basically, choose iterspersample such that
iterspersample * numsamples *
how-long-it-takes-to-compute-f() is slightly less
than how long you are willing to wait for this
leak test.
@return: the slope of the best linear fit, which can be interpreted as 'the
approximate number of Python references created and not
nullified per invocation of f()'
"""
precondition(numsamples > 0, "numsamples is required to be positive.", numsamples)
precondition(iterspersample > 0, "iterspersample is required to be positive.", iterspersample)
try:
sys.gettotalrefcount()
except AttributeError as le:
raise AttributeError(le, "Probably this is not a debug build of Python, so it doesn't have a sys.gettotalrefcount function.")
resiters = [None]*numsamples # values: iters
resnumrefs = [None]*numsamples # values: numrefs
totaliters = 0
for i in range(numsamples):
for j in range(iterspersample):
f(*args, **kwargs)
totaliters = totaliters + iterspersample
resiters[i] = totaliters
gc.collect()
resnumrefs[i] = sys.gettotalrefcount()
# print "totaliters: %s, numrefss: %s" % (resiters[-1], resnumrefs[-1],)
avex = float(reduce(operator.__add__, resiters)) / len(resiters)
avey = float(reduce(operator.__add__, resnumrefs)) / len(resnumrefs)
sxy = reduce(operator.__add__, map(lambda a, avex=avex, avey=avey: (a[0] - avex) * (a[1] - avey), zip(resiters, resnumrefs)))
sxx = reduce(operator.__add__, map(lambda a, avex=avex: (a - avex) ** 2, resiters))
return sxy / sxx
class NotSupportedException(StandardError):
"""
Just an exception class. It is thrown by get_mem_usage if the OS does
not support the operation.
"""
pass
def get_mem_used():
"""
This only works on Linux, and only if the /proc/$PID/statm output is the
same as that in linux kernel 2.6. Also `os.getpid()' must work.
@return: tuple of (res, virt) used by this process
"""
try:
import resource
except ImportError:
raise NotSupportedException()
# sample output from cat /proc/$PID/statm:
# 14317 3092 832 279 0 2108 0
a = os.popen("cat /proc/%s/statm 2>/dev/null" % os.getpid()).read().split() #nosec
if not a:
raise NotSupportedException()
return (int(a[1]) * resource.getpagesize(), int(a[0]) * resource.getpagesize(),)
def get_mem_used_res():
"""
This only works on Linux, and only if the /proc/$PID/statm output is the
same as that in linux kernel 2.6. Also `os.getpid()' must work.
"""
try:
import resource
except ImportError:
raise NotSupportedException()
# sample output from cat /proc/$PID/statm:
# 14317 3092 832 279 0 2108 0
a = os.popen("cat /proc/%s/statm" % os.getpid()).read().split() #nosec
if not len(a) > 1:
raise NotSupportedException()
return int(a[1]) * resource.getpagesize()
def get_mem_usage_virt_and_res():
"""
This only works on Linux, and only if the /proc/$PID/statm output is the
same as that in linux kernel 2.6. Also `os.getpid()' must work.
"""
try:
import resource
except ImportError:
raise NotSupportedException()
# sample output from cat /proc/$PID/statm:
# 14317 3092 832 279 0 2108 0
a = os.popen("cat /proc/%s/statm" % os.getpid()).read().split() #nosec
if not len(a) > 1:
raise NotSupportedException()
return (int(a[0]) * resource.getpagesize(), int(a[1]) * resource.getpagesize(),)
class Measurer(object):
def __init__(self, f, numsamples=2**7, iterspersample=2**4, *args, **kwargs):
"""
@param f a callable; If it returns a deferred then the memory will not
be measured and the next iteration will not be started until the
deferred fires; else the memory will be measured and the next
iteration started when f returns.
"""
self.f = f
self.numsamples = numsamples
self.iterspersample = iterspersample
self.args = args
self.kwargs = kwargs
# from twisted
from twisted.internet import defer
self.d = defer.Deferred()
def when_complete(self):
return self.d
def _invoke(self):
d = self.f(*self.args, **self.kwargs)
# from twisted
from twisted.internet import defer
if isinstance(d, defer.Deferred):
d.addCallback(self._after)
else:
self._after(None)
def start(self):
self.resiters = [None]*self.numsamples # values: iters
self.resmemusage = [None]*self.numsamples # values: memusage
self.totaliters = 0
self.i = 0
self.j = 0
self._invoke()
def _after(self, o):
self.j += 1
if self.j < self.iterspersample:
self._invoke()
return
if self.i < self.numsamples:
self.j = 0
self.i += 1
self.totaliters += self.iterspersample
self.resiters[self.i] = self.totaliters
self.resmemusage[self.i] = get_mem_used_res()
self._invoke()
return
self.d.callback(mathutil.linear_fit_slope(zip(self.resiters, self.resmemusage)))
def measure_mem_leakage(f, numsamples=2**7, iterspersample=2**4, *args, **kwargs):
"""
This does the same thing as measure_obj_leakage() but instead of using
count_all_objects() it uses get_mem_usage(), which is currently
implemented for Linux and barely implemented for Mac OS X.
@param numsamples: recommended: 2**7
@param iterspersample: how many times `f()' should be invoked per sample;
Basically, choose `iterspersample' such that
(iterspersample * numsamples *
how-long-it-takes-to-compute-`f()') is slightly
less than how long you are willing to wait for
this leak test.
@return: the slope of the best linear fit, which can be interpreted as
'the approximate number of system bytes allocated and not freed
per invocation of f()'
"""
precondition(numsamples > 0, "numsamples is required to be positive.", numsamples)
precondition(iterspersample > 0, "iterspersample is required to be positive.", iterspersample)
resiters = [None]*numsamples # values: iters
resmemusage = [None]*numsamples # values: memusage
totaliters = 0
for i in range(numsamples):
for j in range(iterspersample):
f(*args, **kwargs)
totaliters = totaliters + iterspersample
resiters[i] = totaliters
gc.collect()
resmemusage[i] = get_mem_used_res()
# print("totaliters: %s, numobjs: %s" % (resiters[-1], resmemusage[-1],))
avex = float(reduce(operator.__add__, resiters)) / len(resiters)
avey = float(reduce(operator.__add__, resmemusage)) / len(resmemusage)
sxy = reduce(operator.__add__, map(lambda a, avex=avex, avey=avey: (a[0] - avex) * (a[1] - avey), zip(resiters, resmemusage)))
sxx = reduce(operator.__add__, map(lambda a, avex=avex: (a - avex) ** 2, resiters))
if sxx == 0:
return None
return sxy / sxx
def describe_object(o, FunctionType=types.FunctionType, MethodType=types.MethodType):
"""
For human analysis, when humans are attempting to understand where all the
memory is going. Argument o is an object, return value is a string
describing the object.
"""
sl = []
if isinstance(o, FunctionType):
try:
sl.append("<type 'function' %s>" % str(o.func_name))
except:
pass
elif isinstance(o, MethodType):
try:
sl.append("<type 'method' %s>" % str(o.im_func.func_name))
except:
pass
else:
sl.append(str(type(o)))
try:
sl.append(str(len(o)))
except:
pass
return ''.join(sl)
def describe_object_with_dict_details(o):
sl = []
sl.append(str(type(o)))
if isinstance(o, types.FunctionType):
try:
sl.append(str(o.func_name))
except:
pass
elif isinstance(o, types.MethodType):
try:
sl.append(str(o.im_func.func_name))
except:
pass
try:
sl.append(str(len(o)))
except:
pass
if isinstance(o, dict) and o:
sl.append('-')
nd = dictutil.NumDict()
for k, v in o.iteritems():
nd.inc((describe_object(k), describe_object(v),))
k, v = nd.item_with_largest_value()
sl.append("-")
iterator = o.iteritems()
k,v = iterator.next()
sl.append(describe_object(k))
sl.append(":")
sl.append(describe_object(v))
return ''.join(sl)
def describe_dict(o):
sl = ['<dict']
l = len(o)
sl.append(str(l))
if l:
sl.append("-")
iterator = o.iteritems()
firstitem=True
try:
while True:
if firstitem:
firstitem = False
else:
sl.append(", ")
k,v = iterator.next()
sl.append(describe_object(k))
sl.append(": ")
sl.append(describe_object(v))
except StopIteration:
pass
sl.append('>')
return ''.join(sl)
def count_all_objects():
ids = set()
ls = locals()
import inspect
cf = inspect.currentframe()
for o in gc.get_objects():
if o is ids or o is ls or o is cf:
continue
if not id(o) in ids:
ids.add(id(o))
for so in gc.get_referents(o):
if not id(so) in ids:
ids.add(id(so))
return len(ids)
def visit_all_objects(f):
"""
Brian and I *think* that this gets all objects. This is predicated on the
assumption that every object either participates in gc, or is at most one
hop from an object that participates in gc. This was Brian's clever idea.
"""
ids = set()
ls = locals()
import inspect
cf = inspect.currentframe()
for o in gc.get_objects():
if o is ids or o is ls or o is cf:
continue
if not id(o) in ids:
ids.add(id(o))
f(o)
for so in gc.get_referents(o):
if not id(so) in ids:
ids.add(id(so))
f(so)
def get_all_objects():
objs = []
def addit(o):
objs.append(o)
visit_all_objects(addit)
return objs
def describe_all_objects():
import dictutil
d = dictutil.NumDict()
for o in get_all_objects():
d.inc(describe_object(o))
return d
def dump_description_of_object(o, f):
f.write("%x" % (id(o),))
f.write("-")
f.write(describe_object(o))
f.write("\n")
def dump_description_of_object_refs(o, f):
# This holds the ids of all referents that we've already dumped.
dumped = set()
# First, any __dict__ items
try:
itemsiter = o.__dict__.iteritems()
except:
pass
else:
for k, v in itemsiter:
try:
idr = id(v)
if idr not in dumped:
dumped.add(idr)
f.write("%d:"%len(k))
f.write(k)
f.write(",")
f.write("%0x,"%idr)
except:
pass
# Then anything else that gc.get_referents() returns.
for r in gc.get_referents(o):
idr = id(r)
if idr not in dumped:
dumped.add(idr)
f.write("0:,%0x,"%idr)
def dump_descriptions_of_all_objects(f):
ids = set()
ls = locals()
for o in gc.get_objects():
if o is f or o is ids or o is ls:
continue
if not id(o) in ids:
ids.add(id(o))
dump_description_of_object(o, f)
for so in gc.get_referents(o):
if o is f or o is ids or o is ls:
continue
if not id(so) in ids:
ids.add(id(so))
dump_description_of_object(so, f)
ls = None # break reference cycle
return len(ids)
def dump_description_of_object_with_refs(o, f):
f.write("%0x" % (id(o),))
f.write("-")
desc = describe_object(o)
f.write("%d:"%len(desc))
f.write(desc)
f.write(",")
dump_description_of_object_refs(o, f)
f.write("\n")
def dump_descriptions_of_all_objects_with_refs(f):
ids = set()
ls = locals()
for o in gc.get_objects():
if o is f or o is ids or o is ls:
continue
if not id(o) in ids:
ids.add(id(o))
dump_description_of_object_with_refs(o, f)
for so in gc.get_referents(o):
if o is f or o is ids or o is ls:
continue
if not id(so) in ids:
ids.add(id(so))
dump_description_of_object_with_refs(so, f)
ls = None # break reference cycle
return len(ids)
import re
NRE = re.compile("[1-9][0-9]*$")
def undump_descriptions_of_all_objects(inf):
d = {}
for l in inf:
dash=l.find('-')
if dash == -1:
raise l
mo = NRE.search(l)
if mo:
typstr = l[dash+1:mo.start(0)]
num=int(mo.group(0))
if str(num) != mo.group(0):
raise mo.group(0)
else:
typstr = l[dash+1:]
num = None
d[l[:dash]] = (typstr, num,)
return d