| OLD | NEW |
|---|
| (Empty) |
| 1 #!/usr/bin/env python | |
| 2 # Copyright (c) 2012 The Chromium Authors. All rights reserved. | |
| 3 # Use of this source code is governed by a BSD-style license that can be | |
| 4 # found in the LICENSE file. | |
| 5 | |
| 6 """The deep heap profiler script for Chrome.""" | |
| 7 | |
| 8 from datetime import datetime | |
| 9 import json | |
| 10 import os | |
| 11 import re | |
| 12 import subprocess | |
| 13 import sys | |
| 14 import tempfile | |
| 15 | |
| 16 BUCKET_ID = 5 | |
| 17 VIRTUAL = 0 | |
| 18 COMMITTED = 1 | |
| 19 ALLOC_COUNT = 2 | |
| 20 FREE_COUNT = 3 | |
| 21 NULL_REGEX = re.compile('') | |
| 22 PPROF_PATH = os.path.join(os.path.dirname(__file__), | |
| 23 os.pardir, | |
| 24 os.pardir, | |
| 25 'third_party', | |
| 26 'tcmalloc', | |
| 27 'chromium', | |
| 28 'src', | |
| 29 'pprof') | |
| 30 | |
| 31 # Heap Profile Dump versions | |
| 32 | |
| 33 # DUMP_DEEP_1 DOES NOT distinct mmap regions and malloc chunks. | |
| 34 # Their stacktraces DO contain mmap* or tc-* at their tops. | |
| 35 # They should be processed by POLICY_DEEP_1. | |
| 36 DUMP_DEEP_1 = 'DUMP_DEEP_1' | |
| 37 | |
| 38 # DUMP_DEEP_2 DOES distinct mmap regions and malloc chunks. | |
| 39 # Their stacktraces still DO contain mmap* or tc-*. | |
| 40 # They should be processed by POLICY_DEEP_1. | |
| 41 DUMP_DEEP_2 = 'DUMP_DEEP_2' | |
| 42 | |
| 43 # DUMP_DEEP_3 DOES distinct mmap regions and malloc chunks. | |
| 44 # Their stacktraces DO NOT contain mmap* or tc-*. | |
| 45 # They should be processed by POLICY_DEEP_2. | |
| 46 DUMP_DEEP_3 = 'DUMP_DEEP_3' | |
| 47 | |
| 48 # Heap Profile Policy versions | |
| 49 | |
| 50 # POLICY_DEEP_1 DOES NOT include allocation_type columns. | |
| 51 # mmap regions are distincted w/ mmap frames in the pattern column. | |
| 52 POLICY_DEEP_1 = 'POLICY_DEEP_1' | |
| 53 | |
| 54 # POLICY_DEEP_2 DOES include allocation_type columns. | |
| 55 # mmap regions are distincted w/ the allocation_type column. | |
| 56 POLICY_DEEP_2 = 'POLICY_DEEP_2' | |
| 57 | |
| 58 # TODO(dmikurube): Avoid global variables. | |
| 59 address_symbol_dict = {} | |
| 60 appeared_addresses = set() | |
| 61 components = [] | |
| 62 | |
| 63 | |
| 64 class Policy(object): | |
| 65 | |
| 66 def __init__(self, name, mmap, pattern): | |
| 67 self.name = name | |
| 68 self.mmap = mmap | |
| 69 self.condition = re.compile(pattern + r'\Z') | |
| 70 | |
| 71 | |
| 72 def get_component(policy_list, bucket, mmap): | |
| 73 """Returns a component name which a given bucket belongs to. | |
| 74 | |
| 75 Args: | |
| 76 policy_list: A list containing Policy objects. (Parsed policy data by | |
| 77 parse_policy.) | |
| 78 bucket: A Bucket object to be searched for. | |
| 79 mmap: True if searching for a mmap region. | |
| 80 | |
| 81 Returns: | |
| 82 A string representing a component name. | |
| 83 """ | |
| 84 if not bucket: | |
| 85 return 'no-bucket' | |
| 86 if bucket.component: | |
| 87 return bucket.component | |
| 88 | |
| 89 stacktrace = ''.join( | |
| 90 address_symbol_dict[a] + ' ' for a in bucket.stacktrace).strip() | |
| 91 | |
| 92 for policy in policy_list: | |
| 93 if mmap == policy.mmap and policy.condition.match(stacktrace): | |
| 94 bucket.component = policy.name | |
| 95 return policy.name | |
| 96 | |
| 97 assert False | |
| 98 | |
| 99 | |
| 100 class Bucket(object): | |
| 101 | |
| 102 def __init__(self, stacktrace): | |
| 103 self.stacktrace = stacktrace | |
| 104 self.component = '' | |
| 105 | |
| 106 | |
| 107 class Log(object): | |
| 108 | |
| 109 """A class representing one dumped log data.""" | |
| 110 def __init__(self, log_path, buckets): | |
| 111 self.log_path = log_path | |
| 112 with open(self.log_path, mode='r') as log_f: | |
| 113 self.log_lines = log_f.readlines() | |
| 114 self.log_version = '' | |
| 115 sys.stderr.write('parsing a log file:%s\n' % log_path) | |
| 116 self.mmap_stacktrace_lines = [] | |
| 117 self.malloc_stacktrace_lines = [] | |
| 118 self.counters = {} | |
| 119 self.log_time = os.stat(self.log_path).st_mtime | |
| 120 self.parse_log(buckets) | |
| 121 | |
| 122 @staticmethod | |
| 123 def dump_stacktrace_lines(stacktrace_lines, buckets): | |
| 124 """Prints a given stacktrace. | |
| 125 | |
| 126 Args: | |
| 127 stacktrace_lines: A list of strings which are valid as stacktraces. | |
| 128 buckets: A dict mapping bucket ids and their corresponding Bucket | |
| 129 objects. | |
| 130 """ | |
| 131 for l in stacktrace_lines: | |
| 132 words = l.split() | |
| 133 bucket = buckets.get(int(words[BUCKET_ID])) | |
| 134 if not bucket: | |
| 135 continue | |
| 136 for i in range(0, BUCKET_ID - 1): | |
| 137 sys.stdout.write(words[i] + ' ') | |
| 138 for address in bucket.stacktrace: | |
| 139 sys.stdout.write((address_symbol_dict.get(address) or address) + ' ') | |
| 140 sys.stdout.write('\n') | |
| 141 | |
| 142 def dump_stacktrace(self, buckets): | |
| 143 """Prints stacktraces contained in the log. | |
| 144 | |
| 145 Args: | |
| 146 buckets: A dict mapping bucket ids and their corresponding Bucket | |
| 147 objects. | |
| 148 """ | |
| 149 self.dump_stacktrace_lines(self.mmap_stacktrace_lines, buckets) | |
| 150 self.dump_stacktrace_lines(self.malloc_stacktrace_lines, buckets) | |
| 151 | |
| 152 @staticmethod | |
| 153 def accumulate_size_for_pprof(stacktrace_lines, policy_list, buckets, | |
| 154 component_name, mmap): | |
| 155 """Accumulates size of committed chunks and the number of allocated chunks. | |
| 156 | |
| 157 Args: | |
| 158 stacktrace_lines: A list of strings which are valid as stacktraces. | |
| 159 policy_list: A list containing Policy objects. (Parsed policy data by | |
| 160 parse_policy.) | |
| 161 buckets: A dict mapping bucket ids and their corresponding Bucket | |
| 162 objects. | |
| 163 component_name: A name of component for filtering. | |
| 164 mmap: True if searching for a mmap region. | |
| 165 | |
| 166 Returns: | |
| 167 Two integers which are the accumulated size of committed regions and the | |
| 168 number of allocated chunks, respectively. | |
| 169 """ | |
| 170 com_committed = 0 | |
| 171 com_allocs = 0 | |
| 172 for l in stacktrace_lines: | |
| 173 words = l.split() | |
| 174 bucket = buckets.get(int(words[BUCKET_ID])) | |
| 175 if (not bucket or | |
| 176 (component_name and | |
| 177 component_name != get_component(policy_list, bucket, mmap))): | |
| 178 continue | |
| 179 | |
| 180 com_committed += int(words[COMMITTED]) | |
| 181 com_allocs += int(words[ALLOC_COUNT]) - int(words[FREE_COUNT]) | |
| 182 | |
| 183 return com_committed, com_allocs | |
| 184 | |
| 185 @staticmethod | |
| 186 def dump_stacktrace_lines_for_pprof(stacktrace_lines, policy_list, | |
| 187 buckets, component_name, mmap): | |
| 188 """Prints information of stacktrace lines for pprof. | |
| 189 | |
| 190 Args: | |
| 191 stacktrace_lines: A list of strings which are valid as stacktraces. | |
| 192 policy_list: A list containing Policy objects. (Parsed policy data by | |
| 193 parse_policy.) | |
| 194 buckets: A dict mapping bucket ids and their corresponding Bucket | |
| 195 objects. | |
| 196 component_name: A name of component for filtering. | |
| 197 mmap: True if searching for a mmap region. | |
| 198 """ | |
| 199 for l in stacktrace_lines: | |
| 200 words = l.split() | |
| 201 bucket = buckets.get(int(words[BUCKET_ID])) | |
| 202 if (not bucket or | |
| 203 (component_name and | |
| 204 component_name != get_component(policy_list, bucket, mmap))): | |
| 205 continue | |
| 206 | |
| 207 sys.stdout.write('%6d: %8s [%6d: %8s] @' % ( | |
| 208 int(words[ALLOC_COUNT]) - int(words[FREE_COUNT]), | |
| 209 words[COMMITTED], | |
| 210 int(words[ALLOC_COUNT]) - int(words[FREE_COUNT]), | |
| 211 words[COMMITTED])) | |
| 212 for address in bucket.stacktrace: | |
| 213 sys.stdout.write(' ' + address) | |
| 214 sys.stdout.write('\n') | |
| 215 | |
| 216 def dump_for_pprof(self, policy_list, buckets, mapping_lines, component_name): | |
| 217 """Converts the log file so it can be processed by pprof. | |
| 218 | |
| 219 Args: | |
| 220 policy_list: A list containing Policy objects. (Parsed policy data by | |
| 221 parse_policy.) | |
| 222 buckets: A dict mapping bucket ids and their corresponding Bucket | |
| 223 objects. | |
| 224 mapping_lines: A list of strings containing /proc/.../maps. | |
| 225 component_name: A name of component for filtering. | |
| 226 """ | |
| 227 sys.stdout.write('heap profile: ') | |
| 228 com_committed, com_allocs = self.accumulate_size_for_pprof( | |
| 229 self.mmap_stacktrace_lines, policy_list, buckets, component_name, | |
| 230 True) | |
| 231 add_committed, add_allocs = self.accumulate_size_for_pprof( | |
| 232 self.malloc_stacktrace_lines, policy_list, buckets, component_name, | |
| 233 False) | |
| 234 com_committed += add_committed | |
| 235 com_allocs += add_allocs | |
| 236 | |
| 237 sys.stdout.write('%6d: %8s [%6d: %8s] @ heapprofile\n' % ( | |
| 238 com_allocs, com_committed, com_allocs, com_committed)) | |
| 239 | |
| 240 self.dump_stacktrace_lines_for_pprof( | |
| 241 self.mmap_stacktrace_lines, policy_list, buckets, component_name, | |
| 242 True) | |
| 243 self.dump_stacktrace_lines_for_pprof( | |
| 244 self.malloc_stacktrace_lines, policy_list, buckets, component_name, | |
| 245 False) | |
| 246 | |
| 247 sys.stdout.write('MAPPED_LIBRARIES:\n') | |
| 248 for l in mapping_lines: | |
| 249 sys.stdout.write(l) | |
| 250 | |
| 251 @staticmethod | |
| 252 def check_stacktrace_line(stacktrace_line, buckets): | |
| 253 """Checks if a given stacktrace_line is valid as stacktrace. | |
| 254 | |
| 255 Args: | |
| 256 stacktrace_line: A string to be checked. | |
| 257 buckets: A dict mapping bucket ids and their corresponding Bucket | |
| 258 objects. | |
| 259 | |
| 260 Returns: | |
| 261 True if the given stacktrace_line is valid. | |
| 262 """ | |
| 263 words = stacktrace_line.split() | |
| 264 if len(words) < BUCKET_ID + 1: | |
| 265 return False | |
| 266 if words[BUCKET_ID - 1] != '@': | |
| 267 return False | |
| 268 bucket = buckets.get(int(words[BUCKET_ID])) | |
| 269 if bucket: | |
| 270 for address in bucket.stacktrace: | |
| 271 appeared_addresses.add(address) | |
| 272 return True | |
| 273 | |
| 274 @staticmethod | |
| 275 def skip_lines_while(line_number, max_line_number, skipping_condition): | |
| 276 """Increments line_number until skipping_condition(line_number) is false. | |
| 277 """ | |
| 278 while skipping_condition(line_number): | |
| 279 line_number += 1 | |
| 280 if line_number >= max_line_number: | |
| 281 sys.stderr.write('invalid heap profile dump.') | |
| 282 return line_number | |
| 283 return line_number | |
| 284 | |
| 285 def parse_stacktraces_while_valid(self, buckets, log_lines, ln): | |
| 286 """Parses stacktrace lines while the lines are valid. | |
| 287 | |
| 288 Args: | |
| 289 buckets: A dict mapping bucket ids and their corresponding Bucket | |
| 290 objects. | |
| 291 log_lines: A list of lines to be parsed. | |
| 292 ln: An integer representing the starting line number in log_lines. | |
| 293 | |
| 294 Returns: | |
| 295 A pair of a list of valid lines and an integer representing the last | |
| 296 line number in log_lines. | |
| 297 """ | |
| 298 ln = self.skip_lines_while( | |
| 299 ln, len(log_lines), lambda n: not log_lines[n].split()[0].isdigit()) | |
| 300 stacktrace_lines_start = ln | |
| 301 ln = self.skip_lines_while( | |
| 302 ln, len(log_lines), | |
| 303 lambda n: self.check_stacktrace_line(log_lines[n], buckets)) | |
| 304 return (log_lines[stacktrace_lines_start:ln], ln) | |
| 305 | |
| 306 def parse_stacktraces(self, buckets): | |
| 307 """Parses lines in self.log_lines as stacktrace. | |
| 308 | |
| 309 Valid stacktrace lines are stored into self.mmap_stacktrace_lines and | |
| 310 self.malloc_stacktrace_lines. | |
| 311 | |
| 312 Args: | |
| 313 buckets: A dict mapping bucket ids and their corresponding Bucket | |
| 314 objects. | |
| 315 | |
| 316 Returns: | |
| 317 A string representing a version of the stacktrace dump. '' for invalid | |
| 318 dump. | |
| 319 """ | |
| 320 version = '' | |
| 321 | |
| 322 # Skip until an identifiable line. | |
| 323 headers = ('STACKTRACES:\n', 'MMAP_STACKTRACES:\n', 'heap profile: ') | |
| 324 ln = self.skip_lines_while( | |
| 325 0, len(self.log_lines), | |
| 326 lambda n: not self.log_lines[n].startswith(headers)) | |
| 327 | |
| 328 # Identify a version. | |
| 329 if self.log_lines[ln].startswith('heap profile: '): | |
| 330 version = self.log_lines[ln][13:].strip() | |
| 331 if version == DUMP_DEEP_2 or version == DUMP_DEEP_3: | |
| 332 ln = self.skip_lines_while( | |
| 333 ln, len(self.log_lines), | |
| 334 lambda n: self.log_lines[n] != 'MMAP_STACKTRACES:\n') | |
| 335 else: | |
| 336 sys.stderr.write(' invalid heap profile dump version:%s\n' % version) | |
| 337 return '' | |
| 338 elif self.log_lines[ln] == 'STACKTRACES:\n': | |
| 339 version = DUMP_DEEP_1 | |
| 340 elif self.log_lines[ln] == 'MMAP_STACKTRACES:\n': | |
| 341 version = DUMP_DEEP_2 | |
| 342 | |
| 343 if version == DUMP_DEEP_3: | |
| 344 sys.stderr.write(' heap profile dump version: %s\n' % version) | |
| 345 (self.mmap_stacktrace_lines, ln) = self.parse_stacktraces_while_valid( | |
| 346 buckets, self.log_lines, ln) | |
| 347 ln = self.skip_lines_while( | |
| 348 ln, len(self.log_lines), | |
| 349 lambda n: self.log_lines[n] != 'MALLOC_STACKTRACES:\n') | |
| 350 (self.malloc_stacktrace_lines, ln) = self.parse_stacktraces_while_valid( | |
| 351 buckets, self.log_lines, ln) | |
| 352 return version | |
| 353 | |
| 354 elif version == DUMP_DEEP_2: | |
| 355 sys.stderr.write(' heap profile dump version: %s\n' % version) | |
| 356 (self.mmap_stacktrace_lines, ln) = self.parse_stacktraces_while_valid( | |
| 357 buckets, self.log_lines, ln) | |
| 358 ln = self.skip_lines_while( | |
| 359 ln, len(self.log_lines), | |
| 360 lambda n: self.log_lines[n] != 'MALLOC_STACKTRACES:\n') | |
| 361 (self.malloc_stacktrace_lines, ln) = self.parse_stacktraces_while_valid( | |
| 362 buckets, self.log_lines, ln) | |
| 363 self.malloc_stacktrace_lines.extend(self.mmap_stacktrace_lines) | |
| 364 self.mmap_stacktrace_lines = [] | |
| 365 return version | |
| 366 | |
| 367 elif version == DUMP_DEEP_1: | |
| 368 sys.stderr.write(' heap profile dump version: %s\n' % version) | |
| 369 (self.malloc_stacktrace_lines, ln) = self.parse_stacktraces_while_valid( | |
| 370 buckets, self.log_lines, ln) | |
| 371 return version | |
| 372 | |
| 373 else: | |
| 374 sys.stderr.write(' invalid heap profile dump version:%s\n' % version) | |
| 375 return '' | |
| 376 | |
| 377 def parse_global_stats(self): | |
| 378 """Parses lines in self.log_lines as global stats.""" | |
| 379 ln = self.skip_lines_while( | |
| 380 0, len(self.log_lines), | |
| 381 lambda n: self.log_lines[n] != 'GLOBAL_STATS:\n') | |
| 382 | |
| 383 for prefix in ['total', 'file', 'anonymous', 'other', 'mmap', 'tcmalloc']: | |
| 384 ln = self.skip_lines_while( | |
| 385 ln, len(self.log_lines), | |
| 386 lambda n: self.log_lines[n].split()[0] != prefix) | |
| 387 words = self.log_lines[ln].split() | |
| 388 self.counters[prefix + '_virtual'] = int(words[-2]) | |
| 389 self.counters[prefix + '_committed'] = int(words[-1]) | |
| 390 | |
| 391 def parse_log(self, buckets): | |
| 392 self.parse_global_stats() | |
| 393 self.log_version = self.parse_stacktraces(buckets) | |
| 394 | |
| 395 @staticmethod | |
| 396 def accumulate_size_for_policy(stacktrace_lines, | |
| 397 policy_list, buckets, sizes, mmap): | |
| 398 for l in stacktrace_lines: | |
| 399 words = l.split() | |
| 400 bucket = buckets.get(int(words[BUCKET_ID])) | |
| 401 component_match = get_component(policy_list, bucket, mmap) | |
| 402 sizes[component_match] += int(words[COMMITTED]) | |
| 403 | |
| 404 if component_match.startswith('tc-'): | |
| 405 sizes['tc-total-log'] += int(words[COMMITTED]) | |
| 406 elif component_match.startswith('mmap-'): | |
| 407 sizes['mmap-total-log'] += int(words[COMMITTED]) | |
| 408 else: | |
| 409 sizes['other-total-log'] += int(words[COMMITTED]) | |
| 410 | |
| 411 def apply_policy(self, policy_list, buckets, first_log_time): | |
| 412 """Aggregates the total memory size of each component. | |
| 413 | |
| 414 Iterate through all stacktraces and attribute them to one of the components | |
| 415 based on the policy. It is important to apply policy in right order. | |
| 416 | |
| 417 Args: | |
| 418 policy_list: A list containing Policy objects. (Parsed policy data by | |
| 419 parse_policy.) | |
| 420 buckets: A dict mapping bucket ids and their corresponding Bucket | |
| 421 objects. | |
| 422 first_log_time: An integer representing time when the first log is | |
| 423 dumped. | |
| 424 | |
| 425 Returns: | |
| 426 A dict mapping components and their corresponding sizes. | |
| 427 """ | |
| 428 | |
| 429 sys.stderr.write('apply policy:%s\n' % (self.log_path)) | |
| 430 sizes = dict((c, 0) for c in components) | |
| 431 | |
| 432 self.accumulate_size_for_policy(self.mmap_stacktrace_lines, | |
| 433 policy_list, buckets, sizes, True) | |
| 434 self.accumulate_size_for_policy(self.malloc_stacktrace_lines, | |
| 435 policy_list, buckets, sizes, False) | |
| 436 | |
| 437 sizes['mmap-no-log'] = self.counters['mmap_committed'] - sizes[ | |
| 438 'mmap-total-log'] | |
| 439 sizes['mmap-total-record'] = self.counters['mmap_committed'] | |
| 440 sizes['mmap-total-record-vm'] = self.counters['mmap_virtual'] | |
| 441 | |
| 442 sizes['tc-no-log'] = self.counters['tcmalloc_committed'] - sizes[ | |
| 443 'tc-total-log'] | |
| 444 sizes['tc-total-record'] = self.counters['tcmalloc_committed'] | |
| 445 sizes['tc-unused'] = sizes['mmap-tcmalloc'] - self.counters[ | |
| 446 'tcmalloc_committed'] | |
| 447 sizes['tc-total'] = sizes['mmap-tcmalloc'] | |
| 448 | |
| 449 for key, value in { 'total': 'total_committed', | |
| 450 'filemapped': 'file_committed', | |
| 451 'anonymous': 'anonymous_committed', | |
| 452 'other': 'other_committed', | |
| 453 'total-vm': 'total_virtual', | |
| 454 'filemapped-vm': 'file_virtual', | |
| 455 'anonymous-vm': 'anonymous_virtual', | |
| 456 'other-vm': 'other_virtual' }.items(): | |
| 457 if key in sizes: | |
| 458 sizes[key] = self.counters[value] | |
| 459 | |
| 460 if 'unknown' in sizes: | |
| 461 sizes['unknown'] = self.counters['total_committed'] - self.counters[ | |
| 462 'mmap_committed'] | |
| 463 if 'total-exclude-profiler' in sizes: | |
| 464 sizes['total-exclude-profiler'] = self.counters[ | |
| 465 'total_committed'] - sizes['mmap-profiler'] | |
| 466 if 'hour' in sizes: | |
| 467 sizes['hour'] = (self.log_time - first_log_time) / 60.0 / 60.0 | |
| 468 if 'minute' in sizes: | |
| 469 sizes['minute'] = (self.log_time - first_log_time) / 60.0 | |
| 470 if 'second' in sizes: | |
| 471 sizes['second'] = self.log_time - first_log_time | |
| 472 | |
| 473 return sizes | |
| 474 | |
| 475 @staticmethod | |
| 476 def accumulate_size_for_expand(stacktrace_lines, policy_list, buckets, | |
| 477 component_name, depth, sizes, mmap): | |
| 478 for line in stacktrace_lines: | |
| 479 words = line.split() | |
| 480 bucket = buckets.get(int(words[BUCKET_ID])) | |
| 481 component_match = get_component(policy_list, bucket, mmap) | |
| 482 if component_match == component_name: | |
| 483 stacktrace_sequence = '' | |
| 484 for address in bucket.stacktrace[1 : min(len(bucket.stacktrace), | |
| 485 1 + depth)]: | |
| 486 stacktrace_sequence += address_symbol_dict[address] + ' ' | |
| 487 if not stacktrace_sequence in sizes: | |
| 488 sizes[stacktrace_sequence] = 0 | |
| 489 sizes[stacktrace_sequence] += int(words[COMMITTED]) | |
| 490 | |
| 491 def expand(self, policy_list, buckets, component_name, depth): | |
| 492 """Prints all stacktraces in a given component of given depth. | |
| 493 | |
| 494 Args: | |
| 495 policy_list: A list containing Policy objects. (Parsed policy data by | |
| 496 parse_policy.) | |
| 497 buckets: A dict mapping bucket ids and their corresponding Bucket | |
| 498 objects. | |
| 499 component_name: A name of component for filtering. | |
| 500 depth: An integer representing depth to be printed. | |
| 501 """ | |
| 502 sizes = {} | |
| 503 | |
| 504 self.accumulate_size_for_expand( | |
| 505 self.mmap_stacktrace_lines, policy_list, buckets, component_name, | |
| 506 depth, sizes, True) | |
| 507 self.accumulate_size_for_expand( | |
| 508 self.malloc_stacktrace_lines, policy_list, buckets, component_name, | |
| 509 depth, sizes, False) | |
| 510 | |
| 511 sorted_sizes_list = sorted( | |
| 512 sizes.iteritems(), key=(lambda x: x[1]), reverse=True) | |
| 513 total = 0 | |
| 514 for size_pair in sorted_sizes_list: | |
| 515 sys.stdout.write('%10d %s\n' % (size_pair[1], size_pair[0])) | |
| 516 total += size_pair[1] | |
| 517 sys.stderr.write('total: %d\n' % (total)) | |
| 518 | |
| 519 | |
| 520 def update_symbols(symbol_path, mapping_lines, chrome_path): | |
| 521 """Updates address/symbol mapping on memory and in a .symbol cache file. | |
| 522 | |
| 523 It reads cached address/symbol mapping from a .symbol file if it exists. | |
| 524 Then, it resolves unresolved addresses from a Chrome binary with pprof. | |
| 525 Both mappings on memory and in a .symbol cache file are updated. | |
| 526 | |
| 527 Symbol files are formatted as follows: | |
| 528 <Address> <Symbol> | |
| 529 <Address> <Symbol> | |
| 530 <Address> <Symbol> | |
| 531 ... | |
| 532 | |
| 533 Args: | |
| 534 symbol_path: A string representing a path for a .symbol file. | |
| 535 mapping_lines: A list of strings containing /proc/.../maps. | |
| 536 chrome_path: A string representing a path for a Chrome binary. | |
| 537 """ | |
| 538 with open(symbol_path, mode='a+') as symbol_f: | |
| 539 symbol_lines = symbol_f.readlines() | |
| 540 if symbol_lines: | |
| 541 for line in symbol_lines: | |
| 542 items = line.split(None, 1) | |
| 543 address_symbol_dict[items[0]] = items[1].rstrip() | |
| 544 | |
| 545 unresolved_addresses = sorted( | |
| 546 a for a in appeared_addresses if a not in address_symbol_dict) | |
| 547 | |
| 548 if unresolved_addresses: | |
| 549 with tempfile.NamedTemporaryFile( | |
| 550 suffix='maps', prefix="dmprof", mode='w+') as pprof_in: | |
| 551 with tempfile.NamedTemporaryFile( | |
| 552 suffix='symbols', prefix="dmprof", mode='w+') as pprof_out: | |
| 553 for line in mapping_lines: | |
| 554 pprof_in.write(line) | |
| 555 | |
| 556 for address in unresolved_addresses: | |
| 557 pprof_in.write(address + '\n') | |
| 558 | |
| 559 pprof_in.seek(0) | |
| 560 | |
| 561 p = subprocess.Popen( | |
| 562 '%s --symbols %s' % (PPROF_PATH, chrome_path), | |
| 563 shell=True, stdin=pprof_in, stdout=pprof_out) | |
| 564 p.wait() | |
| 565 | |
| 566 pprof_out.seek(0) | |
| 567 symbols = pprof_out.readlines() | |
| 568 symbol_f.seek(0, 2) | |
| 569 for address, symbol in zip(unresolved_addresses, symbols): | |
| 570 stripped_symbol = symbol.strip() | |
| 571 address_symbol_dict[address] = stripped_symbol | |
| 572 symbol_f.write('%s %s\n' % (address, symbol.strip())) | |
| 573 | |
| 574 | |
| 575 def parse_policy(policy_path): | |
| 576 """Parses policy file. | |
| 577 | |
| 578 A policy file contains component's names and their | |
| 579 stacktrace pattern written in regular expression. | |
| 580 Those patterns are matched against each symbols of | |
| 581 each stacktraces in the order written in the policy file | |
| 582 | |
| 583 Args: | |
| 584 policy_path: A path for a policy file. | |
| 585 Returns: | |
| 586 A list containing component's name and its regex object | |
| 587 """ | |
| 588 with open(policy_path, mode='r') as policy_f: | |
| 589 policy_lines = policy_f.readlines() | |
| 590 | |
| 591 policy_version = POLICY_DEEP_1 | |
| 592 if policy_lines[0].startswith('heap profile policy: '): | |
| 593 policy_version = policy_lines[0][21:].strip() | |
| 594 policy_lines.pop(0) | |
| 595 policy_list = [] | |
| 596 | |
| 597 if policy_version == POLICY_DEEP_2 or policy_version == POLICY_DEEP_1: | |
| 598 sys.stderr.write(' heap profile policy version: %s\n' % policy_version) | |
| 599 for line in policy_lines: | |
| 600 if line[0] == '#': | |
| 601 continue | |
| 602 | |
| 603 if policy_version == POLICY_DEEP_2: | |
| 604 (name, allocation_type, pattern) = line.strip().split(None, 2) | |
| 605 mmap = False | |
| 606 if allocation_type == 'mmap': | |
| 607 mmap = True | |
| 608 elif policy_version == POLICY_DEEP_1: | |
| 609 name = line.split()[0] | |
| 610 pattern = line[len(name) : len(line)].strip() | |
| 611 mmap = False | |
| 612 | |
| 613 if pattern != 'default': | |
| 614 policy_list.append(Policy(name, mmap, pattern)) | |
| 615 if components.count(name) == 0: | |
| 616 components.append(name) | |
| 617 | |
| 618 else: | |
| 619 sys.stderr.write(' invalid heap profile policy version: %s\n' % ( | |
| 620 policy_version)) | |
| 621 | |
| 622 return policy_list | |
| 623 | |
| 624 | |
| 625 def main(): | |
| 626 if (len(sys.argv) < 4) or (not (sys.argv[1] in ['--csv', | |
| 627 '--json', | |
| 628 '--expand', | |
| 629 '--list', | |
| 630 '--stacktrace', | |
| 631 '--pprof'])): | |
| 632 sys.stderr.write("""Usage: | |
| 633 %s [options] <chrome-binary> <policy> <profile> [component-name] [depth] | |
| 634 | |
| 635 Options: | |
| 636 --csv Output result in csv format | |
| 637 --json Output result in json format | |
| 638 --stacktrace Convert raw address to symbol names | |
| 639 --list Lists components and their sizes | |
| 640 --expand Show all stacktraces in the specified component | |
| 641 of given depth with their sizes | |
| 642 --pprof Format the profile file so it can be processed | |
| 643 by pprof | |
| 644 | |
| 645 Examples: | |
| 646 dmprof --csv Debug/chrome dmpolicy hprof.12345.0001.heap > result.csv | |
| 647 dmprof --json Debug/chrome dmpolicy hprof.12345.0001.heap > result.json | |
| 648 dmprof --list Debug/chrome dmpolicy hprof.12345.0012.heap | |
| 649 dmprof --expand Debug/chrome dmpolicy hprof.12345.0012.heap tc-webkit 4 | |
| 650 dmprof --pprof Debug/chrome dmpolicy hprof.12345.0012.heap > for_pprof.txt | |
| 651 """ % (sys.argv[0])) | |
| 652 sys.exit(1) | |
| 653 | |
| 654 action = sys.argv[1] | |
| 655 chrome_path = sys.argv[2] | |
| 656 policy_path = sys.argv[3] | |
| 657 log_path = sys.argv[4] | |
| 658 | |
| 659 sys.stderr.write('parsing a policy file\n') | |
| 660 policy_list = parse_policy(policy_path) | |
| 661 | |
| 662 p = re.compile('\.[0-9][0-9][0-9][0-9]\.heap') | |
| 663 prefix = p.sub('', log_path) | |
| 664 symbol_path = prefix + '.symbols' | |
| 665 | |
| 666 sys.stderr.write('parsing the maps file\n') | |
| 667 maps_path = prefix + '.maps' | |
| 668 with open(maps_path, 'r') as maps_f: | |
| 669 maps_lines = maps_f.readlines() | |
| 670 | |
| 671 # Reading buckets | |
| 672 sys.stderr.write('parsing the bucket file\n') | |
| 673 buckets = {} | |
| 674 bucket_count = 0 | |
| 675 n = 0 | |
| 676 while True: | |
| 677 buckets_path = '%s.%04d.buckets' % (prefix, n) | |
| 678 if not os.path.exists(buckets_path): | |
| 679 if n > 10: | |
| 680 break | |
| 681 n += 1 | |
| 682 continue | |
| 683 sys.stderr.write('reading buckets from %s\n' % (buckets_path)) | |
| 684 with open(buckets_path, 'r') as buckets_f: | |
| 685 for l in buckets_f: | |
| 686 words = l.split() | |
| 687 buckets[int(words[0])] = Bucket(words[1:]) | |
| 688 n += 1 | |
| 689 | |
| 690 sys.stderr.write('the number buckets: %d\n' % (bucket_count)) | |
| 691 | |
| 692 log_path_list = [log_path] | |
| 693 | |
| 694 if action in ('--csv', '--json'): | |
| 695 # search for the sequence of files | |
| 696 n = int(log_path[len(log_path) - 9 : len(log_path) - 5]) | |
| 697 n += 1 # skip current file | |
| 698 while True: | |
| 699 p = '%s.%04d.heap' % (prefix, n) | |
| 700 if os.path.exists(p): | |
| 701 log_path_list.append(p) | |
| 702 else: | |
| 703 break | |
| 704 n += 1 | |
| 705 | |
| 706 logs = [Log(path, buckets) for path in log_path_list] | |
| 707 | |
| 708 sys.stderr.write('getting symbols\n') | |
| 709 update_symbols(symbol_path, maps_lines, chrome_path) | |
| 710 | |
| 711 # TODO(dmikurube): Many modes now. Split them into separete functions. | |
| 712 if action == '--stacktrace': | |
| 713 logs[0].dump_stacktrace(buckets) | |
| 714 | |
| 715 elif action == '--csv': | |
| 716 sys.stdout.write(','.join(components)) | |
| 717 sys.stdout.write('\n') | |
| 718 | |
| 719 for log in logs: | |
| 720 component_sizes = log.apply_policy(policy_list, buckets, logs[0].log_time) | |
| 721 s = [] | |
| 722 for c in components: | |
| 723 if c in ('hour', 'minute', 'second'): | |
| 724 s.append('%05.5f' % (component_sizes[c])) | |
| 725 else: | |
| 726 s.append('%05.5f' % (component_sizes[c] / 1024.0 / 1024.0)) | |
| 727 sys.stdout.write(','.join(s)) | |
| 728 sys.stdout.write('\n') | |
| 729 | |
| 730 elif action == '--json': | |
| 731 json_base = { | |
| 732 'version': 'JSON_DEEP_1', | |
| 733 'legends': components, | |
| 734 'snapshots': [], | |
| 735 } | |
| 736 for log in logs: | |
| 737 component_sizes = log.apply_policy(policy_list, buckets, logs[0].log_time) | |
| 738 component_sizes['log_path'] = log.log_path | |
| 739 component_sizes['log_time'] = datetime.fromtimestamp( | |
| 740 log.log_time).strftime('%Y-%m-%d %H:%M:%S') | |
| 741 json_base['snapshots'].append(component_sizes) | |
| 742 json.dump(json_base, sys.stdout, indent=2, sort_keys=True) | |
| 743 | |
| 744 elif action == '--list': | |
| 745 component_sizes = logs[0].apply_policy( | |
| 746 policy_list, buckets, logs[0].log_time) | |
| 747 for c in components: | |
| 748 if c in ['hour', 'minute', 'second']: | |
| 749 sys.stdout.write('%30s %10.3f\n' % (c, component_sizes[c])) | |
| 750 else: | |
| 751 sys.stdout.write('%30s %10.3f\n' % ( | |
| 752 c, component_sizes[c] / 1024.0 / 1024.0)) | |
| 753 | |
| 754 elif action == '--expand': | |
| 755 component_name = sys.argv[5] | |
| 756 depth = sys.argv[6] | |
| 757 logs[0].expand(policy_list, buckets, component_name, int(depth)) | |
| 758 | |
| 759 elif action == '--pprof': | |
| 760 if len(sys.argv) > 5: | |
| 761 logs[0].dump_for_pprof(policy_list, buckets, maps_lines, sys.argv[5]) | |
| 762 else: | |
| 763 logs[0].dump_for_pprof(policy_list, buckets, maps_lines, None) | |
| 764 | |
| 765 | |
| 766 if __name__ == '__main__': | |
| 767 sys.exit(main()) | |
| OLD | NEW |
|---|
Chromium Code Reviews
Issue 9812010:
Breakdown nonprofiled memory regions (f.k.a. 'unknown'), and add new policy files. (Closed)
Base URL: svn://svn.chromium.org/chrome/trunk/src