| OLD | NEW |
|---|
| (Empty) |
| 1 # Copyright 2013 The Chromium Authors. All rights reserved. | |
| 2 # Use of this source code is governed by a BSD-style license that can be | |
| 3 # found in the LICENSE file. | |
| 4 | |
| 5 """Classes and functions related to threading.""" | |
| 6 | |
| 7 import functools | |
| 8 import inspect | |
| 9 import logging | |
| 10 import os | |
| 11 import Queue | |
| 12 import sys | |
| 13 import threading | |
| 14 import time | |
| 15 import traceback | |
| 16 | |
| 17 | |
| 18 class LockWithAssert(object): | |
| 19 """Wrapper around (non recursive) Lock that tracks its owner.""" | |
| 20 | |
| 21 def __init__(self): | |
| 22 self._lock = threading.Lock() | |
| 23 self._owner = None | |
| 24 | |
| 25 def __enter__(self): | |
| 26 self._lock.acquire() | |
| 27 assert self._owner is None | |
| 28 self._owner = threading.current_thread() | |
| 29 | |
| 30 def __exit__(self, _exc_type, _exec_value, _traceback): | |
| 31 self.assert_locked('Releasing unowned lock') | |
| 32 self._owner = None | |
| 33 self._lock.release() | |
| 34 return False | |
| 35 | |
| 36 def assert_locked(self, msg=None): | |
| 37 """Asserts the lock is owned by running thread.""" | |
| 38 assert self._owner == threading.current_thread(), msg | |
| 39 | |
| 40 | |
| 41 class ThreadPoolError(Exception): | |
| 42 """Base class for exceptions raised by ThreadPool.""" | |
| 43 pass | |
| 44 | |
| 45 | |
| 46 class ThreadPoolEmpty(ThreadPoolError): | |
| 47 """Trying to get task result from a thread pool with no pending tasks.""" | |
| 48 pass | |
| 49 | |
| 50 | |
| 51 class ThreadPoolClosed(ThreadPoolError): | |
| 52 """Trying to do something with a closed thread pool.""" | |
| 53 pass | |
| 54 | |
| 55 | |
| 56 class ThreadPool(object): | |
| 57 """Multithreaded worker pool with priority support. | |
| 58 | |
| 59 When the priority of tasks match, it works in strict FIFO mode. | |
| 60 """ | |
| 61 QUEUE_CLASS = Queue.PriorityQueue | |
| 62 | |
| 63 def __init__(self, initial_threads, max_threads, queue_size, prefix=None): | |
| 64 """Immediately starts |initial_threads| threads. | |
| 65 | |
| 66 Arguments: | |
| 67 initial_threads: Number of threads to start immediately. Can be 0 if it is | |
| 68 uncertain that threads will be needed. | |
| 69 max_threads: Maximum number of threads that will be started when all the | |
| 70 threads are busy working. Often the number of CPU cores. | |
| 71 queue_size: Maximum number of tasks to buffer in the queue. 0 for | |
| 72 unlimited queue. A non-zero value may make add_task() | |
| 73 blocking. | |
| 74 prefix: Prefix to use for thread names. Pool's threads will be | |
| 75 named '<prefix>-<thread index>'. | |
| 76 """ | |
| 77 prefix = prefix or 'tp-0x%0x' % id(self) | |
| 78 logging.debug( | |
| 79 'New ThreadPool(%d, %d, %d): %s', initial_threads, max_threads, | |
| 80 queue_size, prefix) | |
| 81 assert initial_threads <= max_threads | |
| 82 # Update this check once 256 cores CPU are common. | |
| 83 assert max_threads <= 256 | |
| 84 | |
| 85 self.tasks = self.QUEUE_CLASS(queue_size) | |
| 86 self._max_threads = max_threads | |
| 87 self._prefix = prefix | |
| 88 | |
| 89 # Used to assign indexes to tasks. | |
| 90 self._num_of_added_tasks_lock = threading.Lock() | |
| 91 self._num_of_added_tasks = 0 | |
| 92 | |
| 93 # Lock that protected everything below (including conditional variable). | |
| 94 self._lock = threading.Lock() | |
| 95 | |
| 96 # Condition 'bool(_outputs) or bool(_exceptions) or _pending_count == 0'. | |
| 97 self._outputs_exceptions_cond = threading.Condition(self._lock) | |
| 98 self._outputs = [] | |
| 99 self._exceptions = [] | |
| 100 | |
| 101 # Number of pending tasks (queued or being processed now). | |
| 102 self._pending_count = 0 | |
| 103 | |
| 104 # List of threads. | |
| 105 self._workers = [] | |
| 106 # Number of threads that are waiting for new tasks. | |
| 107 self._ready = 0 | |
| 108 # Number of threads already added to _workers, but not yet running the loop. | |
| 109 self._starting = 0 | |
| 110 # True if close was called. Forbids adding new tasks. | |
| 111 self._is_closed = False | |
| 112 | |
| 113 for _ in range(initial_threads): | |
| 114 self._add_worker() | |
| 115 | |
| 116 def _add_worker(self): | |
| 117 """Adds one worker thread if there isn't too many. Thread-safe.""" | |
| 118 with self._lock: | |
| 119 if len(self._workers) >= self._max_threads or self._is_closed: | |
| 120 return False | |
| 121 worker = threading.Thread( | |
| 122 name='%s-%d' % (self._prefix, len(self._workers)), target=self._run) | |
| 123 self._workers.append(worker) | |
| 124 self._starting += 1 | |
| 125 logging.debug('Starting worker thread %s', worker.name) | |
| 126 worker.daemon = True | |
| 127 worker.start() | |
| 128 return True | |
| 129 | |
| 130 def add_task(self, priority, func, *args, **kwargs): | |
| 131 """Adds a task, a function to be executed by a worker. | |
| 132 | |
| 133 Arguments: | |
| 134 - priority: priority of the task versus others. Lower priority takes | |
| 135 precedence. | |
| 136 - func: function to run. Can either return a return value to be added to the | |
| 137 output list or be a generator which can emit multiple values. | |
| 138 - args and kwargs: arguments to |func|. Note that if func mutates |args| or | |
| 139 |kwargs| and that the task is retried, see | |
| 140 AutoRetryThreadPool, the retry will use the mutated | |
| 141 values. | |
| 142 | |
| 143 Returns: | |
| 144 Index of the item added, e.g. the total number of enqueued items up to | |
| 145 now. | |
| 146 """ | |
| 147 assert isinstance(priority, int) | |
| 148 assert callable(func) | |
| 149 with self._lock: | |
| 150 if self._is_closed: | |
| 151 raise ThreadPoolClosed('Can not add a task to a closed ThreadPool') | |
| 152 start_new_worker = ( | |
| 153 # Pending task count plus new task > number of available workers. | |
| 154 self.tasks.qsize() + 1 > self._ready + self._starting and | |
| 155 # Enough slots. | |
| 156 len(self._workers) < self._max_threads | |
| 157 ) | |
| 158 self._pending_count += 1 | |
| 159 with self._num_of_added_tasks_lock: | |
| 160 self._num_of_added_tasks += 1 | |
| 161 index = self._num_of_added_tasks | |
| 162 self.tasks.put((priority, index, func, args, kwargs)) | |
| 163 if start_new_worker: | |
| 164 self._add_worker() | |
| 165 return index | |
| 166 | |
| 167 def _run(self): | |
| 168 """Worker thread loop. Runs until a None task is queued.""" | |
| 169 # Thread has started, adjust counters. | |
| 170 with self._lock: | |
| 171 self._starting -= 1 | |
| 172 self._ready += 1 | |
| 173 while True: | |
| 174 try: | |
| 175 task = self.tasks.get() | |
| 176 finally: | |
| 177 with self._lock: | |
| 178 self._ready -= 1 | |
| 179 try: | |
| 180 if task is None: | |
| 181 # We're done. | |
| 182 return | |
| 183 _priority, _index, func, args, kwargs = task | |
| 184 if inspect.isgeneratorfunction(func): | |
| 185 for out in func(*args, **kwargs): | |
| 186 self._output_append(out) | |
| 187 else: | |
| 188 out = func(*args, **kwargs) | |
| 189 self._output_append(out) | |
| 190 except Exception as e: | |
| 191 logging.warning('Caught exception: %s', e) | |
| 192 exc_info = sys.exc_info() | |
| 193 logging.info(''.join(traceback.format_tb(exc_info[2]))) | |
| 194 with self._outputs_exceptions_cond: | |
| 195 self._exceptions.append(exc_info) | |
| 196 self._outputs_exceptions_cond.notifyAll() | |
| 197 finally: | |
| 198 try: | |
| 199 # Mark thread as ready again, mark task as processed. Do it before | |
| 200 # waking up threads waiting on self.tasks.join(). Otherwise they might | |
| 201 # find ThreadPool still 'busy' and perform unnecessary wait on CV. | |
| 202 with self._outputs_exceptions_cond: | |
| 203 self._ready += 1 | |
| 204 self._pending_count -= 1 | |
| 205 if self._pending_count == 0: | |
| 206 self._outputs_exceptions_cond.notifyAll() | |
| 207 self.tasks.task_done() | |
| 208 except Exception as e: | |
| 209 # We need to catch and log this error here because this is the root | |
| 210 # function for the thread, nothing higher will catch the error. | |
| 211 logging.exception('Caught exception while marking task as done: %s', | |
| 212 e) | |
| 213 | |
| 214 def _output_append(self, out): | |
| 215 if out is not None: | |
| 216 with self._outputs_exceptions_cond: | |
| 217 self._outputs.append(out) | |
| 218 self._outputs_exceptions_cond.notifyAll() | |
| 219 | |
| 220 def join(self): | |
| 221 """Extracts all the results from each threads unordered. | |
| 222 | |
| 223 Call repeatedly to extract all the exceptions if desired. | |
| 224 | |
| 225 Note: will wait for all work items to be done before returning an exception. | |
| 226 To get an exception early, use get_one_result(). | |
| 227 """ | |
| 228 # TODO(maruel): Stop waiting as soon as an exception is caught. | |
| 229 self.tasks.join() | |
| 230 with self._outputs_exceptions_cond: | |
| 231 if self._exceptions: | |
| 232 e = self._exceptions.pop(0) | |
| 233 raise e[0], e[1], e[2] | |
| 234 out = self._outputs | |
| 235 self._outputs = [] | |
| 236 return out | |
| 237 | |
| 238 def get_one_result(self): | |
| 239 """Returns the next item that was generated or raises an exception if one | |
| 240 occurred. | |
| 241 | |
| 242 Raises: | |
| 243 ThreadPoolEmpty - no results available. | |
| 244 """ | |
| 245 # Get first available result. | |
| 246 for result in self.iter_results(): | |
| 247 return result | |
| 248 # No results -> tasks queue is empty. | |
| 249 raise ThreadPoolEmpty('Task queue is empty') | |
| 250 | |
| 251 def iter_results(self): | |
| 252 """Yields results as they appear until all tasks are processed.""" | |
| 253 while True: | |
| 254 # Check for pending results. | |
| 255 result = None | |
| 256 with self._outputs_exceptions_cond: | |
| 257 if self._exceptions: | |
| 258 e = self._exceptions.pop(0) | |
| 259 raise e[0], e[1], e[2] | |
| 260 if self._outputs: | |
| 261 # Remember the result to yield it outside of the lock. | |
| 262 result = self._outputs.pop(0) | |
| 263 else: | |
| 264 # No pending tasks -> all tasks are done. | |
| 265 if not self._pending_count: | |
| 266 return | |
| 267 # Some task is queued, wait for its result to appear. | |
| 268 # Use non-None timeout so that process reacts to Ctrl+C and other | |
| 269 # signals, see http://bugs.python.org/issue8844. | |
| 270 self._outputs_exceptions_cond.wait(timeout=5) | |
| 271 continue | |
| 272 yield result | |
| 273 | |
| 274 def close(self): | |
| 275 """Closes all the threads.""" | |
| 276 # Ensure no new threads can be started, self._workers is effectively | |
| 277 # a constant after that and can be accessed outside the lock. | |
| 278 with self._lock: | |
| 279 if self._is_closed: | |
| 280 raise ThreadPoolClosed('Can not close already closed ThreadPool') | |
| 281 self._is_closed = True | |
| 282 for _ in range(len(self._workers)): | |
| 283 # Enqueueing None causes the worker to stop. | |
| 284 self.tasks.put(None) | |
| 285 for t in self._workers: | |
| 286 t.join() | |
| 287 logging.debug( | |
| 288 'Thread pool \'%s\' closed: spawned %d threads total', | |
| 289 self._prefix, len(self._workers)) | |
| 290 | |
| 291 def abort(self): | |
| 292 """Empties the queue. | |
| 293 | |
| 294 To be used when the pool should stop early, like when Ctrl-C was detected. | |
| 295 | |
| 296 Returns: | |
| 297 Number of tasks cancelled. | |
| 298 """ | |
| 299 index = 0 | |
| 300 while True: | |
| 301 try: | |
| 302 self.tasks.get_nowait() | |
| 303 self.tasks.task_done() | |
| 304 index += 1 | |
| 305 except Queue.Empty: | |
| 306 return index | |
| 307 | |
| 308 def __enter__(self): | |
| 309 """Enables 'with' statement.""" | |
| 310 return self | |
| 311 | |
| 312 def __exit__(self, _exc_type, _exc_value, _traceback): | |
| 313 """Enables 'with' statement.""" | |
| 314 self.close() | |
| 315 | |
| 316 | |
| 317 class AutoRetryThreadPool(ThreadPool): | |
| 318 """Automatically retries enqueued operations on exception.""" | |
| 319 INTERNAL_PRIORITY_BITS = (1<<8) - 1 | |
| 320 HIGH, MED, LOW = (1<<8, 2<<8, 3<<8) | |
| 321 | |
| 322 def __init__(self, exceptions, retries, *args, **kwargs): | |
| 323 """ | |
| 324 Arguments: | |
| 325 exceptions: list of exception classes that can be retried on. | |
| 326 retries: maximum number of retries to do. | |
| 327 """ | |
| 328 assert exceptions and all(issubclass(e, Exception) for e in exceptions), ( | |
| 329 exceptions) | |
| 330 assert 1 <= retries <= self.INTERNAL_PRIORITY_BITS | |
| 331 super(AutoRetryThreadPool, self).__init__(*args, **kwargs) | |
| 332 self._swallowed_exceptions = tuple(exceptions) | |
| 333 self._retries = retries | |
| 334 | |
| 335 def add_task(self, priority, func, *args, **kwargs): | |
| 336 """Tasks added must not use the lower priority bits since they are reserved | |
| 337 for retries. | |
| 338 """ | |
| 339 assert (priority & self.INTERNAL_PRIORITY_BITS) == 0 | |
| 340 return super(AutoRetryThreadPool, self).add_task( | |
| 341 priority, | |
| 342 self._task_executer, | |
| 343 priority, | |
| 344 None, | |
| 345 func, | |
| 346 *args, | |
| 347 **kwargs) | |
| 348 | |
| 349 def add_task_with_channel(self, channel, priority, func, *args, **kwargs): | |
| 350 """Tasks added must not use the lower priority bits since they are reserved | |
| 351 for retries. | |
| 352 """ | |
| 353 assert (priority & self.INTERNAL_PRIORITY_BITS) == 0 | |
| 354 return super(AutoRetryThreadPool, self).add_task( | |
| 355 priority, | |
| 356 self._task_executer, | |
| 357 priority, | |
| 358 channel, | |
| 359 func, | |
| 360 *args, | |
| 361 **kwargs) | |
| 362 | |
| 363 def _task_executer(self, priority, channel, func, *args, **kwargs): | |
| 364 """Wraps the function and automatically retry on exceptions.""" | |
| 365 try: | |
| 366 result = func(*args, **kwargs) | |
| 367 if channel is None: | |
| 368 return result | |
| 369 channel.send_result(result) | |
| 370 except self._swallowed_exceptions as e: | |
| 371 # Retry a few times, lowering the priority. | |
| 372 actual_retries = priority & self.INTERNAL_PRIORITY_BITS | |
| 373 if actual_retries < self._retries: | |
| 374 priority += 1 | |
| 375 logging.debug( | |
| 376 'Swallowed exception \'%s\'. Retrying at lower priority %X', | |
| 377 e, priority) | |
| 378 super(AutoRetryThreadPool, self).add_task( | |
| 379 priority, | |
| 380 self._task_executer, | |
| 381 priority, | |
| 382 channel, | |
| 383 func, | |
| 384 *args, | |
| 385 **kwargs) | |
| 386 return | |
| 387 if channel is None: | |
| 388 raise | |
| 389 channel.send_exception(e) | |
| 390 except Exception as e: | |
| 391 if channel is None: | |
| 392 raise | |
| 393 channel.send_exception(e) | |
| 394 | |
| 395 | |
| 396 class Progress(object): | |
| 397 """Prints progress and accepts updates thread-safely.""" | |
| 398 def __init__(self, columns): | |
| 399 """Creates a Progress bar that will updates asynchronously from the worker | |
| 400 threads. | |
| 401 | |
| 402 Arguments: | |
| 403 columns: list of tuple(name, initialvalue), defines both the number of | |
| 404 columns and their initial values. | |
| 405 """ | |
| 406 assert all( | |
| 407 len(c) == 2 and isinstance(c[0], str) and isinstance(c[1], int) | |
| 408 for c in columns), columns | |
| 409 # Members to be used exclusively in the primary thread. | |
| 410 self.use_cr_only = True | |
| 411 self.unfinished_commands = set() | |
| 412 self.start = time.time() | |
| 413 self._last_printed_line = '' | |
| 414 self._columns = [c[1] for c in columns] | |
| 415 self._columns_lookup = dict((c[0], i) for i, c in enumerate(columns)) | |
| 416 # Setting it to True forces a print on the first print_update() call. | |
| 417 self._value_changed = True | |
| 418 | |
| 419 # To be used in all threads. | |
| 420 self._queued_updates = Queue.Queue() | |
| 421 | |
| 422 def update_item(self, name, raw=False, **kwargs): | |
| 423 """Queue information to print out. | |
| 424 | |
| 425 Arguments: | |
| 426 name: string to print out to describe something that was completed. | |
| 427 raw: if True, prints the data without the header. | |
| 428 raw: if True, prints the data without the header. | |
| 429 <kwargs>: argument name is a name of a column. it's value is the increment | |
| 430 to the column, value is usually 0 or 1. | |
| 431 """ | |
| 432 assert isinstance(name, str) | |
| 433 assert isinstance(raw, bool) | |
| 434 assert all(isinstance(v, int) for v in kwargs.itervalues()) | |
| 435 args = [(self._columns_lookup[k], v) for k, v in kwargs.iteritems() if v] | |
| 436 self._queued_updates.put((name, raw, args)) | |
| 437 | |
| 438 def print_update(self): | |
| 439 """Prints the current status.""" | |
| 440 # Flush all the logging output so it doesn't appear within this output. | |
| 441 for handler in logging.root.handlers: | |
| 442 handler.flush() | |
| 443 | |
| 444 got_one = False | |
| 445 while True: | |
| 446 try: | |
| 447 name, raw, args = self._queued_updates.get_nowait() | |
| 448 except Queue.Empty: | |
| 449 break | |
| 450 | |
| 451 for k, v in args: | |
| 452 self._columns[k] += v | |
| 453 self._value_changed = bool(args) | |
| 454 if not name: | |
| 455 # Even if raw=True, there's nothing to print. | |
| 456 continue | |
| 457 | |
| 458 got_one = True | |
| 459 if raw: | |
| 460 # Prints the data as-is. | |
| 461 self._last_printed_line = '' | |
| 462 sys.stdout.write('\n%s\n' % name.strip('\n')) | |
| 463 else: | |
| 464 line, self._last_printed_line = self._gen_line(name) | |
| 465 sys.stdout.write(line) | |
| 466 | |
| 467 if not got_one and self._value_changed: | |
| 468 # Make sure a line is printed in that case where statistics changes. | |
| 469 line, self._last_printed_line = self._gen_line('') | |
| 470 sys.stdout.write(line) | |
| 471 got_one = True | |
| 472 self._value_changed = False | |
| 473 if got_one: | |
| 474 # Ensure that all the output is flushed to prevent it from getting mixed | |
| 475 # with other output streams (like the logging streams). | |
| 476 sys.stdout.flush() | |
| 477 | |
| 478 if self.unfinished_commands: | |
| 479 logging.debug('Waiting for the following commands to finish:\n%s', | |
| 480 '\n'.join(self.unfinished_commands)) | |
| 481 | |
| 482 def _gen_line(self, name): | |
| 483 """Generates the line to be printed.""" | |
| 484 next_line = ('[%s] %6.2fs %s') % ( | |
| 485 self._render_columns(), time.time() - self.start, name) | |
| 486 # Fill it with whitespace only if self.use_cr_only is set. | |
| 487 prefix = '' | |
| 488 if self.use_cr_only and self._last_printed_line: | |
| 489 prefix = '\r' | |
| 490 if self.use_cr_only: | |
| 491 suffix = ' ' * max(0, len(self._last_printed_line) - len(next_line)) | |
| 492 else: | |
| 493 suffix = '\n' | |
| 494 return '%s%s%s' % (prefix, next_line, suffix), next_line | |
| 495 | |
| 496 def _render_columns(self): | |
| 497 """Renders the columns.""" | |
| 498 columns_as_str = map(str, self._columns) | |
| 499 max_len = max(map(len, columns_as_str)) | |
| 500 return '/'.join(i.rjust(max_len) for i in columns_as_str) | |
| 501 | |
| 502 | |
| 503 class QueueWithProgress(Queue.PriorityQueue): | |
| 504 """Implements progress support in join().""" | |
| 505 def __init__(self, progress, *args, **kwargs): | |
| 506 Queue.PriorityQueue.__init__(self, *args, **kwargs) | |
| 507 self.progress = progress | |
| 508 | |
| 509 def task_done(self): | |
| 510 """Contrary to Queue.task_done(), it wakes self.all_tasks_done at each task | |
| 511 done. | |
| 512 """ | |
| 513 with self.all_tasks_done: | |
| 514 try: | |
| 515 unfinished = self.unfinished_tasks - 1 | |
| 516 if unfinished < 0: | |
| 517 raise ValueError('task_done() called too many times') | |
| 518 self.unfinished_tasks = unfinished | |
| 519 # This is less efficient, because we want the Progress to be updated. | |
| 520 self.all_tasks_done.notify_all() | |
| 521 except Exception as e: | |
| 522 logging.exception('task_done threw an exception.\n%s', e) | |
| 523 | |
| 524 def wake_up(self): | |
| 525 """Wakes up all_tasks_done. | |
| 526 | |
| 527 Unlike task_done(), do not substract one from self.unfinished_tasks. | |
| 528 """ | |
| 529 # TODO(maruel): This is highly inefficient, since the listener is awaken | |
| 530 # twice; once per output, once per task. There should be no relationship | |
| 531 # between the number of output and the number of input task. | |
| 532 with self.all_tasks_done: | |
| 533 self.all_tasks_done.notify_all() | |
| 534 | |
| 535 def join(self): | |
| 536 """Calls print_update() whenever possible.""" | |
| 537 self.progress.print_update() | |
| 538 with self.all_tasks_done: | |
| 539 while self.unfinished_tasks: | |
| 540 self.progress.print_update() | |
| 541 # Use a short wait timeout so updates are printed in a timely manner. | |
| 542 # TODO(maruel): Find a way so Progress.queue and self.all_tasks_done | |
| 543 # share the same underlying event so no polling is necessary. | |
| 544 self.all_tasks_done.wait(0.1) | |
| 545 self.progress.print_update() | |
| 546 | |
| 547 | |
| 548 class ThreadPoolWithProgress(ThreadPool): | |
| 549 QUEUE_CLASS = QueueWithProgress | |
| 550 | |
| 551 def __init__(self, progress, *args, **kwargs): | |
| 552 self.QUEUE_CLASS = functools.partial(self.QUEUE_CLASS, progress) | |
| 553 super(ThreadPoolWithProgress, self).__init__(*args, **kwargs) | |
| 554 | |
| 555 def _output_append(self, out): | |
| 556 """Also wakes up the listener on new completed test_case.""" | |
| 557 super(ThreadPoolWithProgress, self)._output_append(out) | |
| 558 self.tasks.wake_up() | |
| 559 | |
| 560 | |
| 561 class DeadlockDetector(object): | |
| 562 """Context manager that can detect deadlocks. | |
| 563 | |
| 564 It will dump stack frames of all running threads if its 'ping' method isn't | |
| 565 called in time. | |
| 566 | |
| 567 Usage: | |
| 568 with DeadlockDetector(timeout=60) as detector: | |
| 569 for item in some_work(): | |
| 570 ... | |
| 571 detector.ping() | |
| 572 ... | |
| 573 | |
| 574 Arguments: | |
| 575 timeout - maximum allowed time between calls to 'ping'. | |
| 576 """ | |
| 577 | |
| 578 def __init__(self, timeout): | |
| 579 self.timeout = timeout | |
| 580 self._thread = None | |
| 581 # Thread stop condition. Also lock for shared variables below. | |
| 582 self._stop_cv = threading.Condition() | |
| 583 self._stop_flag = False | |
| 584 # Time when 'ping' was called last time. | |
| 585 self._last_ping = None | |
| 586 # True if pings are coming on time. | |
| 587 self._alive = True | |
| 588 | |
| 589 def __enter__(self): | |
| 590 """Starts internal watcher thread.""" | |
| 591 assert self._thread is None | |
| 592 self.ping() | |
| 593 self._thread = threading.Thread(name='deadlock-detector', target=self._run) | |
| 594 self._thread.daemon = True | |
| 595 self._thread.start() | |
| 596 return self | |
| 597 | |
| 598 def __exit__(self, *_args): | |
| 599 """Stops internal watcher thread.""" | |
| 600 assert self._thread is not None | |
| 601 with self._stop_cv: | |
| 602 self._stop_flag = True | |
| 603 self._stop_cv.notify() | |
| 604 self._thread.join() | |
| 605 self._thread = None | |
| 606 self._stop_flag = False | |
| 607 | |
| 608 def ping(self): | |
| 609 """Notify detector that main thread is still running. | |
| 610 | |
| 611 Should be called periodically to inform the detector that everything is | |
| 612 running as it should. | |
| 613 """ | |
| 614 with self._stop_cv: | |
| 615 self._last_ping = time.time() | |
| 616 self._alive = True | |
| 617 | |
| 618 def _run(self): | |
| 619 """Loop that watches for pings and dumps threads state if ping is late.""" | |
| 620 with self._stop_cv: | |
| 621 while not self._stop_flag: | |
| 622 # Skipped deadline? Dump threads and switch to 'not alive' state. | |
| 623 if self._alive and time.time() > self._last_ping + self.timeout: | |
| 624 self.dump_threads(time.time() - self._last_ping, True) | |
| 625 self._alive = False | |
| 626 | |
| 627 # Pings are on time? | |
| 628 if self._alive: | |
| 629 # Wait until the moment we need to dump stack traces. | |
| 630 # Most probably some other thread will call 'ping' to move deadline | |
| 631 # further in time. We don't bother to wake up after each 'ping', | |
| 632 # only right before initial expected deadline. | |
| 633 self._stop_cv.wait(self._last_ping + self.timeout - time.time()) | |
| 634 else: | |
| 635 # Skipped some pings previously. Just periodically silently check | |
| 636 # for new pings with some arbitrary frequency. | |
| 637 self._stop_cv.wait(self.timeout * 0.1) | |
| 638 | |
| 639 @staticmethod | |
| 640 def dump_threads(timeout=None, skip_current_thread=False): | |
| 641 """Dumps stack frames of all running threads.""" | |
| 642 all_threads = threading.enumerate() | |
| 643 current_thread_id = threading.current_thread().ident | |
| 644 | |
| 645 # Collect tracebacks: thread name -> traceback string. | |
| 646 tracebacks = {} | |
| 647 | |
| 648 # pylint: disable=W0212 | |
| 649 for thread_id, frame in sys._current_frames().iteritems(): | |
| 650 # Don't dump deadlock detector's own thread, it's boring. | |
| 651 if thread_id == current_thread_id and skip_current_thread: | |
| 652 continue | |
| 653 | |
| 654 # Try to get more informative symbolic thread name. | |
| 655 name = 'untitled' | |
| 656 for thread in all_threads: | |
| 657 if thread.ident == thread_id: | |
| 658 name = thread.name | |
| 659 break | |
| 660 name += ' #%d' % (thread_id,) | |
| 661 tracebacks[name] = ''.join(traceback.format_stack(frame)) | |
| 662 | |
| 663 # Function to print a message. Makes it easier to change output destination. | |
| 664 def output(msg): | |
| 665 logging.warning(msg.rstrip()) | |
| 666 | |
| 667 # Print tracebacks, sorting them by thread name. That way a thread pool's | |
| 668 # threads will be printed as one group. | |
| 669 output('=============== Potential deadlock detected ===============') | |
| 670 if timeout is not None: | |
| 671 output('No pings in last %d sec.' % (timeout,)) | |
| 672 output('Dumping stack frames for all threads:') | |
| 673 for name in sorted(tracebacks): | |
| 674 output('Traceback for \'%s\':\n%s' % (name, tracebacks[name])) | |
| 675 output('===========================================================') | |
| 676 | |
| 677 | |
| 678 class Bit(object): | |
| 679 """Thread safe setable bit.""" | |
| 680 | |
| 681 def __init__(self): | |
| 682 self._lock = threading.Lock() | |
| 683 self._value = False | |
| 684 | |
| 685 def get(self): | |
| 686 with self._lock: | |
| 687 return self._value | |
| 688 | |
| 689 def set(self): | |
| 690 with self._lock: | |
| 691 self._value = True | |
| 692 | |
| 693 | |
| 694 class TaskChannel(object): | |
| 695 """Queue of results of async task execution.""" | |
| 696 | |
| 697 _ITEM_RESULT = 0 | |
| 698 _ITEM_EXCEPTION = 1 | |
| 699 | |
| 700 def __init__(self): | |
| 701 self._queue = Queue.Queue() | |
| 702 | |
| 703 def send_result(self, result): | |
| 704 """Enqueues a result of task execution.""" | |
| 705 self._queue.put((self._ITEM_RESULT, result)) | |
| 706 | |
| 707 def send_exception(self, exc): | |
| 708 """Enqueue an exception raised by a task.""" | |
| 709 assert isinstance(exc, Exception) | |
| 710 self._queue.put((self._ITEM_EXCEPTION, exc)) | |
| 711 | |
| 712 def pull(self): | |
| 713 """Dequeues available result or exception.""" | |
| 714 item_type, value = self._queue.get() | |
| 715 if item_type == self._ITEM_RESULT: | |
| 716 return value | |
| 717 if item_type == self._ITEM_EXCEPTION: | |
| 718 raise value | |
| 719 assert False, 'Impossible queue item type: %r' % item_type | |
| 720 | |
| 721 def wrap_task(self, task): | |
| 722 """Decorator that makes a function push results into this channel.""" | |
| 723 @functools.wraps(task) | |
| 724 def wrapped(*args, **kwargs): | |
| 725 try: | |
| 726 self.send_result(task(*args, **kwargs)) | |
| 727 except Exception as exc: | |
| 728 self.send_exception(exc) | |
| 729 return wrapped | |
| 730 | |
| 731 | |
| 732 def num_processors(): | |
| 733 """Returns the number of processors. | |
| 734 | |
| 735 Python on OSX 10.6 raises a NotImplementedError exception. | |
| 736 """ | |
| 737 try: | |
| 738 # Multiprocessing | |
| 739 import multiprocessing | |
| 740 return multiprocessing.cpu_count() | |
| 741 except: # pylint: disable=W0702 | |
| 742 try: | |
| 743 # Mac OS 10.6 | |
| 744 return int(os.sysconf('SC_NPROCESSORS_ONLN')) # pylint: disable=E1101 | |
| 745 except: | |
| 746 # Some of the windows builders seem to get here. | |
| 747 return 4 | |
| OLD | NEW |
|---|
Chromium Code Reviews
