#!/usr/bin/env python # # Copyright 2012 Facebook # # Licensed under the Apache License, Version 2.0 (the "License"); you may # not use this file except in compliance with the License. You may obtain # a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, WITHOUT # WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the # License for the specific language governing permissions and limitations # under the License. """Utilities for working with threads and ``Futures``. ``Futures`` are a pattern for concurrent programming introduced in Python 3.2 in the `concurrent.futures` package. This package defines a mostly-compatible `Future` class designed for use from coroutines, as well as some utility functions for interacting with the `concurrent.futures` package. """ from __future__ import absolute_import, division, print_function, with_statement import functools import platform import textwrap import traceback import sys from tornado.log import app_log from tornado.stack_context import ExceptionStackContext, wrap from tornado.util import raise_exc_info, ArgReplacer try: from concurrent import futures except ImportError: futures = None # Can the garbage collector handle cycles that include __del__ methods? # This is true in cpython beginning with version 3.4 (PEP 442). _GC_CYCLE_FINALIZERS = (platform.python_implementation() == 'CPython' and sys.version_info >= (3, 4)) class ReturnValueIgnoredError(Exception): pass # This class and associated code in the future object is derived # from the Trollius project, a backport of asyncio to Python 2.x - 3.x class _TracebackLogger(object): """Helper to log a traceback upon destruction if not cleared. This solves a nasty problem with Futures and Tasks that have an exception set: if nobody asks for the exception, the exception is never logged. This violates the Zen of Python: 'Errors should never pass silently. Unless explicitly silenced.' However, we don't want to log the exception as soon as set_exception() is called: if the calling code is written properly, it will get the exception and handle it properly. But we *do* want to log it if result() or exception() was never called -- otherwise developers waste a lot of time wondering why their buggy code fails silently. An earlier attempt added a __del__() method to the Future class itself, but this backfired because the presence of __del__() prevents garbage collection from breaking cycles. A way out of this catch-22 is to avoid having a __del__() method on the Future class itself, but instead to have a reference to a helper object with a __del__() method that logs the traceback, where we ensure that the helper object doesn't participate in cycles, and only the Future has a reference to it. The helper object is added when set_exception() is called. When the Future is collected, and the helper is present, the helper object is also collected, and its __del__() method will log the traceback. When the Future's result() or exception() method is called (and a helper object is present), it removes the the helper object, after calling its clear() method to prevent it from logging. One downside is that we do a fair amount of work to extract the traceback from the exception, even when it is never logged. It would seem cheaper to just store the exception object, but that references the traceback, which references stack frames, which may reference the Future, which references the _TracebackLogger, and then the _TracebackLogger would be included in a cycle, which is what we're trying to avoid! As an optimization, we don't immediately format the exception; we only do the work when activate() is called, which call is delayed until after all the Future's callbacks have run. Since usually a Future has at least one callback (typically set by 'yield From') and usually that callback extracts the callback, thereby removing the need to format the exception. PS. I don't claim credit for this solution. I first heard of it in a discussion about closing files when they are collected. """ __slots__ = ('exc_info', 'formatted_tb') def __init__(self, exc_info): self.exc_info = exc_info self.formatted_tb = None def activate(self): exc_info = self.exc_info if exc_info is not None: self.exc_info = None self.formatted_tb = traceback.format_exception(*exc_info) def clear(self): self.exc_info = None self.formatted_tb = None def __del__(self): if self.formatted_tb: app_log.error('Future exception was never retrieved: %s', ''.join(self.formatted_tb).rstrip()) class Future(object): """Placeholder for an asynchronous result. A ``Future`` encapsulates the result of an asynchronous operation. In synchronous applications ``Futures`` are used to wait for the result from a thread or process pool; in Tornado they are normally used with `.IOLoop.add_future` or by yielding them in a `.gen.coroutine`. `tornado.concurrent.Future` is similar to `concurrent.futures.Future`, but not thread-safe (and therefore faster for use with single-threaded event loops). In addition to ``exception`` and ``set_exception``, methods ``exc_info`` and ``set_exc_info`` are supported to capture tracebacks in Python 2. The traceback is automatically available in Python 3, but in the Python 2 futures backport this information is discarded. This functionality was previously available in a separate class ``TracebackFuture``, which is now a deprecated alias for this class. .. versionchanged:: 4.0 `tornado.concurrent.Future` is always a thread-unsafe ``Future`` with support for the ``exc_info`` methods. Previously it would be an alias for the thread-safe `concurrent.futures.Future` if that package was available and fall back to the thread-unsafe implementation if it was not. .. versionchanged:: 4.1 If a `.Future` contains an error but that error is never observed (by calling ``result()``, ``exception()``, or ``exc_info()``), a stack trace will be logged when the `.Future` is garbage collected. This normally indicates an error in the application, but in cases where it results in undesired logging it may be necessary to suppress the logging by ensuring that the exception is observed: ``f.add_done_callback(lambda f: f.exception())``. """ def __init__(self): self._done = False self._result = None self._exc_info = None self._log_traceback = False # Used for Python >= 3.4 self._tb_logger = None # Used for Python <= 3.3 self._callbacks = [] # Implement the Python 3.5 Awaitable protocol if possible # (we can't use return and yield together until py33). if sys.version_info >= (3, 3): exec(textwrap.dedent(""" def __await__(self): return (yield self) """)) else: # Py2-compatible version for use with cython. def __await__(self): result = yield self # StopIteration doesn't take args before py33, # but Cython recognizes the args tuple. e = StopIteration() e.args = (result,) raise e def cancel(self): """Cancel the operation, if possible. Tornado ``Futures`` do not support cancellation, so this method always returns False. """ return False def cancelled(self): """Returns True if the operation has been cancelled. Tornado ``Futures`` do not support cancellation, so this method always returns False. """ return False def running(self): """Returns True if this operation is currently running.""" return not self._done def done(self): """Returns True if the future has finished running.""" return self._done def _clear_tb_log(self): self._log_traceback = False if self._tb_logger is not None: self._tb_logger.clear() self._tb_logger = None def result(self, timeout=None): """If the operation succeeded, return its result. If it failed, re-raise its exception. This method takes a ``timeout`` argument for compatibility with `concurrent.futures.Future` but it is an error to call it before the `Future` is done, so the ``timeout`` is never used. """ self._clear_tb_log() if self._result is not None: return self._result if self._exc_info is not None: raise_exc_info(self._exc_info) self._check_done() return self._result def exception(self, timeout=None): """If the operation raised an exception, return the `Exception` object. Otherwise returns None. This method takes a ``timeout`` argument for compatibility with `concurrent.futures.Future` but it is an error to call it before the `Future` is done, so the ``timeout`` is never used. """ self._clear_tb_log() if self._exc_info is not None: return self._exc_info[1] else: self._check_done() return None def add_done_callback(self, fn): """Attaches the given callback to the `Future`. It will be invoked with the `Future` as its argument when the Future has finished running and its result is available. In Tornado consider using `.IOLoop.add_future` instead of calling `add_done_callback` directly. """ if self._done: fn(self) else: self._callbacks.append(fn) def set_result(self, result): """Sets the result of a ``Future``. It is undefined to call any of the ``set`` methods more than once on the same object. """ self._result = result self._set_done() def set_exception(self, exception): """Sets the exception of a ``Future.``""" self.set_exc_info( (exception.__class__, exception, getattr(exception, '__traceback__', None))) def exc_info(self): """Returns a tuple in the same format as `sys.exc_info` or None. .. versionadded:: 4.0 """ self._clear_tb_log() return self._exc_info def set_exc_info(self, exc_info): """Sets the exception information of a ``Future.`` Preserves tracebacks on Python 2. .. versionadded:: 4.0 """ self._exc_info = exc_info self._log_traceback = True if not _GC_CYCLE_FINALIZERS: self._tb_logger = _TracebackLogger(exc_info) try: self._set_done() finally: # Activate the logger after all callbacks have had a # chance to call result() or exception(). if self._log_traceback and self._tb_logger is not None: self._tb_logger.activate() self._exc_info = exc_info def _check_done(self): if not self._done: raise Exception("DummyFuture does not support blocking for results") def _set_done(self): self._done = True for cb in self._callbacks: try: cb(self) except Exception: app_log.exception('Exception in callback %r for %r', cb, self) self._callbacks = None # On Python 3.3 or older, objects with a destructor part of a reference # cycle are never destroyed. It's no longer the case on Python 3.4 thanks to # the PEP 442. if _GC_CYCLE_FINALIZERS: def __del__(self): if not self._log_traceback: # set_exception() was not called, or result() or exception() # has consumed the exception return tb = traceback.format_exception(*self._exc_info) app_log.error('Future %r exception was never retrieved: %s', self, ''.join(tb).rstrip()) TracebackFuture = Future if futures is None: FUTURES = Future else: FUTURES = (futures.Future, Future) def is_future(x): return isinstance(x, FUTURES) class DummyExecutor(object): def submit(self, fn, *args, **kwargs): future = TracebackFuture() try: future.set_result(fn(*args, **kwargs)) except Exception: future.set_exc_info(sys.exc_info()) return future def shutdown(self, wait=True): pass dummy_executor = DummyExecutor() def run_on_executor(*args, **kwargs): """Decorator to run a synchronous method asynchronously on an executor. The decorated method may be called with a ``callback`` keyword argument and returns a future. The `.IOLoop` and executor to be used are determined by the ``io_loop`` and ``executor`` attributes of ``self``. To use different attributes, pass keyword arguments to the decorator:: @run_on_executor(executor='_thread_pool') def foo(self): pass .. versionchanged:: 4.2 Added keyword arguments to use alternative attributes. """ def run_on_executor_decorator(fn): executor = kwargs.get("executor", "executor") io_loop = kwargs.get("io_loop", "io_loop") @functools.wraps(fn) def wrapper(self, *args, **kwargs): callback = kwargs.pop("callback", None) future = getattr(self, executor).submit(fn, self, *args, **kwargs) if callback: getattr(self, io_loop).add_future( future, lambda future: callback(future.result())) return future return wrapper if args and kwargs: raise ValueError("cannot combine positional and keyword args") if len(args) == 1: return run_on_executor_decorator(args[0]) elif len(args) != 0: raise ValueError("expected 1 argument, got %d", len(args)) return run_on_executor_decorator _NO_RESULT = object() def return_future(f): """Decorator to make a function that returns via callback return a `Future`. The wrapped function should take a ``callback`` keyword argument and invoke it with one argument when it has finished. To signal failure, the function can simply raise an exception (which will be captured by the `.StackContext` and passed along to the ``Future``). From the caller's perspective, the callback argument is optional. If one is given, it will be invoked when the function is complete with `Future.result()` as an argument. If the function fails, the callback will not be run and an exception will be raised into the surrounding `.StackContext`. If no callback is given, the caller should use the ``Future`` to wait for the function to complete (perhaps by yielding it in a `.gen.engine` function, or passing it to `.IOLoop.add_future`). Usage: .. testcode:: @return_future def future_func(arg1, arg2, callback): # Do stuff (possibly asynchronous) callback(result) @gen.engine def caller(callback): yield future_func(arg1, arg2) callback() .. Note that ``@return_future`` and ``@gen.engine`` can be applied to the same function, provided ``@return_future`` appears first. However, consider using ``@gen.coroutine`` instead of this combination. """ replacer = ArgReplacer(f, 'callback') @functools.wraps(f) def wrapper(*args, **kwargs): future = TracebackFuture() callback, args, kwargs = replacer.replace( lambda value=_NO_RESULT: future.set_result(value), args, kwargs) def handle_error(typ, value, tb): future.set_exc_info((typ, value, tb)) return True exc_info = None with ExceptionStackContext(handle_error): try: result = f(*args, **kwargs) if result is not None: raise ReturnValueIgnoredError( "@return_future should not be used with functions " "that return values") except: exc_info = sys.exc_info() raise if exc_info is not None: # If the initial synchronous part of f() raised an exception, # go ahead and raise it to the caller directly without waiting # for them to inspect the Future. future.result() # If the caller passed in a callback, schedule it to be called # when the future resolves. It is important that this happens # just before we return the future, or else we risk confusing # stack contexts with multiple exceptions (one here with the # immediate exception, and again when the future resolves and # the callback triggers its exception by calling future.result()). if callback is not None: def run_callback(future): result = future.result() if result is _NO_RESULT: callback() else: callback(future.result()) future.add_done_callback(wrap(run_callback)) return future return wrapper def chain_future(a, b): """Chain two futures together so that when one completes, so does the other. The result (success or failure) of ``a`` will be copied to ``b``, unless ``b`` has already been completed or cancelled by the time ``a`` finishes. """ def copy(future): assert future is a if b.done(): return if (isinstance(a, TracebackFuture) and isinstance(b, TracebackFuture) and a.exc_info() is not None): b.set_exc_info(a.exc_info()) elif a.exception() is not None: b.set_exception(a.exception()) else: b.set_result(a.result()) a.add_done_callback(copy)