cpython/Lib/asyncio/windows_events.py

892 lines
31 KiB
Python
Raw Normal View History

"""Selector and proactor event loops for Windows."""
import sys
if sys.platform != 'win32': # pragma: no cover
raise ImportError('win32 only')
import _overlapped
import _winapi
import errno
import math
import msvcrt
import socket
import struct
import time
import weakref
from . import events
from . import base_subprocess
from . import futures
from . import exceptions
from . import proactor_events
from . import selector_events
from . import tasks
from . import windows_utils
from .log import logger
__all__ = (
'SelectorEventLoop', 'ProactorEventLoop', 'IocpProactor',
'DefaultEventLoopPolicy', 'WindowsSelectorEventLoopPolicy',
'WindowsProactorEventLoopPolicy',
)
NULL = _winapi.NULL
INFINITE = _winapi.INFINITE
ERROR_CONNECTION_REFUSED = 1225
ERROR_CONNECTION_ABORTED = 1236
# Initial delay in seconds for connect_pipe() before retrying to connect
CONNECT_PIPE_INIT_DELAY = 0.001
# Maximum delay in seconds for connect_pipe() before retrying to connect
CONNECT_PIPE_MAX_DELAY = 0.100
class _OverlappedFuture(futures.Future):
"""Subclass of Future which represents an overlapped operation.
Cancelling it will immediately cancel the overlapped operation.
"""
def __init__(self, ov, *, loop=None):
super().__init__(loop=loop)
if self._source_traceback:
del self._source_traceback[-1]
self._ov = ov
def _repr_info(self):
info = super()._repr_info()
if self._ov is not None:
state = 'pending' if self._ov.pending else 'completed'
info.insert(1, f'overlapped=<{state}, {self._ov.address:#x}>')
return info
def _cancel_overlapped(self):
if self._ov is None:
return
try:
self._ov.cancel()
except OSError as exc:
context = {
'message': 'Cancelling an overlapped future failed',
'exception': exc,
'future': self,
}
if self._source_traceback:
context['source_traceback'] = self._source_traceback
self._loop.call_exception_handler(context)
self._ov = None
def cancel(self, msg=None):
self._cancel_overlapped()
return super().cancel(msg=msg)
def set_exception(self, exception):
super().set_exception(exception)
self._cancel_overlapped()
def set_result(self, result):
super().set_result(result)
self._ov = None
Issue #23095, asyncio: Rewrite _WaitHandleFuture.cancel() This change fixes a race conditon related to _WaitHandleFuture.cancel() leading to Python crash or "GetQueuedCompletionStatus() returned an unexpected event" logs. Before, the overlapped object was destroyed too early, it was possible that the wait completed whereas the overlapped object was already destroyed. Sometimes, a different overlapped was allocated at the same address, leading to unexpected completition. _WaitHandleFuture.cancel() now waits until the wait is cancelled to clear its reference to the overlapped object. To wait until the cancellation is done, UnregisterWaitEx() is used with an event instead of UnregisterWait(). To wait for this event, a new _WaitCancelFuture class was added. It's a simplified version of _WaitCancelFuture. For example, its cancel() method calls UnregisterWait(), not UnregisterWaitEx(). _WaitCancelFuture should not be cancelled. The overlapped object is kept alive in _WaitHandleFuture until the wait is unregistered. Other changes: * Add _overlapped.UnregisterWaitEx() * Remove fast-path in IocpProactor.wait_for_handle() to immediatly set the result if the wait already completed. I'm not sure that it's safe to call immediatly UnregisterWaitEx() before the completion was signaled. * Add IocpProactor._unregistered() to forget an overlapped which may never be signaled, but may be signaled for the next loop iteration. It avoids to block forever IocpProactor.close() if a wait was cancelled, and it may also avoid some "... unexpected event ..." warnings.
2015-01-21 18:39:51 -04:00
class _BaseWaitHandleFuture(futures.Future):
"""Subclass of Future which represents a wait handle."""
Issue #23095, asyncio: Rewrite _WaitHandleFuture.cancel() This change fixes a race conditon related to _WaitHandleFuture.cancel() leading to Python crash or "GetQueuedCompletionStatus() returned an unexpected event" logs. Before, the overlapped object was destroyed too early, it was possible that the wait completed whereas the overlapped object was already destroyed. Sometimes, a different overlapped was allocated at the same address, leading to unexpected completition. _WaitHandleFuture.cancel() now waits until the wait is cancelled to clear its reference to the overlapped object. To wait until the cancellation is done, UnregisterWaitEx() is used with an event instead of UnregisterWait(). To wait for this event, a new _WaitCancelFuture class was added. It's a simplified version of _WaitCancelFuture. For example, its cancel() method calls UnregisterWait(), not UnregisterWaitEx(). _WaitCancelFuture should not be cancelled. The overlapped object is kept alive in _WaitHandleFuture until the wait is unregistered. Other changes: * Add _overlapped.UnregisterWaitEx() * Remove fast-path in IocpProactor.wait_for_handle() to immediatly set the result if the wait already completed. I'm not sure that it's safe to call immediatly UnregisterWaitEx() before the completion was signaled. * Add IocpProactor._unregistered() to forget an overlapped which may never be signaled, but may be signaled for the next loop iteration. It avoids to block forever IocpProactor.close() if a wait was cancelled, and it may also avoid some "... unexpected event ..." warnings.
2015-01-21 18:39:51 -04:00
def __init__(self, ov, handle, wait_handle, *, loop=None):
super().__init__(loop=loop)
if self._source_traceback:
del self._source_traceback[-1]
Issue #23095, asyncio: Rewrite _WaitHandleFuture.cancel() This change fixes a race conditon related to _WaitHandleFuture.cancel() leading to Python crash or "GetQueuedCompletionStatus() returned an unexpected event" logs. Before, the overlapped object was destroyed too early, it was possible that the wait completed whereas the overlapped object was already destroyed. Sometimes, a different overlapped was allocated at the same address, leading to unexpected completition. _WaitHandleFuture.cancel() now waits until the wait is cancelled to clear its reference to the overlapped object. To wait until the cancellation is done, UnregisterWaitEx() is used with an event instead of UnregisterWait(). To wait for this event, a new _WaitCancelFuture class was added. It's a simplified version of _WaitCancelFuture. For example, its cancel() method calls UnregisterWait(), not UnregisterWaitEx(). _WaitCancelFuture should not be cancelled. The overlapped object is kept alive in _WaitHandleFuture until the wait is unregistered. Other changes: * Add _overlapped.UnregisterWaitEx() * Remove fast-path in IocpProactor.wait_for_handle() to immediatly set the result if the wait already completed. I'm not sure that it's safe to call immediatly UnregisterWaitEx() before the completion was signaled. * Add IocpProactor._unregistered() to forget an overlapped which may never be signaled, but may be signaled for the next loop iteration. It avoids to block forever IocpProactor.close() if a wait was cancelled, and it may also avoid some "... unexpected event ..." warnings.
2015-01-21 18:39:51 -04:00
# Keep a reference to the Overlapped object to keep it alive until the
# wait is unregistered
self._ov = ov
self._handle = handle
self._wait_handle = wait_handle
Issue #23095, asyncio: Rewrite _WaitHandleFuture.cancel() This change fixes a race conditon related to _WaitHandleFuture.cancel() leading to Python crash or "GetQueuedCompletionStatus() returned an unexpected event" logs. Before, the overlapped object was destroyed too early, it was possible that the wait completed whereas the overlapped object was already destroyed. Sometimes, a different overlapped was allocated at the same address, leading to unexpected completition. _WaitHandleFuture.cancel() now waits until the wait is cancelled to clear its reference to the overlapped object. To wait until the cancellation is done, UnregisterWaitEx() is used with an event instead of UnregisterWait(). To wait for this event, a new _WaitCancelFuture class was added. It's a simplified version of _WaitCancelFuture. For example, its cancel() method calls UnregisterWait(), not UnregisterWaitEx(). _WaitCancelFuture should not be cancelled. The overlapped object is kept alive in _WaitHandleFuture until the wait is unregistered. Other changes: * Add _overlapped.UnregisterWaitEx() * Remove fast-path in IocpProactor.wait_for_handle() to immediatly set the result if the wait already completed. I'm not sure that it's safe to call immediatly UnregisterWaitEx() before the completion was signaled. * Add IocpProactor._unregistered() to forget an overlapped which may never be signaled, but may be signaled for the next loop iteration. It avoids to block forever IocpProactor.close() if a wait was cancelled, and it may also avoid some "... unexpected event ..." warnings.
2015-01-21 18:39:51 -04:00
# Should we call UnregisterWaitEx() if the wait completes
# or is cancelled?
self._registered = True
def _poll(self):
# non-blocking wait: use a timeout of 0 millisecond
return (_winapi.WaitForSingleObject(self._handle, 0) ==
_winapi.WAIT_OBJECT_0)
def _repr_info(self):
info = super()._repr_info()
info.append(f'handle={self._handle:#x}')
Issue #23095, asyncio: Rewrite _WaitHandleFuture.cancel() This change fixes a race conditon related to _WaitHandleFuture.cancel() leading to Python crash or "GetQueuedCompletionStatus() returned an unexpected event" logs. Before, the overlapped object was destroyed too early, it was possible that the wait completed whereas the overlapped object was already destroyed. Sometimes, a different overlapped was allocated at the same address, leading to unexpected completition. _WaitHandleFuture.cancel() now waits until the wait is cancelled to clear its reference to the overlapped object. To wait until the cancellation is done, UnregisterWaitEx() is used with an event instead of UnregisterWait(). To wait for this event, a new _WaitCancelFuture class was added. It's a simplified version of _WaitCancelFuture. For example, its cancel() method calls UnregisterWait(), not UnregisterWaitEx(). _WaitCancelFuture should not be cancelled. The overlapped object is kept alive in _WaitHandleFuture until the wait is unregistered. Other changes: * Add _overlapped.UnregisterWaitEx() * Remove fast-path in IocpProactor.wait_for_handle() to immediatly set the result if the wait already completed. I'm not sure that it's safe to call immediatly UnregisterWaitEx() before the completion was signaled. * Add IocpProactor._unregistered() to forget an overlapped which may never be signaled, but may be signaled for the next loop iteration. It avoids to block forever IocpProactor.close() if a wait was cancelled, and it may also avoid some "... unexpected event ..." warnings.
2015-01-21 18:39:51 -04:00
if self._handle is not None:
state = 'signaled' if self._poll() else 'waiting'
Issue #23095, asyncio: Rewrite _WaitHandleFuture.cancel() This change fixes a race conditon related to _WaitHandleFuture.cancel() leading to Python crash or "GetQueuedCompletionStatus() returned an unexpected event" logs. Before, the overlapped object was destroyed too early, it was possible that the wait completed whereas the overlapped object was already destroyed. Sometimes, a different overlapped was allocated at the same address, leading to unexpected completition. _WaitHandleFuture.cancel() now waits until the wait is cancelled to clear its reference to the overlapped object. To wait until the cancellation is done, UnregisterWaitEx() is used with an event instead of UnregisterWait(). To wait for this event, a new _WaitCancelFuture class was added. It's a simplified version of _WaitCancelFuture. For example, its cancel() method calls UnregisterWait(), not UnregisterWaitEx(). _WaitCancelFuture should not be cancelled. The overlapped object is kept alive in _WaitHandleFuture until the wait is unregistered. Other changes: * Add _overlapped.UnregisterWaitEx() * Remove fast-path in IocpProactor.wait_for_handle() to immediatly set the result if the wait already completed. I'm not sure that it's safe to call immediatly UnregisterWaitEx() before the completion was signaled. * Add IocpProactor._unregistered() to forget an overlapped which may never be signaled, but may be signaled for the next loop iteration. It avoids to block forever IocpProactor.close() if a wait was cancelled, and it may also avoid some "... unexpected event ..." warnings.
2015-01-21 18:39:51 -04:00
info.append(state)
if self._wait_handle is not None:
info.append(f'wait_handle={self._wait_handle:#x}')
return info
Issue #23095, asyncio: Rewrite _WaitHandleFuture.cancel() This change fixes a race conditon related to _WaitHandleFuture.cancel() leading to Python crash or "GetQueuedCompletionStatus() returned an unexpected event" logs. Before, the overlapped object was destroyed too early, it was possible that the wait completed whereas the overlapped object was already destroyed. Sometimes, a different overlapped was allocated at the same address, leading to unexpected completition. _WaitHandleFuture.cancel() now waits until the wait is cancelled to clear its reference to the overlapped object. To wait until the cancellation is done, UnregisterWaitEx() is used with an event instead of UnregisterWait(). To wait for this event, a new _WaitCancelFuture class was added. It's a simplified version of _WaitCancelFuture. For example, its cancel() method calls UnregisterWait(), not UnregisterWaitEx(). _WaitCancelFuture should not be cancelled. The overlapped object is kept alive in _WaitHandleFuture until the wait is unregistered. Other changes: * Add _overlapped.UnregisterWaitEx() * Remove fast-path in IocpProactor.wait_for_handle() to immediatly set the result if the wait already completed. I'm not sure that it's safe to call immediatly UnregisterWaitEx() before the completion was signaled. * Add IocpProactor._unregistered() to forget an overlapped which may never be signaled, but may be signaled for the next loop iteration. It avoids to block forever IocpProactor.close() if a wait was cancelled, and it may also avoid some "... unexpected event ..." warnings.
2015-01-21 18:39:51 -04:00
def _unregister_wait_cb(self, fut):
# The wait was unregistered: it's not safe to destroy the Overlapped
# object
self._ov = None
def _unregister_wait(self):
Issue #23095, asyncio: Rewrite _WaitHandleFuture.cancel() This change fixes a race conditon related to _WaitHandleFuture.cancel() leading to Python crash or "GetQueuedCompletionStatus() returned an unexpected event" logs. Before, the overlapped object was destroyed too early, it was possible that the wait completed whereas the overlapped object was already destroyed. Sometimes, a different overlapped was allocated at the same address, leading to unexpected completition. _WaitHandleFuture.cancel() now waits until the wait is cancelled to clear its reference to the overlapped object. To wait until the cancellation is done, UnregisterWaitEx() is used with an event instead of UnregisterWait(). To wait for this event, a new _WaitCancelFuture class was added. It's a simplified version of _WaitCancelFuture. For example, its cancel() method calls UnregisterWait(), not UnregisterWaitEx(). _WaitCancelFuture should not be cancelled. The overlapped object is kept alive in _WaitHandleFuture until the wait is unregistered. Other changes: * Add _overlapped.UnregisterWaitEx() * Remove fast-path in IocpProactor.wait_for_handle() to immediatly set the result if the wait already completed. I'm not sure that it's safe to call immediatly UnregisterWaitEx() before the completion was signaled. * Add IocpProactor._unregistered() to forget an overlapped which may never be signaled, but may be signaled for the next loop iteration. It avoids to block forever IocpProactor.close() if a wait was cancelled, and it may also avoid some "... unexpected event ..." warnings.
2015-01-21 18:39:51 -04:00
if not self._registered:
return
Issue #23095, asyncio: Rewrite _WaitHandleFuture.cancel() This change fixes a race conditon related to _WaitHandleFuture.cancel() leading to Python crash or "GetQueuedCompletionStatus() returned an unexpected event" logs. Before, the overlapped object was destroyed too early, it was possible that the wait completed whereas the overlapped object was already destroyed. Sometimes, a different overlapped was allocated at the same address, leading to unexpected completition. _WaitHandleFuture.cancel() now waits until the wait is cancelled to clear its reference to the overlapped object. To wait until the cancellation is done, UnregisterWaitEx() is used with an event instead of UnregisterWait(). To wait for this event, a new _WaitCancelFuture class was added. It's a simplified version of _WaitCancelFuture. For example, its cancel() method calls UnregisterWait(), not UnregisterWaitEx(). _WaitCancelFuture should not be cancelled. The overlapped object is kept alive in _WaitHandleFuture until the wait is unregistered. Other changes: * Add _overlapped.UnregisterWaitEx() * Remove fast-path in IocpProactor.wait_for_handle() to immediatly set the result if the wait already completed. I'm not sure that it's safe to call immediatly UnregisterWaitEx() before the completion was signaled. * Add IocpProactor._unregistered() to forget an overlapped which may never be signaled, but may be signaled for the next loop iteration. It avoids to block forever IocpProactor.close() if a wait was cancelled, and it may also avoid some "... unexpected event ..." warnings.
2015-01-21 18:39:51 -04:00
self._registered = False
wait_handle = self._wait_handle
self._wait_handle = None
try:
_overlapped.UnregisterWait(wait_handle)
except OSError as exc:
if exc.winerror != _overlapped.ERROR_IO_PENDING:
context = {
'message': 'Failed to unregister the wait handle',
'exception': exc,
'future': self,
}
if self._source_traceback:
context['source_traceback'] = self._source_traceback
self._loop.call_exception_handler(context)
return
# ERROR_IO_PENDING means that the unregister is pending
self._unregister_wait_cb(None)
def cancel(self, msg=None):
self._unregister_wait()
return super().cancel(msg=msg)
def set_exception(self, exception):
self._unregister_wait()
Issue #23095, asyncio: Rewrite _WaitHandleFuture.cancel() This change fixes a race conditon related to _WaitHandleFuture.cancel() leading to Python crash or "GetQueuedCompletionStatus() returned an unexpected event" logs. Before, the overlapped object was destroyed too early, it was possible that the wait completed whereas the overlapped object was already destroyed. Sometimes, a different overlapped was allocated at the same address, leading to unexpected completition. _WaitHandleFuture.cancel() now waits until the wait is cancelled to clear its reference to the overlapped object. To wait until the cancellation is done, UnregisterWaitEx() is used with an event instead of UnregisterWait(). To wait for this event, a new _WaitCancelFuture class was added. It's a simplified version of _WaitCancelFuture. For example, its cancel() method calls UnregisterWait(), not UnregisterWaitEx(). _WaitCancelFuture should not be cancelled. The overlapped object is kept alive in _WaitHandleFuture until the wait is unregistered. Other changes: * Add _overlapped.UnregisterWaitEx() * Remove fast-path in IocpProactor.wait_for_handle() to immediatly set the result if the wait already completed. I'm not sure that it's safe to call immediatly UnregisterWaitEx() before the completion was signaled. * Add IocpProactor._unregistered() to forget an overlapped which may never be signaled, but may be signaled for the next loop iteration. It avoids to block forever IocpProactor.close() if a wait was cancelled, and it may also avoid some "... unexpected event ..." warnings.
2015-01-21 18:39:51 -04:00
super().set_exception(exception)
def set_result(self, result):
self._unregister_wait()
Issue #23095, asyncio: Rewrite _WaitHandleFuture.cancel() This change fixes a race conditon related to _WaitHandleFuture.cancel() leading to Python crash or "GetQueuedCompletionStatus() returned an unexpected event" logs. Before, the overlapped object was destroyed too early, it was possible that the wait completed whereas the overlapped object was already destroyed. Sometimes, a different overlapped was allocated at the same address, leading to unexpected completition. _WaitHandleFuture.cancel() now waits until the wait is cancelled to clear its reference to the overlapped object. To wait until the cancellation is done, UnregisterWaitEx() is used with an event instead of UnregisterWait(). To wait for this event, a new _WaitCancelFuture class was added. It's a simplified version of _WaitCancelFuture. For example, its cancel() method calls UnregisterWait(), not UnregisterWaitEx(). _WaitCancelFuture should not be cancelled. The overlapped object is kept alive in _WaitHandleFuture until the wait is unregistered. Other changes: * Add _overlapped.UnregisterWaitEx() * Remove fast-path in IocpProactor.wait_for_handle() to immediatly set the result if the wait already completed. I'm not sure that it's safe to call immediatly UnregisterWaitEx() before the completion was signaled. * Add IocpProactor._unregistered() to forget an overlapped which may never be signaled, but may be signaled for the next loop iteration. It avoids to block forever IocpProactor.close() if a wait was cancelled, and it may also avoid some "... unexpected event ..." warnings.
2015-01-21 18:39:51 -04:00
super().set_result(result)
class _WaitCancelFuture(_BaseWaitHandleFuture):
"""Subclass of Future which represents a wait for the cancellation of a
_WaitHandleFuture using an event.
"""
def __init__(self, ov, event, wait_handle, *, loop=None):
super().__init__(ov, event, wait_handle, loop=loop)
self._done_callback = None
def cancel(self):
raise RuntimeError("_WaitCancelFuture must not be cancelled")
def set_result(self, result):
super().set_result(result)
if self._done_callback is not None:
self._done_callback(self)
def set_exception(self, exception):
super().set_exception(exception)
Issue #23095, asyncio: Rewrite _WaitHandleFuture.cancel() This change fixes a race conditon related to _WaitHandleFuture.cancel() leading to Python crash or "GetQueuedCompletionStatus() returned an unexpected event" logs. Before, the overlapped object was destroyed too early, it was possible that the wait completed whereas the overlapped object was already destroyed. Sometimes, a different overlapped was allocated at the same address, leading to unexpected completition. _WaitHandleFuture.cancel() now waits until the wait is cancelled to clear its reference to the overlapped object. To wait until the cancellation is done, UnregisterWaitEx() is used with an event instead of UnregisterWait(). To wait for this event, a new _WaitCancelFuture class was added. It's a simplified version of _WaitCancelFuture. For example, its cancel() method calls UnregisterWait(), not UnregisterWaitEx(). _WaitCancelFuture should not be cancelled. The overlapped object is kept alive in _WaitHandleFuture until the wait is unregistered. Other changes: * Add _overlapped.UnregisterWaitEx() * Remove fast-path in IocpProactor.wait_for_handle() to immediatly set the result if the wait already completed. I'm not sure that it's safe to call immediatly UnregisterWaitEx() before the completion was signaled. * Add IocpProactor._unregistered() to forget an overlapped which may never be signaled, but may be signaled for the next loop iteration. It avoids to block forever IocpProactor.close() if a wait was cancelled, and it may also avoid some "... unexpected event ..." warnings.
2015-01-21 18:39:51 -04:00
if self._done_callback is not None:
self._done_callback(self)
class _WaitHandleFuture(_BaseWaitHandleFuture):
def __init__(self, ov, handle, wait_handle, proactor, *, loop=None):
super().__init__(ov, handle, wait_handle, loop=loop)
self._proactor = proactor
self._unregister_proactor = True
self._event = _overlapped.CreateEvent(None, True, False, None)
self._event_fut = None
def _unregister_wait_cb(self, fut):
if self._event is not None:
_winapi.CloseHandle(self._event)
self._event = None
self._event_fut = None
# If the wait was cancelled, the wait may never be signalled, so
# it's required to unregister it. Otherwise, IocpProactor.close() will
# wait forever for an event which will never come.
#
# If the IocpProactor already received the event, it's safe to call
# _unregister() because we kept a reference to the Overlapped object
# which is used as a unique key.
Issue #23095, asyncio: Rewrite _WaitHandleFuture.cancel() This change fixes a race conditon related to _WaitHandleFuture.cancel() leading to Python crash or "GetQueuedCompletionStatus() returned an unexpected event" logs. Before, the overlapped object was destroyed too early, it was possible that the wait completed whereas the overlapped object was already destroyed. Sometimes, a different overlapped was allocated at the same address, leading to unexpected completition. _WaitHandleFuture.cancel() now waits until the wait is cancelled to clear its reference to the overlapped object. To wait until the cancellation is done, UnregisterWaitEx() is used with an event instead of UnregisterWait(). To wait for this event, a new _WaitCancelFuture class was added. It's a simplified version of _WaitCancelFuture. For example, its cancel() method calls UnregisterWait(), not UnregisterWaitEx(). _WaitCancelFuture should not be cancelled. The overlapped object is kept alive in _WaitHandleFuture until the wait is unregistered. Other changes: * Add _overlapped.UnregisterWaitEx() * Remove fast-path in IocpProactor.wait_for_handle() to immediatly set the result if the wait already completed. I'm not sure that it's safe to call immediatly UnregisterWaitEx() before the completion was signaled. * Add IocpProactor._unregistered() to forget an overlapped which may never be signaled, but may be signaled for the next loop iteration. It avoids to block forever IocpProactor.close() if a wait was cancelled, and it may also avoid some "... unexpected event ..." warnings.
2015-01-21 18:39:51 -04:00
self._proactor._unregister(self._ov)
self._proactor = None
super()._unregister_wait_cb(fut)
def _unregister_wait(self):
if not self._registered:
return
self._registered = False
wait_handle = self._wait_handle
self._wait_handle = None
Issue #23095, asyncio: Rewrite _WaitHandleFuture.cancel() This change fixes a race conditon related to _WaitHandleFuture.cancel() leading to Python crash or "GetQueuedCompletionStatus() returned an unexpected event" logs. Before, the overlapped object was destroyed too early, it was possible that the wait completed whereas the overlapped object was already destroyed. Sometimes, a different overlapped was allocated at the same address, leading to unexpected completition. _WaitHandleFuture.cancel() now waits until the wait is cancelled to clear its reference to the overlapped object. To wait until the cancellation is done, UnregisterWaitEx() is used with an event instead of UnregisterWait(). To wait for this event, a new _WaitCancelFuture class was added. It's a simplified version of _WaitCancelFuture. For example, its cancel() method calls UnregisterWait(), not UnregisterWaitEx(). _WaitCancelFuture should not be cancelled. The overlapped object is kept alive in _WaitHandleFuture until the wait is unregistered. Other changes: * Add _overlapped.UnregisterWaitEx() * Remove fast-path in IocpProactor.wait_for_handle() to immediatly set the result if the wait already completed. I'm not sure that it's safe to call immediatly UnregisterWaitEx() before the completion was signaled. * Add IocpProactor._unregistered() to forget an overlapped which may never be signaled, but may be signaled for the next loop iteration. It avoids to block forever IocpProactor.close() if a wait was cancelled, and it may also avoid some "... unexpected event ..." warnings.
2015-01-21 18:39:51 -04:00
try:
_overlapped.UnregisterWaitEx(wait_handle, self._event)
Issue #23095, asyncio: Rewrite _WaitHandleFuture.cancel() This change fixes a race conditon related to _WaitHandleFuture.cancel() leading to Python crash or "GetQueuedCompletionStatus() returned an unexpected event" logs. Before, the overlapped object was destroyed too early, it was possible that the wait completed whereas the overlapped object was already destroyed. Sometimes, a different overlapped was allocated at the same address, leading to unexpected completition. _WaitHandleFuture.cancel() now waits until the wait is cancelled to clear its reference to the overlapped object. To wait until the cancellation is done, UnregisterWaitEx() is used with an event instead of UnregisterWait(). To wait for this event, a new _WaitCancelFuture class was added. It's a simplified version of _WaitCancelFuture. For example, its cancel() method calls UnregisterWait(), not UnregisterWaitEx(). _WaitCancelFuture should not be cancelled. The overlapped object is kept alive in _WaitHandleFuture until the wait is unregistered. Other changes: * Add _overlapped.UnregisterWaitEx() * Remove fast-path in IocpProactor.wait_for_handle() to immediatly set the result if the wait already completed. I'm not sure that it's safe to call immediatly UnregisterWaitEx() before the completion was signaled. * Add IocpProactor._unregistered() to forget an overlapped which may never be signaled, but may be signaled for the next loop iteration. It avoids to block forever IocpProactor.close() if a wait was cancelled, and it may also avoid some "... unexpected event ..." warnings.
2015-01-21 18:39:51 -04:00
except OSError as exc:
if exc.winerror != _overlapped.ERROR_IO_PENDING:
Issue #23095, asyncio: Rewrite _WaitHandleFuture.cancel() This change fixes a race conditon related to _WaitHandleFuture.cancel() leading to Python crash or "GetQueuedCompletionStatus() returned an unexpected event" logs. Before, the overlapped object was destroyed too early, it was possible that the wait completed whereas the overlapped object was already destroyed. Sometimes, a different overlapped was allocated at the same address, leading to unexpected completition. _WaitHandleFuture.cancel() now waits until the wait is cancelled to clear its reference to the overlapped object. To wait until the cancellation is done, UnregisterWaitEx() is used with an event instead of UnregisterWait(). To wait for this event, a new _WaitCancelFuture class was added. It's a simplified version of _WaitCancelFuture. For example, its cancel() method calls UnregisterWait(), not UnregisterWaitEx(). _WaitCancelFuture should not be cancelled. The overlapped object is kept alive in _WaitHandleFuture until the wait is unregistered. Other changes: * Add _overlapped.UnregisterWaitEx() * Remove fast-path in IocpProactor.wait_for_handle() to immediatly set the result if the wait already completed. I'm not sure that it's safe to call immediatly UnregisterWaitEx() before the completion was signaled. * Add IocpProactor._unregistered() to forget an overlapped which may never be signaled, but may be signaled for the next loop iteration. It avoids to block forever IocpProactor.close() if a wait was cancelled, and it may also avoid some "... unexpected event ..." warnings.
2015-01-21 18:39:51 -04:00
context = {
'message': 'Failed to unregister the wait handle',
'exception': exc,
'future': self,
}
if self._source_traceback:
context['source_traceback'] = self._source_traceback
self._loop.call_exception_handler(context)
return
# ERROR_IO_PENDING is not an error, the wait was unregistered
self._event_fut = self._proactor._wait_cancel(self._event,
self._unregister_wait_cb)
class PipeServer(object):
"""Class representing a pipe server.
This is much like a bound, listening socket.
"""
def __init__(self, address):
self._address = address
self._free_instances = weakref.WeakSet()
# initialize the pipe attribute before calling _server_pipe_handle()
# because this function can raise an exception and the destructor calls
# the close() method
self._pipe = None
self._accept_pipe_future = None
self._pipe = self._server_pipe_handle(True)
def _get_unconnected_pipe(self):
# Create new instance and return previous one. This ensures
# that (until the server is closed) there is always at least
# one pipe handle for address. Therefore if a client attempt
# to connect it will not fail with FileNotFoundError.
tmp, self._pipe = self._pipe, self._server_pipe_handle(False)
return tmp
def _server_pipe_handle(self, first):
# Return a wrapper for a new pipe handle.
if self.closed():
return None
flags = _winapi.PIPE_ACCESS_DUPLEX | _winapi.FILE_FLAG_OVERLAPPED
if first:
flags |= _winapi.FILE_FLAG_FIRST_PIPE_INSTANCE
h = _winapi.CreateNamedPipe(
self._address, flags,
_winapi.PIPE_TYPE_MESSAGE | _winapi.PIPE_READMODE_MESSAGE |
_winapi.PIPE_WAIT,
_winapi.PIPE_UNLIMITED_INSTANCES,
windows_utils.BUFSIZE, windows_utils.BUFSIZE,
_winapi.NMPWAIT_WAIT_FOREVER, _winapi.NULL)
pipe = windows_utils.PipeHandle(h)
self._free_instances.add(pipe)
return pipe
def closed(self):
return (self._address is None)
def close(self):
if self._accept_pipe_future is not None:
self._accept_pipe_future.cancel()
self._accept_pipe_future = None
# Close all instances which have not been connected to by a client.
if self._address is not None:
for pipe in self._free_instances:
pipe.close()
self._pipe = None
self._address = None
self._free_instances.clear()
__del__ = close
class _WindowsSelectorEventLoop(selector_events.BaseSelectorEventLoop):
"""Windows version of selector event loop."""
class ProactorEventLoop(proactor_events.BaseProactorEventLoop):
"""Windows version of proactor event loop using IOCP."""
def __init__(self, proactor=None):
if proactor is None:
proactor = IocpProactor()
super().__init__(proactor)
def run_forever(self):
try:
assert self._self_reading_future is None
self.call_soon(self._loop_self_reading)
super().run_forever()
finally:
if self._self_reading_future is not None:
ov = self._self_reading_future._ov
self._self_reading_future.cancel()
# self_reading_future was just cancelled so if it hasn't been
# finished yet, it never will be (it's possible that it has
# already finished and its callback is waiting in the queue,
# where it could still happen if the event loop is restarted).
# Unregister it otherwise IocpProactor.close will wait for it
# forever
if ov is not None:
self._proactor._unregister(ov)
self._self_reading_future = None
async def create_pipe_connection(self, protocol_factory, address):
f = self._proactor.connect_pipe(address)
pipe = await f
protocol = protocol_factory()
trans = self._make_duplex_pipe_transport(pipe, protocol,
extra={'addr': address})
return trans, protocol
async def start_serving_pipe(self, protocol_factory, address):
server = PipeServer(address)
2013-11-01 18:20:55 -03:00
def loop_accept_pipe(f=None):
pipe = None
try:
if f:
pipe = f.result()
server._free_instances.discard(pipe)
if server.closed():
# A client connected before the server was closed:
# drop the client (close the pipe) and exit
pipe.close()
return
protocol = protocol_factory()
self._make_duplex_pipe_transport(
pipe, protocol, extra={'addr': address})
pipe = server._get_unconnected_pipe()
if pipe is None:
return
f = self._proactor.accept_pipe(pipe)
except OSError as exc:
if pipe and pipe.fileno() != -1:
self.call_exception_handler({
'message': 'Pipe accept failed',
'exception': exc,
'pipe': pipe,
})
pipe.close()
elif self._debug:
logger.warning("Accept pipe failed on pipe %r",
pipe, exc_info=True)
except exceptions.CancelledError:
if pipe:
pipe.close()
else:
server._accept_pipe_future = f
f.add_done_callback(loop_accept_pipe)
2013-11-01 18:20:55 -03:00
self.call_soon(loop_accept_pipe)
return [server]
async def _make_subprocess_transport(self, protocol, args, shell,
stdin, stdout, stderr, bufsize,
extra=None, **kwargs):
waiter = self.create_future()
transp = _WindowsSubprocessTransport(self, protocol, args, shell,
stdin, stdout, stderr, bufsize,
asyncio: sync with Tulip Issue #23347: send_signal(), kill() and terminate() methods of BaseSubprocessTransport now check if the transport was closed and if the process exited. Issue #23347: Refactor creation of subprocess transports. Changes on BaseSubprocessTransport: * Add a wait() method to wait until the child process exit * The constructor now accepts an optional waiter parameter. The _post_init() coroutine must not be called explicitly anymore. It makes subprocess transports closer to other transports, and it gives more freedom if we want later to change completly how subprocess transports are created. * close() now kills the process instead of kindly terminate it: the child process may ignore SIGTERM and continue to run. Call explicitly terminate() and wait() if you want to kindly terminate the child process. * close() now logs a warning in debug mode if the process is still running and needs to be killed * _make_subprocess_transport() is now fully asynchronous again: if the creation of the transport failed, wait asynchronously for the process eixt. Before the wait was synchronous. This change requires close() to *kill*, and not terminate, the child process. * Remove the _kill_wait() method, replaced with a more agressive close() method. It fixes _make_subprocess_transport() on error. BaseSubprocessTransport.close() calls the close() method of pipe transports, whereas _kill_wait() closed directly pipes of the subprocess.Popen object without unregistering file descriptors from the selector (which caused severe bugs). These changes simplifies the code of subprocess.py.
2015-01-29 19:05:19 -04:00
waiter=waiter, extra=extra,
**kwargs)
try:
await waiter
except (SystemExit, KeyboardInterrupt):
raise
except BaseException:
transp.close()
await transp._wait()
raise
return transp
class IocpProactor:
"""Proactor implementation using IOCP."""
def __init__(self, concurrency=INFINITE):
self._loop = None
self._results = []
self._iocp = _overlapped.CreateIoCompletionPort(
_overlapped.INVALID_HANDLE_VALUE, NULL, 0, concurrency)
self._cache = {}
self._registered = weakref.WeakSet()
Issue #23095, asyncio: Rewrite _WaitHandleFuture.cancel() This change fixes a race conditon related to _WaitHandleFuture.cancel() leading to Python crash or "GetQueuedCompletionStatus() returned an unexpected event" logs. Before, the overlapped object was destroyed too early, it was possible that the wait completed whereas the overlapped object was already destroyed. Sometimes, a different overlapped was allocated at the same address, leading to unexpected completition. _WaitHandleFuture.cancel() now waits until the wait is cancelled to clear its reference to the overlapped object. To wait until the cancellation is done, UnregisterWaitEx() is used with an event instead of UnregisterWait(). To wait for this event, a new _WaitCancelFuture class was added. It's a simplified version of _WaitCancelFuture. For example, its cancel() method calls UnregisterWait(), not UnregisterWaitEx(). _WaitCancelFuture should not be cancelled. The overlapped object is kept alive in _WaitHandleFuture until the wait is unregistered. Other changes: * Add _overlapped.UnregisterWaitEx() * Remove fast-path in IocpProactor.wait_for_handle() to immediatly set the result if the wait already completed. I'm not sure that it's safe to call immediatly UnregisterWaitEx() before the completion was signaled. * Add IocpProactor._unregistered() to forget an overlapped which may never be signaled, but may be signaled for the next loop iteration. It avoids to block forever IocpProactor.close() if a wait was cancelled, and it may also avoid some "... unexpected event ..." warnings.
2015-01-21 18:39:51 -04:00
self._unregistered = []
self._stopped_serving = weakref.WeakSet()
def _check_closed(self):
if self._iocp is None:
raise RuntimeError('IocpProactor is closed')
def __repr__(self):
info = ['overlapped#=%s' % len(self._cache),
'result#=%s' % len(self._results)]
if self._iocp is None:
info.append('closed')
return '<%s %s>' % (self.__class__.__name__, " ".join(info))
def set_loop(self, loop):
self._loop = loop
def select(self, timeout=None):
if not self._results:
self._poll(timeout)
tmp = self._results
self._results = []
try:
return tmp
finally:
# Needed to break cycles when an exception occurs.
tmp = None
def _result(self, value):
fut = self._loop.create_future()
fut.set_result(value)
return fut
@staticmethod
def finish_socket_func(trans, key, ov):
try:
return ov.getresult()
except OSError as exc:
if exc.winerror in (_overlapped.ERROR_NETNAME_DELETED,
_overlapped.ERROR_OPERATION_ABORTED):
raise ConnectionResetError(*exc.args)
else:
raise
def recv(self, conn, nbytes, flags=0):
self._register_with_iocp(conn)
ov = _overlapped.Overlapped(NULL)
try:
if isinstance(conn, socket.socket):
ov.WSARecv(conn.fileno(), nbytes, flags)
else:
ov.ReadFile(conn.fileno(), nbytes)
except BrokenPipeError:
return self._result(b'')
2013-11-01 18:20:55 -03:00
return self._register(ov, conn, self.finish_socket_func)
def recv_into(self, conn, buf, flags=0):
self._register_with_iocp(conn)
ov = _overlapped.Overlapped(NULL)
try:
if isinstance(conn, socket.socket):
ov.WSARecvInto(conn.fileno(), buf, flags)
else:
ov.ReadFileInto(conn.fileno(), buf)
except BrokenPipeError:
return self._result(0)
return self._register(ov, conn, self.finish_socket_func)
def recvfrom(self, conn, nbytes, flags=0):
self._register_with_iocp(conn)
ov = _overlapped.Overlapped(NULL)
try:
ov.WSARecvFrom(conn.fileno(), nbytes, flags)
except BrokenPipeError:
return self._result((b'', None))
return self._register(ov, conn, self.finish_socket_func)
def recvfrom_into(self, conn, buf, flags=0):
self._register_with_iocp(conn)
ov = _overlapped.Overlapped(NULL)
try:
ov.WSARecvFromInto(conn.fileno(), buf, flags)
except BrokenPipeError:
return self._result((0, None))
def finish_recv(trans, key, ov):
try:
return ov.getresult()
except OSError as exc:
if exc.winerror in (_overlapped.ERROR_NETNAME_DELETED,
_overlapped.ERROR_OPERATION_ABORTED):
raise ConnectionResetError(*exc.args)
else:
raise
return self._register(ov, conn, finish_recv)
def sendto(self, conn, buf, flags=0, addr=None):
self._register_with_iocp(conn)
ov = _overlapped.Overlapped(NULL)
ov.WSASendTo(conn.fileno(), buf, flags, addr)
return self._register(ov, conn, self.finish_socket_func)
def send(self, conn, buf, flags=0):
self._register_with_iocp(conn)
ov = _overlapped.Overlapped(NULL)
if isinstance(conn, socket.socket):
ov.WSASend(conn.fileno(), buf, flags)
else:
ov.WriteFile(conn.fileno(), buf)
2013-11-01 18:20:55 -03:00
return self._register(ov, conn, self.finish_socket_func)
def accept(self, listener):
self._register_with_iocp(listener)
conn = self._get_accept_socket(listener.family)
ov = _overlapped.Overlapped(NULL)
ov.AcceptEx(listener.fileno(), conn.fileno())
2013-11-01 18:20:55 -03:00
def finish_accept(trans, key, ov):
ov.getresult()
# Use SO_UPDATE_ACCEPT_CONTEXT so getsockname() etc work.
buf = struct.pack('@P', listener.fileno())
conn.setsockopt(socket.SOL_SOCKET,
_overlapped.SO_UPDATE_ACCEPT_CONTEXT, buf)
conn.settimeout(listener.gettimeout())
return conn, conn.getpeername()
2013-11-01 18:20:55 -03:00
async def accept_coro(future, conn):
# Coroutine closing the accept socket if the future is cancelled
try:
await future
except exceptions.CancelledError:
conn.close()
raise
future = self._register(ov, listener, finish_accept)
coro = accept_coro(future, conn)
tasks.ensure_future(coro, loop=self._loop)
return future
def connect(self, conn, address):
if conn.type == socket.SOCK_DGRAM:
# WSAConnect will complete immediately for UDP sockets so we don't
# need to register any IOCP operation
_overlapped.WSAConnect(conn.fileno(), address)
fut = self._loop.create_future()
fut.set_result(None)
return fut
self._register_with_iocp(conn)
# The socket needs to be locally bound before we call ConnectEx().
try:
_overlapped.BindLocal(conn.fileno(), conn.family)
except OSError as e:
if e.winerror != errno.WSAEINVAL:
raise
# Probably already locally bound; check using getsockname().
if conn.getsockname()[1] == 0:
raise
ov = _overlapped.Overlapped(NULL)
ov.ConnectEx(conn.fileno(), address)
2013-11-01 18:20:55 -03:00
def finish_connect(trans, key, ov):
ov.getresult()
# Use SO_UPDATE_CONNECT_CONTEXT so getsockname() etc work.
conn.setsockopt(socket.SOL_SOCKET,
_overlapped.SO_UPDATE_CONNECT_CONTEXT, 0)
return conn
2013-11-01 18:20:55 -03:00
return self._register(ov, conn, finish_connect)
def sendfile(self, sock, file, offset, count):
self._register_with_iocp(sock)
ov = _overlapped.Overlapped(NULL)
offset_low = offset & 0xffff_ffff
offset_high = (offset >> 32) & 0xffff_ffff
ov.TransmitFile(sock.fileno(),
msvcrt.get_osfhandle(file.fileno()),
offset_low, offset_high,
count, 0, 0)
return self._register(ov, sock, self.finish_socket_func)
def accept_pipe(self, pipe):
self._register_with_iocp(pipe)
ov = _overlapped.Overlapped(NULL)
connected = ov.ConnectNamedPipe(pipe.fileno())
if connected:
# ConnectNamePipe() failed with ERROR_PIPE_CONNECTED which means
# that the pipe is connected. There is no need to wait for the
# completion of the connection.
return self._result(pipe)
2013-11-01 18:20:55 -03:00
def finish_accept_pipe(trans, key, ov):
ov.getresult()
return pipe
2013-11-01 18:20:55 -03:00
return self._register(ov, pipe, finish_accept_pipe)
async def connect_pipe(self, address):
delay = CONNECT_PIPE_INIT_DELAY
while True:
# Unfortunately there is no way to do an overlapped connect to
# a pipe. Call CreateFile() in a loop until it doesn't fail with
# ERROR_PIPE_BUSY.
try:
handle = _overlapped.ConnectPipe(address)
break
except OSError as exc:
if exc.winerror != _overlapped.ERROR_PIPE_BUSY:
raise
# ConnectPipe() failed with ERROR_PIPE_BUSY: retry later
delay = min(delay * 2, CONNECT_PIPE_MAX_DELAY)
await tasks.sleep(delay)
return windows_utils.PipeHandle(handle)
def wait_for_handle(self, handle, timeout=None):
"""Wait for a handle.
Return a Future object. The result of the future is True if the wait
completed, or False if the wait did not complete (on timeout).
"""
Issue #23095, asyncio: Rewrite _WaitHandleFuture.cancel() This change fixes a race conditon related to _WaitHandleFuture.cancel() leading to Python crash or "GetQueuedCompletionStatus() returned an unexpected event" logs. Before, the overlapped object was destroyed too early, it was possible that the wait completed whereas the overlapped object was already destroyed. Sometimes, a different overlapped was allocated at the same address, leading to unexpected completition. _WaitHandleFuture.cancel() now waits until the wait is cancelled to clear its reference to the overlapped object. To wait until the cancellation is done, UnregisterWaitEx() is used with an event instead of UnregisterWait(). To wait for this event, a new _WaitCancelFuture class was added. It's a simplified version of _WaitCancelFuture. For example, its cancel() method calls UnregisterWait(), not UnregisterWaitEx(). _WaitCancelFuture should not be cancelled. The overlapped object is kept alive in _WaitHandleFuture until the wait is unregistered. Other changes: * Add _overlapped.UnregisterWaitEx() * Remove fast-path in IocpProactor.wait_for_handle() to immediatly set the result if the wait already completed. I'm not sure that it's safe to call immediatly UnregisterWaitEx() before the completion was signaled. * Add IocpProactor._unregistered() to forget an overlapped which may never be signaled, but may be signaled for the next loop iteration. It avoids to block forever IocpProactor.close() if a wait was cancelled, and it may also avoid some "... unexpected event ..." warnings.
2015-01-21 18:39:51 -04:00
return self._wait_for_handle(handle, timeout, False)
def _wait_cancel(self, event, done_callback):
fut = self._wait_for_handle(event, None, True)
# add_done_callback() cannot be used because the wait may only complete
# in IocpProactor.close(), while the event loop is not running.
fut._done_callback = done_callback
return fut
def _wait_for_handle(self, handle, timeout, _is_cancel):
self._check_closed()
if timeout is None:
ms = _winapi.INFINITE
else:
# RegisterWaitForSingleObject() has a resolution of 1 millisecond,
# round away from zero to wait *at least* timeout seconds.
ms = math.ceil(timeout * 1e3)
# We only create ov so we can use ov.address as a key for the cache.
ov = _overlapped.Overlapped(NULL)
Issue #23095, asyncio: Rewrite _WaitHandleFuture.cancel() This change fixes a race conditon related to _WaitHandleFuture.cancel() leading to Python crash or "GetQueuedCompletionStatus() returned an unexpected event" logs. Before, the overlapped object was destroyed too early, it was possible that the wait completed whereas the overlapped object was already destroyed. Sometimes, a different overlapped was allocated at the same address, leading to unexpected completition. _WaitHandleFuture.cancel() now waits until the wait is cancelled to clear its reference to the overlapped object. To wait until the cancellation is done, UnregisterWaitEx() is used with an event instead of UnregisterWait(). To wait for this event, a new _WaitCancelFuture class was added. It's a simplified version of _WaitCancelFuture. For example, its cancel() method calls UnregisterWait(), not UnregisterWaitEx(). _WaitCancelFuture should not be cancelled. The overlapped object is kept alive in _WaitHandleFuture until the wait is unregistered. Other changes: * Add _overlapped.UnregisterWaitEx() * Remove fast-path in IocpProactor.wait_for_handle() to immediatly set the result if the wait already completed. I'm not sure that it's safe to call immediatly UnregisterWaitEx() before the completion was signaled. * Add IocpProactor._unregistered() to forget an overlapped which may never be signaled, but may be signaled for the next loop iteration. It avoids to block forever IocpProactor.close() if a wait was cancelled, and it may also avoid some "... unexpected event ..." warnings.
2015-01-21 18:39:51 -04:00
wait_handle = _overlapped.RegisterWaitWithQueue(
handle, self._iocp, ov.address, ms)
Issue #23095, asyncio: Rewrite _WaitHandleFuture.cancel() This change fixes a race conditon related to _WaitHandleFuture.cancel() leading to Python crash or "GetQueuedCompletionStatus() returned an unexpected event" logs. Before, the overlapped object was destroyed too early, it was possible that the wait completed whereas the overlapped object was already destroyed. Sometimes, a different overlapped was allocated at the same address, leading to unexpected completition. _WaitHandleFuture.cancel() now waits until the wait is cancelled to clear its reference to the overlapped object. To wait until the cancellation is done, UnregisterWaitEx() is used with an event instead of UnregisterWait(). To wait for this event, a new _WaitCancelFuture class was added. It's a simplified version of _WaitCancelFuture. For example, its cancel() method calls UnregisterWait(), not UnregisterWaitEx(). _WaitCancelFuture should not be cancelled. The overlapped object is kept alive in _WaitHandleFuture until the wait is unregistered. Other changes: * Add _overlapped.UnregisterWaitEx() * Remove fast-path in IocpProactor.wait_for_handle() to immediatly set the result if the wait already completed. I'm not sure that it's safe to call immediatly UnregisterWaitEx() before the completion was signaled. * Add IocpProactor._unregistered() to forget an overlapped which may never be signaled, but may be signaled for the next loop iteration. It avoids to block forever IocpProactor.close() if a wait was cancelled, and it may also avoid some "... unexpected event ..." warnings.
2015-01-21 18:39:51 -04:00
if _is_cancel:
f = _WaitCancelFuture(ov, handle, wait_handle, loop=self._loop)
else:
f = _WaitHandleFuture(ov, handle, wait_handle, self,
loop=self._loop)
if f._source_traceback:
del f._source_traceback[-1]
def finish_wait_for_handle(trans, key, ov):
# Note that this second wait means that we should only use
# this with handles types where a successful wait has no
# effect. So events or processes are all right, but locks
# or semaphores are not. Also note if the handle is
# signalled and then quickly reset, then we may return
# False even though we have not timed out.
return f._poll()
self._cache[ov.address] = (f, ov, 0, finish_wait_for_handle)
return f
def _register_with_iocp(self, obj):
# To get notifications of finished ops on this objects sent to the
# completion port, were must register the handle.
if obj not in self._registered:
self._registered.add(obj)
_overlapped.CreateIoCompletionPort(obj.fileno(), self._iocp, 0, 0)
# XXX We could also use SetFileCompletionNotificationModes()
# to avoid sending notifications to completion port of ops
# that succeed immediately.
def _register(self, ov, obj, callback):
self._check_closed()
# Return a future which will be set with the result of the
# operation when it completes. The future's value is actually
# the value returned by callback().
f = _OverlappedFuture(ov, loop=self._loop)
if f._source_traceback:
del f._source_traceback[-1]
if not ov.pending:
# The operation has completed, so no need to postpone the
# work. We cannot take this short cut if we need the
# NumberOfBytes, CompletionKey values returned by
# PostQueuedCompletionStatus().
try:
value = callback(None, None, ov)
except OSError as e:
f.set_exception(e)
else:
f.set_result(value)
# Even if GetOverlappedResult() was called, we have to wait for the
# notification of the completion in GetQueuedCompletionStatus().
# Register the overlapped operation to keep a reference to the
# OVERLAPPED object, otherwise the memory is freed and Windows may
# read uninitialized memory.
# Register the overlapped operation for later. Note that
# we only store obj to prevent it from being garbage
# collected too early.
self._cache[ov.address] = (f, ov, obj, callback)
return f
Issue #23095, asyncio: Rewrite _WaitHandleFuture.cancel() This change fixes a race conditon related to _WaitHandleFuture.cancel() leading to Python crash or "GetQueuedCompletionStatus() returned an unexpected event" logs. Before, the overlapped object was destroyed too early, it was possible that the wait completed whereas the overlapped object was already destroyed. Sometimes, a different overlapped was allocated at the same address, leading to unexpected completition. _WaitHandleFuture.cancel() now waits until the wait is cancelled to clear its reference to the overlapped object. To wait until the cancellation is done, UnregisterWaitEx() is used with an event instead of UnregisterWait(). To wait for this event, a new _WaitCancelFuture class was added. It's a simplified version of _WaitCancelFuture. For example, its cancel() method calls UnregisterWait(), not UnregisterWaitEx(). _WaitCancelFuture should not be cancelled. The overlapped object is kept alive in _WaitHandleFuture until the wait is unregistered. Other changes: * Add _overlapped.UnregisterWaitEx() * Remove fast-path in IocpProactor.wait_for_handle() to immediatly set the result if the wait already completed. I'm not sure that it's safe to call immediatly UnregisterWaitEx() before the completion was signaled. * Add IocpProactor._unregistered() to forget an overlapped which may never be signaled, but may be signaled for the next loop iteration. It avoids to block forever IocpProactor.close() if a wait was cancelled, and it may also avoid some "... unexpected event ..." warnings.
2015-01-21 18:39:51 -04:00
def _unregister(self, ov):
"""Unregister an overlapped object.
Call this method when its future has been cancelled. The event can
already be signalled (pending in the proactor event queue). It is also
safe if the event is never signalled (because it was cancelled).
"""
self._check_closed()
Issue #23095, asyncio: Rewrite _WaitHandleFuture.cancel() This change fixes a race conditon related to _WaitHandleFuture.cancel() leading to Python crash or "GetQueuedCompletionStatus() returned an unexpected event" logs. Before, the overlapped object was destroyed too early, it was possible that the wait completed whereas the overlapped object was already destroyed. Sometimes, a different overlapped was allocated at the same address, leading to unexpected completition. _WaitHandleFuture.cancel() now waits until the wait is cancelled to clear its reference to the overlapped object. To wait until the cancellation is done, UnregisterWaitEx() is used with an event instead of UnregisterWait(). To wait for this event, a new _WaitCancelFuture class was added. It's a simplified version of _WaitCancelFuture. For example, its cancel() method calls UnregisterWait(), not UnregisterWaitEx(). _WaitCancelFuture should not be cancelled. The overlapped object is kept alive in _WaitHandleFuture until the wait is unregistered. Other changes: * Add _overlapped.UnregisterWaitEx() * Remove fast-path in IocpProactor.wait_for_handle() to immediatly set the result if the wait already completed. I'm not sure that it's safe to call immediatly UnregisterWaitEx() before the completion was signaled. * Add IocpProactor._unregistered() to forget an overlapped which may never be signaled, but may be signaled for the next loop iteration. It avoids to block forever IocpProactor.close() if a wait was cancelled, and it may also avoid some "... unexpected event ..." warnings.
2015-01-21 18:39:51 -04:00
self._unregistered.append(ov)
def _get_accept_socket(self, family):
s = socket.socket(family)
s.settimeout(0)
return s
def _poll(self, timeout=None):
if timeout is None:
ms = INFINITE
elif timeout < 0:
raise ValueError("negative timeout")
else:
# GetQueuedCompletionStatus() has a resolution of 1 millisecond,
# round away from zero to wait *at least* timeout seconds.
ms = math.ceil(timeout * 1e3)
if ms >= INFINITE:
raise ValueError("timeout too big")
while True:
status = _overlapped.GetQueuedCompletionStatus(self._iocp, ms)
if status is None:
Issue #23095, asyncio: Rewrite _WaitHandleFuture.cancel() This change fixes a race conditon related to _WaitHandleFuture.cancel() leading to Python crash or "GetQueuedCompletionStatus() returned an unexpected event" logs. Before, the overlapped object was destroyed too early, it was possible that the wait completed whereas the overlapped object was already destroyed. Sometimes, a different overlapped was allocated at the same address, leading to unexpected completition. _WaitHandleFuture.cancel() now waits until the wait is cancelled to clear its reference to the overlapped object. To wait until the cancellation is done, UnregisterWaitEx() is used with an event instead of UnregisterWait(). To wait for this event, a new _WaitCancelFuture class was added. It's a simplified version of _WaitCancelFuture. For example, its cancel() method calls UnregisterWait(), not UnregisterWaitEx(). _WaitCancelFuture should not be cancelled. The overlapped object is kept alive in _WaitHandleFuture until the wait is unregistered. Other changes: * Add _overlapped.UnregisterWaitEx() * Remove fast-path in IocpProactor.wait_for_handle() to immediatly set the result if the wait already completed. I'm not sure that it's safe to call immediatly UnregisterWaitEx() before the completion was signaled. * Add IocpProactor._unregistered() to forget an overlapped which may never be signaled, but may be signaled for the next loop iteration. It avoids to block forever IocpProactor.close() if a wait was cancelled, and it may also avoid some "... unexpected event ..." warnings.
2015-01-21 18:39:51 -04:00
break
ms = 0
err, transferred, key, address = status
try:
f, ov, obj, callback = self._cache.pop(address)
except KeyError:
if self._loop.get_debug():
self._loop.call_exception_handler({
'message': ('GetQueuedCompletionStatus() returned an '
'unexpected event'),
'status': ('err=%s transferred=%s key=%#x address=%#x'
% (err, transferred, key, address)),
})
# key is either zero, or it is used to return a pipe
# handle which should be closed to avoid a leak.
if key not in (0, _overlapped.INVALID_HANDLE_VALUE):
_winapi.CloseHandle(key)
continue
if obj in self._stopped_serving:
f.cancel()
# Don't call the callback if _register() already read the result or
# if the overlapped has been cancelled
elif not f.done():
try:
value = callback(transferred, key, ov)
except OSError as e:
f.set_exception(e)
self._results.append(f)
else:
f.set_result(value)
self._results.append(f)
finally:
f = None
2017-12-13 11:50:16 -04:00
# Remove unregistered futures
Issue #23095, asyncio: Rewrite _WaitHandleFuture.cancel() This change fixes a race conditon related to _WaitHandleFuture.cancel() leading to Python crash or "GetQueuedCompletionStatus() returned an unexpected event" logs. Before, the overlapped object was destroyed too early, it was possible that the wait completed whereas the overlapped object was already destroyed. Sometimes, a different overlapped was allocated at the same address, leading to unexpected completition. _WaitHandleFuture.cancel() now waits until the wait is cancelled to clear its reference to the overlapped object. To wait until the cancellation is done, UnregisterWaitEx() is used with an event instead of UnregisterWait(). To wait for this event, a new _WaitCancelFuture class was added. It's a simplified version of _WaitCancelFuture. For example, its cancel() method calls UnregisterWait(), not UnregisterWaitEx(). _WaitCancelFuture should not be cancelled. The overlapped object is kept alive in _WaitHandleFuture until the wait is unregistered. Other changes: * Add _overlapped.UnregisterWaitEx() * Remove fast-path in IocpProactor.wait_for_handle() to immediatly set the result if the wait already completed. I'm not sure that it's safe to call immediatly UnregisterWaitEx() before the completion was signaled. * Add IocpProactor._unregistered() to forget an overlapped which may never be signaled, but may be signaled for the next loop iteration. It avoids to block forever IocpProactor.close() if a wait was cancelled, and it may also avoid some "... unexpected event ..." warnings.
2015-01-21 18:39:51 -04:00
for ov in self._unregistered:
self._cache.pop(ov.address, None)
self._unregistered.clear()
def _stop_serving(self, obj):
# obj is a socket or pipe handle. It will be closed in
# BaseProactorEventLoop._stop_serving() which will make any
# pending operations fail quickly.
self._stopped_serving.add(obj)
def close(self):
if self._iocp is None:
# already closed
return
# Cancel remaining registered operations.
for fut, ov, obj, callback in list(self._cache.values()):
if fut.cancelled():
# Nothing to do with cancelled futures
pass
elif isinstance(fut, _WaitCancelFuture):
# _WaitCancelFuture must not be cancelled
pass
else:
try:
fut.cancel()
except OSError as exc:
if self._loop is not None:
context = {
'message': 'Cancelling a future failed',
'exception': exc,
'future': fut,
}
if fut._source_traceback:
context['source_traceback'] = fut._source_traceback
self._loop.call_exception_handler(context)
# Wait until all cancelled overlapped complete: don't exit with running
# overlapped to prevent a crash. Display progress every second if the
# loop is still running.
msg_update = 1.0
start_time = time.monotonic()
next_msg = start_time + msg_update
while self._cache:
if next_msg <= time.monotonic():
logger.debug('%r is running after closing for %.1f seconds',
self, time.monotonic() - start_time)
next_msg = time.monotonic() + msg_update
# handle a few events, or timeout
self._poll(msg_update)
self._results = []
_winapi.CloseHandle(self._iocp)
self._iocp = None
def __del__(self):
self.close()
class _WindowsSubprocessTransport(base_subprocess.BaseSubprocessTransport):
def _start(self, args, shell, stdin, stdout, stderr, bufsize, **kwargs):
self._proc = windows_utils.Popen(
args, shell=shell, stdin=stdin, stdout=stdout, stderr=stderr,
bufsize=bufsize, **kwargs)
2013-11-01 18:20:55 -03:00
def callback(f):
returncode = self._proc.poll()
self._process_exited(returncode)
2013-11-01 18:20:55 -03:00
f = self._loop._proactor.wait_for_handle(int(self._proc._handle))
f.add_done_callback(callback)
SelectorEventLoop = _WindowsSelectorEventLoop
class WindowsSelectorEventLoopPolicy(events.BaseDefaultEventLoopPolicy):
_loop_factory = SelectorEventLoop
class WindowsProactorEventLoopPolicy(events.BaseDefaultEventLoopPolicy):
_loop_factory = ProactorEventLoop
DefaultEventLoopPolicy = WindowsProactorEventLoopPolicy