# -*- coding: utf-8 -*-
# Copyright (C) 2013 Yahoo! Inc. All Rights Reserved.
#
# 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.
import abc
import collections
from multiprocessing import managers
import os
import pickle
import threading
import futurist
from oslo_utils import excutils
from oslo_utils import reflection
from oslo_utils import timeutils
from oslo_utils import uuidutils
import six
from six.moves import queue as compat_queue
from taskflow import logging
from taskflow import task as task_atom
from taskflow.types import failure
from taskflow.types import notifier
from taskflow.utils import threading_utils
# Execution and reversion outcomes.
EXECUTED = 'executed'
REVERTED = 'reverted'
# See http://bugs.python.org/issue1457119 for why this is so complex...
_PICKLE_ERRORS = [pickle.PickleError, TypeError]
try:
import cPickle as _cPickle
_PICKLE_ERRORS.append(_cPickle.PickleError)
except ImportError:
pass
_PICKLE_ERRORS = tuple(_PICKLE_ERRORS)
_SEND_ERRORS = (IOError, EOFError)
_UPDATE_PROGRESS = task_atom.EVENT_UPDATE_PROGRESS
# Message types/kind sent from worker/child processes...
_KIND_COMPLETE_ME = 'complete_me'
_KIND_EVENT = 'event'
LOG = logging.getLogger(__name__)
def _execute_retry(retry, arguments):
try:
result = retry.execute(**arguments)
except Exception:
result = failure.Failure()
return (EXECUTED, result)
def _revert_retry(retry, arguments):
try:
result = retry.revert(**arguments)
except Exception:
result = failure.Failure()
return (REVERTED, result)
def _execute_task(task, arguments, progress_callback=None):
with notifier.register_deregister(task.notifier,
_UPDATE_PROGRESS,
callback=progress_callback):
try:
task.pre_execute()
result = task.execute(**arguments)
except Exception:
# NOTE(imelnikov): wrap current exception with Failure
# object and return it.
result = failure.Failure()
finally:
task.post_execute()
return (EXECUTED, result)
def _revert_task(task, arguments, result, failures, progress_callback=None):
arguments = arguments.copy()
arguments[task_atom.REVERT_RESULT] = result
arguments[task_atom.REVERT_FLOW_FAILURES] = failures
with notifier.register_deregister(task.notifier,
_UPDATE_PROGRESS,
callback=progress_callback):
try:
task.pre_revert()
result = task.revert(**arguments)
except Exception:
# NOTE(imelnikov): wrap current exception with Failure
# object and return it.
result = failure.Failure()
finally:
task.post_revert()
return (REVERTED, result)
class _ViewableSyncManager(managers.SyncManager):
"""Manager that exposes its state as methods."""
def is_shutdown(self):
return self._state.value == managers.State.SHUTDOWN
def is_running(self):
return self._state.value == managers.State.STARTED
class _Channel(object):
"""Helper wrapper around a multiprocessing queue used by a worker."""
def __init__(self, queue, identity):
self._queue = queue
self._identity = identity
self._sent_messages = collections.defaultdict(int)
self._pid = None
@property
def sent_messages(self):
return self._sent_messages
def put(self, message):
# NOTE(harlowja): this is done in late in execution to ensure that this
# happens in the child process and not the parent process (where the
# constructor is called).
if self._pid is None:
self._pid = os.getpid()
message.update({
'sent_on': timeutils.utcnow(),
'sender': {
'pid': self._pid,
'id': self._identity,
},
})
if 'body' not in message:
message['body'] = {}
try:
self._queue.put(message)
except _PICKLE_ERRORS:
LOG.warn("Failed serializing message %s", message, exc_info=True)
return False
except _SEND_ERRORS:
LOG.warn("Failed sending message %s", message, exc_info=True)
return False
else:
self._sent_messages[message['kind']] += 1
return True
class _WaitWorkItem(object):
"""The piece of work that will executed by a process executor.
This will call the target function, then wait until the tasks emitted
events/items have been depleted before offically being finished.
NOTE(harlowja): this is done so that the task function will *not* return
until all of its notifications have been proxied back to its originating
task. If we didn't do this then the executor would see this task as done
and then potentially start tasks that are successors of the task that just
finished even though notifications are still left to be sent from the
previously finished task...
"""
def __init__(self, channel, barrier,
func, task, *args, **kwargs):
self._channel = channel
self._barrier = barrier
self._func = func
self._task = task
self._args = args
self._kwargs = kwargs
def _on_finish(self):
sent_events = self._channel.sent_messages.get(_KIND_EVENT, 0)
if sent_events:
message = {
'created_on': timeutils.utcnow(),
'kind': _KIND_COMPLETE_ME,
}
if self._channel.put(message):
watch = timeutils.StopWatch()
watch.start()
self._barrier.wait()
LOG.trace("Waited %s seconds until task '%s' %s emitted"
" notifications were depleted", watch.elapsed(),
self._task, sent_events)
def __call__(self):
args = self._args
kwargs = self._kwargs
try:
return self._func(self._task, *args, **kwargs)
finally:
self._on_finish()
class _EventSender(object):
"""Sends event information from a child worker process to its creator."""
def __init__(self, channel):
self._channel = channel
def __call__(self, event_type, details):
message = {
'created_on': timeutils.utcnow(),
'kind': _KIND_EVENT,
'body': {
'event_type': event_type,
'details': details,
},
}
self._channel.put(message)
class _Target(object):
"""An immutable helper object that represents a target of a message."""
def __init__(self, task, barrier, identity):
self.task = task
self.barrier = barrier
self.identity = identity
# Counters used to track how many message 'kinds' were proxied...
self.dispatched = collections.defaultdict(int)
def __repr__(self):
return "<%s at 0x%x targeting '%s' with identity '%s'>" % (
reflection.get_class_name(self), id(self),
self.task, self.identity)
class _Dispatcher(object):
"""Dispatches messages received from child worker processes."""
# When the run() method is busy (typically in a thread) we want to set
# these so that the thread can know how long to sleep when there is no
# active work to dispatch.
_SPIN_PERIODICITY = 0.01
def __init__(self, dispatch_periodicity=None):
if dispatch_periodicity is None:
dispatch_periodicity = self._SPIN_PERIODICITY
if dispatch_periodicity <= 0:
raise ValueError("Provided dispatch periodicity must be greater"
" than zero and not '%s'" % dispatch_periodicity)
self._targets = {}
self._dead = threading.Event()
self._dispatch_periodicity = dispatch_periodicity
self._stop_when_empty = False
def register(self, identity, target):
self._targets[identity] = target
def deregister(self, identity):
try:
target = self._targets.pop(identity)
except KeyError:
pass
else:
# Just incase set the barrier to unblock any worker...
target.barrier.set()
if LOG.isEnabledFor(logging.TRACE):
LOG.trace("Dispatched %s messages %s to target '%s' during"
" the lifetime of its existence in the dispatcher",
sum(six.itervalues(target.dispatched)),
dict(target.dispatched), target)
def reset(self):
self._stop_when_empty = False
self._dead.clear()
if self._targets:
leftover = set(six.iterkeys(self._targets))
while leftover:
self.deregister(leftover.pop())
def interrupt(self):
self._stop_when_empty = True
self._dead.set()
def _dispatch(self, message):
if LOG.isEnabledFor(logging.TRACE):
LOG.trace("Dispatching message %s (it took %s seconds"
" for it to arrive for processing after being"
" sent)", message,
timeutils.delta_seconds(message['sent_on'],
timeutils.utcnow()))
try:
kind = message['kind']
sender = message['sender']
body = message['body']
except (KeyError, ValueError, TypeError):
LOG.warn("Badly formatted message %s received", message,
exc_info=True)
return
target = self._targets.get(sender['id'])
if target is None:
# Must of been removed...
return
if kind == _KIND_COMPLETE_ME:
target.dispatched[kind] += 1
target.barrier.set()
elif kind == _KIND_EVENT:
task = target.task
target.dispatched[kind] += 1
task.notifier.notify(body['event_type'], body['details'])
else:
LOG.warn("Unknown message '%s' found in message from sender"
" %s to target '%s'", kind, sender, target)
def run(self, queue):
watch = timeutils.StopWatch(duration=self._dispatch_periodicity)
while (not self._dead.is_set() or
(self._stop_when_empty and self._targets)):
watch.restart()
leftover = watch.leftover()
while leftover:
try:
message = queue.get(timeout=leftover)
except compat_queue.Empty:
break
else:
self._dispatch(message)
leftover = watch.leftover()
leftover = watch.leftover()
if leftover:
self._dead.wait(leftover)
[docs]class SerialRetryExecutor(object):
"""Executes and reverts retries."""
def __init__(self):
self._executor = futurist.SynchronousExecutor()
[docs] def start(self):
"""Prepare to execute retries."""
self._executor.restart()
[docs] def stop(self):
"""Finalize retry executor."""
self._executor.shutdown()
[docs] def execute_retry(self, retry, arguments):
"""Schedules retry execution."""
fut = self._executor.submit(_execute_retry, retry, arguments)
fut.atom = retry
return fut
[docs] def revert_retry(self, retry, arguments):
"""Schedules retry reversion."""
fut = self._executor.submit(_revert_retry, retry, arguments)
fut.atom = retry
return fut
@six.add_metaclass(abc.ABCMeta)
[docs]class TaskExecutor(object):
"""Executes and reverts tasks.
This class takes task and its arguments and executes or reverts it.
It encapsulates knowledge on how task should be executed or reverted:
right now, on separate thread, on another machine, etc.
"""
@abc.abstractmethod
[docs] def execute_task(self, task, task_uuid, arguments,
progress_callback=None):
"""Schedules task execution."""
@abc.abstractmethod
[docs] def revert_task(self, task, task_uuid, arguments, result, failures,
progress_callback=None):
"""Schedules task reversion."""
[docs] def start(self):
"""Prepare to execute tasks."""
[docs] def stop(self):
"""Finalize task executor."""
[docs]class SerialTaskExecutor(TaskExecutor):
"""Executes tasks one after another."""
def __init__(self):
self._executor = futurist.SynchronousExecutor()
def start(self):
self._executor.restart()
def stop(self):
self._executor.shutdown()
def execute_task(self, task, task_uuid, arguments, progress_callback=None):
fut = self._executor.submit(_execute_task,
task, arguments,
progress_callback=progress_callback)
fut.atom = task
return fut
def revert_task(self, task, task_uuid, arguments, result, failures,
progress_callback=None):
fut = self._executor.submit(_revert_task,
task, arguments, result, failures,
progress_callback=progress_callback)
fut.atom = task
return fut
[docs]class ParallelTaskExecutor(TaskExecutor):
"""Executes tasks in parallel.
Submits tasks to an executor which should provide an interface similar
to concurrent.Futures.Executor.
"""
constructor_options = [
('max_workers', lambda v: v if v is None else int(v)),
]
"""
Optional constructor keyword arguments this executor supports. These will
typically be passed via engine options (by a engine user) and converted
into the correct type before being sent into this
classes ``__init__`` method.
"""
def __init__(self, executor=None, max_workers=None):
self._executor = executor
self._max_workers = max_workers
self._own_executor = executor is None
@abc.abstractmethod
def _create_executor(self, max_workers=None):
"""Called when an executor has not been provided to make one."""
def _submit_task(self, func, task, *args, **kwargs):
fut = self._executor.submit(func, task, *args, **kwargs)
fut.atom = task
return fut
def execute_task(self, task, task_uuid, arguments, progress_callback=None):
return self._submit_task(_execute_task, task, arguments,
progress_callback=progress_callback)
def revert_task(self, task, task_uuid, arguments, result, failures,
progress_callback=None):
return self._submit_task(_revert_task, task, arguments, result,
failures, progress_callback=progress_callback)
def start(self):
if self._own_executor:
self._executor = self._create_executor(
max_workers=self._max_workers)
def stop(self):
if self._own_executor:
self._executor.shutdown(wait=True)
self._executor = None
[docs]class ParallelThreadTaskExecutor(ParallelTaskExecutor):
"""Executes tasks in parallel using a thread pool executor."""
def _create_executor(self, max_workers=None):
return futurist.ThreadPoolExecutor(max_workers=max_workers)
[docs]class ParallelProcessTaskExecutor(ParallelTaskExecutor):
"""Executes tasks in parallel using a process pool executor.
NOTE(harlowja): this executor executes tasks in external processes, so that
implies that tasks that are sent to that external process are pickleable
since this is how the multiprocessing works (sending pickled objects back
and forth) and that the bound handlers (for progress updating in
particular) are proxied correctly from that external process to the one
that is alive in the parent process to ensure that callbacks registered in
the parent are executed on events in the child.
"""
constructor_options = [
('max_workers', lambda v: v if v is None else int(v)),
('dispatch_periodicity', lambda v: v if v is None else float(v)),
]
"""
Optional constructor keyword arguments this executor supports. These will
typically be passed via engine options (by a engine user) and converted
into the correct type before being sent into this
classes ``__init__`` method.
"""
def __init__(self, executor=None, max_workers=None,
dispatch_periodicity=None):
super(ParallelProcessTaskExecutor, self).__init__(
executor=executor, max_workers=max_workers)
self._manager = _ViewableSyncManager()
self._dispatcher = _Dispatcher(
dispatch_periodicity=dispatch_periodicity)
# Only created after starting...
self._worker = None
self._queue = None
def _create_executor(self, max_workers=None):
return futurist.ProcessPoolExecutor(max_workers=max_workers)
def start(self):
if threading_utils.is_alive(self._worker):
raise RuntimeError("Worker thread must be stopped via stop()"
" before starting/restarting")
super(ParallelProcessTaskExecutor, self).start()
# These don't seem restartable; make a new one...
if self._manager.is_shutdown():
self._manager = _ViewableSyncManager()
if not self._manager.is_running():
self._manager.start()
self._dispatcher.reset()
self._queue = self._manager.Queue()
self._worker = threading_utils.daemon_thread(self._dispatcher.run,
self._queue)
self._worker.start()
def stop(self):
self._dispatcher.interrupt()
super(ParallelProcessTaskExecutor, self).stop()
if threading_utils.is_alive(self._worker):
self._worker.join()
self._worker = None
self._queue = None
self._dispatcher.reset()
self._manager.shutdown()
self._manager.join()
def _rebind_task(self, task, clone, channel, progress_callback=None):
# Creates and binds proxies for all events the task could receive
# so that when the clone runs in another process that this task
# can recieve the same notifications (thus making it look like the
# the notifications are transparently happening in this process).
needed = set()
for (event_type, listeners) in task.notifier.listeners_iter():
if listeners:
needed.add(event_type)
if progress_callback is not None:
needed.add(_UPDATE_PROGRESS)
if needed:
sender = _EventSender(channel)
for event_type in needed:
clone.notifier.register(event_type, sender)
def _submit_task(self, func, task, *args, **kwargs):
"""Submit a function to run the given task (with given args/kwargs).
NOTE(harlowja): Adjust all events to be proxies instead since we want
those callbacks to be activated in this process, not in the child,
also since typically callbacks are functors (or callables) we can
not pickle those in the first place...
To make sure people understand how this works, the following is a
lengthy description of what is going on here, read at will:
So to ensure that we are proxying task triggered events that occur
in the executed subprocess (which will be created and used by the
thing using the multiprocessing based executor) we need to establish
a link between that process and this process that ensures that when a
event is triggered in that task in that process that a corresponding
event is triggered on the original task that was requested to be ran
in this process.
To accomplish this we have to create a copy of the task (without
any listeners) and then reattach a new set of listeners that will
now instead of calling the desired listeners just place messages
for this process (a dispatcher thread that is created in this class)
to dispatch to the original task (using a common queue + per task
sender identity/target that is used and associated to know which task
to proxy back too, since it is possible that there many be *many*
subprocess running at the same time, each running a different task
and using the same common queue to submit messages back to).
Once the subprocess task has finished execution, the executor will
then trigger a callback that will remove the task + target from the
dispatcher (which will stop any further proxying back to the original
task).
"""
progress_callback = kwargs.pop('progress_callback', None)
clone = task.copy(retain_listeners=False)
identity = uuidutils.generate_uuid()
target = _Target(task, self._manager.Event(), identity)
channel = _Channel(self._queue, identity)
self._rebind_task(task, clone, channel,
progress_callback=progress_callback)
def register():
if progress_callback is not None:
task.notifier.register(_UPDATE_PROGRESS, progress_callback)
self._dispatcher.register(identity, target)
def deregister():
if progress_callback is not None:
task.notifier.deregister(_UPDATE_PROGRESS, progress_callback)
self._dispatcher.deregister(identity)
register()
work = _WaitWorkItem(channel, target.barrier,
func, clone, *args, **kwargs)
try:
fut = self._executor.submit(work)
except RuntimeError:
with excutils.save_and_reraise_exception():
deregister()
fut.atom = task
fut.add_done_callback(lambda fut: deregister())
return fut