def get_backend(backend_uri=None):
    tmp_dir = None
    if not backend_uri:
        if not backend_uri:
            tmp_dir = tempfile.mkdtemp()
            backend_uri = "file:///%s" % tmp_dir
    try:
        backend = backends.fetch(_make_conf(backend_uri))
    except exceptions.NotFound as e:
        # Fallback to one that will work if the provided backend is not found.
        if not tmp_dir:
            tmp_dir = tempfile.mkdtemp()
            backend_uri = "file:///%s" % tmp_dir
            LOG.exception("Falling back to file backend using temporary"
                          " directory located at: %s", tmp_dir)
            backend = backends.fetch(_make_conf(backend_uri))
        else:
            raise e
    try:
        # Ensure schema upgraded before we continue working.
        with contextlib.closing(backend.get_connection()) as conn:
            conn.upgrade()
        yield backend
    finally:
        # Make sure to cleanup the temporary path if one was created for us.
        if tmp_dir:
            rm_path(tmp_dir)

def save_factory_details(flow_detail, flow_factory, factory_args, factory_kwargs, backend=None):
    """Saves the given factories reimportable attributes into the flow detail.

    This function saves the factory name, arguments, and keyword arguments
    into the given flow details object  and if a backend is provided it will
    also ensure that the backend saves the flow details after being updated.

    :param flow_detail: FlowDetail that holds state of the flow to load
    :param flow_factory: function or string: function that creates the flow
    :param factory_args: list or tuple of factory positional arguments
    :param factory_kwargs: dict of factory keyword arguments
    :param backend: storage backend to use or configuration
    """
    if not factory_args:
        factory_args = []
    if not factory_kwargs:
        factory_kwargs = {}
    factory_name, _factory_fun = _fetch_validate_factory(flow_factory)
    factory_data = {"factory": {"name": factory_name, "args": factory_args, "kwargs": factory_kwargs}}
    if not flow_detail.meta:
        flow_detail.meta = factory_data
    else:
        flow_detail.meta.update(factory_data)
    if backend is not None:
        if isinstance(backend, dict):
            backend = p_backends.fetch(backend)
        with contextlib.closing(backend.get_connection()) as conn:
            conn.update_flow_details(flow_detail)

 def _create_engine(**kwargs):
     flow = lf.Flow('test-flow').add(utils.DummyTask())
     backend = backends.fetch({'connection': 'memory'})
     flow_detail = pu.create_flow_detail(flow, backend=backend)
     options = kwargs.copy()
     return engine.WorkerBasedActionEngine(flow, flow_detail,
                                           backend, options)

def load(flow, store=None, flow_detail=None, book=None,
         engine_conf=None, backend=None,
         namespace=ENGINES_NAMESPACE, engine=ENGINE_DEFAULT, **kwargs):
    """Load a flow into an engine.

    This function creates and prepares an engine to run the provided flow. All
    that is left after this returns is to run the engine with the
    engines :py:meth:`~taskflow.engines.base.Engine.run` method.

    Which engine to load is specified via the ``engine`` parameter. It
    can be a string that names the engine type to use, or a string that
    is a URI with a scheme that names the engine type to use and further
    options contained in the URI's host, port, and query parameters...

    Which storage backend to use is defined by the backend parameter. It
    can be backend itself, or a dictionary that is passed to
    :py:func:`~taskflow.persistence.backends.fetch` to obtain a
    viable backend.

    :param flow: flow to load
    :param store: dict -- data to put to storage to satisfy flow requirements
    :param flow_detail: FlowDetail that holds the state of the flow (if one is
        not provided then one will be created for you in the provided backend)
    :param book: LogBook to create flow detail in if flow_detail is None
    :param engine_conf: engine type or URI and options (**deprecated**)
    :param backend: storage backend to use or configuration that defines it
    :param namespace: driver namespace for stevedore (or empty for default)
    :param engine: string engine type or URI string with scheme that contains
                   the engine type and any URI specific components that will
                   become part of the engine options.
    :param kwargs: arbitrary keyword arguments passed as options (merged with
                   any extracted ``engine`` and ``engine_conf`` options),
                   typically used for any engine specific options that do not
                   fit as any of the existing arguments.
    :returns: engine
    """

    kind, options = _extract_engine(engine_conf=engine_conf,
                                    engine=engine, **kwargs)

    if isinstance(backend, dict):
        backend = p_backends.fetch(backend)

    if flow_detail is None:
        flow_detail = p_utils.create_flow_detail(flow, book=book,
                                                 backend=backend)

    LOG.debug('Looking for %r engine driver in %r', kind, namespace)
    try:
        mgr = stevedore.driver.DriverManager(
            namespace, kind,
            invoke_on_load=True,
            invoke_args=(flow, flow_detail, backend, options))
        engine = mgr.driver
    except RuntimeError as e:
        raise exc.NotFound("Could not find engine '%s'" % (kind), e)
    else:
        if store:
            engine.storage.inject(store)
        return engine

 def upgrade_backend(self, persistence_backend):
     try:
         backend = backends.fetch(persistence_backend)
         with contextlib.closing(backend.get_connection()) as conn:
             conn.upgrade()
     except exceptions.NotFound as e:
         raise e

def main():
    # Need to share the same backend, so that data can be shared...
    persistence_conf = {
        'connection': 'memory',
    }
    saver = persistence.fetch(persistence_conf)
    with contextlib.closing(saver.get_connection()) as conn:
        # This ensures that the needed backend setup/data directories/schema
        # upgrades and so on... exist before they are attempted to be used...
        conn.upgrade()
    fc1 = fake_client.FakeClient()
    # Done like this to share the same client storage location so the correct
    # zookeeper features work across clients...
    fc2 = fake_client.FakeClient(storage=fc1.storage)
    entities = [
        generate_reviewer(fc1, saver),
        generate_conductor(fc2, saver),
    ]
    for t, stopper in entities:
        t.start()
    try:
        watch = timeutils.StopWatch(duration=RUN_TIME)
        watch.start()
        while not watch.expired():
            time.sleep(0.1)
    finally:
        for t, stopper in reversed(entities):
            stopper()
            t.join()

def load_from_factory(flow_factory, factory_args=None, factory_kwargs=None,
                      store=None, book=None, engine_conf=None, backend=None,
                      namespace=ENGINES_NAMESPACE, engine=ENGINE_DEFAULT,
                      **kwargs):
    """Loads a flow from a factory function into an engine.

    Gets flow factory function (or name of it) and creates flow with
    it. Then, the flow is loaded into an engine with the :func:`load() <load>`
    function, and the factory function fully qualified name is saved to flow
    metadata so that it can be later resumed.

    :param flow_factory: function or string: function that creates the flow
    :param factory_args: list or tuple of factory positional arguments
    :param factory_kwargs: dict of factory keyword arguments

    Further arguments are interpreted as for :func:`load() <load>`.

    :returns: engine
    """

    _factory_name, factory_fun = _fetch_validate_factory(flow_factory)
    if not factory_args:
        factory_args = []
    if not factory_kwargs:
        factory_kwargs = {}
    flow = factory_fun(*factory_args, **factory_kwargs)
    if isinstance(backend, dict):
        backend = p_backends.fetch(backend)
    flow_detail = p_utils.create_flow_detail(flow, book=book, backend=backend)
    save_factory_details(flow_detail,
                         flow_factory, factory_args, factory_kwargs,
                         backend=backend)
    return load(flow=flow, store=store, flow_detail=flow_detail, book=book,
                engine_conf=engine_conf, backend=backend, namespace=namespace,
                engine=engine, **kwargs)

def execute_flow(flow):
    """
    Create all necessary prerequisites like task database and thread pool and
    execute TaskFlow flow.
    :param flow: TaskFlow flow instance
    """
    backend = backends.fetch({
        'connection': 'sqlite:///' + TASK_DATABASE_FILE,
        'isolation_level': 'SERIALIZABLE'
    })
    executor = futurist.ThreadPoolExecutor(max_workers=MAX_WORKERS)
    conn = backend.get_connection()
    logbook, flow_detail = _ensure_db_initialized(conn, flow)
    engine = engines.load(
        flow, flow_detail=flow_detail, backend=backend, book=logbook,
        engine='parallel', executor=executor)

    engine.compile()
    _workaround_reverted_reset(flow_detail)
    try:
        engine.run()
    except exceptions.WrappedFailure as wf:
        for failure in wf:
            if failure.exc_info is not None:
                traceback.print_exception(*failure.exc_info)
            else:
                print failure

def run_poster():
    # This just posts a single job and then ends...
    print("Starting poster with pid: %s" % ME)
    my_name = "poster-%s" % ME
    persist_backend = persistence_backends.fetch(PERSISTENCE_URI)
    with contextlib.closing(persist_backend):
        with contextlib.closing(persist_backend.get_connection()) as conn:
            conn.upgrade()
        job_backend = job_backends.fetch(my_name, JB_CONF,
                                         persistence=persist_backend)
        job_backend.connect()
        with contextlib.closing(job_backend):
            # Create information in the persistence backend about the
            # unit of work we want to complete and the factory that
            # can be called to create the tasks that the work unit needs
            # to be done.
            lb = models.LogBook("post-from-%s" % my_name)
            fd = models.FlowDetail("song-from-%s" % my_name,
                                   uuidutils.generate_uuid())
            lb.add(fd)
            with contextlib.closing(persist_backend.get_connection()) as conn:
                conn.save_logbook(lb)
            engines.save_factory_details(fd, make_bottles,
                                         [HOW_MANY_BOTTLES], {},
                                         backend=persist_backend)
            # Post, and be done with it!
            jb = job_backend.post("song-from-%s" % my_name, book=lb)
            print("Posted: %s" % jb)
            print("Goodbye...")

 def test_entrypoint(self):
     # Test that the entrypoint fetching also works (even with dialects)
     # using the same configuration we used in setUp() but not using
     # the impl_sqlalchemy SQLAlchemyBackend class directly...
     with contextlib.closing(backends.fetch(self.db_conf)) as backend:
         with contextlib.closing(backend.get_connection()):
             pass

 def test_file_persistence_entry_point(self):
     conf = {
         'connection': 'file:',
         'path': self.path
     }
     with contextlib.closing(backends.fetch(conf)) as be:
         self.assertIsInstance(be, impl_dir.DirBackend)

def get_backend():
    try:
        backend_uri = sys.argv[1]
    except Exception:
        backend_uri = 'sqlite://'
    backend = backends.fetch({'connection': backend_uri})
    backend.get_connection().upgrade()
    return backend

 def test_dir_persistence_entry_point(self):
     conf = {
         'connection': 'dir:',
         'path': self.path
     }
     backend = backends.fetch(conf)
     self.assertIsInstance(backend, impl_dir.DirBackend)
     backend.close()

def load(flow, store=None, flow_detail=None, book=None,
         engine_conf=None, backend=None, namespace=ENGINES_NAMESPACE):
    """Load flow into engine.

    This function creates and prepares engine to run the
    flow. All that is left is to run the engine with 'run()' method.

    Which engine to load is specified in 'engine_conf' parameter. It
    can be a string that names engine type or a dictionary which holds
    engine type (with 'engine' key) and additional engine-specific
    configuration (for example, executor for multithreaded engine).

    Which storage backend to use is defined by backend parameter. It
    can be backend itself, or a dictionary that is passed to
    taskflow.persistence.backends.fetch to obtain backend.

    :param flow: flow to load
    :param store: dict -- data to put to storage to satisfy flow requirements
    :param flow_detail: FlowDetail that holds the state of the flow (if one is
        not provided then one will be created for you in the provided backend)
    :param book: LogBook to create flow detail in if flow_detail is None
    :param engine_conf: engine type and configuration configuration
    :param backend: storage backend to use or configuration
    :param namespace: driver namespace for stevedore (default is fine
       if you don't know what is it)
    :returns: engine
    """

    if engine_conf is None:
        engine_conf = {'engine': 'default'}

    # NOTE(imelnikov): this allows simpler syntax.
    if isinstance(engine_conf, six.string_types):
        engine_conf = {'engine': engine_conf}

    if isinstance(backend, dict):
        backend = p_backends.fetch(backend)

    if flow_detail is None:
        flow_detail = p_utils.create_flow_detail(flow, book=book,
                                                 backend=backend)

    mgr = stevedore.driver.DriverManager(
        namespace, engine_conf['engine'],
        invoke_on_load=True,
        invoke_kwds={
            'conf': engine_conf.copy(),
            'flow': flow,
            'flow_detail': flow_detail,
            'backend': backend
        })
    engine = mgr.driver
    if store:
        engine.storage.inject(store)
    return engine

def run_conductor(only_run_once=False):
    # This continuously consumers until its stopped via ctrl-c or other
    # kill signal...
    event_watches = {}

    # This will be triggered by the conductor doing various activities
    # with engines, and is quite nice to be able to see the various timing
    # segments (which is useful for debugging, or watching, or figuring out
    # where to optimize).
    def on_conductor_event(cond, event, details):
        print("Event '%s' has been received..." % event)
        print("Details = %s" % details)
        if event.endswith("_start"):
            w = timing.StopWatch()
            w.start()
            base_event = event[0:-len("_start")]
            event_watches[base_event] = w
        if event.endswith("_end"):
            base_event = event[0:-len("_end")]
            try:
                w = event_watches.pop(base_event)
                w.stop()
                print("It took %0.3f seconds for event '%s' to finish"
                      % (w.elapsed(), base_event))
            except KeyError:
                pass
        if event == 'running_end' and only_run_once:
            cond.stop()

    print("Starting conductor with pid: %s" % ME)
    my_name = "conductor-%s" % ME
    persist_backend = persistence_backends.fetch(PERSISTENCE_URI)
    with contextlib.closing(persist_backend):
        with contextlib.closing(persist_backend.get_connection()) as conn:
            conn.upgrade()
        job_backend = job_backends.fetch(my_name, JB_CONF,
                                         persistence=persist_backend)
        job_backend.connect()
        with contextlib.closing(job_backend):
            cond = conductor_backends.fetch('blocking', my_name, job_backend,
                                            persistence=persist_backend)
            on_conductor_event = functools.partial(on_conductor_event, cond)
            cond.notifier.register(cond.notifier.ANY, on_conductor_event)
            # Run forever, and kill -9 or ctrl-c me...
            try:
                cond.run()
            finally:
                cond.stop()
                cond.wait()

def load_taskflow_into_engine(action, nested_flow,
                              process_what):
    book = None
    backend = None
    if PERSISTENCE_BACKEND:
        backend = backends.fetch(PERSISTENCE_BACKEND)
        with contextlib.closing(backend.get_connection()) as conn:
            try:
                book = conn.get_logbook(process_what['notification_uuid'])
            except exceptions.NotFound:
                pass
            if book is None:
                book = models.LogBook(action,
                                      process_what['notification_uuid'])

    return taskflow.engines.load(nested_flow, store=process_what,
                                 backend=backend, book=book)

def load_from_factory(
    flow_factory, factory_args=None, factory_kwargs=None, store=None, book=None, engine_conf=None, backend=None
):
    """Load flow from factory function into engine

    Gets flow factory function (or name of it) and creates flow with
    it. Then, flow is loaded into engine with load(), and factory
    function fully qualified name is saved to flow metadata so that
    it can be later resumed with resume.

    :param flow_factory: function or string: function that creates the flow
    :param factory_args: list or tuple of factory positional arguments
    :param factory_kwargs: dict of factory keyword arguments
    :param store: dict -- data to put to storage to satisfy flow requirements
    :param book: LogBook to create flow detail in
    :param engine_conf: engine type and configuration configuration
    :param backend: storage backend to use or configuration
    :returns: engine
    """

    if isinstance(flow_factory, six.string_types):
        factory_fun = importutils.import_class(flow_factory)
        factory_name = flow_factory
    else:
        factory_fun = flow_factory
        factory_name = reflection.get_callable_name(flow_factory)
        try:
            reimported = importutils.import_class(factory_name)
            assert reimported == factory_fun
        except (ImportError, AssertionError):
            raise ValueError("Flow factory %r is not reimportable by name %s" % (factory_fun, factory_name))

    args = factory_args or []
    kwargs = factory_kwargs or {}
    flow = factory_fun(*args, **kwargs)
    factory_data = dict(name=factory_name, args=args, kwargs=kwargs)

    if isinstance(backend, dict):
        backend = p_backends.fetch(backend)
    flow_detail = p_utils.create_flow_detail(flow, book=book, backend=backend, meta={"factory": factory_data})
    return load(flow=flow, flow_detail=flow_detail, store=store, book=book, engine_conf=engine_conf, backend=backend)

def main():
    persistence = persistence_backends.fetch({
        'connection': 'sqlite:////tmp/taskflow.db'

    })

    board = HypernodeJobBoard('my-board', {
        "hosts": "localhost",
    }, persistence=persistence)

    # board = job_backends.fetch("my-board", {
    #     "board": "zookeeper",
    #     "hosts": "localhost",
    #     "path": "/jobboard",
    # }, persistence=persistence)
    board.connect()

    # conductor = conductors.fetch("blocking", "executor 1", board, engine="parallel", wait_timeout=.1)
    conductor = AsyncConductor("async", board, engine="parallel")

    with contextlib.closing(board):
        conductor.run()

def load_from_factory(flow_factory, factory_args=None, factory_kwargs=None,
                      store=None, book=None, engine_conf=None, backend=None,
                      namespace=ENGINES_NAMESPACE, **kwargs):
    """Loads a flow from a factory function into an engine.

    Gets flow factory function (or name of it) and creates flow with
    it. Then, flow is loaded into engine with load(), and factory
    function fully qualified name is saved to flow metadata so that
    it can be later resumed with resume.

    :param flow_factory: function or string: function that creates the flow
    :param factory_args: list or tuple of factory positional arguments
    :param factory_kwargs: dict of factory keyword arguments
    :param store: dict -- data to put to storage to satisfy flow requirements
    :param book: LogBook to create flow detail in
    :param engine_conf: engine type and configuration configuration
    :param backend: storage backend to use or configuration
    :param namespace: driver namespace for stevedore (default is fine
       if you don't know what is it)
    :returns: engine
    """

    _factory_name, factory_fun = _fetch_validate_factory(flow_factory)
    if not factory_args:
        factory_args = []
    if not factory_kwargs:
        factory_kwargs = {}
    flow = factory_fun(*factory_args, **factory_kwargs)
    if isinstance(backend, dict):
        backend = p_backends.fetch(backend)
    flow_detail = p_utils.create_flow_detail(flow, book=book, backend=backend)
    save_factory_details(flow_detail,
                         flow_factory, factory_args, factory_kwargs,
                         backend=backend)
    return load(flow=flow, store=store, flow_detail=flow_detail, book=book,
                engine_conf=engine_conf, backend=backend, namespace=namespace,
                **kwargs)

 def test_postgres_persistence_entry_point(self):
     uri = _get_connect_string('postgres', USER, PASSWD, database=DATABASE)
     conf = {'connection': uri}
     with contextlib.closing(backends.fetch(conf)) as be:
         self.assertIsInstance(be, impl_sqlalchemy.SQLAlchemyBackend)