def test_socket_manager_recv_safe_not_ready(wait_until_ready):
("SocketManager().recv_safe() should return None when the socket is not ready")
# Background: wait_until_ready is mocked to return None
wait_until_ready.return_value = None
# Given a zmq mock
zmq = Mock()
# And a context
context = Mock()
# And a serializer
serializer = Mock(name='serializer')
# And a socket manager
manager = SocketManager(zmq, context, serialization_backend=serializer)
# When I call .recv_safe()
result = manager.recv_safe('foobar')
# Then it should never had anything to unpack
serializer.unpack.called.should.be.false
# And the result should be None
result.should.be.none
def test_socket_manager_recv_safe(wait_until_ready):
("SocketManager().recv_safe() should deserialize "
"after receiving, using the configured backend")
# Background: wait_until_ready is mocked to return the socket
wait_until_ready.side_effect = lambda name, *args: manager.sockets[name]
# Given a zmq mock
zmq = Mock()
# And a context
context = Mock()
# And a serializer
serializer = Mock(name='serializer')
# And a socket manager
manager = SocketManager(zmq, context, serialization_backend=serializer)
# And a socket
socket = manager.create('foobar', zmq.REP)
# When I call .recv_safe()
result = manager.recv_safe('foobar')
# Then it should have unpacked the payload after receiving
serializer.unpack.assert_called_once_with(socket.recv.return_value)
# And the result should be the unpacked value
result.should.equal(serializer.unpack.return_value)
def test_socket_manager_recv_safe_not_ready(wait_until_ready):
("SocketManager().recv_safe() should return None when the socket is not ready"
)
# Background: wait_until_ready is mocked to return None
wait_until_ready.return_value = None
# Given a zmq mock
zmq = Mock()
# And a context
context = Mock()
# And a serializer
serializer = Mock(name='serializer')
# And a socket manager
manager = SocketManager(zmq, context, serialization_backend=serializer)
# When I call .recv_safe()
result = manager.recv_safe('foobar')
# Then it should never had anything to unpack
serializer.unpack.called.should.be.false
# And the result should be None
result.should.be.none
def test_socket_manager_recv_safe(wait_until_ready):
("SocketManager().recv_safe() should deserialize "
"after receiving, using the configured backend")
# Background: wait_until_ready is mocked to return the socket
wait_until_ready.side_effect = lambda name, *args: manager.sockets[name]
# Given a zmq mock
zmq = Mock()
# And a context
context = Mock()
# And a serializer
serializer = Mock(name='serializer')
# And a socket manager
manager = SocketManager(zmq, context, serialization_backend=serializer)
# And a socket
socket = manager.create('foobar', zmq.REP)
# When I call .recv_safe()
result = manager.recv_safe('foobar')
# Then it should have unpacked the payload after receiving
serializer.unpack.assert_called_once_with(socket.recv.return_value)
# And the result should be the unpacked value
result.should.equal(serializer.unpack.return_value)
class GraphClient(object):
def __init__(self, request_connect_addr='tcp://127.0.0.1:6000'):
self.context = zmq.Context()
self.sockets = SocketManager(zmq, self.context)
self.sockets.create('query', zmq.REQ)
def request(self, data):
self.sockets.send_safe('query', data)
return self.sockets.recv_safe('query')
class GraphClient(object):
def __init__(self, request_connect_addr='tcp://127.0.0.1:6000'):
self.context = zmq.Context()
self.sockets = SocketManager(zmq, self.context)
self.sockets.ensure_and_connect('query', zmq.REQ, request_connect_addr, zmq.POLLIN | zmq.POLLOUT)
print 'connected to', request_connect_addr
def request(self, data):
self.sockets.send_safe('query', data)
result = self.sockets.recv_safe('query')
return result
def query(self, string):
return self.request({'query': string})
class GraphServer(object):
def __init__(self, sleep_interval=3, pool_size=8):
self.context = zmq.Context()
self.sockets = SocketManager(zmq, self.context)
self.sleep_interval = sleep_interval
self.pool_size = pool_size
self.pool = Pool(pool_size)
self.allowed_to_run = Event()
self.allowed_to_run.set()
def should_run(self):
return self.allowed_to_run.is_set()
def listener_coroutine(self, socket_name, reply_bind_addr):
# TODO: DEALER/ROUTER + internal multiplexing of request ids?
self.sockets.ensure_and_bind(socket_name, zmq.REP, reply_bind_addr,
zmq.POLLIN | zmq.POLLOUT)
self.sockets.set_socket_option(socket_name, zmq.IDENTITY, socket_name)
while self.should_run():
if not self.process_request(socket_name):
logger.debug('waiting %s second%s', self.sleep_interval,
self.sleep_interval == 1 and '' or 's')
def process_request(self, socket_name):
data = self.sockets.recv_safe(socket_name, timeout=self.sleep_interval)
if data:
logger.info('processed request')
pprint(data)
query = data['query']
result = schema.execute(query)
if result.errors:
payload = {'errors': "\n".join(map(unicode, result.errors))}
else:
payload = {'result': json.dumps(result.data, indent=2)}
self.sockets.send_safe(socket_name, payload)
return True
return False
def run(self, reply_bind_addr='tcp://*:6000'):
logger.info('listening on %s', reply_bind_addr)
self.pool.spawn(self.listener_coroutine, 'reply1', reply_bind_addr)
self.pool.join(raise_error=True)
def stop(self):
self.allowed_to_run.clear()
self.sockets.close('reply1')
self.pool.kill()
class Step(object):
def __init__(self, pull_bind_address='tcp://127.0.0.1', subscriber_connect_address='tcp://127.0.0.1:6000', concurrency=100, timeout=1):
self.context = zmq.Context()
self.sockets = SocketManager(zmq, self.context)
self.sockets.create('pull-in', zmq.PULL)
# self.sockets.set_socket_option('pull-in', zmq.RCVHWM, concurrency)
self.sockets.create('events', zmq.PUB)
self.name = self.__class__.__name__
self.subscriber_connect_address = subscriber_connect_address
self._allowed_to_run = True
self.pool = gevent.pool.Pool(concurrency + 1)
self.timeout = timeout
self.pull_bind_address = pull_bind_address
self.id = str(uuid.uuid4())
self.logger = self.sockets.get_logger('events', 'logs', 'logs')
def listen(self):
_, self.address = self.sockets.bind_to_random_port('pull-in', zmq.POLLIN, local_address=self.pull_bind_address)
gevent.sleep(1)
def connect(self):
self.sockets.connect('events', self.subscriber_connect_address, zmq.POLLOUT)
gevent.sleep(1)
self.notify_available()
def notify_available(self):
self.send_event('available', self.to_dict())
def should_run(self):
gevent.sleep(0.1)
return self._allowed_to_run
def send_event(self, name, data):
self.sockets.publish_safe('events', name, data)
def dispatch(self, job):
try:
start = time.time()
job['job_started_at'] = start
self.send_event('job:started', job)
job['instructions'] = self.execute(job['instructions'])
job['job_finished_at'] = time.time()
self.logger.info("done processing %s", job)
except Exception as e:
job['job_finished_at'] = time.time()
job['error'] = serialized_exception(e)
self.send_event('job:error', job)
job['success'] = False
self.send_event('job:failed', job)
self.logger.exception('failed to execute job {id}'.format(**dict(job)))
finally:
end = time.time()
job['end'] = end
if job.get('instructions'):
self.send_event('job:success', job)
else:
self.send_event('job:failed', job)
def to_dict(self):
return dict(
id=self.id,
address=self.address,
job_type=self.job_type,
step_name=self.name,
)
def loop(self):
self.listen()
self.connect()
self.logger.info('listening for jobs on %s', self.address)
while self.should_run():
if self.pool.free_count() == 0:
self.logger.info('waiting for an execution slot')
self.pool.wait_available()
job = self.sockets.recv_safe('pull-in')
if job:
self.logger.info('received job')
self.pool.spawn(self.dispatch, job)
else:
self.notify_available()
gevent.sleep(1)
class Pipeline(object):
def __init__(self, name, steps=[]):
self.name = name
self.actions = Speaker(
'actions',
[
'available',
'failed',
'started',
'success',
'metric',
'error',
'logs',
]
)
self.steps = [s.job_type for s in steps]
self.total_steps = len(steps)
self.context = zmq.Context()
self.sockets = SocketManager(zmq, self.context)
self.sockets.create('step-events', zmq.SUB)
self.sockets.create('jobs-in', zmq.PULL)
for step in self.steps:
self.sockets.create(step, zmq.PUSH)
for action in self.actions.actions.keys():
self.bind_action(action)
self.total_actions = len(self.actions.actions)
self.pool = gevent.pool.Pool(self.total_actions ** (self.total_steps + 1))
self.greenlets = []
self._allowed_to_run = True
self.default_interval = 0.1
self.backend = StorageBackend()
self.logger = logging.getLogger('pipeline')
def on_started(self, event):
worker = Worker.from_event(event)
self.logger.info('%s [%s] started to process a job', worker.job_type, worker.id)
def on_available(self, event):
worker = Worker.from_event(event)
if self.backend.register_worker(worker):
self.sockets.connect(worker.job_type, worker.address, zmq.POLLOUT)
self.logger.info('connected to worker: [%s]', dict(worker))
def on_failed(self, event):
worker = Worker.from_event(event)
self.logger.warning('%s [%s] failed', worker.job_type, worker.id)
def on_success(self, event):
worker = Worker.from_event(event)
self.logger.info('%s [%s] success', worker.job_type, worker.id)
self.enqueue_next_job(event.data)
def on_metric(self, event):
self.logger.info(' '.join([event.topic, event.data]))
def on_error(self, event):
worker = Worker.from_event(event)
self.logger.warning('%s [%s] errored: %s', worker.job_type, worker.id, event)
def on_logs(self, event):
msg = event.data.pop('msg', None)
if msg:
self.logger.debug(msg)
def enqueue_next_job(self, data):
result = data.pop('instructions')
job = Job.from_dict(data)
job['instructions'] = result
step_index = self.steps.index(job.type)
try:
next_job_type = self.steps[step_index + 1]
except IndexError:
next_job_type = None
if next_job_type:
self.logger.info("enqueuing next job: %s", next_job_type)
job['job_type'] = next_job_type
self.backend.enqueue_job(job)
def bind_action(self, name, method=None):
action = getattr(self.actions, name, None)
if not action:
raise KeyError('undefined action: {0}'.format(name))
method = method or getattr(self, 'on_{0}'.format(name), None)
if not method:
raise TypeError('{0} does not have method {1}(self, topic, data)'.format(self.__class__, name))
action(lambda _, event: self.spawn(method, event))
def should_run(self):
return self._allowed_to_run
def listen(self, subscriber_bind_address='tcp://127.0.0.1:6000', pull_bind_address='tcp://127.0.0.1:7000'):
self.sockets.bind('step-events', subscriber_bind_address, zmq.POLLIN)
self.sockets.bind('jobs-in', pull_bind_address, zmq.POLLIN)
self.logger.info('listening for events on %s', subscriber_bind_address)
self.logger.info('listening for instructions on %s', pull_bind_address)
def route_event(self, event):
if not event:
return
ROUTES = {
re.compile(r'available'): self.actions.available,
re.compile(r'failed'): self.actions.failed,
re.compile(r'success'): self.actions.success,
re.compile(r'started'): self.actions.started,
re.compile(r'metric'): self.actions.metric,
re.compile(r'logs'): self.actions.logs,
re.compile(r'error'): self.actions.error,
}
matched = False
for regex, action in ROUTES.items():
if regex.search(event.topic):
action.shout(event)
matched = True
if not matched:
print 'unmatched event', event.topic, event.data
def drain_jobs_in(self):
while self.should_run():
data = self.sockets.recv_safe('jobs-in')
if not data:
gevent.sleep(0)
continue
job = Job.new(data)
self.backend.enqueue_job(job)
gevent.sleep(0)
def drain_jobs_out(self):
iteration = -1
while self.should_run():
iteration += 1
index = iteration % len(self.steps)
job_type = self.steps[index]
worker = self.backend.get_next_available_worker_for_type(job_type)
if not worker:
gevent.sleep(0)
continue
job = self.backend.dequeue_job_of_type(job_type)
if not job:
gevent.sleep(0)
continue
self.sockets.send_safe(worker.job_type, job.to_dict())
gevent.sleep(0)
def spawn(self, *args, **kw):
self.greenlets.append(
self.pool.spawn(*args, **kw)
)
def idle(self):
gevent.sleep(0)
def loop(self):
self.listen()
self.spawn(self.drain_events)
self.spawn(self.drain_jobs_in)
self.spawn(self.drain_jobs_out)
while self.should_run():
gevent.sleep(5)
def drain_events(self):
# drain events
while self.should_run():
event = self.sockets.recv_event_safe('step-events')
if event:
self.route_event(event)
gevent.sleep(0)
else:
self.idle()
