def test_socket_manager_recv_event_safe_missing_socket(wait_until_ready, set_topic):
("SocketManager().recv_event_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)
# And a socket
socket = manager.create('foobar', zmq.REP)
socket.recv_multipart.return_value = b'a fake topic', 'the-raw-payload'
# When I call .recv_event_safe()
result = manager.recv_event_safe('foobar', topic=b'helloworld')
# And the result should be None
result.should.be.none
def test_socket_manager_recv_event_safe_missing_socket(wait_until_ready,
set_topic):
("SocketManager().recv_event_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)
# And a socket
socket = manager.create('foobar', zmq.REP)
socket.recv_multipart.return_value = b'a fake topic', 'the-raw-payload'
# When I call .recv_event_safe()
result = manager.recv_event_safe('foobar', topic=b'helloworld')
# And the result should be None
result.should.be.none
def test_socket_manager_recv_event_safe(wait_until_ready, set_topic):
("SocketManager().recv_event_safe() should set the topic "
"on the given socket, wait for data to become available "
"and return it")
# 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)
socket.recv_multipart.return_value = b'a fake topic', 'the-raw-payload'
# When I call .recv_event_safe()
result = manager.recv_event_safe('foobar', topic=b'helloworld')
# Then it should have unpacked the payload after receiving
serializer.unpack.assert_called_once_with('the-raw-payload')
# And the result should be the unpacked value
result.should.be.an(Event)
def test_socket_manager_recv_event_safe(wait_until_ready, set_topic):
("SocketManager().recv_event_safe() should set the topic "
"on the given socket, wait for data to become available "
"and return it")
# 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)
socket.recv_multipart.return_value = b'a fake topic', 'the-raw-payload'
# When I call .recv_event_safe()
result = manager.recv_event_safe('foobar', topic=b'helloworld')
# Then it should have unpacked the payload after receiving
serializer.unpack.assert_called_once_with('the-raw-payload')
# And the result should be the unpacked value
result.should.be.an(Event)
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()
