this is better.

separate workers, failures and success monitors
This commit is contained in:
Ilan Steemers
2015-06-15 20:53:42 +02:00
parent 8c0a26e0b4
commit 317101d821
2 changed files with 81 additions and 64 deletions
+79 -64
View File
@@ -3,111 +3,125 @@ from time import sleep
from multiprocessing import Queue, Process, current_process, cpu_count from multiprocessing import Queue, Process, current_process, cpu_count
import sys import sys
import signal import signal
from uuid import uuid4
import ujson as json
import redis import redis
from django.conf import settings from django.conf import settings
try: r = redis.StrictRedis(decode_responses=True)
import cPickle as pickle
except ImportError:
import pickle
r = redis.StrictRedis()
secret = settings.SECRET_KEY secret = settings.SECRET_KEY
prefix = 'django_q' prefix = 'django_q'
q_list = '{}:q'.format(prefix) q_list = '{}:q'.format(prefix)
def test(): def test():
for i in range(20): for i in range(4):
defer('testq.tasks.multiply', 5, i) defer(u'testq.tasks.multiply', 5, i)
def defer(func, *args, **kwargs): def defer(func, *args, **kwargs):
# serialize the func pack = json.dumps([uuid4().urn, func, args, kwargs])
task = {u'func': func, u'args': args, u'kwargs': kwargs}
pack = pickle.dumps(task)
r.rpush(q_list, pack) r.rpush(q_list, pack)
class Worker(object): class Worker(object):
def __init__(self): def __init__(self):
self.running = True self.running = True
self.stable_size = cpu_count() self.stable_size = cpu_count() - 2
self.stable = [] self.stable = []
self.task_queue = Queue() self.task_queue = Queue()
self.done_queue = Queue() self.done_queue = Queue()
self.fail_queue = Queue()
self.failure_monitor_pid = None
self.success_monitor_pid = None
# Spawn work horses # Spawn work horses
for i in range(self.stable_size): for i in range(self.stable_size):
self._spawn_horse() self.spawn_horse()
# Spawn monitor # Spawn monitors
self._spawn_monitor() self.spawn_success_monitor()
self.spawn_failure_monitor()
# Attach signal handler # Attach signal handler
signal.signal(signal.SIGTERM, self.sighandler) signal.signal(signal.SIGTERM, self.sig_handler)
signal.signal(signal.SIGINT, self.sighandler) signal.signal(signal.SIGINT, self.sig_handler)
# Keep popping Redis # Keep popping Redis
while self.running: while self.running:
self.work() self.stable_boy()
sleep(0.2) sleep(0.2)
task = r.lpop(q_list)
if task:
self.task_queue.put(task)
def _spawn_horse(self): def spawn_process(self, target, *args):
# This is just for PyCharm to not crash. Ignore it. # This is just for PyCharm to not crash. Ignore it.
if not hasattr(sys.stdin, 'close'): if not hasattr(sys.stdin, 'close'):
def dummy_close(): def dummy_close():
pass pass
sys.stdin.close = dummy_close sys.stdin.close = dummy_close
p = Process(target=target, args=args)
p = Process(target=self.horse, args=(self.task_queue, self.done_queue))
self.stable.append(p) self.stable.append(p)
p.start() p.start()
return p.pid
def _spawn_monitor(self): def spawn_horse(self):
# This is just for PyCharm to not crash. Ignore it. self.spawn_process(self.horse, self.task_queue, self.done_queue, self.fail_queue)
if not hasattr(sys.stdin, 'close'):
def dummy_close():
pass
sys.stdin.close = dummy_close def spawn_success_monitor(self):
self.success_monitor_pid = self.spawn_process(self.success_monitor, self.done_queue)
self.mon = Process(target=self.monitor, args=(self.done_queue,)) def spawn_failure_monitor(self):
self.mon.start() self.failure_monitor_pid = self.spawn_process(self.failure_monitor, self.fail_queue)
def reincarnate(self, pid):
if pid == self.success_monitor_pid:
self.spawn_success_monitor()
elif pid == self.failure_monitor_pid:
self.spawn_failure_monitor()
else:
self.spawn_horse()
@staticmethod @staticmethod
def monitor(queue_in): def success_monitor(done_queue):
print("Monitor started at {}".format(current_process().pid))
for result in iter(queue_in.get, 'STOP'):
task = result['task']
res = result['result']
print("{} - {}".format(task['func'], res))
print("Monitor stopped")
@staticmethod
def horse(queue_in, queue_out):
name = current_process().name name = current_process().name
print(name, 'Ready for work at {}'.format(current_process().pid)) print("{} monitoring successes at {}".format(name, current_process().pid))
for pack in iter(queue_in.get, 'STOP'): for result in iter(done_queue.get, 'STOP'):
task = pickle.loads(pack) task = result[0]
func = task['func'] res = result[1]
print("Success [{}:{} - {}]".format(task[1], task[0], res))
print("{} stopped".format(name))
@staticmethod
def failure_monitor(fail_queue):
name = current_process().name
print("{} monitoring failures at {}".format(name, current_process().pid))
for result in iter(fail_queue.get, 'STOP'):
task = result[0]
e = result[1]
print("Failure [{}:{} - {}]".format(task[1], task[0], e))
print("{} stopped".format(name))
@staticmethod
def horse(task_queue, done_queue, fail_queue):
name = current_process().name
print('{} ready for work at {}'.format(name, current_process().pid))
for pack in iter(task_queue.get, 'STOP'):
task = json.loads(pack)
uid = task[0]
func = task[1]
module, func = func.rsplit('.', 1) module, func = func.rsplit('.', 1)
args = task['args'] args = task[2]
kwargs = task['kwargs'] kwargs = task[3]
print(name, 'Starting Task {}'.format(func)) print(name, 'Processing [{}:{}]'.format(func, uid))
try: try:
m = importlib.import_module(module) m = importlib.import_module(module)
f = getattr(m, func) f = getattr(m, func)
result = f(*args, **kwargs) result = f(*args, **kwargs)
queue_out.put({'task': task, 'result': result}) done_queue.put((task, result))
print(name, 'Finished JTask {}'.format(func)) except TypeError as e:
except TypeError: fail_queue.put((task, e))
print('job failed')
# TODO log failure to django
print(name, 'Stopped') print(name, 'Stopped')
def work(self):
self.stable_boy()
self.task_queue.put(r.brpop(q_list))
def stable_boy(self): def stable_boy(self):
# Check if all the horses are alive # Check if all the horses are alive
for p in list(self.stable): for p in list(self.stable):
@@ -116,22 +130,23 @@ class Worker(object):
p.terminate() p.terminate()
self.stable.remove(p) self.stable.remove(p)
# Replace it with a fresh one # Replace it with a fresh one
self._spawn_horse() self.reincarnate(p.pid)
def stop(self): def stop(self):
# Send the STOP signal to the stable # Send the STOP signal to the stable
self.running = False self.running = False
print('Stopping queue') print('Stopping')
for i in range(self.stable_size):
self.task_queue.put('STOP')
# Optional: Delete everything in the queue and then add STOP
self.done_queue.put('STOP')
# Wait for all the workers to finish the queue # Wait for all the workers to finish the queue
for p in self.stable: for p in self.stable:
if p.pid == self.failure_monitor_pid:
self.fail_queue.put('STOP')
elif p.pid == self.success_monitor_pid:
self.done_queue.put('STOP')
else:
self.task_queue.put('STOP')
p.join() p.join()
self.mon.join()
print('All horses stopped.')
print('Goodbye. Have a wonderful time.') print('Goodbye. Have a wonderful time.')
def sighandler(self, signum, frame): def sig_handler(self, signum, frame):
self.stop() self.stop()
+2
View File
@@ -1,3 +1,5 @@
ujson
simplejson
Django==1.8.2 Django==1.8.2
hiredis==0.2.0 hiredis==0.2.0
pyzmq==14.6.0 pyzmq==14.6.0