class Test(unittest.TestCase):
def setUp(self):
self.dir_ = tempfile.mkdtemp()
self.addr = '127.0.0.1'
self.addrs = [self.addr, ]
def tearDown(self):
try:
self.mq.shutdown()
finally:
shutil.rmtree(self.dir_)
def testMqProxy(self):
self.mq = MessageQueue(self.dir_, None, self.addr, self.addrs,
copies=0, n_priorities=1)
self.proxy = MpMessageQueueClient(self.mq.new_connection('0'))
try:
test_obj = Url('http://qinxuye.me')
self.proxy.put(test_obj, )
self.assertEqual(self.proxy.get(), test_obj)
finally:
self.mq.shutdown()
def testAddNode(self):
data = range(100)
new_port = random.randint(10000, 30000)
new_node = 'localhost:%s' % new_port
new_rpc_server = ColaRPCServer(('localhost', new_port))
thd = threading.Thread(target=new_rpc_server.serve_forever)
thd.setDaemon(True)
thd.start()
new_dir = tempfile.mkdtemp()
ns = list(self.nodes)
ns.append(new_node)
new_mq = MessageQueue(new_dir, new_rpc_server, new_node, ns)
try:
self.mq0.add_node(new_node)
self.mq1.add_node(new_node)
self.mq2.add_node(new_node)
self.mq0.put(data)
self.assertEqual(data, sorted(self.mq0.get(size=100)))
finally:
try:
new_rpc_server.shutdown()
new_mq.shutdown()
finally:
shutil.rmtree(new_dir)
def testAddNode(self):
data = range(100)
new_port = random.randint(10000, 30000)
new_node = 'localhost:%s' % new_port
new_rpc_server = ColaRPCServer(('localhost', new_port))
thd = threading.Thread(target=new_rpc_server.serve_forever)
thd.setDaemon(True)
thd.start()
new_dir = tempfile.mkdtemp()
ns = list(self.nodes)
ns.append(new_node)
new_mq = MessageQueue(new_dir, new_rpc_server, new_node, ns)
try:
self.mq0.add_node(new_node)
self.mq1.add_node(new_node)
self.mq2.add_node(new_node)
self.mq0.put(data)
self.assertEqual(data, sorted(self.mq0.get(size=100)))
finally:
try:
new_rpc_server.shutdown()
new_mq.shutdown()
finally:
shutil.rmtree(new_dir)
class Test(unittest.TestCase):
def setUp(self):
self.dir_ = tempfile.mkdtemp()
self.addr = '127.0.0.1'
self.addrs = [self.addr, ]
def tearDown(self):
try:
self.mq.shutdown()
finally:
shutil.rmtree(self.dir_)
def testMqProxy(self):
self.mq = MessageQueue(self.dir_, None, self.addr, self.addrs,
copies=0, n_priorities=1)
self.proxy = MpMessageQueueClient(self.mq.new_connection('0'))
try:
test_obj = Url(u'http://qinxuye.me/三星')
self.proxy.put(test_obj, )
self.assertEqual(self.proxy.get(), test_obj)
test_obj = u'三星'
self.proxy.put(test_obj, )
self.assertEqual(self.proxy.get(), test_obj)
finally:
self.mq.shutdown()
def testMqProxy(self):
self.mq = MessageQueue(self.dir_, None, self.addr, self.addrs,
copies=0, n_priorities=1)
self.proxy = MpMessageQueueClient(self.mq.new_connection('0'))
try:
test_obj = Url(u'http://qinxuye.me/三星')
self.proxy.put(test_obj, )
self.assertEqual(self.proxy.get(), test_obj)
test_obj = u'三星'
self.proxy.put(test_obj, )
self.assertEqual(self.proxy.get(), test_obj)
finally:
self.mq.shutdown()
def init_mq(self):
mq_store_dir = os.path.join(self.root, 'store')
mq_backup_dir = os.path.join(self.root, 'backup')
if not os.path.exists(mq_store_dir):
os.makedirs(mq_store_dir)
if not os.path.exists(mq_backup_dir):
os.makedirs(mq_backup_dir)
self.mq = MessageQueue(self.nodes,
self.local,
self.rpc_server,
copies=self.copies)
self.mq.init_store(mq_store_dir,
mq_backup_dir,
verify_exists_hook=self._init_bloom_filter())
def init_mq(self):
mq_dir = os.path.join(self.working_dir, 'mq')
copies = self.job_desc.settings.job.copies
n_priorities = self.job_desc.settings.job.priorities
kw = {
'app_name': self.job_name,
'copies': copies,
'n_priorities': n_priorities,
'deduper': self.deduper
}
self.mq = MessageQueue(mq_dir, self.rpc_server, self.ctx.worker_addr,
self.ctx.addrs[:], **kw)
# register shutdown callback
self.shutdown_callbacks.append(self.mq.shutdown)
class MessageQueueClient(object):
def __init__(self, nodes, copies=1):
self.nodes = nodes
self.hash_ring = HashRing(self.nodes)
self.copies = max(min(len(self.nodes) - 1, copies), 0)
self.mq = MessageQueue(nodes, copies=copies)
def put(self, objs):
self.mq.put(objs)
def get(self):
for n in self.nodes:
obj = self.mq._get(n)
if obj is not None:
return obj
class MessageQueueClient(object):
def __init__(self, nodes, copies=1):
self.nodes = nodes
self.hash_ring = HashRing(self.nodes)
self.copies = max(min(len(self.nodes)-1, copies), 0)
self.mq = MessageQueue(nodes, copies=copies)
def put(self, objs):
self.mq.put(objs)
def get(self):
for n in self.nodes:
obj = self.mq._get(n)
if obj is not None:
return obj
def init_mq(self):
mq_store_dir = os.path.join(self.root, "store")
mq_backup_dir = os.path.join(self.root, "backup")
if not os.path.exists(mq_store_dir):
os.makedirs(mq_store_dir)
if not os.path.exists(mq_backup_dir):
os.makedirs(mq_backup_dir)
self.mq = MessageQueue(self.nodes, self.local, self.rpc_server, copies=self.copies)
self.mq.init_store(mq_store_dir, mq_backup_dir, verify_exists_hook=self._init_bloom_filter())
def init_mq(self):
mq_dir = os.path.join(self.working_dir, 'mq')
copies = self.job_desc.settings.job.copies
n_priorities = self.job_desc.settings.job.priorities
kw = {'app_name': self.job_name, 'copies': copies,
'n_priorities': n_priorities, 'deduper': self.deduper}
self.mq = MessageQueue(mq_dir, self.rpc_server, self.ctx.worker_addr,
self.ctx.addrs[:], **kw)
# register shutdown callback
self.shutdown_callbacks.append(self.mq.shutdown)
def testMqProxy(self):
self.mq = MessageQueue(self.dir_, None, self.addr, self.addrs,
copies=0, n_priorities=1)
self.proxy = MpMessageQueueClient(self.mq.new_connection('0'))
try:
test_obj = Url('http://qinxuye.me')
self.proxy.put(test_obj, )
self.assertEqual(self.proxy.get(), test_obj)
finally:
self.mq.shutdown()
def init_mq(self,
nodes,
local_node,
loc,
verify_exists_hook=None,
copies=1):
mq_store_dir = os.path.join(loc, 'store')
mq_backup_dir = os.path.join(loc, 'backup')
if not os.path.exists(mq_store_dir):
os.mkdir(mq_store_dir)
if not os.path.exists(mq_backup_dir):
os.mkdir(mq_backup_dir)
# MQ relative
self.mq = MessageQueue(nodes,
local_node,
self.rpc_server,
copies=copies)
self.mq.init_store(mq_store_dir,
mq_backup_dir,
verify_exists_hook=verify_exists_hook)
def init_mq(self):
mq_store_dir = os.path.join(self.root, 'store')
mq_backup_dir = os.path.join(self.root, 'backup')
if not os.path.exists(mq_store_dir):
os.makedirs(mq_store_dir)
if not os.path.exists(mq_backup_dir):
os.makedirs(mq_backup_dir)
self.mq = MessageQueue(self.nodes, self.local, self.rpc_server,
copies=self.copies)
self.mq.init_store(mq_store_dir, mq_backup_dir,
verify_exists_hook=self._init_bloom_filter())
self.redismq = redis.Redis(host=REDIS_HOST, port=REDIS_PORT)
def setUp(self):
ports = tuple([random.randint(10000, 30000) for _ in range(3)])
self.nodes = ['localhost:%s'%port for port in ports]
self.dirs = [tempfile.mkdtemp() for _ in range(len(ports))]
self.size = len(ports)
for i in range(self.size):
setattr(self, 'rpc_server%s'%i, ColaRPCServer(('localhost', ports[i])))
setattr(self, 'mq%s'%i,
MessageQueue(self.dirs[i], getattr(self, 'rpc_server%s'%i),
self.nodes[i], self.nodes[:])
)
thd = threading.Thread(target=getattr(self, 'rpc_server%s'%i).serve_forever)
thd.setDaemon(True)
thd.start()
self.client = MessageQueueClient(self.nodes)
def init_mq(self, nodes, local_node, loc,
verify_exists_hook=None, copies=1):
mq_store_dir = os.path.join(loc, 'store')
mq_backup_dir = os.path.join(loc, 'backup')
if not os.path.exists(mq_store_dir):
os.mkdir(mq_store_dir)
if not os.path.exists(mq_backup_dir):
os.mkdir(mq_backup_dir)
# MQ relative
self.mq = MessageQueue(
nodes,
local_node,
self.rpc_server,
copies=copies
)
self.mq.init_store(mq_store_dir, mq_backup_dir,
verify_exists_hook=verify_exists_hook)
def setUp(self):
ports = (11111, 11211, 11311)
self.nodes = ['localhost:%s' % port for port in ports]
self.dirs = [tempfile.mkdtemp() for _ in range(2 * len(ports))]
self.size = len(ports)
for i in range(self.size):
setattr(self, 'rpc_server%s' % i,
ColaRPCServer(('localhost', ports[i])))
setattr(
self, 'mq%s' % i,
MessageQueue(self.nodes[:], self.nodes[i],
getattr(self, 'rpc_server%s' % i)))
getattr(self, 'mq%s' % i).init_store(self.dirs[2 * i],
self.dirs[2 * i + 1])
thd = threading.Thread(target=getattr(self, 'rpc_server%s' %
i).serve_forever)
thd.setDaemon(True)
thd.start()
self.client = MessageQueueClient(self.nodes)
class BasicWorkerJobLoader(JobLoader):
def __init__(self, job, data_dir, context=None, logger=None,
local=None, nodes=None, copies=1, force=False):
self.job = job
ctx = context or self.job.context
self.local = local
if self.local is None:
host, port = get_ip(), ctx.job.port
self.local = '%s:%s' % (host, port)
else:
host, port = tuple(self.local.split(':', 1))
self.nodes = nodes
if self.nodes is None:
self.nodes = [self.local]
self.logger = logger
self.info_logger = get_logger(
name='cola_worker_info_%s'%self.job.real_name)
super(BasicWorkerJobLoader, self).__init__(
self.job, data_dir, self.local,
context=ctx, copies=copies, force=force)
# instances count that run at the same time
self.instances = max(min(self.ctx.job.instances, MAX_THREADS_SIZE), 1)
# excecutings
self.executings = []
# exception times that continously throw
self.error_times = 0
# budget
self.budget = 0
# counter
self.pages_size = 0
# lock when not stopped
self.stop_lock = threading.Lock()
self.stop_lock.acquire()
self.check()
# init rpc server
self.init_rpc_server()
# init message queue
self.init_mq()
# register signal
signal.signal(signal.SIGINT, self.signal_handler)
signal.signal(signal.SIGTERM, self.signal_handler)
self.rpc_server.register_function(self.stop, name='stop')
self.rpc_server.register_function(self.add_node, name='add_node')
self.rpc_server.register_function(self.remove_node, name='remove_node')
self.rpc_server.register_function(self.run, name='run')
self.rpc_server.register_function(self.pages, name='pages')
def _init_bloom_filter(self):
size = self.job.context.job.size
base = 1 if not self.job.is_bundle else 1000
bloom_filter_file = os.path.join(self.root, 'bloomfilter')
if not os.path.exists(bloom_filter_file):
if size > 0:
bloom_filter_size = size*10*base
else:
bloom_filter_size = UNLIMIT_BLOOM_FILTER_CAPACITY
else:
if size > 0:
bloom_filter_size = size*2*base
else:
bloom_filter_size = UNLIMIT_BLOOM_FILTER_CAPACITY
return FileBloomFilter(bloom_filter_file, bloom_filter_size)
def init_mq(self):
mq_store_dir = os.path.join(self.root, 'store')
mq_backup_dir = os.path.join(self.root, 'backup')
if not os.path.exists(mq_store_dir):
os.makedirs(mq_store_dir)
if not os.path.exists(mq_backup_dir):
os.makedirs(mq_backup_dir)
self.mq = MessageQueue(self.nodes, self.local, self.rpc_server,
copies=self.copies)
self.mq.init_store(mq_store_dir, mq_backup_dir,
verify_exists_hook=self._init_bloom_filter())
self.redismq = redis.Redis(host=REDIS_HOST, port=REDIS_PORT)
def _release_stop_lock(self):
try:
self.stop_lock.release()
except:
pass
def check(self):
env_legal = self.check_env(force=self.force)
if not env_legal:
raise JobWorkerRunning('There has been a running job worker.')
def finish(self):
if self.logger is not None:
self.logger.info('Finish visiting pages count: %s' % self.pages_size)
self.stopped = True
self.mq.shutdown()
try:
for handler in self.logger.handlers:
handler.close()
finally:
super(BasicWorkerJobLoader, self).finish()
def complete(self, obj):
if self.logger is not None:
self.logger.info('Finish %s' % obj)
if obj in self.executings:
self.executings.remove(obj)
if self.ctx.job.size <= 0:
return True
return False
def error(self, obj):
if obj in self.executings:
self.executings.remove(obj)
def stop(self):
try:
# self.mq.put(self.executings, force=True)
self.redismq.rpush(REDIS_UID, *self.executings)
super(BasicWorkerJobLoader, self).stop()
finally:
self._release_stop_lock()
def signal_handler(self, signum, frame):
self.stop()
def _login(self, opener):
if self.job.login_hook is not None:
if 'login' not in self.ctx.job or \
not isinstance(self.ctx.job.login, list):
# raise ConfigurationError('If login_hook set, config files must contains `login`')
setattr(self.ctx.job, 'login', [])
#get a new account from redis
account = json.loads(self.redismq.blpop(REDIS_WEIBO_ACCOUNT)[1])
self.ctx.job.login.append(account)
kw = random.choice(self.ctx.job.login)
login_result = self.job.login_hook(opener, **kw)
if isinstance(login_result, tuple) and len(login_result) == 2:
self.logger.error('login fail, reason: %s' % login_result[1])
return login_result[0]
elif not login_result:
self.logger.error('login fail')
return login_result
return True
def _log_error(self, obj, err):
if self.logger is not None:
self.logger.error('Error when get bundle: %s' % obj)
self.logger.exception(err)
if self.job.debug:
raise err
def _require_budget(self, count):
raise NotImplementedError
def pages(self):
return self.pages_size
def apply(self):
raise NotImplementedError
def _execute_bundle(self, obj, opener=None):
bundle = self.job.unit_cls(obj)
urls = bundle.urls()
url = None
try:
while len(urls) > 0 and not self.stopped:
url = urls.pop(0)
self.info_logger.info('get %s url: %s' % (bundle.label, url))
try:
parser_cls, options = self.job.url_patterns.get_parser(url, options=True)
except TypeError:
continue
if parser_cls is not None:
self._require_budget()
self.pages_size += 1
next_urls, bundles = parser_cls(opener, url, bundle=bundle, logger=self.logger,
**options).parse()
next_urls = list(self.job.url_patterns.matches(next_urls))
next_urls.extend(urls)
urls = next_urls
if bundles:
# self.mq.put([str(b) for b in bundles if b.force is False])
# self.mq.put([str(b) for b in bundles if b.force is True], force=True)
self.redismq.rpush(REDIS_UID, *[str(b) for b in bundles if b.force is False])
# self.redismq.rpush(REDIS_KEY, [str(b) for b in bundles if b.force is True])
if hasattr(opener, 'close'):
opener.close()
self.error_times = 0
except LoginFailure, e:
if not self._login(opener):
self.error_times += 1
self._log_error(obj, e)
self.error(obj)
except Exception, e:
self.error_times += 1
if self.logger is not None and url is not None:
self.logger.error('Error when fetch url: %s' % url)
self._log_error(obj, e)
self.error(obj)
def __init__(self, nodes, copies=1):
self.nodes = nodes
self.hash_ring = HashRing(self.nodes)
self.copies = max(min(len(self.nodes) - 1, copies), 0)
self.mq = MessageQueue(nodes, copies=copies)
class Job(object):
def __init__(self,
ctx,
job_def_path,
job_name,
job_desc=None,
working_dir=None,
rpc_server=None,
manager=None,
job_offset=0):
self.status = NOTSTARTED
self.ctx = ctx
self.shutdown_callbacks = []
self.stopped = multiprocessing.Event()
self.nonsuspend = multiprocessing.Event()
self.nonsuspend.set()
self.job_def_path = job_def_path
self.job_name = job_name
self.working_dir = working_dir or os.path.join(self.ctx.working_dir,
self.job_name)
self.logger = get_logger(name='cola_job' + str(time.time()))
self.job_desc = job_desc or import_job_desc(job_def_path)
self.settings = self.job_desc.settings
self.is_bundle = self.settings.job.mode == 'bundle'
self.rpc_server = rpc_server
self.n_instances = self.job_desc.settings.job.instances
self.n_containers = min(get_cpu_count(), max(self.n_instances, 1))
self.job_offset = job_offset
self.is_multi_process = self.n_containers > 1
self.processes = []
self.idle_statuses = manager.list([False] * self.n_containers)
self.manager = manager
if not os.path.exists(self.working_dir):
os.makedirs(self.working_dir)
self.inited = False
self._register_rpc()
def _register_rpc(self):
if self.rpc_server:
self.prefix = get_rpc_prefix(app_name=self.job_name, prefix='job')
self.rpc_server.register_function(self.shutdown,
name='shutdown',
prefix=self.prefix)
if self.ctx.is_local_mode:
self.rpc_server.register_function(lambda: [
self.job_name,
],
name='get_jobs')
def init_deduper(self):
deduper_cls = import_module(self.settings.job.components.deduper.cls)
base = 1 if not self.is_bundle else 1000
size = self.job_desc.settings.job.size
capacity = UNLIMIT_BLOOM_FILTER_CAPACITY
if size > 0:
capacity = max(base * size * 10, capacity)
params = dict(self.settings.job.components.deduper)
del params['cls']
deduper_cls = deduper_cls if not self.is_multi_process \
else getattr(self.manager, deduper_cls.__name__)
self.deduper = deduper_cls(self.working_dir, capacity, **params)
# register shutdown callback
self.shutdown_callbacks.append(self.deduper.shutdown)
def init_mq(self):
mq_dir = os.path.join(self.working_dir, 'mq')
copies = self.job_desc.settings.job.copies
n_priorities = self.job_desc.settings.job.priorities
kw = {
'app_name': self.job_name,
'copies': copies,
'n_priorities': n_priorities,
'deduper': self.deduper
}
self.mq = MessageQueue(mq_dir, self.rpc_server, self.ctx.worker_addr,
self.ctx.addrs[:], **kw)
# register shutdown callback
self.shutdown_callbacks.append(self.mq.shutdown)
def _init_function_servers(self):
budget_dir = os.path.join(self.working_dir, 'budget')
budget_cls = BudgetApplyServer if not self.is_multi_process \
else self.manager.budget_server
self.budget_server = budget_cls(budget_dir, self.settings, None,
self.job_name)
if self.rpc_server:
BudgetApplyServer.register_rpc(self.budget_server,
self.rpc_server,
app_name=self.job_name)
self.shutdown_callbacks.append(self.budget_server.shutdown)
counter_dir = os.path.join(self.working_dir, 'counter')
counter_cls = CounterServer if not self.is_multi_process \
else self.manager.counter_server
self.counter_server = counter_cls(counter_dir, self.settings, None,
self.job_name)
if self.rpc_server:
CounterServer.register_rpc(self.counter_server,
self.rpc_server,
app_name=self.job_name)
self.shutdown_callbacks.append(self.counter_server.shutdown)
speed_dir = os.path.join(self.working_dir, 'speed')
speed_cls = SpeedControlServer if not self.is_multi_process \
else self.manager.speed_server
self.speed_server = speed_cls(speed_dir, self.settings, None,
self.job_name, self.counter_server,
self.ctx.ips)
if self.rpc_server:
SpeedControlServer.register_rpc(self.speed_server,
self.rpc_server,
app_name=self.job_name)
self.shutdown_callbacks.append(self.speed_server.shutdown)
def init_functions(self):
if self.ctx.is_local_mode:
self._init_function_servers()
self.counter_arg = self.counter_server
self.budget_arg = self.budget_server
self.speed_arg = self.speed_server
else:
self.counter_arg, self.budget_arg, self.speed_arg = \
tuple([self.ctx.master_addr for _ in range(3)])
def init(self):
if self.inited:
return
self.lock_file = os.path.join(self.working_dir, 'lock')
if os.path.exists(self.lock_file):
raise JobRunning('The job has already started')
open(self.lock_file, 'w').close()
self.init_deduper()
self.init_mq()
self.init_functions()
self.inited = True
self.status = RUNNING
def run(self, block=False):
self.init()
try:
self.processes = run_containers(
self.n_containers,
self.n_instances,
self.working_dir,
self.job_def_path,
self.job_name,
self.ctx.env,
self.mq,
self.counter_arg,
self.budget_arg,
self.speed_arg,
self.stopped,
self.nonsuspend,
self.idle_statuses,
is_multi_process=self.is_multi_process,
is_local=self.ctx.is_local_mode,
master_ip=self.ctx.master_ip,
offset=self.job_offset)
if block:
self.wait_for_stop()
finally:
if os.path.exists(self.lock_file):
os.remove(self.lock_file)
def wait_for_stop(self):
[process.join() for process in self.processes]
def stop_running(self):
if 'main' not in multiprocessing.current_process().name.lower():
return
self.stopped.set()
self.wait_for_stop()
def clear_running(self):
if 'main' not in multiprocessing.current_process().name.lower():
return
try:
# output counters
if self.ctx.is_local_mode:
self.logger.debug('Counters during running:')
self.logger.debug(
pprint.pformat(self.counter_server.output(), width=1))
self.logger.debug('Processing shutting down')
for cb in self.shutdown_callbacks:
cb()
if self.ctx.is_local_mode is True and hasattr(self, 'manager'):
try:
self.manager.shutdown()
except socket.error:
pass
self.status = FINISHED
self.logger.debug('Shutdown finished')
finally:
if os.path.exists(self.lock_file):
os.remove(self.lock_file)
def shutdown(self):
if 'main' not in multiprocessing.current_process().name.lower():
return
try:
self.stop_running()
finally:
self.clear_running()
def get_status(self):
if self.ctx.is_local_mode and self.status == RUNNING:
if self.budget_server.get_status() == ALLFINISHED and \
self.settings.job.inc is False:
return FINISHED
if all(list(self.idle_statuses)):
return IDLE
return self.status
def suspend(self):
self.nonsuspend.clear()
def resume(self):
self.nonsuspend.set()
def add_node(self, node):
if hasattr(self, 'mq'):
self.mq.add_node(node)
def remove_node(self, node):
if hasattr(self, 'mq'):
self.mq.remove_node(node)
class JobLoader(object):
def __init__(self,
job,
rpc_server,
mq=None,
logger=None,
master=None,
context=None):
self.job = job
self.rpc_server = rpc_server
self.mq = mq
self.master = master
self.logger = logger
# If stop
self.stopped = False
self.ctx = context or self.job.context
self.instances = max(min(self.ctx.job.instances, MAX_THREADS_SIZE), 1)
self.size = self.ctx.job.size
self.budget = 0
# The execute unit
self.executing = None
# register signal
signal.signal(signal.SIGINT, self.signal_handler)
signal.signal(signal.SIGTERM, self.signal_handler)
rpc_server.register_function(self.stop, name='stop')
rpc_server.register_function(self.add_node, name='add_node')
rpc_server.register_function(self.remove_node, name='remove_node')
rpc_server.register_function(self.run, name='run')
def init_mq(self,
nodes,
local_node,
loc,
verify_exists_hook=None,
copies=1):
mq_store_dir = os.path.join(loc, 'store')
mq_backup_dir = os.path.join(loc, 'backup')
if not os.path.exists(mq_store_dir):
os.mkdir(mq_store_dir)
if not os.path.exists(mq_backup_dir):
os.mkdir(mq_backup_dir)
# MQ relative
self.mq = MessageQueue(nodes,
local_node,
self.rpc_server,
copies=copies)
self.mq.init_store(mq_store_dir,
mq_backup_dir,
verify_exists_hook=verify_exists_hook)
def stop(self):
self.stopped = True
if self.executing is not None:
self.mq.put(self.executing)
self.finish()
def signal_handler(self, signum, frame):
self.stop()
def complete(self, obj):
if self.logger is not None:
self.logger.info('Finish %s' % obj)
if self.ctx.job.size <= 0:
return False
self.executing = None
if self.master is not None:
return client_call(self.master, 'complete', obj)
else:
self.size -= 1
# sth to log
if self.size <= 0:
self.stopped = True
return self.stopped
def finish(self):
self.mq.shutdown()
self.stopped = True
def _require_budget(self):
if self.master is None or self.ctx.job.limits == 0:
return
if self.budget > 0:
self.budget -= 1
return
while self.budget == 0 and not self.stopped:
self.budget = client_call(self.master, 'require', BUDGET_REQUIRE)
def _log(self, obj, err):
if self.logger is not None:
self.logger.info('Error when get bundle: %s' % obj)
self.logger.exception(err)
if self.job.debug:
raise err
def _login(self, opener):
if self.job.login_hook is not None:
if 'login' not in self.ctx.job or \
not isinstance(self.ctx.job.login, list):
raise ConfigurationError(
'If login_hook set, config files must contains `login`')
kw = random.choice(self.ctx.job.login)
login_success = self.job.login_hook(opener, **kw)
if not login_success:
self.logger.info('login fail')
return login_success
def _execute(self, obj, opener=None):
if opener is None:
opener = self.job.opener_cls()
if self.job.is_bundle:
bundle = self.job.unit_cls(obj)
urls = bundle.urls()
try:
while len(urls) > 0 and not self.stopped:
url = urls.pop(0)
parser_cls = self.job.url_patterns.get_parser(url)
if parser_cls is not None:
self._require_budget()
next_urls, bundles = parser_cls(opener,
url,
bundle=bundle).parse()
next_urls = list(
self.job.url_patterns.matches(next_urls))
next_urls.extend(urls)
urls = next_urls
if bundles:
self.mq.put([str(bundle) for bundle in bundles])
except LoginFailure:
if not self._login(opener):
return
except Exception, e:
self._log(obj, e)
else:
def __init__(self, nodes, copies=1):
self.nodes = nodes
self.hash_ring = HashRing(self.nodes)
self.copies = max(min(len(self.nodes)-1, copies), 0)
self.mq = MessageQueue(nodes, copies=copies)
class JobLoader(object):
def __init__(self, job, rpc_server,
mq=None, logger=None, master=None, context=None):
self.job = job
self.rpc_server = rpc_server
self.mq = mq
self.master = master
self.logger = logger
# If stop
self.stopped = False
self.ctx = context or self.job.context
self.instances = max(min(self.ctx.job.instances, MAX_THREADS_SIZE), 1)
self.size =self.ctx.job.size
self.budget = 0
# The execute unit
self.executing = None
# register signal
signal.signal(signal.SIGINT, self.signal_handler)
signal.signal(signal.SIGTERM, self.signal_handler)
rpc_server.register_function(self.stop, name='stop')
rpc_server.register_function(self.add_node, name='add_node')
rpc_server.register_function(self.remove_node, name='remove_node')
rpc_server.register_function(self.run, name='run')
def init_mq(self, nodes, local_node, loc,
verify_exists_hook=None, copies=1):
mq_store_dir = os.path.join(loc, 'store')
mq_backup_dir = os.path.join(loc, 'backup')
if not os.path.exists(mq_store_dir):
os.mkdir(mq_store_dir)
if not os.path.exists(mq_backup_dir):
os.mkdir(mq_backup_dir)
# MQ relative
self.mq = MessageQueue(
nodes,
local_node,
self.rpc_server,
copies=copies
)
self.mq.init_store(mq_store_dir, mq_backup_dir,
verify_exists_hook=verify_exists_hook)
def stop(self):
self.stopped = True
if self.executing is not None:
self.mq.put(self.executing)
self.finish()
def signal_handler(self, signum, frame):
self.stop()
def complete(self, obj):
if self.logger is not None:
self.logger.info('Finish %s' % obj)
if self.ctx.job.size <= 0:
return False
self.executing = None
if self.master is not None:
return client_call(self.master, 'complete', obj)
else:
self.size -= 1
# sth to log
if self.size <= 0:
self.stopped = True
return self.stopped
def finish(self):
self.mq.shutdown()
self.stopped = True
def _require_budget(self):
if self.master is None or self.ctx.job.limits == 0:
return
if self.budget > 0:
self.budget -= 1
return
while self.budget == 0 and not self.stopped:
self.budget = client_call(self.master, 'require', BUDGET_REQUIRE)
def _log(self, obj, err):
if self.logger is not None:
self.logger.error('Error when get bundle: %s' % obj)
self.logger.exception(err)
if self.job.debug:
raise err
def _login(self, opener):
if self.job.login_hook is not None:
if 'login' not in self.ctx.job or \
not isinstance(self.ctx.job.login, list):
raise ConfigurationError('If login_hook set, config files must contains `login`')
kw = random.choice(self.ctx.job.login)
login_success = self.job.login_hook(opener, **kw)
if not login_success:
self.logger.info('login fail')
return login_success
def _execute(self, obj, opener=None):
if opener is None:
opener = self.job.opener_cls()
if self.job.is_bundle:
bundle = self.job.unit_cls(obj)
urls = bundle.urls()
try:
while len(urls) > 0 and not self.stopped:
url = urls.pop(0)
self.logger.info('get %s url: %s' % (bundle.label, url))
parser_cls = self.job.url_patterns.get_parser(url)
if parser_cls is not None:
self._require_budget()
next_urls, bundles = parser_cls(opener, url, bundle=bundle).parse()
next_urls = list(self.job.url_patterns.matches(next_urls))
next_urls.extend(urls)
urls = next_urls
if bundles:
self.mq.put([str(b) for b in bundles])
except LoginFailure:
if not self._login(opener):
return
except Exception, e:
self._log(obj, e)
else:
class BasicWorkerJobLoader(JobLoader):
def __init__(self,
job,
data_dir,
context=None,
logger=None,
local=None,
nodes=None,
copies=1,
force=False):
self.job = job
ctx = context or self.job.context
self.local = local
if self.local is None:
host, port = get_ip(), ctx.job.port
self.local = '%s:%s' % (host, port)
else:
host, port = tuple(self.local.split(':', 1))
self.nodes = nodes
if self.nodes is None:
self.nodes = [self.local]
self.logger = logger
self.info_logger = get_logger(name='cola_worker_info_%s' %
self.job.real_name)
super(BasicWorkerJobLoader, self).__init__(self.job,
data_dir,
self.local,
context=ctx,
copies=copies,
force=force)
# instances count that run at the same time
self.instances = max(min(self.ctx.job.instances, MAX_THREADS_SIZE), 1)
# excecutings
self.executings = []
# exception times that continously throw
self.error_times = 0
# budget
self.budget = 0
self.check()
# init rpc server
self.init_rpc_server()
# init message queue
self.init_mq()
# register signal
signal.signal(signal.SIGINT, self.signal_handler)
signal.signal(signal.SIGTERM, self.signal_handler)
self.rpc_server.register_function(self.stop, name='stop')
self.rpc_server.register_function(self.add_node, name='add_node')
self.rpc_server.register_function(self.remove_node, name='remove_node')
self.rpc_server.register_function(self.run, name='run')
def _init_bloom_filter(self):
size = self.job.context.job.size
base = 1 if not self.job.is_bundle else 1000
bloom_filter_file = os.path.join(self.root, 'bloomfilter')
if not os.path.exists(bloom_filter_file):
if size > 0:
bloom_filter_size = size * 10 * base
else:
bloom_filter_size = UNLIMIT_BLOOM_FILTER_CAPACITY
else:
if size > 0:
bloom_filter_size = size * 2 * base
else:
bloom_filter_size = UNLIMIT_BLOOM_FILTER_CAPACITY
return FileBloomFilter(bloom_filter_file, bloom_filter_size)
def init_mq(self):
mq_store_dir = os.path.join(self.root, 'store')
mq_backup_dir = os.path.join(self.root, 'backup')
if not os.path.exists(mq_store_dir):
os.makedirs(mq_store_dir)
if not os.path.exists(mq_backup_dir):
os.makedirs(mq_backup_dir)
self.mq = MessageQueue(self.nodes,
self.local,
self.rpc_server,
copies=self.copies)
self.mq.init_store(mq_store_dir,
mq_backup_dir,
verify_exists_hook=self._init_bloom_filter())
def check(self):
env_legal = self.check_env(force=self.force)
if not env_legal:
raise JobWorkerRunning('There has been a running job worker.')
def finish(self):
self.stopped = True
self.mq.shutdown()
try:
for handler in self.logger.handlers:
handler.close()
finally:
super(BasicWorkerJobLoader, self).finish()
def complete(self, obj):
if self.logger is not None:
self.logger.info('Finish %s' % obj)
if obj in self.executings:
self.executings.remove(obj)
if self.ctx.job.size <= 0:
return True
return False
def error(self, obj):
if obj in self.executings:
self.executings.remove(obj)
def stop(self):
self.mq.put(self.executings, force=True)
super(BasicWorkerJobLoader, self).stop()
def signal_handler(self, signum, frame):
self.stop()
def _login(self, opener):
if self.job.login_hook is not None:
if 'login' not in self.ctx.job or \
not isinstance(self.ctx.job.login, list):
raise ConfigurationError(
'If login_hook set, config files must contains `login`')
kw = random.choice(self.ctx.job.login)
login_result = self.job.login_hook(opener, **kw)
if isinstance(login_result, tuple) and len(login_result) == 2:
self.logger.error('login fail, reason: %s' % login_result[1])
return login_result[0]
elif not login_result:
self.logger.error('login fail')
return login_result
return True
def _log_error(self, obj, err):
if self.logger is not None:
self.logger.error('Error when get bundle: %s' % obj)
self.logger.exception(err)
if self.job.debug:
raise err
def _require_budget(self, count):
raise NotImplementedError
def apply(self):
raise NotImplementedError
def _execute_bundle(self, obj, opener=None):
bundle = self.job.unit_cls(obj)
urls = bundle.urls()
url = None
try:
while len(urls) > 0 and not self.stopped:
url = urls.pop(0)
self.info_logger.info('get %s url: %s' % (bundle.label, url))
parser_cls, options = self.job.url_patterns.get_parser(
url, options=True)
if parser_cls is not None:
self._require_budget()
next_urls, bundles = parser_cls(opener,
url,
bundle=bundle,
logger=self.logger,
**options).parse()
next_urls = list(self.job.url_patterns.matches(next_urls))
next_urls.extend(urls)
urls = next_urls
if bundles:
self.mq.put(
[str(b) for b in bundles if b.force is False])
self.mq.put(
[str(b) for b in bundles if b.force is True],
force=True)
if hasattr(opener, 'close'):
opener.close()
self.error_times = 0
except LoginFailure, e:
if not self._login(opener):
self.error_times += 1
self._log_error(obj, e)
self.error(obj)
except Exception, e:
self.error_times += 1
if self.logger is not None and url is not None:
self.logger.error('Error when fetch url: %s' % url)
self._log_error(obj, e)
self.error(obj)
class Job(object):
def __init__(self, ctx, job_def_path, job_name,
job_desc=None, working_dir=None, rpc_server=None,
manager=None, job_offset=0):
self.status = NOTSTARTED
self.ctx = ctx
self.shutdown_callbacks = []
self.stopped = multiprocessing.Event()
self.nonsuspend = multiprocessing.Event()
self.nonsuspend.set()
self.job_def_path = job_def_path
self.job_name = job_name
self.working_dir = working_dir or os.path.join(self.ctx.working_dir,
self.job_name)
self.logger = get_logger(name='cola_job'+str(time.time()))
self.job_desc = job_desc or import_job_desc(job_def_path)
self.settings = self.job_desc.settings
self.is_bundle = self.settings.job.mode == 'bundle'
self.rpc_server = rpc_server
self.n_instances = self.job_desc.settings.job.instances
self.n_containers = min(get_cpu_count(), max(self.n_instances, 1))
self.job_offset = job_offset
self.is_multi_process = self.n_containers > 1
self.processes = []
self.idle_statuses = manager.list([False] * self.n_containers)
self.manager = manager
if not os.path.exists(self.working_dir):
os.makedirs(self.working_dir)
self.inited = False
self._register_rpc()
def _register_rpc(self):
if self.rpc_server:
self.prefix = get_rpc_prefix(app_name=self.job_name,
prefix='job')
self.rpc_server.register_function(self.shutdown, name='shutdown',
prefix=self.prefix)
if self.ctx.is_local_mode:
self.rpc_server.register_function(lambda: [self.job_name, ],
name='get_jobs')
def init_deduper(self):
deduper_cls = import_module(self.settings.job.components.deduper.cls)
base = 1 if not self.is_bundle else 1000
size = self.job_desc.settings.job.size
capacity = UNLIMIT_BLOOM_FILTER_CAPACITY
if size > 0:
capacity = max(base * size * 10, capacity)
params = dict(self.settings.job.components.deduper)
del params['cls']
deduper_cls = deduper_cls if not self.is_multi_process \
else getattr(self.manager, deduper_cls.__name__)
self.deduper = deduper_cls(self.working_dir, capacity, **params)
# register shutdown callback
self.shutdown_callbacks.append(self.deduper.shutdown)
def init_mq(self):
mq_dir = os.path.join(self.working_dir, 'mq')
copies = self.job_desc.settings.job.copies
n_priorities = self.job_desc.settings.job.priorities
kw = {'app_name': self.job_name, 'copies': copies,
'n_priorities': n_priorities, 'deduper': self.deduper}
self.mq = MessageQueue(mq_dir, self.rpc_server, self.ctx.worker_addr,
self.ctx.addrs[:], **kw)
# register shutdown callback
self.shutdown_callbacks.append(self.mq.shutdown)
def _init_function_servers(self):
budget_dir = os.path.join(self.working_dir, 'budget')
budget_cls = BudgetApplyServer if not self.is_multi_process \
else self.manager.budget_server
self.budget_server = budget_cls(budget_dir, self.settings,
None, self.job_name)
if self.rpc_server:
BudgetApplyServer.register_rpc(self.budget_server, self.rpc_server,
app_name=self.job_name)
self.shutdown_callbacks.append(self.budget_server.shutdown)
counter_dir = os.path.join(self.working_dir, 'counter')
counter_cls = CounterServer if not self.is_multi_process \
else self.manager.counter_server
self.counter_server = counter_cls(counter_dir, self.settings,
None, self.job_name)
if self.rpc_server:
CounterServer.register_rpc(self.counter_server, self.rpc_server,
app_name=self.job_name)
self.shutdown_callbacks.append(self.counter_server.shutdown)
speed_dir = os.path.join(self.working_dir, 'speed')
speed_cls = SpeedControlServer if not self.is_multi_process \
else self.manager.speed_server
self.speed_server = speed_cls(speed_dir, self.settings,
None, self.job_name,
self.counter_server, self.ctx.ips)
if self.rpc_server:
SpeedControlServer.register_rpc(self.speed_server, self.rpc_server,
app_name=self.job_name)
self.shutdown_callbacks.append(self.speed_server.shutdown)
def init_functions(self):
if self.ctx.is_local_mode:
self._init_function_servers()
self.counter_arg = self.counter_server
self.budget_arg = self.budget_server
self.speed_arg = self.speed_server
else:
self.counter_arg, self.budget_arg, self.speed_arg = \
tuple([self.ctx.master_addr for _ in range(3)])
def init(self):
if self.inited:
return
self.lock_file = os.path.join(self.working_dir, 'lock')
if os.path.exists(self.lock_file):
raise JobRunning('The job has already started')
open(self.lock_file, 'w').close()
self.init_deduper()
self.init_mq()
self.init_functions()
self.inited = True
self.status = RUNNING
def run(self, block=False):
self.init()
try:
self.processes = run_containers(
self.n_containers, self.n_instances, self.working_dir,
self.job_def_path, self.job_name, self.ctx.env, self.mq,
self.counter_arg, self.budget_arg, self.speed_arg,
self.stopped, self.nonsuspend, self.idle_statuses,
is_multi_process=self.is_multi_process,
is_local=self.ctx.is_local_mode, master_ip=self.ctx.master_ip,
offset=self.job_offset)
if block:
self.wait_for_stop()
finally:
if os.path.exists(self.lock_file):
os.remove(self.lock_file)
def wait_for_stop(self):
[process.join() for process in self.processes]
def stop_running(self):
if 'main' not in multiprocessing.current_process().name.lower():
return
self.stopped.set()
self.wait_for_stop()
def clear_running(self):
if 'main' not in multiprocessing.current_process().name.lower():
return
try:
# output counters
if self.ctx.is_local_mode:
self.logger.debug('Counters during running:')
self.logger.debug(pprint.pformat(self.counter_server.output(),
width=1))
self.logger.debug('Processing shutting down')
for cb in self.shutdown_callbacks:
cb()
if self.ctx.is_local_mode is True and hasattr(self, 'manager'):
try:
self.manager.shutdown()
except socket.error:
pass
self.status = FINISHED
self.logger.debug('Shutdown finished')
finally:
if os.path.exists(self.lock_file):
os.remove(self.lock_file)
def shutdown(self):
if 'main' not in multiprocessing.current_process().name.lower():
return
try:
self.stop_running()
finally:
self.clear_running()
def get_status(self):
if self.ctx.is_local_mode and self.status == RUNNING:
if self.budget_server.get_status() == ALLFINISHED and \
self.settings.job.inc is False:
return FINISHED
if all(list(self.idle_statuses)):
return IDLE
return self.status
def suspend(self):
self.nonsuspend.clear()
def resume(self):
self.nonsuspend.set()
def add_node(self, node):
if hasattr(self, 'mq'):
self.mq.add_node(node)
def remove_node(self, node):
if hasattr(self, 'mq'):
self.mq.remove_node(node)
class BasicWorkerJobLoader(JobLoader):
def __init__(self, job, data_dir, context=None, logger=None, local=None, nodes=None, copies=1, force=False):
self.job = job
ctx = context or self.job.context
self.local = local
if self.local is None:
host, port = get_ip(), ctx.job.port
self.local = "%s:%s" % (host, port)
else:
host, port = tuple(self.local.split(":", 1))
self.nodes = nodes
if self.nodes is None:
self.nodes = [self.local]
self.logger = logger
self.info_logger = get_logger(name="cola_worker_info_%s" % self.job.real_name)
super(BasicWorkerJobLoader, self).__init__(
self.job, data_dir, self.local, context=ctx, copies=copies, force=force
)
# instances count that run at the same time
self.instances = max(min(self.ctx.job.instances, MAX_THREADS_SIZE), 1)
# excecutings
self.executings = []
# exception times that continously throw
self.error_times = 0
# budget
self.budget = 0
self.check()
# init rpc server
self.init_rpc_server()
# init message queue
self.init_mq()
# register signal
signal.signal(signal.SIGINT, self.signal_handler)
signal.signal(signal.SIGTERM, self.signal_handler)
self.rpc_server.register_function(self.stop, name="stop")
self.rpc_server.register_function(self.add_node, name="add_node")
self.rpc_server.register_function(self.remove_node, name="remove_node")
self.rpc_server.register_function(self.run, name="run")
def _init_bloom_filter(self):
size = self.job.context.job.size
base = 1 if not self.job.is_bundle else 1000
bloom_filter_file = os.path.join(self.root, "bloomfilter")
if not os.path.exists(bloom_filter_file):
if size > 0:
bloom_filter_size = size * 10 * base
else:
bloom_filter_size = UNLIMIT_BLOOM_FILTER_CAPACITY
else:
if size > 0:
bloom_filter_size = size * 2 * base
else:
bloom_filter_size = UNLIMIT_BLOOM_FILTER_CAPACITY
return FileBloomFilter(bloom_filter_file, bloom_filter_size)
def init_mq(self):
mq_store_dir = os.path.join(self.root, "store")
mq_backup_dir = os.path.join(self.root, "backup")
if not os.path.exists(mq_store_dir):
os.makedirs(mq_store_dir)
if not os.path.exists(mq_backup_dir):
os.makedirs(mq_backup_dir)
self.mq = MessageQueue(self.nodes, self.local, self.rpc_server, copies=self.copies)
self.mq.init_store(mq_store_dir, mq_backup_dir, verify_exists_hook=self._init_bloom_filter())
def check(self):
env_legal = self.check_env(force=self.force)
if not env_legal:
raise JobWorkerRunning("There has been a running job worker.")
def finish(self):
self.stopped = True
self.mq.shutdown()
try:
for handler in self.logger.handlers:
handler.close()
finally:
super(BasicWorkerJobLoader, self).finish()
def complete(self, obj):
if self.logger is not None:
self.logger.info("Finish %s" % obj)
if obj in self.executings:
self.executings.remove(obj)
if self.ctx.job.size <= 0:
return True
return False
def error(self, obj):
if obj in self.executings:
self.executings.remove(obj)
def stop(self):
self.mq.put(self.executings, force=True)
super(BasicWorkerJobLoader, self).stop()
def signal_handler(self, signum, frame):
self.stop()
def _login(self, opener):
if self.job.login_hook is not None:
if "login" not in self.ctx.job or not isinstance(self.ctx.job.login, list):
raise ConfigurationError("If login_hook set, config files must contains `login`")
kw = random.choice(self.ctx.job.login)
login_success = self.job.login_hook(opener, **kw)
if not login_success:
self.logger.info("login fail")
return login_success
return True
def _log_error(self, obj, err):
if self.logger is not None:
self.logger.error("Error when get bundle: %s" % obj)
self.logger.exception(err)
if self.job.debug:
raise err
def _require_budget(self, count):
raise NotImplementedError
def apply(self):
raise NotImplementedError
def _execute_bundle(self, obj, opener=None):
bundle = self.job.unit_cls(obj)
urls = bundle.urls()
try:
while len(urls) > 0 and not self.stopped:
url = urls.pop(0)
self.info_logger.info("get %s url: %s" % (bundle.label, url))
parser_cls = self.job.url_patterns.get_parser(url)
if parser_cls is not None:
self._require_budget()
next_urls, bundles = parser_cls(opener, url, bundle=bundle).parse()
next_urls = list(self.job.url_patterns.matches(next_urls))
next_urls.extend(urls)
urls = next_urls
if bundles:
self.mq.put([str(b) for b in bundles])
self.error_times = 0
except LoginFailure, e:
if not self._login(opener):
self.error_times += 1
self._log_error(obj, e)
self.error(obj)
except Exception, e:
self.error_times += 1
self._log_error(obj, e)
self.error(obj)
