def create_consumer(self, topic, proxy, fanout=False):
# Register with matchmaker.
_get_matchmaker().register(topic, CONF.rpc_zmq_host)
# Subscription scenarios
if fanout:
sock_type = zmq.SUB
subscribe = ("", fanout)[type(fanout) == str]
topic = "fanout~" + topic.split(".", 1)[0]
else:
sock_type = zmq.PULL
subscribe = None
topic = ".".join((topic.split(".", 1)[0], CONF.rpc_zmq_host))
if topic in self.topics:
LOG.info(_("Skipping topic registration. Already registered."))
return
# Receive messages from (local) proxy
inaddr = "ipc://%s/zmq_topic_%s" % (CONF.rpc_zmq_ipc_dir, topic)
LOG.debug(_("Consumer is a zmq.%s"), ["PULL", "SUB"][sock_type == zmq.SUB])
self.reactor.register(proxy, inaddr, sock_type, subscribe=subscribe, in_bind=False)
self.topics.append(topic)
def ssh_execute(ssh, cmd, process_input=None, addl_env=None, check_exit_code=True):
LOG.debug(_("Running cmd (SSH): %s"), cmd)
if addl_env:
raise InvalidArgumentError(_("Environment not supported over SSH"))
if process_input:
# This is (probably) fixable if we need it...
raise InvalidArgumentError(_("process_input not supported over SSH"))
stdin_stream, stdout_stream, stderr_stream = ssh.exec_command(cmd)
channel = stdout_stream.channel
# NOTE(justinsb): This seems suspicious...
# ...other SSH clients have buffering issues with this approach
stdout = stdout_stream.read()
stderr = stderr_stream.read()
stdin_stream.close()
exit_status = channel.recv_exit_status()
# exit_status == -1 if no exit code was returned
if exit_status != -1:
LOG.debug(_("Result was %s") % exit_status)
if check_exit_code and exit_status != 0:
raise ProcessExecutionError(exit_code=exit_status, stdout=stdout, stderr=stderr, cmd=cmd)
return (stdout, stderr)
def run_periodic_tasks(self, context, raise_on_error=False):
"""Tasks to be run at a periodic interval."""
idle_for = DEFAULT_INTERVAL
for task_name, task in self._periodic_tasks:
full_task_name = '.'.join([self.__class__.__name__, task_name])
now = timeutils.utcnow()
spacing = self._periodic_spacing[task_name]
last_run = self._periodic_last_run[task_name]
# If a periodic task is _nearly_ due, then we'll run it early
if spacing is not None and last_run is not None:
due = last_run + datetime.timedelta(seconds=spacing)
if not timeutils.is_soon(due, 0.2):
idle_for = min(idle_for, timeutils.delta_seconds(now, due))
continue
if spacing is not None:
idle_for = min(idle_for, spacing)
LOG.debug(_("Running periodic task %(full_task_name)s"),
{"full_task_name": full_task_name})
self._periodic_last_run[task_name] = timeutils.utcnow()
try:
task(self, context)
except Exception as e:
if raise_on_error:
raise
LOG.exception(_("Error during %(full_task_name)s: %(e)s"),
{"full_task_name": full_task_name, "e": e})
time.sleep(0)
return idle_for
def _wait_for_exit_or_signal(self):
status = None
signo = 0
LOG.debug(_('Full set of CONF:'))
CONF.log_opt_values(LOG, std_logging.DEBUG)
try:
super(ServiceLauncher, self).wait()
except SignalExit as exc:
signame = _signo_to_signame(exc.signo)
LOG.info(_('Caught %s, exiting'), signame)
status = exc.code
signo = exc.signo
except SystemExit as exc:
status = exc.code
finally:
self.stop()
if rpc:
try:
rpc.cleanup()
except Exception:
# We're shutting down, so it doesn't matter at this point.
LOG.exception(_('Exception during rpc cleanup.'))
return status, signo
def __call__(self, message_data):
"""Consumer callback to call a method on a proxy object.
Parses the message for validity and fires off a thread to call the
proxy object method.
Message data should be a dictionary with two keys:
method: string representing the method to call
args: dictionary of arg: value
Example: {'method': 'echo', 'args': {'value': 42}}
"""
# It is important to clear the context here, because at this point
# the previous context is stored in local.store.context
if hasattr(local.store, 'context'):
del local.store.context
rpc_common._safe_log(LOG.debug, _('received %s'), message_data)
self.msg_id_cache.check_duplicate_message(message_data)
ctxt = unpack_context(self.conf, message_data)
method = message_data.get('method')
args = message_data.get('args', {})
version = message_data.get('version')
namespace = message_data.get('namespace')
if not method:
LOG.warn(_('no method for message: %s') % message_data)
ctxt.reply(_('No method for message: %s') % message_data,
connection_pool=self.connection_pool)
return
self.pool.spawn_n(self._process_data, ctxt, version, method,
namespace, args)
def publisher(waiter):
LOG.info(_("Creating proxy for topic: %s"), topic)
try:
# The topic is received over the network,
# don't trust this input.
if self.badchars.search(topic) is not None:
emsg = _("Topic contained dangerous characters.")
LOG.warn(emsg)
raise RPCException(emsg)
out_sock = ZmqSocket("ipc://%s/zmq_topic_%s" % (ipc_dir, topic), sock_type, bind=True)
except RPCException:
waiter.send_exception(*sys.exc_info())
return
self.topic_proxy[topic] = eventlet.queue.LightQueue(CONF.rpc_zmq_topic_backlog)
self.sockets.append(out_sock)
# It takes some time for a pub socket to open,
# before we can have any faith in doing a send() to it.
if sock_type == zmq.PUB:
eventlet.sleep(0.5)
waiter.send(True)
while True:
data = self.topic_proxy[topic].get()
out_sock.send(data, copy=False)
def consume(self, sock):
# TODO(ewindisch): use zero-copy (i.e. references, not copying)
data = sock.recv()
LOG.debug(_("CONSUMER RECEIVED DATA: %s"), data)
proxy = self.proxies[sock]
if data[2] == "cast": # Legacy protocol
packenv = data[3]
ctx, msg = _deserialize(packenv)
request = rpc_common.deserialize_msg(msg)
ctx = RpcContext.unmarshal(ctx)
elif data[2] == "impl_zmq_v2":
packenv = data[4:]
msg = unflatten_envelope(packenv)
request = rpc_common.deserialize_msg(msg)
# Unmarshal only after verifying the message.
ctx = RpcContext.unmarshal(data[3])
else:
LOG.error(_("ZMQ Envelope version unsupported or unknown."))
return
self.pool.spawn_n(self.process, proxy, ctx, request)
def __init__(self, addr, zmq_type, bind=True, subscribe=None):
self.sock = _get_ctxt().socket(zmq_type)
self.addr = addr
self.type = zmq_type
self.subscriptions = []
# Support failures on sending/receiving on wrong socket type.
self.can_recv = zmq_type in (zmq.PULL, zmq.SUB)
self.can_send = zmq_type in (zmq.PUSH, zmq.PUB)
self.can_sub = zmq_type in (zmq.SUB,)
# Support list, str, & None for subscribe arg (cast to list)
do_sub = {list: subscribe, str: [subscribe], type(None): []}[type(subscribe)]
for f in do_sub:
self.subscribe(f)
str_data = {"addr": addr, "type": self.socket_s(), "subscribe": subscribe, "bind": bind}
LOG.debug(_("Connecting to %(addr)s with %(type)s"), str_data)
LOG.debug(_("-> Subscribed to %(subscribe)s"), str_data)
LOG.debug(_("-> bind: %(bind)s"), str_data)
try:
if bind:
self.sock.bind(addr)
else:
self.sock.connect(addr)
except Exception:
raise RPCException(_("Could not open socket."))
def _start_child(self, wrap):
if len(wrap.forktimes) > wrap.workers:
# Limit ourselves to one process a second (over the period of
# number of workers * 1 second). This will allow workers to
# start up quickly but ensure we don't fork off children that
# die instantly too quickly.
if time.time() - wrap.forktimes[0] < wrap.workers:
LOG.info(_('Forking too fast, sleeping'))
time.sleep(1)
wrap.forktimes.pop(0)
wrap.forktimes.append(time.time())
pid = os.fork()
if pid == 0:
launcher = self._child_process(wrap.service)
while True:
self._child_process_handle_signal()
status, signo = self._child_wait_for_exit_or_signal(launcher)
if not _is_sighup(signo):
break
launcher.restart()
os._exit(status)
LOG.info(_('Started child %d'), pid)
wrap.children.add(pid)
self.children[pid] = wrap
return pid
def wait(self):
"""Loop waiting on children to die and respawning as necessary."""
LOG.debug(_('Full set of CONF:'))
CONF.log_opt_values(LOG, std_logging.DEBUG)
while True:
self.handle_signal()
self._respawn_children()
if self.sigcaught:
signame = _signo_to_signame(self.sigcaught)
LOG.info(_('Caught %s, stopping children'), signame)
if not _is_sighup(self.sigcaught):
break
for pid in self.children:
os.kill(pid, signal.SIGHUP)
self.running = True
self.sigcaught = None
for pid in self.children:
try:
os.kill(pid, signal.SIGTERM)
except OSError as exc:
if exc.errno != errno.ESRCH:
raise
# Wait for children to die
if self.children:
LOG.info(_('Waiting on %d children to exit'), len(self.children))
while self.children:
self._wait_child()
def _wait_child(self):
try:
# Don't block if no child processes have exited
pid, status = os.waitpid(0, os.WNOHANG)
if not pid:
return None
except OSError as exc:
if exc.errno not in (errno.EINTR, errno.ECHILD):
raise
return None
if os.WIFSIGNALED(status):
sig = os.WTERMSIG(status)
LOG.info(_('Child %(pid)d killed by signal %(sig)d'),
dict(pid=pid, sig=sig))
else:
code = os.WEXITSTATUS(status)
LOG.info(_('Child %(pid)s exited with status %(code)d'),
dict(pid=pid, code=code))
if pid not in self.children:
LOG.warning(_('pid %d not in child list'), pid)
return None
wrap = self.children.pop(pid)
wrap.children.remove(pid)
return wrap
def _connect(self, params):
"""Connect to rabbit. Re-establish any queues that may have
been declared before if we are reconnecting. Exceptions should
be handled by the caller.
"""
if self.connection:
LOG.info(_("Reconnecting to AMQP server on "
"%(hostname)s:%(port)d") % params)
try:
self.connection.release()
except self.connection_errors:
pass
# Setting this in case the next statement fails, though
# it shouldn't be doing any network operations, yet.
self.connection = None
self.connection = kombu.connection.BrokerConnection(**params)
self.connection_errors = self.connection.connection_errors
if self.memory_transport:
# Kludge to speed up tests.
self.connection.transport.polling_interval = 0.0
self.consumer_num = itertools.count(1)
self.connection.connect()
self.channel = self.connection.channel()
# work around 'memory' transport bug in 1.1.3
if self.memory_transport:
self.channel._new_queue('ae.undeliver')
for consumer in self.consumers:
consumer.reconnect(self.channel)
LOG.info(_('Connected to AMQP server on %(hostname)s:%(port)d') %
params)
def _process_data(self, ctxt, version, method, namespace, args):
"""Process a message in a new thread.
If the proxy object we have has a dispatch method
(see rpc.dispatcher.RpcDispatcher), pass it the version,
method, and args and let it dispatch as appropriate. If not, use
the old behavior of magically calling the specified method on the
proxy we have here.
"""
ctxt.update_store()
try:
rval = self.proxy.dispatch(ctxt, version, method, namespace,
**args)
# Check if the result was a generator
if inspect.isgenerator(rval):
for x in rval:
ctxt.reply(x, None, connection_pool=self.connection_pool)
else:
ctxt.reply(rval, None, connection_pool=self.connection_pool)
# This final None tells multicall that it is done.
ctxt.reply(ending=True, connection_pool=self.connection_pool)
except rpc_common.ClientException as e:
LOG.debug(_('Expected exception during message handling (%s)') %
e._exc_info[1])
ctxt.reply(None, e._exc_info,
connection_pool=self.connection_pool,
log_failure=False)
except Exception:
# sys.exc_info() is deleted by LOG.exception().
exc_info = sys.exc_info()
LOG.error(_('Exception during message handling'),
exc_info=exc_info)
ctxt.reply(None, exc_info, connection_pool=self.connection_pool)
def _multi_send(method, context, topic, msg, timeout=None, envelope=False, _msg_id=None):
"""Wraps the sending of messages.
Dispatches to the matchmaker and sends message to all relevant hosts.
"""
conf = CONF
LOG.debug(_("%(msg)s") % {"msg": " ".join(map(pformat, (topic, msg)))})
queues = _get_matchmaker().queues(topic)
LOG.debug(_("Sending message(s) to: %s"), queues)
# Don't stack if we have no matchmaker results
if not queues:
LOG.warn(_("No matchmaker results. Not casting."))
# While not strictly a timeout, callers know how to handle
# this exception and a timeout isn't too big a lie.
raise rpc_common.Timeout(_("No match from matchmaker."))
# This supports brokerless fanout (addresses > 1)
for queue in queues:
(_topic, ip_addr) = queue
_addr = "tcp://%s:%s" % (ip_addr, conf.rpc_zmq_port)
if method.__name__ == "_cast":
eventlet.spawn_n(method, _addr, context, _topic, msg, timeout, envelope, _msg_id)
return
return method(_addr, context, _topic, msg, timeout, envelope)
def inner(*args, **kwargs):
with lock(name, lock_file_prefix, external, lock_path):
LOG.debug(_('Got semaphore / lock "%(function)s"'),
{'function': f.__name__})
return f(*args, **kwargs)
LOG.debug(_('Semaphore / lock released "%(function)s"'),
{'function': f.__name__})
def _error_callback(exc):
if isinstance(exc, qpid_exceptions.Empty):
LOG.debug(_('Timed out waiting for RPC response: %s') %
str(exc))
raise rpc_common.Timeout()
else:
LOG.exception(_('Failed to consume message from queue: %s') %
str(exc))
def _error_callback(exc):
if isinstance(exc, socket.timeout):
LOG.debug(_('Timed out waiting for RPC response: %s') %
str(exc))
raise rpc_common.Timeout()
else:
LOG.exception(_('Failed to consume message from queue: %s') %
str(exc))
info['do_consume'] = True
def _process_data(self, message_data):
msg_id = message_data.pop('_msg_id', None)
waiter = self._call_waiters.get(msg_id)
if not waiter:
LOG.warn(_('No calling threads waiting for msg_id : %(msg_id)s'
', message : %(data)s'), {'msg_id': msg_id,
'data': message_data})
LOG.warn(_('_call_waiters: %s') % str(self._call_waiters))
else:
waiter.put(message_data)
def notify(context, publisher_id, event_type, priority, payload):
"""Sends a notification using the specified driver
:param publisher_id: the source worker_type.host of the message
:param event_type: the literal type of event (ex. Instance Creation)
:param priority: patterned after the enumeration of Python logging
levels in the set (DEBUG, WARN, INFO, ERROR, CRITICAL)
:param payload: A python dictionary of attributes
Outgoing message format includes the above parameters, and appends the
following:
message_id
a UUID representing the id for this notification
timestamp
the GMT timestamp the notification was sent at
The composite message will be constructed as a dictionary of the above
attributes, which will then be sent via the transport mechanism defined
by the driver.
Message example::
{'message_id': str(uuid.uuid4()),
'publisher_id': 'compute.host1',
'timestamp': timeutils.utcnow(),
'priority': 'WARN',
'event_type': 'compute.create_instance',
'payload': {'instance_id': 12, ... }}
"""
if priority not in log_levels:
raise BadPriorityException(
_('%s not in valid priorities') % priority)
# Ensure everything is JSON serializable.
payload = jsonutils.to_primitive(payload, convert_instances=True)
msg = dict(message_id=str(uuid.uuid4()),
publisher_id=publisher_id,
event_type=event_type,
priority=priority,
payload=payload,
timestamp=str(timeutils.utcnow()))
for driver in _get_drivers():
try:
driver.notify(context, msg)
except Exception as e:
LOG.exception(_("Problem '%(e)s' attempting to "
"send to notification system. "
"Payload=%(payload)s")
% dict(e=e, payload=payload))
def reconnect(self):
"""Handles reconnecting and re-establishing queues.
Will retry up to self.max_retries number of times.
self.max_retries = 0 means to retry forever.
Sleep between tries, starting at self.interval_start
seconds, backing off self.interval_stepping number of seconds
each attempt.
"""
attempt = 0
while True:
params = self.params_list[attempt % len(self.params_list)]
attempt += 1
try:
self._connect(params)
return
except (IOError, self.connection_errors) as e:
pass
except Exception as e:
# NOTE(comstud): Unfortunately it's possible for amqplib
# to return an error not covered by its transport
# connection_errors in the case of a timeout waiting for
# a protocol response. (See paste link in LP888621)
# So, we check all exceptions for 'timeout' in them
# and try to reconnect in this case.
if 'timeout' not in str(e):
raise
log_info = {}
log_info['err_str'] = str(e)
log_info['max_retries'] = self.max_retries
log_info.update(params)
if self.max_retries and attempt == self.max_retries:
msg = _('Unable to connect to AMQP server on '
'%(hostname)s:%(port)d after %(max_retries)d '
'tries: %(err_str)s') % log_info
LOG.error(msg)
raise rpc_common.RPCException(msg)
if attempt == 1:
sleep_time = self.interval_start or 1
elif attempt > 1:
sleep_time += self.interval_stepping
if self.interval_max:
sleep_time = min(sleep_time, self.interval_max)
log_info['sleep_time'] = sleep_time
LOG.error(_('AMQP server on %(hostname)s:%(port)d is '
'unreachable: %(err_str)s. Trying again in '
'%(sleep_time)d seconds.') % log_info)
time.sleep(sleep_time)
def register(self, proxy, in_addr, zmq_type_in, in_bind=True, subscribe=None):
LOG.info(_("Registering reactor"))
if zmq_type_in not in (zmq.PULL, zmq.SUB):
raise RPCException("Bad input socktype")
# Items push in.
inq = ZmqSocket(in_addr, zmq_type_in, bind=in_bind, subscribe=subscribe)
self.proxies[inq] = proxy
self.sockets.append(inq)
LOG.info(_("In reactor registered"))
def cast(conf, context, topic, msg, connection_pool):
"""Sends a message on a topic without waiting for a response."""
LOG.debug(_('Making asynchronous cast on %s...'), topic)
_add_unique_id(msg)
pack_context(msg, context)
with ConnectionContext(conf, connection_pool) as conn:
conn.topic_send(topic, rpc_common.serialize_msg(msg))
def __exit__(self, exc_type, exc_val, exc_tb):
try:
self.unlock()
self.lockfile.close()
except IOError:
LOG.exception(_("Could not release the acquired lock `%s`"),
self.fname)
def fanout_cast(conf, context, topic, msg, connection_pool):
"""Sends a message on a fanout exchange without waiting for a response."""
LOG.debug(_('Making asynchronous fanout cast...'))
_add_unique_id(msg)
pack_context(msg, context)
with ConnectionContext(conf, connection_pool) as conn:
conn.fanout_send(topic, rpc_common.serialize_msg(msg))
def inner_func(*args, **kwargs):
last_log_time = 0
last_exc_message = None
exc_count = 0
while True:
try:
return infunc(*args, **kwargs)
except Exception as exc:
this_exc_message = six.u(str(exc))
if this_exc_message == last_exc_message:
exc_count += 1
else:
exc_count = 1
# Do not log any more frequently than once a minute unless
# the exception message changes
cur_time = int(time.time())
if (cur_time - last_log_time > 60 or
this_exc_message != last_exc_message):
logging.exception(
_('Unexpected exception occurred %d time(s)... '
'retrying.') % exc_count)
last_log_time = cur_time
last_exc_message = this_exc_message
exc_count = 0
# This should be a very rare event. In case it isn't, do
# a sleep.
time.sleep(1)
def start(self):
super(Service, self).start()
self.conn = rpc.create_connection(new=True)
LOG.debug(_("Creating Consumer connection for Service %s") %
self.topic)
dispatcher = rpc_dispatcher.RpcDispatcher([self.manager],
self.serializer)
# Share this same connection for these Consumers
self.conn.create_consumer(self.topic, dispatcher, fanout=False)
node_topic = '%s.%s' % (self.topic, self.host)
self.conn.create_consumer(node_topic, dispatcher, fanout=False)
self.conn.create_consumer(self.topic, dispatcher, fanout=True)
# Hook to allow the manager to do other initializations after
# the rpc connection is created.
if callable(getattr(self.manager, 'initialize_service_hook', None)):
self.manager.initialize_service_hook(self)
# Consume from all consumers in a thread
self.conn.consume_in_thread()
def initialize_if_enabled():
backdoor_locals = {
'exit': _dont_use_this, # So we don't exit the entire process
'quit': _dont_use_this, # So we don't exit the entire process
'fo': _find_objects,
'pgt': _print_greenthreads,
'pnt': _print_nativethreads,
}
if CONF.backdoor_port is None:
return None
start_port, end_port = _parse_port_range(str(CONF.backdoor_port))
# NOTE(johannes): The standard sys.displayhook will print the value of
# the last expression and set it to __builtin__._, which overwrites
# the __builtin__._ that gettext sets. Let's switch to using pprint
# since it won't interact poorly with gettext, and it's easier to
# read the output too.
def displayhook(val):
if val is not None:
pprint.pprint(val)
sys.displayhook = displayhook
sock = _listen('localhost', start_port, end_port, eventlet.listen)
# In the case of backdoor port being zero, a port number is assigned by
# listen(). In any case, pull the port number out here.
port = sock.getsockname()[1]
LOG.info(_('Eventlet backdoor listening on %(port)s for process %(pid)d') %
{'port': port, 'pid': os.getpid()})
eventlet.spawn_n(eventlet.backdoor.backdoor_server, sock,
locals=backdoor_locals)
return port
def deprecated(self, msg, *args, **kwargs):
stdmsg = _("Deprecated: %s") % msg
if CONF.fatal_deprecations:
self.critical(stdmsg, *args, **kwargs)
raise DeprecatedConfig(msg=stdmsg)
else:
self.warn(stdmsg, *args, **kwargs)
def __init__(self, info=None, topic=None, method=None):
"""Initiates Timeout object.
:param info: Extra info to convey to the user
:param topic: The topic that the rpc call was sent to
:param rpc_method_name: The name of the rpc method being
called
"""
self.info = info
self.topic = topic
self.method = method
super(Timeout, self).__init__(
None,
info=info or _('<unknown>'),
topic=topic or _('<unknown>'),
method=method or _('<unknown>'))
def __init__(cls, names, bases, dict_):
"""Metaclass that allows us to collect decorated periodic tasks."""
super(_PeriodicTasksMeta, cls).__init__(names, bases, dict_)
# NOTE(sirp): if the attribute is not present then we must be the base
# class, so, go ahead an initialize it. If the attribute is present,
# then we're a subclass so make a copy of it so we don't step on our
# parent's toes.
try:
cls._periodic_tasks = cls._periodic_tasks[:]
except AttributeError:
cls._periodic_tasks = []
try:
cls._periodic_last_run = cls._periodic_last_run.copy()
except AttributeError:
cls._periodic_last_run = {}
try:
cls._periodic_spacing = cls._periodic_spacing.copy()
except AttributeError:
cls._periodic_spacing = {}
for value in cls.__dict__.values():
if getattr(value, '_periodic_task', False):
task = value
name = task.__name__
if task._periodic_spacing < 0:
LOG.info(_('Skipping periodic task %(task)s because '
'its interval is negative'),
{'task': name})
continue
if not task._periodic_enabled:
LOG.info(_('Skipping periodic task %(task)s because '
'it is disabled'),
{'task': name})
continue
# A periodic spacing of zero indicates that this task should
# be run every pass
if task._periodic_spacing == 0:
task._periodic_spacing = None
cls._periodic_tasks.append((name, task))
cls._periodic_spacing[name] = task._periodic_spacing
cls._periodic_last_run[name] = task._periodic_last_run