def transport_task(key, queue, args=(), kwargs={}):
# payload that will be sent to broker
payload = {
'args': args,
'kwargs': kwargs,
}
# retry policy for declaring the queue
retry_policy = {
'interval_start': 0,
'interval_step': 2,
'interval_max': 60,
'max_retries': 30
}
with Connection(os.environ.get('BROKER_URL')) as connection:
with connection.channel() as channel:
producer = Producer(channel)
producer.publish(payload,
retry=True,
retry_policy=retry_policy,
exchange=exchange,
routing_key=queue.routing_key,
declare=[queue])
producer.release()
class Emitter(object):
"""Helper class for writing robust AMQP producers.
An instance of this class will connect to the given AMQP broker,
and allow you to send messages to an exchange on that broker with
a given routing key, optionally also declaring a queue and and
binding it to the exchange with that routing key. Upon connection
errors (either upon initial connection or message send) it will
retry several times before giving up, and if a connection is
interrupted it will recover gracefully on the next attempt to send
a message.
Short-lived connections
-----------------------
To create a short-lived connection which automatically
releases/disconnects when you've finished, just use an instance of
this class as a context manager; for example::
url = broker_url('guest', 'guest', 'localhost')
with Emitter(url, 'ex_name', 'key') as emitter:
emitter.send_message({'status': 'success'}, serializer='json')
Note the lack of any explicit `connect()` call here:
`send_message()` calls it if it hasn't been called already, so you
never need to call it explicitly (though you can).
Note also the lack of any queue_name parameter in the Emitter
constructor call: the message will be sent to the given exchange
with the given routing key, but no queue declaration/binding takes
place here (compare with the next example).
Long-lived connections
----------------------
Alternatively, you can just create an instance of this class and
have it live for as long as you need it, in which case the
connection will remain open until the instance falls out of scope
or you call its `release()` method.
To have an instance live for the lifetime of an application, a
helpful pattern is to put it in a module-level global and lazily
declare it on first use, e.g.::
# Module-level global for the instance; initially nothing here.
_emitter = None
def send_message(msg, serializer):
global _emitter
if _emitter is None:
url = broker_url('guest', 'guest', 'localhost')
_emitter = Emitter(url, 'ex_name', 'key', 'q_name')
_emitter.send_message(msg, serializer)
Then you can just call this module's `send_message()` function
whenever you need to send a message: it will connect to the broker
on the first call, keep the connection open between calls, and
attempt to reconnect on the next call if the connection ever goes
away.
Note also that in this example we do pass a queue_name to the
Emitter constructor: upon connection, the Emitter will ensure that
the named queue is declared, and bound to the given exchange with
the given routing key.
Customisation via subclassing
-----------------------------
You may wish to use a custom subclass to override certain
behaviour around error handling. In particular, you can override:
* Various `CONN_*` and `SEND_*` class variables, which control
connection/send retry behaviour. E.g. `SEND_MAX_RETRIES`
defines how many times to retry sending a message before giving
up.
* The `conn_errback()` callback method, which is called upon an
error while connecting.
* The `send_errback()` callback method, which is called upon an
error while sending a message.
* The `errback()` callback method, which (by default) is just what
`conn_errback()` and `send_errback()` call - so override just this
to change them both. The default version just logs the error.
For example, here's a class which retries connecting forever;
waits up to 10 seconds between retries; only retries message sends
once; and logs the full traceback for connect errors::
class CustomEmitter(Emitter):
CONN_MAX_RETRIES = None
CONN_INTERVAL_MAX = 10
SEND_MAX_RETRIES = 1
def conn_errback(self, ex, interval):
logger.error('Error: %r', ex, exc_info=1)
logger.info('Retry in %s seconds.', interval)
Note the following edge cases for `*_MAX_RETRIES`:
* `None` - for both `CONN_` and `SEND_` this means "try forever".
* 0 - for `CONN_` this means "try forever" but for `SEND_` this
means "retry once", i.e. the same as `SEND_MAX_RETRIES = 1`;
this seems to be a kombu bug.
* Negative values - for `CONN_` this means "just try once/no
retries", but for `SEND_` it just breaks things (it will never
recover the connection); this seems to be a kombu bug.
"""
CONN_MAX_RETRIES = 2 # Max retries when attempting to connect.
CONN_INTERVAL_START = 2 # Seconds to wait between connect retries, initially.
CONN_INTERVAL_STEP = 2 # Increase time between connect retries by this amount.
CONN_INTERVAL_MAX = 4 # Maximum time between connect retries.
SEND_MAX_RETRIES = 3 # Max retries when attempting to send.
SEND_INTERVAL_START = 1 # Seconds to wait between send retries, initially.
SEND_INTERVAL_STEP = 1 # Increase time between send retries by this amount.
SEND_INTERVAL_MAX = 1 # Maximum time between send retries.
def __init__(self,
url,
exchange_name,
routing_key,
queue_name=None,
exchange_type='direct'):
logger.debug('Initialising {}'.format(self.__class__.__name__))
self.url = url
self.exchange_name = exchange_name
self.routing_key = routing_key
self.queue_name = queue_name
self.exchange_type = exchange_type
self._connection = None
self._producer = None
def conn_errback(self, ex, interval):
"""Callback called upon connection error."""
self.errback(ex, interval)
def send_errback(self, ex, interval):
"""Callback called upon send error."""
self.errback(ex, interval)
def errback(self, ex, interval):
"""Default callback called upon connection or send error."""
logger.info('Error: {} - {}'.format(ex.__class__.__name__, str(ex)))
logger.info('Retry in %s seconds.', interval)
def connect(self):
"""Connect to broker and possibly ensure exchange/queue/routing declared.
In case of errors, this will retry connecting several times
(controlled by the `CONN_*` class variables), and gracefully
recover if possible.
If it never succeeds, the exception raised by the final
attempt is re-raised.
"""
logger.debug('Connecting to broker at: {}'.format(self.url))
self._connection = Connection(self.url)
# Kombu interprets interval_max incorrectly; work around that.
interval_max = self.CONN_INTERVAL_MAX - self.CONN_INTERVAL_STEP
self._connection.ensure_connection(
errback=self.conn_errback,
max_retries=self.CONN_MAX_RETRIES,
interval_start=self.CONN_INTERVAL_START,
interval_step=self.CONN_INTERVAL_STEP,
interval_max=interval_max)
exchange = Exchange(name=self.exchange_name, type=self.exchange_type)
channel = self._connection.channel()
self._producer = Producer(channel=channel,
exchange=exchange,
routing_key=self.routing_key)
if self.queue_name:
# Bind/declare queue.
queue = Queue(name=self.queue_name,
exchange=exchange,
routing_key=self.routing_key)
queue = queue(channel) # Bind queue
logger.debug('Declaring queue {}, on exchange {} at {}'.format(
self.queue_name, self.exchange_name, self.url))
queue.declare()
def release(self):
"""Disconnect/release."""
self._producer.release()
self._connection.release()
def __enter__(self):
"""Context manager entry: connect to the broker."""
self.connect()
return self
def __exit__(self, exc_type, exc_value, traceback):
"""Context manager exit: disconnect/release."""
self.release()
def send_message(self, message, serializer='json', headers=None):
"""Send a message with retries (and connect to broker if necessary).
In case of errors, this will retry sending several times
(controlled by the `SEND_*` class variables), and gracefully
recover if possible.
If it never succeeds, the exception raised by the final
attempt is re-raised.
You can use the headers argument to pass in any custom headers. It
is a dictionary {"a-custom-header": "a-custom-value"}.
"""
logger.debug("Sending message...")
if self._producer is None:
self.connect()
# Kombu interprets interval_max incorrectly; work around that.
interval_max = self.SEND_INTERVAL_MAX - self.SEND_INTERVAL_STEP
publish = self._connection.ensure(
self._producer,
self._producer.publish,
errback=self.send_errback,
max_retries=self.SEND_MAX_RETRIES,
interval_start=self.SEND_INTERVAL_START,
interval_step=self.SEND_INTERVAL_STEP,
interval_max=interval_max)
publish(message, serializer=serializer, headers=headers)
