Domain event broker

This library provides a shallow layer on top of RabbitMQ topic exchanges for publishing and receiving domain events. Publisher and subscriber need not know about each other and can be started and stopped in any order. Each subscriber controls their own retry policy, whether they need a durable queue for the time they are down, or a dead-letter queue in case there is an error in the subscriber.

Configuration

This library needs to connect to RabbitMQ. By default, a local instance of RabbitMQ is used. This can be changed by passing an amqp URL to publish_domain_event or when instantiating Publisher or Subscriber:

from domain_event_broker import Subscriber
subscriber = Subscriber('amqp://user:password@rabbitmq-host/domain-events')

Integrations

Django

This library can be configured via your Django settings. Add domain_event_broker.django to your INSTALLED_APPS and set the DOMAIN_EVENT_BROKER in your settings:

INSTALLED_APPS = (
    'domain_event_broker.django',
    )

DOMAIN_EVENT_BROKER = 'amqp://user:password@rabbitmq-host/domain-events'

Sending events

Events can be sent by calling publish_domain_event:

from domain_event_broker import publish_domain_event
publish_domain_event('user.registered', {'user_id': user.id})

Domain events are sent immediately. When emitting domain events from within a database transaction, it's recommended to defer publishing until the transaction is committed. Using a commit hook avoids spurious domain events if a transaction is rolled back after an error.

Receiving events

Subscribers can listen to one or more domain events - controlled via the binding keys. Binding keys may contain wildcards. A queue will be created for each subscriber. RabbitMQ takes care of routing only the relevant events to this queue.

This script will receive all events that are sent in the user domain:

from domain_event_broker import Subscriber

def handle_user_event(event):
    print event

subscriber = Subscriber()
subscriber.register(handle_user_event, 'printer', ['user.*'])
subscriber.start_consuming()

Retry policy

If there is a problem consuming a message - for example a web service is down - the subscriber can raise an error to retry handling the event after the given delay:

from domain_event_broker import Subscriber

def sync_user_data(event):
    try:
        publish_to_service(event)
    except ServiceIsDown:
        raise Retry(5.0 ** event.retries) # 1s, 5s, 25s

subscriber = Subscriber()
subscriber.register(sync_user_data, 'sync_data', ['user.*'], max_retries=3)
subscriber.start_consuming()

The delayed retries are bound to the consumer, not the event. If max_retries is exceeded, the event will be dropped or dead-lettered.

Development

Make sure you have RabbitMQ installed locally for testing.

• create virtualenv and activate it
• run pip install -r requirements.txt -r dev_requirements.txt -e .
• the only external dependency (so far) is pika

Architecture

There's

• Generic domain events: domain_event_broker.events
• The transport, via rabbitmq: domain_event_broker.transport

Testing

Testing is done with py.test.