Distributed Messaging for OpenStack
This page was created to outline the core requirements for messaging in OpenStack and how to achieve those requirements in a reliable, performant and resilient way.
Core Design Principles
- Fully distributed design?
- Easy failover
- Scalable == Federatable
RabbitMQ and Qpid are troublesome. 0MQ doesn't have the support that it needs. I think we need something pure python and much simpler than the current messaging infra to take care of all this stuff. A pure-python easily understood implementation will attract maintainers and people interested in extending things.
- Cloud failures along availability zones
- One to One (host to host or service to service)
- One to Many