[rabbitmq-discuss] Getting Started with RabbitMQ
opendev at gmail.com
Thu Feb 19 13:04:38 GMT 2009
2009/2/18 Alexis Richardson <alexis.richardson at cohesiveft.com>:
> Thanks again for this wikipedia work, it is great for AMQP.
> Speaking for the RabbitMQ team, we have no current plans to implement
> AMQP 0-10. Instead we're moving to 0-9-1, which is a very small
> change from 0-8 and will lead to interop with other brokers. From
> there we shall attend AMQP 1.0.
Sounds like a good plan.
> Speaking for the AMQP Working Group: there is 1.0 work under way,
> which among others things aims to refactor and simplify 0-10. It is
> hoped that drafts and even release candidates will be published for
> review by summertime. We've learnt in the past that it is not prudent
> to put hard dates on GA release, because baking protocols takes time
> and care.
I've taken a look. Sounds like a pretty big deal with exchanges and
bindings going out of the window.
> How about: The main technology introduced in 0-10 and subsequently, is
> the 'session' which abstracts a bilateral conversation across
> failures. Flow control and distributed transactions should also be
> mentioned, and more rigrorous support for JMS.
> IMO - These concepts are all important, but do leave the 0-10 reader
> craving a simpler approach.
I think sessions are actually make things simpler e.g. coupling the
lifespan of a session to the lifespan of auto deleting, exclusive
queues for instance.
IMO we should just sketch a short history of the AMQP specifications
and make a note that the article is about version 0.10. This probably
means (among other things) we should put sessions into the model.
> BTW, JMS has annoying and lovely impedance mismatches with AMQP. It
> truly is a hairy thing.
I never did use JMS so I can't really tell.
> Yes, that is a good point. Arguably the whole idea of a 'message
> protocol' is to standardise how parties get out of trouble.
And this is where things get a bit involved. Did the specs involve
that much between revisions? If not it would make sense to just write
about the 0.10 way of handling errors.
> Maybe ... I'd hold back from turning the wikipedia page into a new
> spec though ;-)
Of course. We should put our efforts into making the description
_simple_ not complete.
How should we proceed? Do we just assume 0.10 for everything, sketch
out the rough differences between 0.8, 0.9 ... and make the model
complete (including transactions, sessions etc.)?
More information about the rabbitmq-discuss