Open Stack

On Mon Sep 8 15:18:35 UTC 2014, Doug Hellmann wrote:
>On Sep 8, 2014, at 10:35 AM, Antonio Messina <antonio.s.messina at gmail.com> wrote:
>
>> Hi All,
>>
>> We tested briefly ZeroMQ with Havana last year, but we couldn't find
>> any good documentation on how to implement it, and we were not able to
>> get it working. We also got the impression that the support was not at
>> all mature, so we decided to use RabbitMQ instead.
>>
>> However, I must say that the broker-less design of ZeroMQ is very
>> appealing, and we would like to give it a try, assuming
>> 1) the documentation is improved
>> 2) there is some assurance that support for ZeroMQ is not going to be dropped.
>>
>> I can help with 1) if there is someone that knows a bit of the
>> internals and can bootstrap me, because I have no first hand
>> experience on how message queues are used in OpenStack, and little
>> experience with ZeroMQ.
>
>Unfortunately, the existing Oslo team doesn’t have a lot of
>experience with ZeroMQ either (hence this thread). It sounds like Li
>Ma’s team has made it work, though, so maybe you could work
>together. We should prioritize documentation and then functional
>testing, I think.
>
> About 2), well this is a decision for the developers, but IMHO there
> *should* be support for ZeroMQ in OpenStack: its broker-less
> architecture would eliminate a SPoF (the message broker), could ease
> the deployment (especially in HA setup) and grant very high
> performance.
>
>I agree, it would be useful to support it. This is purely a resource
>allocation problem for me. I don't have anyone willing to do the work
>needed to ensure the driver is functional and can be deployed sanely
>(although maybe I’ve found a couple of volunteers now :-).
>
>There is another effort going on to support AMQP 1.0, which (as I
>understand it) includes similar broker-less deployment options. Before
>we decide whether to invest in ZeroMQ for that reason alone, it would
>be useful to know if AMQP 1.0 support makes potential ZeroMQ support
>less interesting.
>

While the AMQP 1.0 protocol permits it, the current implementation of
the new driver does not support broker-less point-to-point - yet.

I'm planning on adding that support to the AMQP 1.0 driver in the
future.  I have yet to spend any time ramping up on the existing
brokerless support implemented by the zmq driver, so forgive my
ignorance, but I'm hoping to leverage what is there if it makes sense.

If it doesn't make sense, and the existing code is zmq specific, then
I'd be interested in working with the zmq folks to help develop a more
generic implementation that functions across both drivers.

>Doug


-- 
Ken Giusti  (kgiusti at gmail.com)