[rabbitmq-discuss] producer/consumer and every n-th message

[Michael Klishin]

2011/7/8 Torsten Curdt <tcurdt at vafer.org>

> Hey,
>
> Let's say I have n producers sending messages to an exchange and m
> consumers consuming from a single queue bound to that exchange. When a
> consumer does not acknowledges a message (and dies), the message
> should go back into the queue for another consumer to pick it up.
>
> That should be easy to implement I assume?
>

Yes, AMQP 0.9.1 has this feature built-in. See tutorial 2:

http://www.rabbitmq.com/getstarted.html

and a longer example with multiple consumers in the Ruby amqp gem guide:

http://rubydoc.info/github/ruby-amqp/amqp/master/file/docs/Queues.textile#Message_acknowledgements


> I also assume for this scenario a direct exchange would do?
> How bad would be the contention n and m are fairly large?
> ...or do I just need to have enough broker?
>
>
Sorry, what are n and m? Message acknowledgements have very low overhead and
redeliveries typically do not happen
often enough to create any contention.


> I would also like to be able to "tap into" the stream of messages
> flowing through the exchange. I assume I could use a fanout exchange
> instead. Have the normal queue with the m consumers and a second queue
> with a single consumer "tapping in".
>
> As the throughput of message can potentially be very high (which is
> why there are so many consumers) of course a single consumer cannot
> drain the second queue. So my idea was to only care about every n-th
> message.
>
> Is there a away to tell the fanout consumer to deliver every message
> to Q1 but only every 10th message to Q2?
> How else could one do this?
>
>
Every solution I can think of will require producers to keep track of
message sequences. Take a look at
http://www.rabbitmq.com/extensions.html#sender-selected-distribution.

-- 
MK

http://github.com/michaelklishin
http://twitter.com/michaelklishin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rabbitmq.com/pipermail/rabbitmq-discuss/attachments/20110708/1abc7cc5/attachment.htm>