On 9/5/07, jgunz <[EMAIL PROTECTED]> wrote: > I realized after I sent my original message that I didn't specify what type > of clustering I was referring to. I think I understand how the routing rules > would work in a master/slave setup but I'm still a little foggy on how they > would function in a store and forward network of brokers. My goal in > clustering would be to scale the brokering (and separately the processing) > across multiple machines so I could just throw more CPUs at the problem if I > need to later. > > Ideally I'd like to be able to add brokers or consumers without having to > configure each one specially (or at least only having to do so minimally).
You typically only need a few brokers (usually one is enough :). Though the processing of the messages can often be costly, so if you've alot of heavy processing to do, you might wanna run a cluster of processors processing messages. ActiveMQ is a great high performance reliable load balancer and is a very good choice for this purpose thanks to Queues. So you might find its easier to do your Camel processing using a spring.xml and wrapping that up with Spring-OSGi to make an OSGi bundle and deploying it on as many machines as you need. So basically make an OSGi bundle with your META-INF/spring/camel-context.xml in it along with whatever jars you need; then just deploy on your OSGi containers. In terms of the ActiveMQ configuration; its usually just a single URL which you can either hardwire in your spring.xml or parameterize using the spring property resolving stuff; if you've only a few brokers you can just list them using the failover URL... http://activemq.apache.org/failover-transport-reference.html If you need tons of brokers; you could use multicast discovery, so JVMs discover the available brokers. > So if I setup an ActiveMQ installation, if I wanted to improve the > throughput or capacity I could basically take the same installation and put > it on another machine and that's it. Agreed. > If I did that and I distributed the > same routing rules to each, would that cause weird message duplication as > messages were forwarded between brokers and each of them duplicated the > application of the routing rule? If you duplicate the Camel processing code to multiple JVMs (using OSGi or a stand alone main in a shell script or WAR etc) and you use queues to balance the load then you'll have reliable processing of messages without duplication (as queues guarantee only one consumer processes a message). If you are dealing with any unreliable kinds of messages or protocols, there's always the Idempotent Consumer to help... http://activemq.apache.org/camel/idempotent-consumer.html > In regards to OSGi, your suggestion is music to my ears =) That's precisely > what I was thinking would be an ideal solution. OSGi could potentially allow > a consumer application to install a routing rule in the broker so I didn't > have to statically configure them before deployment. > > Thanks for the reply, I'm stoked to get rolling on using some of these great > features. My pleasure :) -- James ------- http://macstrac.blogspot.com/
