This is very true for more reasons than those you describe. I am
presenting on activemq at apachecon in a couple of weeks and I will
probably talk about this a bit.
JMX can only give you a very limited number of useful metrics. JMX is
also almost useless in a more complex topology.
I would separate the metric collection from reporting.
$0.02,
Hadrian
On 04/01/2015 10:38 PM, Kevin Burton wrote:
A couple things I wanted to talk about which are somewhat related.
JMX is somewhat slow. Further, at least from my perspective, it’s somewhat
dated. All the cool kinds have fancy REST API endpoints except for
ActiveMQ.
I think we’re going to migrate to our own embedded activemq using our own
infra and tooling which means I could setup a REST API endpoint using JSON.
I just did a benchmark using an embedded broker and it took 8000ms to get
queue metadata on 10k queues vs 15ms to do it inside its own VM.
The cool thing is that ALL the JMX infra is there. So I would really be
mirroring the JMX protocol, just in HTTP.
Now, I guess i could roll my own, but this seems silly. I think others
would want this too at some point and at least this would end up in a more
documented protocol.
This is where I wind up talking about the 5.x series post 6.0. I assume
5.x will still be maintained by the community considering a lot of people
are still using it in production. But I guess this is OSS so it’s all
based on merit.
I think maybe a sub-module in ActiveMQ for a REST API could be warranted.
I could get ours working and then throw it over the fence in a rough form
at some point. At least just a collection of servlets that could be used
with Jetty.
