Cluster Name in JBossEAP6.2

[kuldeep11]

Hi,

I have query regarding the Jgroup Clustering setup.

When we configure cluster(e.g UDP based), it print the message as below :

GMS: address={jboss.node.name}/web, cluster=web, physical address=0.0.0.0:55200

In this message from where it pick up the cluster name i.e, "web".

Thanks!

[inspector]

From the name of the infinispan cache container "web". There should also be a similar message for the container "ejb".

[kuldeep11]

thank you.

So, this means that it will use the cache container name.

I have below queries:

1.) Which will be picked up, if we have multiple cache container.

2.) Is there any configuration where we can specify which to use as default.

I can see only one message in logs which is printing one out of multiple cache containers.

Thanks!

[inspector]

The cluster name you observed is a name on jgroups level. This is quite low level. Every cache container that is not local will have a jgroups channel with a corresponding log message similar to the one you observed.

Maybe in your case only the web cache container was started. I'm not sure but I guess the ejb container is only started when you have @Clustered stateful session beans.

Speaking in general: You should not change that name. If you want to have multiple clusters running on the same network you should use different multicast addresses in case of the default UDP stack. Have a look at you socket bindings:

<socket-binding name="jgroups-udp" port="55200" multicast-address="${jboss.default.multicast.address:230.0.0.4}" multicast-port="45688"/>

You could also keep your clusters apart by using a different discovery protocol in your JGroups stack.

[wdfink]

What is the reason behind this question?

In most of the cases you should not care about the names.

If you have different clusters you should set the mcast address and port to separate it (as inspector mentioned)

[kuldeep11]

Ok, So reason behind the question is that In my case I can see the different Name(instead of "web"). So, curious to know that from where it is being picked up?

As per inspector It will be from defined cache containers and we have multiple cache containers. So how it is being decided that which one to use?

GMS: address={jboss.node.name}/abc, cluster=abc, physical address=0.0.0.0:55200

Thanks!

[inspector]

Such a name could be set by pretty much anybody who opens a jgroups channel. In standard full-ha profile this could be various infinispan cache containers and as far as I know, also hornetq. If i've got you right, you have an actual log message where the name is _not_ "web" or "ejb". Could you please provide your exact log message?

I just started an eap 6 on my local machine with a <distributable/> web application and multiple @Clustered EJBs. And my log looks like this:

13:24:38,115 INFO  [stdout] (ServerService Thread Pool -- 55)

13:24:38,116 INFO  [stdout] (ServerService Thread Pool -- 55) -------------------------------------------------------------------

13:24:38,116 INFO  [stdout] (ServerService Thread Pool -- 55) GMS: address=isi/web, cluster=web, physical address=127.0.0.1:55200

13:24:38,116 INFO  [stdout] (ServerService Thread Pool -- 55) -------------------------------------------------------------------

13:24:38,122 INFO  [stdout] (ServerService Thread Pool -- 59)

13:24:38,123 INFO  [stdout] (ServerService Thread Pool -- 59) -------------------------------------------------------------------

13:24:38,124 INFO  [stdout] (ServerService Thread Pool -- 59) GMS: address=isi/ejb, cluster=ejb, physical address=127.0.0.1:55200

13:24:38,128 INFO  [stdout] (ServerService Thread Pool -- 59) -------------------------------------------------------------------

Regards!

[kuldeep11]

Below is what I am getting in logs:

11-03-2015 15:58:44,473 INFO  (stdout) [ServerService Thread Pool -- 54]

11-03-2015 15:58:44,474 INFO  (stdout) [ServerService Thread Pool -- 54] -------------------------------------------------------------------

11-03-2015 15:58:44,474 INFO  (stdout) [ServerService Thread Pool -- 54] GMS: address=node1/switchyard, cluster=switchyard, physical address=0.0.0.0:55300

11-03-2015 15:58:44,474 INFO  (stdout) [ServerService Thread Pool -- 54] -------------------------------------------------------------------

Thanks!

[belaban]

Note that, regardless of the naming issue, a physical address of 0.0.0.0:55200 won't work in JGroups and your cluster won't form. If you give option -b 0.0.0.0 to Wildfly, make sure to use -Djgroups.bind_addr=x.x.x.x to point to a valid interface.

[inspector]

Ahh I see. I guess it's switchyard who is starting a JGroups channel. I wasn't aware switchyard uses JGroups. I guess you use fuse service works or the community switchyard.

By the way Bela's comment is quite important.

[kuldeep11]

Yes, we are using FSW with switchyard and using JGroup configuration from xml as below.

        <socket-binding name="jgroups-udp" port="55200" multicast-address="${jboss.default.multicast.address:230.0.0.4}" multicast-port="45688"/>

My question is still there, where we configure this or where it is being picked up .

[inspector]

Well, I'm not that deep into FSW. Maybe there is some place where you can configure that name. But as elaborated on earlier you shouldn't. Could be that even some FSW internal logic depends on that name. I'd compare the name of a JGroups channel to the jndi name of an ejb. It's application internal stuff that you maybe could touch, but you never should do that if you're not the application's developer.

I'm still not sure though, why you want to change that name. The only reason that I see would be to prevent separate clusters from forming a big cluster. And for that reason I'd configure different multicast addresses for JGroups.

Maybe you should read a little bit about JGroups. A good starting point is the project page JGroups - Overview.

[kuldeep11]

Thank You Bela and Inspector. I am agree with all of you and currently going through JGroups only. I was just curious to know about the this name.

Anyway thanks all for your views .

Thank!!