Different partitions, same cluster?

[a.novarini]

Hi all,

I'm having a strange problem with clustering on JBoss 5.1.0.GA.

Basically, I'm running the 'all' configuration, specifying the parameter -g (but the same with -Djboss.partition.name) with different names for each node.

One of the examples is: two jboss instances on the same machine.

Node1: ./run.sh -c all -Djboss.partition.name=FirstPartition

Node2: ./run.sh -c all-node2 -Djboss.service.binding.set=ports-01 -Djboss.messaging.ServerPeerID=1 -Djboss.partition.name=SecondPartition

all and all-node2 are the same, all-node2 is just a copy of all

The result log is, on the node2, but it's the same for node1:

12:27:38,762 INFO  [GroupMember] New Members : 2 ([127.0.0.1:54061, 127.0.0.1:62242])

12:27:38,762 INFO  [GroupMember] All Members : 2 ([127.0.0.1:54061, 127.0.0.1:62242])

12:27:38,892 WARN  [TCP] failed to join /224.0.0.75:7500 on en1: java.net.SocketException: Address already in use

12:27:38,76212:27:38,762 INFO  [GroupMember] New Members : 2 ([127.0.0.1:54061, 127.0.0.1:62242])

12:27:38,762 INFO  [GroupMember] All Members : 2 ([127.0.0.1:54061, 127.0.0.1:62242])

12:27:38,892 WARN  [TCP] failed to join /224.0.0.75:7500 on en1: java.net.SocketException: Address already in use INFO  [GroupMember] New Members : 2 ([127.0.0.1:54061, 127.0.0.1:62242])

12:27:38,762 INFO  [GroupMember] All Members : 2 ([127.0.0.1:54061, 127.0.0.1:62242])

12:27:38,892 WARN  [TCP] failed to join /224.0.0.75:7500 on en1: java.net.SocketException: Address already in use

What we did is to change the udp address for one partition, but I'm sure it worked fine with previous versions, and I am sure this is something so evident that there must be something "outside" jboss that I don't get right.

Do you have any advice on it?

Thanks a lot

Ale

[belaban]

Set enable_diagnostics="false" in your JGroups config file. Or pick a different diagnostics address / or port

[a.novarini]

Thanks.

Actually what you suggested is related to the error message

failed to join /224.0.0.75:7500 on en1: java.net.SocketException: Address already in use

But my main concern is about the two nodes with different partition names, that are in the same group, something it shouldn't happen unless there is something really wrong.

What I described, is happening with different nodes on different machines and with different nodes on the same machine AND different networks AND different operating systems (MacOs X and Windows)

So, what there is in common is just the jboss version (5.1.0.GA) and the 'all' instance, everything is different, but with the same result.

I'm puzzled, so do you have any idea?

Thanks a lot

Ale

[belaban]

Ah, I see...

Go to jboss-5.1.0/bin and run ./probe.sh -addr 224.0.0.75

It'll dump information about all instances. You'll see that instances do *not* cluster, except for the JBossMessaging cluster instances. Apparently JBM doesn't honor jboss.partition.name. I suggest ping the JBM folks for details...

[a.novarini]

Thanks for the help.

So, are you saying you're experiencing the same behavior on your environment also?

Regards

Ale

[belaban]

Yes, but only the JBM channels form a cluster, the others don't

[wdfink]

That sounds like an issue that we have with the EAP.

There is a JIRA initiate by this https://jira.jboss.org/jira/browse/JBMESSAGING-1627

The workaround is to set

-g <partition name> -u <mcast address>

[a.novarini]

Ok then, for now we'll change the mcast address; not a big deal, just to be sure that we didn't mess up our configurations.

Thanks to all

Ale