Problem with AutoDiscovery in JBoss-4.0.2 (on RH EL 3.0)

[sinus]

Hello,

we have two separate clusters running where a client from the first cluster wants to look up a bean in HA-JNDI from the second one.
After an upgrade from 4.0.0 to 4.0.2, the AutoDiscovery feature doesn't work anymore.

JBoss 4.0.0 starts up with this message:
2005-06-06 09:51:31,359 INFO [org.jboss.ha.jndi.DetachedHANamingService$AutomaticDiscovery] Listening on /192.168.0.1:1102, group=230.0.0.4, HA-JNDI address=192.168.0.1:1100

netstat -a -n gives:
udp 0 0 0.0.0.0:1102 0.0.0.0:*

and everything is fine.

JBoss 4.0.2 comes up with that message:
2005-06-06 09:25:23,868 INFO [org.jboss.ha.jndi.DetachedHANamingService$AutomaticDiscovery] Listening on /0.0.0.0:1102, group=230.0.0.4, HA-JNDI address=192.168.0.1:1100

netstat -a -n gives:
udp 0 0 192.168.0.1:1102 0.0.0.0:*

The (new) attribute AutoDiscoveryBindAddress in cluster-service.xml is set to:

<attribute name="AutoDiscoveryBindAddress">${jboss.bind.address}</attribute>

Changing this attribute has no impact.
A network trace shows the client sending a UDP packet "GET_ADDRESS:TestPartition" to 230.0.0.4:1102, but no server answeres.
Clustering is working fine. OS is RH EL 3.0.

Any hints are really appreciated.

Thanks,
Mathias.

[sinus]

This problem seems to correspond with bug report JBAS-1843:
http://jira.jboss.com/jira/browse/JBAS-1843

Mathias

[starksm64]

Yes. What does the multicast routing table look like for the 192.168.0.1 interface?

[sinus]

Multicasting was ok and the server answered to a "ping 230.0.0.4".
The problem persisted on SuSE Linux 9.1 kernel 2.6.4-52-default, but not on WinXP.

I had a look in the source org.jboss.ha.jndi.DetachedHANamingService.java, class AutomaticDiscovery, method start() and found two issues.
The original code:

stopping = false;
 // Use the jndi bind address if there is no discovery address
 if (discoveryBindAddress != null)
 discoveryBindAddress = bindAddress;
 InetSocketAddress bindAddr = new InetSocketAddress(discoveryBindAddress,
 adGroupPort);
 socket = new MulticastSocket(bindAddr);
 socket.setTimeToLive(autoDiscoveryTTL);
 group = InetAddress.getByName(adGroupAddress);
 socket.joinGroup(group);

First, discoveryBindAddress should likely be tested as equal to null. It makes no sense to set it to bindAddr if one already exists.
Second, there seems to be a bug in the constructor of MulticastSocket when instantiated with an InetAddress. I changed the code to first instantiate with a port number and then set the interface in a second call.
Changed code:

stopping = false;
 // Use the jndi bind address if there is no discovery address
 if (discoveryBindAddress == null)
 discoveryBindAddress = bindAddress;
 socket = new MulticastSocket(adGroupPort);
 socket.setInterface(discoveryBindAddress);
 socket.setTimeToLive(autoDiscoveryTTL);
 group = InetAddress.getByName(adGroupAddress);
 socket.joinGroup(group);

This code works on the Linux server. I had not the time to test other platforms. Please check this out.

Mathias