HA-JNDI not working as expected

[michael.daleiden]

I have two machines clustered using JBoss 3.2.2RC1. One is Windoze 2000 and one is HP/UX. Clustering and farming are working fine (nodes see each other, the partition manager reports the existence of both members, and modules are correctly farmed between the two nodes).

My problem is that the HA-JNDI does not appear to be working as expected. The following code creates the initial context for the HA-JNDI:

Properties jndiProps = new Properties();
jndiProps.put(Context.INITIAL_CONTEXT_FACTORY, "org.jnp.interfaces.NamingContextFactory");
jndiProps.put(Context.URL_PKG_PREFIXES, "org.jboss.naming:org.jnp.interfaces");
jndiProps.put(Context.PROVIDER_URL, "");
jndiProps.put("jnp.partitionName", "DefaultPartition");

Then, I have code that binds new objects into the HA-JNDI tree during server startup (custom MBeans perform this binding operation on startup), using the node name as part of the name for the object:

String hostName = InetAddress.getLocalHost().getHostName();
String key = "jmx/device/" + hostName + "/xxxxxx";
Util.rebind(jndiContext, key, obj);

No exceptions or errors are reported by this code when binding objects to the HA-JNDI tree from both nodes, so I surmised that the objects were successfully bound.

In my application, I then use the following code to list the contents of the HA-JNDI tree (using the same context as described above):

for (NamingEnumeration enumObj = jndiContext.list("jmx/device"); enumObj.hasMore();) {
 NameClassPair ncHandle = (NameClassPair)enumObj.next();
 System.out.println("Found context: 'jmx/device/' + ncHandle.getName() + "'...");
}

Depending on which server the app is run from (winnode or hpnode), the above code shows only the bindings for that server. For example, if I run the app from winnode, the output shows as follows:

"Found context: 'jmx/device/winnode'..."

If run from hpnode, it shows:

"Found context: 'jmx/device/hpnode'..."

It should show the following, since both nodes are up and running and objects have been registered on both nodes using the HA-JNDI context:

"Found context: 'jmx/device/winnode'..."
"Found context: 'jmx/device/hpnode'..."

So, I'm a bit confused as to why the objects bound into HA-JNDI by one server are not being seen by the other server when the HA-JNDI tree is queried.

Any ideas?

[michael.daleiden]

Well, I found part of the problem. It appears that you must be careful with JNDI context names. The local JNDI already contains a top-level context named jmx, which confuses the HA-JNDI logic. The logic first scans the HA-JNDI for the context and, when it does not initially find it, it performs a lookupLocally, which finds the local JNDI context. From that point on, it binds the remaining subcontexts to the local JNDI context, instead of the HA-JNDI context.

When I changed the naming of the HA-JNDI context so that it uses a unique top-level context, HA-JNDI appears to create the appropriate subcontexts in the HA-JNDI tree, as well as bind the object to the lowest-level subcontext. However, the list function still only shows the objects that are bound from the local server.

I turned up the logging for the cluster components and observed that the correct calls (HAJNDI._rebind()) were being propagated across the cluster to the various HA-JNDI instances in order to replicate the bindings, but as I noted, the bindings do not seem to be replicated at all.

I further tried to reduce the number of variables by clustering two Windows machines that are on the same subnet (to eliminate any potential multicast propagation and firewall issues), but received the exact same result.

I am at a complete loss to understand what the problem is. It is URGENT that I figure out how to resolve this issue ASAP, as we are gearing up for a pilot customer launch and I need to have clustering working properly.

[slaboure]

can you please post your description on sf.net as a bug and assign it to slaboure.

sacha

[michael.daleiden]

I have not done this (posted a bug) before. What do I need to do?