[eclipselink-users] Cache Coordination in Glassfish

[wintermute@xxxxxxxxxxxxxxxxxx]

Hi list,

it's been quite a while -- some of you may remember I was trying to  
get EclipseLink cache coordination to work in Glassfish.

The good news is: When I got back to this, I did it. I'll attach the  
piece of code that does the trick.
The bad news: There's one more issue.

Most of my changes propagate to the cluster, but as soon as a  
particular entity is involved, I see exceptions in the log. The only  
thing special about the entity is that it uses an enum constant  
defined in an Interface elswhere, but certainly in the reach of the  
application classloader. Here's the exception:

MergeChangeSetCommand command failed due to: Local Exception Stack:
Exception [EclipseLink-12002] (Eclipse Persistence Services - 1.0.2  
(Build 20081024)):  
org.eclipse.persistence.exceptions.CommunicationException
Exception Description: Unable to propagate changes to Service 
[BookpacCacheCoordination, 8f4d39c0-d1bb-4530-8c54-a5b3542d68ff,  
rmi://192.168.23.21:38686].
Internal Exception: java.lang.ClassNotFoundException:  
com.bookpac.archive.core.IDocumentList$ListType
        at  
org.eclipse.persistence.exceptions.CommunicationException.unableToPropag 
ateChanges(CommunicationException.java:68)
        at  
org.eclipse.persistence.sessions.coordination.MergeChangeSetCommand.getC 
hangeSet(MergeChangeSetCommand.java:47)
[...]
        at java.lang.Thread.run(Thread.java:619)
Caused by: java.lang.ClassNotFoundException:  
com.bookpac.archive.core.IDocumentList$ListType
        at com.sun.appserv.server.util.ASURLClassLoader.loadClass 
(ASURLClassLoader.java:129)
[...]


ListType is an enum, defined in IDocumentList. It's there, and the  
class is in a jar known to persistence.xml, in fact, in the same jar  
as the other entities that do get propagated.

Is there a general problem with enums? Or do they need to be defined  
in the JPA entity that uses them?

Maybe someone knows something about this Classloading thing with Enums?

Thanks in advance,

rv


----

This is the code I'm using:

	public void customize(Session session) throws Exception {
						
		RemoteCommandManager rcm = new RemoteCommandManager 
((CommandProcessor)session);

		rcm.getDiscoveryManager().setMulticastGroupAddress("226.1.2.3"); // 
$NON-NLS-1$
		rcm.getDiscoveryManager().setMulticastPort(3122);
		rcm.getDiscoveryManager().setAnnouncementDelay(1000*5);

		rcm.setChannel("CacheCoordination"); //$NON-NLS-1$
		rcm.setShouldPropagateAsynchronously(true);

		rcm.getTransportManager().setNamingServiceType 
(TransportManager.REGISTRY_NAMING_SERVICE);
		rcm.getTransportManager().setUserName("admin"); //$NON-NLS-1$
		rcm.getTransportManager().setPassword("adminadmin"); //$NON-NLS-1$
		rcm.setUrl("rmi://$HOST:38686"); //$NON-NLS-1$

		rcm.setServerPlatform(session.getServerPlatform());
		((DatabaseSession)session).setCommandManager(rcm);
		((DatabaseSession)session).setShouldPropagateChanges(true);
		rcm.initialize();
	}

This is the customize Method of an EclipseLink session customizer,  
obviously.
The important bit is the rcm.setUrl(...) line -- 38686 is the port  
you want to be using for RMI in Glassfish clusters.

One thing that hat confused me was the announcement delay: I thougt  
announcement would be done periodically, but it's not. Nodes announce  
excactly once by default, when they start up. So there's a chance of  
nodes missing each other if one of them is starting up slowly and  
doesn't hear the other's announcements. The announcement delay should  
be larger than the maximum difference between startup times of the  
nodes -- unfortunately, there's no way of knowing this. To make sure  
I don't end up with a slow node that doesn't have its cache updated  
without me noticing in production, I added some more announcements.