Urgent: please help - Object Hash Code of RMI stubs in a cluster

Question

Hello all,
I'm trying to find out for a week now if members of a cluster can get out of synch
regarding "Object Hash Code" of RMI stubs in the JNDI.
After binding an RMI object to one server (no rmic performed since the object
is in the classpath of both servers and the service is pinned), the JNDI mapping
looked identical on both servers.
However, while one server went down for a few hours, the other server has changed
the "Object Hash Code" to a new one (the binding name stayed constant). When the
server went up again it tried to lookup the object under the same binding, then
found it but failed to execute remote methods on it. An error message says that
the object was garbage collected:
java.rmi.NoSuchObjectException: Unable to dispatch request to Remote object with
id: '347'. The object has been garbage collected. at weblogic.rjvm.RJVMImpl.dispatchRequest(RJVMImpl.java:766)
at weblogic.rjvm.RJVMImpl.dispatch(RJVMImpl.java:738) at weblogic.rjvm.ConnectionManagerServer.handleRJVM(ConnectionManagerServer.java:207)
at weblogic.rjvm.ConnectionManager.dispatch(ConnectionManager.java:777) at weblogic.rjvm.t3.T3JVMConnection.dispatch(T3JVMConnection.java:508)
at weblogic.socket.NTSocketMuxer.processSockets(NTSocketMuxer.java:664) at weblogic.socket.SocketReaderRequest.execute(SocketReaderRequest.java:23)
at weblogic.kernel.ExecuteThread.execute(ExecuteThread.java:213) at weblogic.kernel.ExecuteThread.run(ExecuteThread.java:189)
Why isn't the new JNDI mapping replicated to the new booting server? Should I
use specific rmic flags? How can I maintain the JNDI trees identical on all clustered
servers?
I'm using two managed WLS 7.02 servers running on win2000 servers.
Dani

Answer 1

Hi Andy,
Thank you for responding.
The binding code looks like that:
m_ctx = new InitialContext();
int index = getNextAvailableRepositoryJNDIIndex();
m_ctx.bind(
NODE_PREFIX+index,
this
where this is the Remote implementation.
How will I make the object hash code constant or have all views remain constant
when the GC changes its reference?
Dani
Andy Piper <[email protected]> wrote:
"Daniel Gordon" <[email protected]> writes:
I'm trying to find out for a week now if members of a cluster can getout of synch
regarding "Object Hash Code" of RMI stubs in the JNDI.The hash code can probably change since the stub represents an
aggregate view of the cluster. But this may be a red-herring. What
does your code look like that binds the rmi object? It may be that DGC
is biting you.
andy