Premature Garbage Collection of Remote Objects

Question

I have a problem with distributed garbage collection and RMI. In CSPoker we try to support RMI for client-server communication. The client looks up the server through the registry, calls the login method and receives back a Remote Object reference called ServerContext. The client can register Remote even listener Objects with the server.
In both directions we occasionally see:
java.rmi.NoSuchObjectException: no such object in table
The behaviour is completely unpredictable. Some developers never see it, others see it all the time. Even on the same machine, sometimes it failes immediately
and sometimes it fails after a while. One person running the Sun JRE 6 on Windows has this poblem most frequently.
Can this be a bug or do we misinterpret something? The way we see it, as long as the client leases the Remote Object it shouldn't be GCed. All tests are run on localhost so no network partition can occur.
The problem would probably be solved if we hold static references to all Remote Objects in the local JVM. This is not an option because we want the Objects to be GCed once the lease expires.
Here are 2 debugging outputs of server runs with RMI logging enabled:
http://cspoker.pastebin.com/f27a7d6fc
As you can see, the lease requests are coming through but a NoSuchObjectException is thrown and the Object is finalized.
http://cspoker.pastebin.com/f732797d2

Answer 1

Thanks for the quick reply.
For instance:
11:56:02,427 DEBUG (org.cspoker.common.api.shared.context.ForwardingRemoteServerContext.java:72) [finalize] - Garbage collecting old context: org.cspoker.server.rmi.export.ExportingServerContext@15a8767
07.11.2008 11:56:02 sun.rmi.server.UnicastServerRef logCall
FEINER: RMI TCP Connection(2)-10.0.4.106: [10.0.4.106: sun.rmi.transport.DGCImpl[0:0:0, 2]: java.rmi.dgc.Lease dirty(java.rmi.server.ObjID[], long, java.rmi.dgc.Lease)]
07.11.2008 11:56:03 sun.rmi.transport.Transport serviceCall
FEIN: RMI TCP Connection(2)-10.0.4.106: [10.0.4.106] exception:
java.rmi.NoSuchObjectException: no such object in table
        at sun.rmi.transport.Transport.serviceCall(Transport.java:129)
        at sun.rmi.transport.tcp.TCPTransport.handleMessages(TCPTransport.java:466)
        at sun.rmi.transport.tcp.TCPTransport$ConnectionHandler.run(TCPTransport.java:707)
        at java.lang.Thread.run(Thread.java:595)The context Object is exported and returned to the client here :
return (ExternalRemoteServerContext) UnicastRemoteObject.exportObject(context, 0);Then [the object|http://code.google.com/p/cspoker/source/browse/trunk/server/rmi/src/main/java/org/cspoker/server/rmi/export/ExportingServerContext.java?r=1023] that was just exported is GCed.
11:56:02,427 DEBUG (org.cspoker.common.api.shared.context.ForwardingRemoteServerContext.java:72) [finalize] - Garbage collecting old context: org.cspoker.server.rmi.export.ExportingServerContext@15a8767Then a lease request comes in:
FEINER: RMI TCP Connection(2)-10.0.4.106: [10.0.4.106: sun.rmi.transport.DGCImpl[0:0:0, 2]: java.rmi.dgc.Lease dirty(java.rmi.server.ObjID[], long, java.rmi.dgc.Lease)]Then the client calls a method on the remote object and the NoSuchObjectException is thrown.
The behaviour is non-deterministic in any case but yes, on the Linux JVM I'm not seeing the problem, somebody on Windows gets it all the time.