JSP+RMI = socket leak (with tomcat 5)

Question

Tried this in the RMI forum, no answer, so let's try here...
When I use RMI from a JSP servlet, I have a "socket leak" problem. When I create an RMI unicast remote object, the RMI library (of course) opens up a server/listening socket. In a normal application, I can unbind the remote object, but then the next object I create will re-use the same server socket, so the server socket never closes, but only one is ever created, so it's not a big deal.
Now for the problem: When I use the Tomcat manager to unload my webapp, the socket is still there. Sure, fine. But then when I reload the webapp, instead of re-using the same server socket, a new one is created! Furthermore, if I try to use explicit port numbers to force the server socket to be reused, I get "port already in use" errors. It seems that somehow tomcat won't let RMI recycle its entry points in between webapp restarts. Does anybody know how I can either:
a) Get RMI to re-use the same ports
b) Get RMI to completely shut down its ports when the webapp is unloaded (this would be even better)
Thanks. It's irritating to see 10 open but unused server sockets on my system if I have to restart the webapp.

Answer 1

Ah, I left out something: I construct the RMI object a servlet context listener in the contextInitialized() call, and I call UnicastRemoteObject.unexportObject() in the contextDestroyed() call. I added log messages, I am definitely correctly unbinding the object. The problem isn't that it is failing to be unbound - the problem is that the RMI system is refusing to reuse the same port after it is unbound. After unbinding, the port is still open, but it is never reused, each time I re-instantiate the remote object, it is making a new server socket and leaving the old one sitting open.
As for why I'm doing this, it is a bit like a homegrown EJB, but just so small that this was the simplest way. I have a largish non-web based application, then I have a few JSP pages that do database queries to provide information to browsers. The JSP pages are not at all central to the application though, and just show some database info. The JSP system caches some SQL data, and I use RMI to tell JSP when to flush cached data due to database changes. It's all very simple, and it works fantastic, except for the issue that every time I reload the webapp I leave these server sockets hanging open. Even that is just an irritation, I don't reload much except when I'm testing, but it's a big enough irritation that I would really like to fix it.