Cay, great summary. A few comments on "Nested exception is..." section. When a system exceptin (e.g. NPE in your case) occurred in EJB, the container is required to log it and rethrow a EJBException. This is the nature of any distributed computing. Users really need to look at both client side errors and server side logs. From the client side error (e.g., RemoteException, ServerException, etc), users can know the root cause is at the server side. I'd suggest appclient container print out something like: "please refer to server.log for server-side errors..." I feel the same pain every time I see the long stacktrace from appclient container for any errors. They are not well ordered. Normally people would expect the outer-most exceptions first, and the inner-most exceptions last. But from the appclent exceptions I've got, it's hard to tell which one wraps which. It seems all these appclient exceptions are chained like this, from outer-most to inner-most: 1. java.lang.RuntimeException 2. java.lang.reflect.InvocationTargetException 3. javax.ejb.EJBException 4. java.rmi.ServerException 5. java.rmi.RemoteException A general suggestion on logging wrapper exceptions like InvocationTargetException and PrivilegedActionException. IMHO, It adds no value to log them. Just logging their getCause() should be sufficient. Why would a user care if you are using reflection, or doPriv block? These exceptions only expose implementation details, and clutter logs. For the same reason, I wouldn't rethrow a InvocationTargetException and PrivilegedActionException to the client (caller) code. But I'm less sure about this since appserver is very complex software. Just my 2 cents.

Posted by Cheng Fang on July 08, 2006 at 02:46 PM PDT #