Applets fail to deserailize messages in a timely manner after a network failure triggers a reconnect using the FailoverTransport
In ClassLoadingAwareObjectInputStream.java the load() method makes a call to Class.forName().
For the primitive types (int, boolean, etc.) this would result in a call similar to:
Class.forName("int", false, loader); //Where loader is the Applet2ClassLoader
Since Applet2ClassLoader is a URLClassLoader and "int.class" is not in the jar cache it pulled down from the server at the start of application, it is going to try and go to the server to resolve this class.
In the event of a network failure, this will result in the ClassLoader having to wait for the socket timeout. (see stacktrace at link) Once this socket timeout occurs, the load() method then attempts to lookup the class in the primitive HashMap that is statically initialized. This returns the class for the int and the deserialization continues on.
At first it seemed like the messages were failing to be received but it turned out they were just taking a very long time to be deserialized. This problem can be avoided by changing the order in which ClassLoadingAwareObjectInputStream tries to resolve the class.
Here is the change I made to the load() method: http://activemq.2283324.n4.nabble.com/Applet-Class-Loader-Problems-td4671835.html