The test
sun/rmi/transport/tcp/DeadCachedConnection.java
gets an unused port, starts a registory (in a vm), ping the registry, kills it then restarts it.
When this test is run on a mac test machine (in the Mach 5 mesos pool), the ports are not release soon enough for the restart and this fail every time with a port already in use error.
A small sleep between killing and restarting the registry (I arbitrarily put in a 5 second sleep) allows this to pass.
        // Kill and restart the registry
        System.err.println("Killing registry...");
        DeadCachedConnection.killRegistry();
        System.err.println("sleeping 5 seconds...");
        Thread.sleep (5000);
        System.err.println("Restarting registry...");
        DeadCachedConnection.makeRegistry(regport);
This test passes on mac in mach5 which does not use Mesos, but since we're heading that way for resource sharing, I think this would be a good idea to do now.