Message10817

re the thread dump:

Note that it's deadlocking here in PyType.fromClass:

    public static synchronized PyType fromClass(Class<?> c, boolean hardRef) {
        if (class_to_type == null) {
            class_to_type = new MapMaker().weakKeys().weakValues().makeMap();
            addFromClass(PyType.class, null);
        }

with respect to ending a call to a method (initChannel) defined in a Java superclass (ChannelInitializer), which is running in a Netty thread pool.

So I don't think this is a problem in this patch per se, just something that is stressing things in regrtest.

I don't like that singleton setup at all, we should be using the "Initialization On Demand Holder" idiom instead (http://www.cs.umd.edu/~pugh/java/memoryModel/jsr-133-faq.html#dcl). Most likely that's the root cause here.