Colocation and local optimization - JBAS-57

[adrian.brock]

This relates to the following JIRA task:
http://jira.jboss.com/jira/browse/JBAS-57

Since the colocation check in HAJRMP isn't redudant, this needs
fixing in two ways:

a) It needs applying to the other HA invokers
b) It needs moving into the interceptor chain where the policy can be overridden when
required.

Design:
1) Move the colocation hashmap into a separate static class maintained by
ProxyFactoryHA
2) Rather than creating a whole new set of InvokerInterceptors that understand
the colocation map, I propose to update the default InvokerInterceptor.isLocal()
to also check the colocation map, not just the GUID.
This will be done in a protected method so that subclasses may override the
behaviour as required.

[starksm64]

If there is going to be a check for an entry in the collocation map, then effectively the GUID becomes irrelevant. Other cleanup than could be done is removal of the unused remoteInvoker instance variable.

[adrian.brock]

Only if the non HA invoker deployment also maintained the colocation info.

[starksm64]

Ok, I thought you were indicating that the base InvokerInterceptor was going to be checking for existence in the collocation map which would require the non-ha proxy factories to maintain this info. Its a less intrusive change if the GUID continues to be a marker for non-ha collocation and only ha proxy factories maintain the explicit collocation mapping. This would require one new HAInvokerInterceptor as far as I can see.

[adrian.brock]

Actually implementing this, there is no need for a Colocation object.
This object already exists in the form of org.jboss.system.Registry
(it should needs adding to the client jar where it will be empty on a client).

Also, looking at org.jboss.proxy.ejb.ProxyFactory
This looks wrong. It creates and binds the proxy in start(), but doesn't remove it
until destroy(). I can see circumstances where it or a subclass/dependency
needs some service that is no longer there.
Anything in done start() should be reversed in stop()!

[starksm64]

So are you ready to pull the commit action on this?