[Design of JBoss Web Services] - Re: WS-RM Prototype Ready for Preview
by richard.opalka@jboss.com
"thomas.diesler(a)jboss.com" wrote :
|
| On the contrary, if the RM Sender is coupled with the client proxy or the RM Receiver is coupled with the Endpoint, it would in my opinion defeat the purpose of RM. In other words, if the RM Receiver is only available if the Endpoint is available, the client might as well talk to the Endpoint directly.
|
| Using your implementation, would it (eventually) be possible to deploy the RM Sender/Receiver as separate HA components (i.e. on dedicated boxes)?
Hi Thomas,
you're absolutely right. The fact that both RM Sender / Receiver are coupled with client proxy/endpoint defeat the purpose of RM.
The reason I decided to provide the initial prototype this way was to support WS-Security as well.
If we will want to support WS-Security and decouple both RM Sender / Receiver we need to move WS-Security one level lower (that is to move it from handler to transport marshallers).
Other option is to forget about WS-RM + WS-Security combination and to decouple the RM Sender/Receiver.
View the original post : http://www.jboss.com/index.html?module=bb&op=viewtopic&p=4114969#4114969
Reply to the post : http://www.jboss.com/index.html?module=bb&op=posting&mode=reply&p=4114969
16 years, 4 months
[Design of JBoss Web Services] - Re: WS-RM Prototype Ready for Preview
by thomas.diesler@jboss.com
Hi Richard,
while working with the RM implementation I was wondering about the Quality of Service (QoS) aspect of WS-RM. In my understanding the RM Sender/Receiver must be high available (HA) components such that the Client can (always) communicate with the RM Sender and likewise the RM Sender can always communicate with the RM Receiver.
| Client <--> RM Sender <-------> RM Receiver <--> Endpoint
| QoS QoS
|
On the contrary, if the RM Sender is coupled with the client proxy or the RM Receiver is coupled with the Endpoint, it would in my opinion defeat the purpose of RM. In other words, if the RM Receiver is only available if the Endpoint is available, the client might as well talk to the Endpoint directly.
Using your implementation, would it (eventually) be possible to deploy the RM Sender/Receiver as separate HA components (i.e. on dedicated boxes)?
View the original post : http://www.jboss.com/index.html?module=bb&op=viewtopic&p=4114954#4114954
Reply to the post : http://www.jboss.com/index.html?module=bb&op=posting&mode=reply&p=4114954
16 years, 4 months
[Design of POJO Server] - Re: ProfileServiceUnitTestCase value/type misuse
by alesj
"scott.stark(a)jboss.org" wrote : There was the question of primative types vs SimpleMetaType and whether the ManagedProperty should be accepting primatives and autoboxing these into SimpleValues. I think we should look into doing this.
|
| The question also came up of enforcing a check against the ManagedProperty.getMetaType().isValue(Object) for the values passed into ManagedProperty.setValue(Object). Could be done with aop, but I don't think we can assume jboss aop is available/used in the admin client. It could also be done via a wrapping proxy that imposed this behavior, as well as the autoboxing.
|
Looking at this piece of horrible code - which I added :-( - we really must clear up the confusion about this values/types.
| // TODO - decide what is the type of value when we set it
| // since currently get always returns MetaValue, but we see some cases
| // where plain Serializable is enough
| Serializable serializable;
| MetaValue metaValue;
| Object value = prop.getValue();
| if (value instanceof MetaValue)
| {
| metaValue = (MetaValue)value;
| MetaType metaType = metaValue.getMetaType();
| if (metaType.isSimple())
| serializable = ((SimpleValue)metaValue).getValue();
| else if (metaType.isGeneric())
| serializable = ((GenericValue)metaValue).getValue();
| else
| serializable = null;
| }
| else
| {
| serializable = (Serializable)value;
| metaValue = MetaValueFactory.getInstance().create(value);
| }
|
| if(serializable != null)
| ctxProp.setValue(serializable);
| // todo - should this also dispatch to runtime component?
| Object componentName = getComponentName(ctxProp);
| if (componentName != null)
| dispatcher.set(componentName, ctxProp.getName(), metaValue);
|
I would expect a ManagedProperty to take MetaValue (so that it can also take non-serializable 'composite' values) and a runtime dispatcher to take plain object value.
View the original post : http://www.jboss.com/index.html?module=bb&op=viewtopic&p=4114940#4114940
Reply to the post : http://www.jboss.com/index.html?module=bb&op=posting&mode=reply&p=4114940
16 years, 4 months
[Design of EJB 3.0] - New Project to Validate External API
by ALRubinger
With the addition of the EJB3 External API, we've clearly-defined a contract for Bean Providers such that this artifact and the JavaEE EJB3 API should suffice for compile-time constraints in developing EJB3 on JBoss.
However, we currently do not have tests to ensure that the API is complete, that all required libraries for the compile-time classpath are in ext-api.
It'd be nice if we could move the current EJB3 Testsuite and have it not depend on core, but it tests internals (rightfully so!).
So, I propose the addition of a new, small test project that contains definitions of arbitrary EJBs, using each of the classes in ext-api. If this project compiles, ext-api is valid and complete. Any time we introduce a dependency outside, the test project won't compile.
Thoughts, or more simple ways of validating ext-api?
S,
ALR
View the original post : http://www.jboss.com/index.html?module=bb&op=viewtopic&p=4114919#4114919
Reply to the post : http://www.jboss.com/index.html?module=bb&op=posting&mode=reply&p=4114919
16 years, 4 months