cxf-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Dan Diephouse" <...@envoisolutions.com>
Subject Re: When should we close the handlers in CXF?
Date Thu, 18 Jan 2007 13:55:38 GMT
I think the registering of actions to be run at the end of the chain is
good.

Another possibility is to add a close(Message) method to the Interceptor
which gets called at the end of the chain. If we did that I would think we
might want to get rid of the handleFault method as cleanup could be done in
close().  (Eoghan - I'm actually suggesting we move closer to the JAX-WS
APIs! ;-))

Thoughts?

- Dan

On 1/18/07, Glynn, Eoghan <eoghan.glynn@iona.com> wrote:
>
>
>
>
> > -----Original Message-----
> > From: Unreal Jiang [mailto:sinbad_jia@yahoo.com]
> > Sent: 18 January 2007 11:44
> > To: cxf-dev@incubator.apache.org
> > Subject: RE: When should we close the handlers in CXF?
> >
> > Hi Eoghan,
> >   I think those two approach are work fine.
> >
> >   The first approach is only for handlers process,
> >   The second approach can do some clean-up works not only for
> > handlers  but interceptors,  but if we use runnable object
> > for  TerminalAction,  the order of handlers or interceptors
> > will be  hard to ensure.
>
> In most cases, a FIFO ordering of the terminal actions would be fine,
> i.e. the order in which the terminal actions are executed would reflect
> the order in which the interceptors were traversed. If you needed more
> fine-grained control, maybe InterceptorChain.addTerminalAction() could
> be replaced by something like ...
>
> interface InterceptorChain {
>     List<Runnbale> getTerminalActions();
> }
>
> ... so that the code submitting the terminal action could control
> ordering with respect to previously submitted terminal actions. Going
> even further that this (e.g. something akin to the
> PhaseInterceptor.getBefore/After() business) would I think be overkill
> without a specific ordering-sensitive usecase.
>
> However, a closer look at the JAX-WS HandlerChainInvoker code suggests
> that only one of the LogicalHandlerInterceptor and
> SOAPHandlerInterceptor will actually need to submit a terminal action.
> So with only a *single* terminal action concerned with closing handlers,
> ordering shouldn't be an issue here.
>
> This is because the same HandlerChainInvoker instance is shared by these
> two interceptors, and the code that calls handleMessage/Fault() on the
> individual Handlers also adds each of these to a separate list
> (closeHandlers) of handlers for which close() should be called.
>
> Only a single call to HandlerChainInvoker.mepComplete() is then actually
> required to ensure that *all* the traversed handlers are close()d in the
> correct order.
>
> So maybe the simplest approach would be be to submit the terminal action
> in AbstractJAXWSHandlerInterceptor.getInvoker(), i.e.:
>
>     protected HandlerChainInvoker getInvoker(final T message) {
>         HandlerChainInvoker invoker =
>             message.getExchange().get(HandlerChainInvoker.class);
>         if (null == invoker) {
>             invoker = new HandlerChainInvoker(binding.getHandlerChain(),
>                                               isOutbound(message));
>             message.getExchange().put(HandlerChainInvoker.class,
> invoker);
>
>             // submit a *single* terminal action for entire handler
> chain
>             message.getInterceptorChain().addTerminalAction(new
> Runnable() {
>                 public void run() {
>                     mepComplete(message);
>                 }
>             }
>         }
>         //...
>     }
>
> Cheers,
> Eoghan
>
> >   So I incline to  the second approach, but we should use
> > some other way to instead of runnable object.
> >
> >   Regards
> >   Unreal
> >
> > "Liu, Jervis" <jliu@iona.com> wrote:I  would vote for the
> > second approach. When its there, we can probably use  the
> > similiar approach to remove the sub-chain (interceptor chain
> > reentrance) wherever it is possible.
> >
> > ________________________________
> >
> > From: Glynn, Eoghan [mailto:eoghan.glynn@iona.com]
> > Sent: Wed 1/17/2007 9:42 PM
> > To: cxf-dev@incubator.apache.org
> > Subject: RE: When should we close the handlers in CXF?
> >
> >
> >
> >
> >
> > > -----Original Message-----
> > > From: Glynn, Eoghan [mailto:eoghan.glynn@iona.com]
> > > Sent: 17 January 2007 12:34
> > > To: cxf-dev@incubator.apache.org
> > > Subject: RE: When should we close the handlers in CXF?
> > >
> > >
> > > Hi Unreal,
> > >
> > > One point to note is that all the other JAX-WS Handler touch points
> > > are driven through a set of interceptors, each wrapping a
> > chain of a
> > > particular Handler type (logical, protocol etc.).
> > >
> > > So for example the SOAPHanderInterceptor takes care of calling the
> > > handleMessage/Fault() methods of any SOAPHandlers in the chain.
> > > Similarly there's a *separate* LogicalHandlerInterceptor that
> > > traverses the chain of LogicalHandlers. I'm guessing you
> > already know
> > > all this ...
> > >
> > > But the point is that it would be a good idea to maintain
> > the pattern
> > > of wrapper-interceptor calling out to JAX-WS Handler, and
> > obviously it
> > > would be badness to for example put this JAXWS-specific
> > logic into the
> > > ClientImpl code.
> > >
> > > However, because Handler.close() should only be called at the very
> > > *end* of the interceptor chain tarversal, and because we currently
> > > have
> > > *multiple* interceptors wrapping the JAX-WS Handler chains
> > of various
> > > types, the close() call should not be made from within the existing
> > > wrapper interceptors. Otherwise we'd end up with for
> > example close()
> > > called prematurely on the SOAPHandlers *before* the LogicalHandlers
> > > have even been traversed (inbound on the client-side).
> > >
> > > So we'd need a *single* new wrapper interceptor, positioned
> > at the end
> > > of the in & fault-in interceptor chains, that's responsible for
> > > calling
> > > close() on all types of handler. This could be driven via a pattern
> > > similar to the
> > > LogicalHandlerInterceptor.onCompletion() method (e.g. the new
> > > interceptor walks back along the chain to find the
> > > LogicalHandlerInterceptor & SOAPHandlerInterceptor and calls
> > > onCompletion() on these).
> >
> > On second thoughts, maybe a cleaner may of doing this would
> > be allow an interceptor to register some sort of terminal
> > action with the InterceptorChain to be executed when the
> > chain traversal is complete, e.g.
> >
> > public interface InterceptorChain {
> >     void addTerminalAction(Runnable r);
> >
> >     //...
> > }
> >
> > Or alternatively take the Runnable as a return value from
> > Interceptor.handleMessage/Fault().
> >
> > Then in the InterceptorChain impl, run all the
> > TerminalAction(s) from a finally block, e.g.
> >
> > public class PhaseInterceptorChain {
> >    public boolean doIntercept(Message m) {
> >        try {
> >            while (interceptorIterator.hasNext()) {
> >                interceptorIterator.next().handleMessage(m);
> >            }
> >        } finally {
> >            for (Runnable r : terminalActions) {
> >                r.run();
> >            }
> >        }
> >    }
> > }
> >
> > Then for example the
> > LogicalHandlerInterceptor.handleMessage() would end with some
> > logic like:
> >
> >    if (isRequestor(message) && (isOneway(message) ||
> > !isOutbound(message))) {
> >       message.getInterceptorChain().addTerminalAction(new Runnable() {
> >           public void run() {
> >               getInvoker(message).mepComplete(message);
> >           }
> >       }
> >    }
> >
> > Similarly for SOAPHandlerInterceptor etc.
> >
> > Cheers,
> > Eoghan
> >
> >
> >
> >
> >
> > ---------------------------------
> > Cheap Talk? Check out Yahoo! Messenger's low PC-to-Phone call rates.
> >
>



-- 
Dan Diephouse
Envoi Solutions
http://envoisolutions.com | http://netzooid.com/blog

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message