Re: Server Response Policy

[Sergey Beryozkin]

Hi

I reckon it would be useful if it were possible to explicitly state in the CXF 
WS-Policy feature that a given policy
expression is inbound/outbound only.
Perhaps another option is to provide reusable policy interceptors which could simply (and blindly) assert some policies using 
preconfigured custom policy namespaces (ex : get an AssertionInfo matching a configured policy namespace and assert it), such that 
such interceptors can be reused in cases whereby it does not make much sense to enforce some policies on the inbound or outbound 
chain...

Cheers, Sergey

Well I can definitely work around the problem, just by configuring my  interceptors to check off the same assertions on the 
outbound side.   This requires though, that I stick a reference to some request state  information on the exchange, and that I 
ensure that an OutInterceptor  is installed before between the policy interceptors on the outbound  side.

My solution seems a little hokey to me -- I'd rather that the  AssertionInfo objects I've already ticked off as asserted be the 
same  instances on the response channel, but that's not the case (I have not  looked yet to see where a copy of the 
AssertionInfoMap is made --  maybe that's the real issue here, as opposed to the same effective  policy on the inbound and 
outbound sides).

-Fred

On Aug 8, 2008, at 11:55 AM, Daniel Kulp wrote:

Hmm..  I'm not really sure, but this MAY be correct behavior.

I'm thinking something like the MTOM policy.   For that, there has  to be
something on the outgoing side to see if the MTOM policy is there  and turn on
MTOM if so.

It's somewhat like some policies are "enforced" on the incoming chain,
and "configured" on the outgoing chains.   Thus, the policy should  be able to
handle both sides of that.

Dan



On Thursday 07 August 2008 3:25:31 pm Fred Dushin wrote:
I'm having trouble with the CXF policy framework, which is causing a
little bit of grief.  I think this is a developer, as opposed to a
user issue, as I /think/ it points to a bug in the policy framework.
If not, I can migrate the conversation to the users list.

What I'm finding is that a CXF server that has policy defined at the
endpoint (I'm using the CXF policy feature, whereby you can reference
a WS-Policy expression in Spring) is effectively treating the
operative policy in the server on the outbound interceptor chain the
same way as it does on the inbound interceptor chain.

Let me be more concrete.  Assume the following config fragment in
Spring:

    <jaxws:endpoint ... >
        <jaxws:features>
            <cxfp:policies>
                <wsp:PolicyReference URI="#MyPolicy"/>
            </cxfp:policies>
        </jaxws:features>
    </jaxws:endpoint>
    ...
    <wsp:Policy wsu:Id="MyPolicy">
        ...
    </wsp:Policy>

So on the inbound server side, I definitely get this as the effective
policy, and I can do policy enforcement by setting the asserted flag
on various assertions selected (in the case of inbound server,
everything is selected).  Note that I have interceptors installed in
the runtime to do this enforcement.

However, on the outbound response, I'm getting the same effective
policy, and I can see how it's getting set -- precisely in the
following stacktrace (pardon the wrap):

Thread [btpool2-1] (Suspended)

InitializingPolicyEngine
(PolicyEngineImpl).getEndpointPolicy(EndpointInfo, EndpointPolicy,
boolean, Assertor) line: 220

InitializingPolicyEngine
(PolicyEngineImpl).getServerEndpointPolicy(EndpointInfo, Destination)
line: 214
EffectivePolicyImpl.initialisePolicy(EndpointInfo,
BindingOperationInfo, PolicyEngineImpl, boolean) line: 101
EffectivePolicyImpl.initialise(EndpointInfo, BindingOperationInfo,
PolicyEngineImpl, Assertor, boolean) line: 79

InitializingPolicyEngine
(PolicyEngineImpl).getEffectiveServerResponsePolicy(EndpointInfo,
BindingOperationInfo, Destination) line: 171
ServerPolicyOutInterceptor.handle(Message) line: 77

ServerPolicyOutInterceptor
(AbstractPolicyInterceptor).handleMessage(Message) line: 56
PhaseInterceptorChain.doIntercept(Message) line: 221
OutgoingChainInterceptor.handleMessage(Message) line: 74
PhaseInterceptorChain.doIntercept(Message) line: 221
ChainInitiationObserver.onMessage(Message) line: 78
JettyHTTPDestination.serviceRequest(ServletContext,
HttpServletRequest, HttpServletResponse) line: 278
JettyHTTPDestination.doService(ServletContext, HttpServletRequest,
HttpServletResponse) line: 252
JettyHTTPHandler.handle(String, HttpServletRequest,
HttpServletResponse, int) line: 70
ContextHandler.handle(String, HttpServletRequest,
HttpServletResponse, int) line: 726
ContextHandlerCollection.handle(String, HttpServletRequest,
HttpServletResponse, int) line: 206
Server(HandlerWrapper).handle(String, HttpServletRequest,
HttpServletResponse, int) line: 152
Server.handle(HttpConnection) line: 324
HttpConnection.handleRequest() line: 505
HttpConnection$RequestHandler.content(Buffer) line: 842
HttpParser.parseNext() line: 730
HttpParser.parseAvailable() line: 205
HttpConnection.handle() line: 380
SslSocketConnector$SslConnection(SocketConnector$Connection).run()
line: 228
SslSocketConnector$SslConnection.run() line: 620
BoundedThreadPool$PoolThread.run() line: 450

But in this case, there are no interceptors installed in the response
interceptor chain.  As a consequence, the request "fails" (in a  manner
described below) with a "None of the policy alternatives can be
satisfied" message, which is set in the
PolicyVerificationOutInterceptor.

Note, however, that the SOAP:Fault is actually embedded in the  response:

<soap:Envelope xmlns:soap="http://schemas.xmlsoap.org/soap/ envelope/">
    <soap:Body>
         <sayHiResponse xmlns="http://acme.com/greeter/types";>
            <responseType>Bonjour tony</responseType>
         </sayHiResponse>
         <soap:Fault>
             <faultcode>soap:Server</faultcode>
             <faultstring>None of the policy alternatives can be
satisfied.</faultstring>
         </soap:Fault>
    </soap:Body>
</soap:Envelope>

Oops.  Something seems awry with the phase in which the fault is
written.

But regardless, should the effective policy on the response be the
same as the effective policy on the request?  Or should policy
assertion implementors code their interceptors to handle the response
chain, as well as the request?

-Fred



--
Daniel Kulp
[EMAIL PROTECTED]
http://www.dankulp.com/blog


----------------------------
IONA Technologies PLC (registered in Ireland)
Registered Number: 171387
Registered Address: The IONA Building, Shelbourne Road, Dublin 4, Ireland