Re: ISSUE (3639) Which policy alternative was selected?

Daniel

Thanks for taking the time to write an explanation,  which makes  
sense to me when every message visibly uses MTOM.

The problem which I was referring to, and which  my bug report  
referred to, is the following:

1. Even when MTOM is engaged it does not always make sense to use an  
MTOM attachment on every message.

For example, a SOAP message may only require a simple amount of XML  
within the SOAP body element, so nothing will be serialized for that  
message using MTOM, i.e. there will be no multipart message on the wire.

2. However, in this message exchange pattern some messages will make  
sense to use MTOM and serialize some material as an attachment.

3. The first message in the exchange may be what is mentioned in #1,  
thus the first requestor message may not appear to use MTOM

4. Thus I argue that the requestor and provider need to determine  
whether MTOM is in use by a means other than looking at the first  
message in this case.

5. Thus in this case, optional policy should not be used with MTOM,  
since the policy alternative should require MTOM, yet not every  
message will visibly demonstrate MTOM - the alternative chosen  
indicates MTOM MUST be engaged, yet this doesn't mean every message  
will have a serialized attachment.

6. Thus I suggest changing the optional examples to something other  
than MTOM, where examination of the first message is not used to  
determine compliance.

Does this set of statements make sense? I think my key statement is  
#5, and the key question is that when MTOM is engaged, you aren't  
always going to see a multipart on every message, are you?

regards, Frederick

Frederick Hirsch
Nokia


On Nov 2, 2006, at 1:25 PM, ext Daniel Roth wrote:

> Hi Frederick,
>
> If an assertion is marked optional then the provider will not fault  
> if the corresponding behavior is not engaged.  The MTOM assertion  
> defines its behavior as:
>
> "/wsoma:OptimizedMimeSerialization
> A policy assertion that specifies that MTOM [MTOM] MUST be used in  
> messages sent to the Web service.  It also specifies that responses  
> from the Web service MUST be optimized using MTOM, i.e. that the  
> messages must be sent using the application/xop+xml mime type."
>
> So, if you send a request using MTOM you MUST also send a response  
> using MTOM.  If the request sends a message that does not use MTOM,  
> the provider should respond with a message that does not use MTOM.
>
> The provider can tell from the message if MTOM is in use, so I  
> think it's fine to use wsp:Optional with the MTOM assertion.
>
> I hope this helps.
>
> Daniel Roth
>
> -----Original Message-----
> From: public-ws-policy-request@w3.org [mailto:public-ws-policy- 
> request@w3.org] On Behalf Of Frederick Hirsch
> Sent: Wednesday, November 01, 2006 5:52 AM
> To: ext Sergey Beryozkin
> Cc: Frederick Hirsch; Yalcinalp, Umit; Sanka Samaranyake; Fabian  
> Ritzmann; Ashok Malhotra; public-ws-policy@w3.org
> Subject: Re: ISSUE (3639) Which policy alternative was selected?
>
>
> Can someone please explain to me what it means to have an optional
> MTOM assertion on a single endpoint, thus appropriate to all messages?
>
> The MTOM assertion Asir referenced says that the assertion means that
> all requests and responses MUST use MTOM.
>
> Thus if this assertion is optional on a given endpoint, and the
> requestor sends a request that does not use (or need) MTOM for that
> request, how does the provider know whether or not to use MTOM on the
> response, assuming it would be appropriate when using MTOM?
> (requestor retrieved provider policy and chose consistent alternative
> to use, but no explicit agreement with provider exists)
>
> I would think that for a case like this the endpoint should not offer
> a policy alternative to this assertion. Is this a potential
> guideline? Is this what Daniel said before?
>
> regards, Frederick
>
> Frederick Hirsch
> Nokia
>
>
> On Oct 31, 2006, at 12:42 PM, ext Sergey Beryozkin wrote:
>
>>
>> Hi all,
>>
>> Can you please clarify a couple of things...
>> I don't understand why a service provider needs to know which
>> policy alternative was selected.
>> I feel what Daniel suggested about attaching policies to different
>> endpoints should be an adequate approach for most cases, because
>> this is what endpoints are for in the first place.
>> If two alternatives can be meaninfully/usefully created in a scope
>> of a single endpoint then it's fine (security related
>> alternatives, etc), this is also probably useful when nested
>> polices are used.
>>
>> I've read the proposal for optional assertions and the suggestion
>> to mark with wsp:optional something a provider will not always do
>> and to enable optional behaviours out-of-band, perhaps by letting a
>> provider know which policy alternative was selected.
>> I'm confused. For example :
>>
>> <custom:MTOM wsp:optional/>
>>
>> it says custom:foo is something a provider won't always do and it
>> will do only if it's told somehow by a requester to do it.
>>
>> So here we go :
>>
>> <Policy>
>>   <ExactlyOnce>
>>   <!-- Alt1 -->
>>   <All>
>>     <custom:bar/>
>>   </All>
>>   <!-- Alt2 -->
>>   <All>
>>     <custom:MTOM/>
>>     <custom:bar/>
>>   </All>
>>   </ExactlyOnce>
>> </Policy>
>>
>> By selecting the second alternative and by the virtue of sending
>> MTOM-serialized messages the requester is saying to a provider that
>> it needs to load its <custom:MTOM/> handler. I don't see a reason
>> for some more out-of-band indication that alternative2 was selected.
>>
>> In what cases one would want to explicitly tell the provider which
>> alternative was selected ?
>>
>> Thanks, Sergey
>>
>>
>>
>>
>>
>> I agree. We should definitely acknowledge that this is a problem
>> even if
>> we may choose not to tackle it in an interoperable way in this
>> version.
>>
>> Please see that the proposal for optional assertions that I sent a
>> while
>> back also contains some language for out of band mechanisms in
>> order to
>> make messages self describing (which includes the option of an
>> additional protocol).
>>
>> --umit
>>
>>
>>> -----Original Message-----
>>> From: public-ws-policy-request@w3.org
>>> [mailto:public-ws-policy-request@w3.org] On Behalf Of Sanka
>>> Samaranyake
>>> Sent: Tuesday, Oct 10, 2006 12:55 PM
>>> To: Fabian Ritzmann
>>> Cc: Ashok Malhotra; public-ws-policy@w3.org
>>> Subject: Re: ISSUE (3639) Which policy alternative was selected?
>>>
>>>
>>> -----BEGIN PGP SIGNED MESSAGE-----
>>> Hash: SHA1
>>>
>>> Fabian Ritzmann wrote:
>>>>
>>>> Ashok Malhotra wrote:
>>>>> My original motivation in raising this issue was to provide a
>>>>> rationale for why we
>>>>> wanted a pointer from the message to the policy (alternative) that
>>>>> was applied to it.  (There are
>>>>> other reasons why such a pointer may be useful, for example if the
>>>>> policy changes
>>>>> during the course of a long-running transaction, or to indicate
>>>>> policies or assertions which do not affect the wire format of
>>>>> messages - bug 3789.)
>>>>>
>>>>> At the f2f in Bellevue the WG said:
>>>>> 1. You can add such a pointer to a msg using the SOAP
>>> extensibility
>>>>> mechanism but
>>>>> 2. The WG did not want to standardize such a header as it raised
>>>>> all manner of questions such "shd this
>>>>> be the first header."
>>>>>
>>>>> Subsequently, Dan Roth told the WG
>>>>>
>>> http://lists.w3.org/Archives/Public/public-ws-policy/2006Oct/
>>> 0043.html
>>>>> that Microsoft products used the following solutions:
>>>>> 1. In case multiple alternatives apply, create an endpoint that
>>>>> supports exactly one policy alternative.
>>>>> 2. Use an out-of-band mechanism to convey which alternative was
>>>>> selected.  In effect, this uses an out-of-band
>>>>> mechanism instead of the pointer in the message that I wanted.
>>>>>
>>>>> So, at this point I am willing to agree to close the issue with no
>>>>> action.
>>>>> If others feel differently please propose the solution you
>>> would like.
>>>>>
>>>>
>>>> Regarding the suggested solutions:
>>>>
>>>> 1. Web services are meant to be interoperable. I don't think that
>>>> any product-specific solution can satisfy that requirement.
>>>>
>>>> 2. Irrespectively of whether an in-band or out-of-band mechanism is
>>>> chosen, it would be beneficial to have standardized solutions.
>>>>
>>>> Having said that, I believe that designing a solution would take
>>>> more time and effort than would be good for our schedule. I'd like
>>>> to see this issue deferred to V.next.
>>>>
>>>> Fabian
>>>>
>>>
>>> I believe as policies are get used in more and more real life
>>> scenarios, it is likely that we might end up with policy  
>>> alternatives
>>> that are not distinguish able just looking at the message itself.  
>>> And
>>> I feel that, irrespectively of whether it is in-band or out-of-band,
>>> there should be way to indicate to the server, which alternative the
>>> client has picked. I am ok if with differing this  to the next
>>> version
>>> as far as  it  is remains noted.
>>>
>>> Sanka
>>>
>>> - --
>>> Sanka Samaranayake
>>> WSO2 Inc.
>>> T:+94-77-3506382 F:+94-11-2424304
>>>
>>> http://sankas.blogspot.com/
>>> http://www.wso2.net/
>>> -----BEGIN PGP SIGNATURE-----
>>> Version: GnuPG v1.4.2.1 (GNU/Linux)
>>>
>>> iD8DBQFFK/qJ/Hd0ETKdgNIRAmtvAJ9A//SEe+02heUIpqj6ba35k0ypzgCdE6M5
>>> GiWeAJJim2cGpqHoqPe3U7A=
>>> =VrOI
>>> -----END PGP SIGNATURE-----
>>>
>>>
>>>
>>>
>>
>>
>
>

Received on Monday, 6 November 2006 23:23:57 UTC