RE: [ws-rx] PR issue 09

["Durand, Jacques R." <JDurand@us.fujitsu.com>]

Paul:

2 levels of concern here, one of them I think needs and can be addressed
by this TC:

(a)- a minimal set of DAs, their representation and how they are
communicated. We know this will be needed, whether specified in or out
of the TC. This would define QoS-level interoperability.

(b)- in case these DAs and/or their parameters are communicated via
extensibility points, support in the protocol to at least signal the
expected behavior of the other RM module - in particular if not
understood.

So (b) seems to be within reach, see inline <JD>.

Thanks,
Jacques



-----Original Message-----
From: Paul Fremantle [mailto:paul@wso2.com] 
Sent: Monday, October 30, 2006 11:04 PM
To: Durand, Jacques R.
Cc: BAYES_00
Subject: Re: [ws-rx] PR issue 09

Jacques

Let us take a concrete example. Suppose there was a third-party defined 
an extension to CreateSequence that specified InOrderExactlyOnce.

If the RMD specifies this, as we know from extensive discussion, there 
is no change to the wire protocol. Therefore there cannot be interop 
problems even if the client doesn't understand the extension.

<JD> agree.

If the specifier wants the RMS to be able to force this, then they need 
to specify the extension correctly. In general we do not have 
client-side policies, so the extension would need to be specified so 
that the RMD responded to say it had accepted that model. If the RMD 
didn't understand the extension, it won't respond.

<JD> works for messages received by RMD - note that we have extensible
elements going the other way too.

At one point we discussed making any extensions of CS mustUnderstand. 
Would that help allay the concerns of this group?

<JD> I remember that and the problem was something that looked like an
extension of the current mustUnderstand SOAP model in a way that
presented difficulties for several of us, (if I remember: interaction
with SOAP model, extensions may be not just elements but attributes,
etc.) Maybe that needs be revisited - On our side, we considered another
general alternative where a single element (e.g. called "mustSupport")
could be added to either body or headers, that contains a list of XPaths
on elements/attributes that must be supported (either in extensibility
points but not restricted to... also possibly targeting a subset of
their values), and return a specific fault if not. Powerful, though a
bit more costly to process.  I am willing to spend more time on this
problem with interested folks.
 

Paul

Durand, Jacques R. wrote:
> Paul:
>
> Let me reword one of the concerns I see in the eAC comment:
>
> - because DAs are not specified - although often expected, as our
> charter recognizes - some WS-RM implementers may be led to establish a
> proprietary way to do it that requires use of extensibility points in
> the RM protocol, for signaling these DAs or parameters of these DAs.
> This way could become a de-facto standard for a significant part of
the
> market (just an hypothesis, but quite possible).
>
> - if this happens, there would be interop issues with other RM
endpoints
> that may not support these extensions, when you look at the bigger
> picture (RM protocol + DAs) even though you may still have
> interoperability at RM protocol level alone, since an RMD/RMS may
ignore
> any extension. In case these other implementations want to align with
> the practice, besides the trouble in upgrading already deployed
> products, there might be unexpected IP issues.
>
> So the worst interoperability scenario for those who need to
communicate
> DA data, would be if sometime later this comes to affect
implementations
> of the lower protocol layer - RMS/RMD.
>
> Short of specifying DAs somewhere (and the way to communicate them) I
> see no good solution for preventing this scenario except getting rid
of
> extensibility points... 
>
> Jacques
>
>
>
> -----Original Message-----
> From: Paul Fremantle [mailto:paul@wso2.com] 
> Sent: Thursday, October 26, 2006 4:23 PM
> To: Durand, Jacques R.; ws-rx@lists.oasis-open.org
> Subject: Re: [ws-rx] PR issue 09
>
> Jacques
>
> I understand what you are saying. I'm trying to understand how we
would 
> prove something so obvious :-)
>
> Our protocol ensures that the messages are correctly transmitted to
the 
> RMD together with a message number, which increases by one.
>
> So even if we did define this in our specification, the text would
sound
>
> pretty lame:
>
> i.e.
> "To achieve the "ordered delivery" delivery assurance, the RMD must 
> deliver the messages in the order of the MessageNumbers."
> "To achieve the AtMostOnce delivery assurance, the RMD must only
deliver
>
> one message with a given MessageNumber"
> "To achieve the AtLeastOnce delivery assurance, the RMD must ensure
that
>
> it delivers each transmitted message with a given MessageNumber."
>
> As regards your second point, I have some sympathy for that.
>
> Paul
>
>
> Durand, Jacques R. wrote:
>   
>> I think the issue is not so much "how can I implement my DAs on top
of
>>     
>
>   
>> this protocol" . Many folks in eAC are quite experimented with RM and

>> have known sequence numbers way before WS-RX started.
>>
>> But without going as far as bringing back the DAs, at a minimum it 
>> would be helpful to demonstrate the following, either in the spec 
>> (appendix) or in a companion doc:
>>
>> - whatever DAs (among most popular ones) are defined on top of this 
>> protocol, and assuming both sides are aware of which DA is being used

>> (communicated out of band), then the protocol as defined is
sufficient
>>     
>
>   
>> to *enable* the DAs and does not need additional interoperability 
>> tightening or extensions when actual DAs are implemented. Were it 
>> otherwise, it would mean that proprietary extensions to the protocol 
>> are needed that would introduce both interop and IP issues.
>>
>> Now that probably isn't enough to make everyone happy. Standard DAs
to
>>     
>
>   
>> choose from, along with their parameters, are still expected from
some
>>     
>
>   
>> users and are considered as part of the interop equation. But
wherever
>>     
>
>   
>> these are defined - either wsrmp or elsewhere - it is important to 
>> show first that this has no bearing on the wsrm protocol layer and
its
>>     
>
>   
>> implementations, i.e. this layer can be considered stable.
>>
>> -Jacques
>>
>>     
>
>   

-- 
Paul Fremantle
VP/Technology and Partnerships, WSO2 
OASIS WS-RX TC Co-chair

http://bloglines.com/blog/paulfremantle
paul@wso2.com
(646) 290 8050

"Oxygenating the Web Service Platform", www.wso2.com