RE: too many extension points (was Re: [Bug 6692] New: Remove Mode from the specification)

Gil,

My basic problem with pushing the mode concept up to the header is that since mode is a fundamental concept associated with Eventing and is at least an internal state associated with eventing implementations, then its manipulation should be part of the message body, not part of the message header.  If it is not, then where will it end?  If you wanted, you could put all the subscribe elements, attributes, etc in the soap header and not have any body at all.

While that is clearly silly, it does beg the discussion on where the line should actually be drawn.  You are coming from one side which says that mode is an extension, thus you should use the extension mechanism already defined in SOAP (using mustUnderstand=1) and an extension element in the body.  That makes sense from one perspective.

But you can look at setting the mode attribute as setting a specific eventing value, much in the same way as NotifyTo sets the value of the EPR to send events to.  Why is setting the mode any less setting a value than setting NotifyTo, or Filter, or ... ?  Why not put NotifyTo in the header?  Just because mode is a flag and you can thus implement it in a header using a neat mustUnderstand concept, does not mean that is the right place to put it.

First tell me why the value associated with mode should be set in the header and not in the message body.

--Geoff


From: public-ws-resource-access-request@w3.org [mailto:public-ws-resource-access-request@w3.org] On Behalf Of Gilbert Pilz
Sent: Thursday, April 09, 2009 12:33 PM
To: David Snelling
Cc: Chou, Wu (Wu); public-ws-resource-access@w3.org; Doug Davis; Bob Freund; Li, Li (Li)
Subject: Re: too many extension points (was Re: [Bug 6692] New: Remove Mode from the specification)

Another place where @Mode is inferior is in the dynamic support of extensions. For example, suppose I am an Event Sink that, for whatever reasons, would prefer to acknowledge Notifications but would still be willing to create a subscription even if the Event Source didn't support acks (you can substitute any flavor of extension here). Using @Mode you would have to:

1.) send a Subscribe request with the "PushWithAcks" mode
2.) handle the resulting "mode not understood" fault
3.) send a Subscribe request w/out the "PushWithAcks" mode

You have to do this because of @Mode's "fault on non-understood" semantics. Using standard, XML extension mechanisms and our new, "ignore extensions you don't understand" model, you could define an extension to wse:Subscribe that meant "PushWithAcks" and a corresponding extension to wse:SubscribeResponse that meant "AcceptPushWithAcks". Now all you have to do to support the optional use of acks is to

1.) send a Subscribe request with the "PushWithAcks" extension
2.) look at the SubscribeResponse - if it doesn't contain a "AcceptPushWithAcks" extension then the Event Source  didn't understand the request to accept/process acks for Notifications

Same result as above with one less round trip. Of course, as has been mentioned numerous times before, if the Event Sink wanted the "don't create the subscription if you don't understand this extension" behavior, it could include a "UsingPushWithAcks" SOAP Header with @mustUnderstand="1".

- gp

David Snelling wrote:
Gil,

Thanks for this discussion. It also brought something back to my mind that has worried me about Mode. If in the end we keep it, it will need to be enhanced to provide even the minimal required support. Reasoning follows, very incomplete at this stage, but would form the basis of a new issue should we retain Mode.

So assuming Mode stays, we should encourage (at least SHOLD) ALL extensions to the semantics of Eventing to use the same mechanism. As soon as this logical step takes place, meeting the simple queuing use case becomes problematic not to mention more interesting cases. Where do we put queuing limits etc? The Mode attribute is only a URI. As Gil points out there are a number of places to stuff extension information, but to support Mode properly, we would need a <wse:PutModeExtensionDataHere/> element so people could use the extension points consistently.

I don't want this can of worms opened at all, as the specification starts to become very complicated (folks from the WS-Notification WG will remember painful discussions when this kind of bloat was discussed there). Eventing is a really good simple specification with clean semantics. Retaining Mode in (a clean and functional manner) will probably change all that.

Please let's get rid of Mode.


On 03 Apr 2009, at 21:31, Gilbert Pilz wrote:


Let's look at the schema for the wse:Subscribe message and it's included DeliveryType element (I've taken the liberty of highlighting the existing extension points - you need to view this as HTML to see what I'm talking about):

<xs:element name="Subscribe">

  <xs:complexType>

    <xs:sequence>

      <xs:element name="EndTo" type="wsa:EndpointReferenceType"



                  minOccurs="0" />

      <xs:element name="Delivery" type="tns:DeliveryType" />

      <xs:element name="Expires" type="tns:ExpirationType"



                  minOccurs="0" />

      <xs:element name="Filter" type="tns:FilterType"



                  minOccurs="0" />

      <xs:any namespace="##other" processContents="lax"



              minOccurs="0" maxOccurs="unbounded" />

    </xs:sequence>

    <xs:anyAttribute namespace="##other" processContents="lax" />

  </xs:complexType>

</xs:element>



<xs:complexType name="DeliveryType" mixed="true">

  <xs:sequence>

    <xs:element ref="wse:NotifyTo" minOccurs="0" maxOccurs="1" />

    <xs:any namespace="##any" processContents="lax"

            minOccurs="0" maxOccurs="unbounded" />

  </xs:sequence>

  <xs:attribute name="Mode" type="xs:anyURI" use="optional"

                default="http://schemas.xmlsoap.org/ws/2004/08/eventing/DeliveryModes/Push" />

  <xs:anyAttribute namespace="##other" processContents="lax" />

</xs:complexType>


If I was a developer interested in extending WS-Eventing, I would find this situation confusing. Ignoring the fact that the wse:NotifyTo element is an EPR and can itself be extended, there are five (5) different ways I can extend the wse:Subscribe message (or, if you count element/attribute extension of the same element as a "single way", there are three (3)). What makes this even more confusing is that not all of these extension points have the same processing rules. The XML extension points (in black-bold) are subject to the following, recently accepted, language:

The elements defined in this specification MAY be extended at the points

indicated by their outlines and schema. Implementations MAY add child elements

and/or attributes at the indicated extension points but MUST NOT contradict the

semantics of the parent and/or owner, respectively. If a receiver does not

recognize an extension, the receiver SHOULD ignore that extension. Senders MAY

indicate the presence of an extension that has to be understood through the use

of a corresponding SOAP Header with a soap:mustUnderstand attribute with the

value "1".


Whereas the WS-Eventing specific extension point (in blue-bold) is subject to the following:
If the event source does not support the requested delivery mode, the request MUST fail, and the event source MAY generate a wse:DeliveryModeRequestedUnavailable fault indicating that the requested delivery mode is not supported.
I think the common reaction to this situation would be "WTF<http://thedailywtf.com/Default.aspx>?" This needs to be simplified so that, if you want to extend WS-Eventing, you aren't required to navigate through a maze of different extension points that may or may not result in different processing behavior (i.e. "What is the difference between extending /wse:Subscibe/wse:Delivery or just /wse:Subscribe?").

Clearly the "Mode" attribute, with its conflicting semantics, should be removed. In addition to this, we need to remove the "Delivery" element and it's extension points (obviously we would move the wse:NotifyTo element up into the wse:Subscribe element). What we would be left with would look like this:

<xs:element name="Subscribe">

  <xs:complexType>

    <xs:sequence>

      <xs:element ref="wse:NotifyTo" minOccurs="0" maxOccurs="1" />

      <xs:element name="EndTo" type="wsa:EndpointReferenceType"



                  minOccurs="0" />

      <xs:element name="Expires" type="tns:ExpirationType"



                  minOccurs="0" />

      <xs:element name="Filter" type="tns:FilterType"



                  minOccurs="0" />

      <xs:any namespace="##other" processContents="lax"



              minOccurs="0" maxOccurs="unbounded" />

    </xs:sequence>

    <xs:anyAttribute namespace="##other" processContents="lax" />

  </xs:complexType>

</xs:element>
What you have now is really two ways to extend the wse:Subscribe request. You can extend the wse:Subscribe message (through either an attribute or an element) if you want an extension that is specific to WS-Eventing, or you can extend the NotifyTo EPR if you want an extension that operates at the WS-Addressing level (i.e. across any technology that uses WS-Addressing).

- gp

Chou, Wu (Wu) wrote:
Doug,

Mode attribute in WS-E is to specify the additional required event delivery behavior or required application/business logic extensions that the event source must understand/support before being allowed to accept the event subscription.

My understanding is: "push_event_thru_proxy" is for all events delivery from the Subscribe. As such, it covers SubscriptionEnd event. The Mode attribute also allows to specify fine grained application logic to control the event delivery.  If you want, more fine grained event delivery logic/procedure can be easily added to Mode attribute to control the event delivery, e.g.  "push_thru_proxy_for_subscriptionend_event_only", "push_thru_proxy_for_non_subscriptionend_event", etc.

Mode attribute in WS-E provides a standard extension point to add application specific logic and behavior to event delivery, so that it can enforce the required procedure/logic for business applications when using WS-Eventing for event delivery.

Thanks,

- Wu Chou.

>Wu,
 > since its too easy for things to get lost in these long messages, let me
focus on one particular aspect of this.

>Surely the idea of using a proxy
for sending messages is not WS-Eventing specific.  How do you support
sending any other asynchronous message (non-NotifyTo messages) thru a
proxy?
>Need a clarification on th, e.g. a general question, a question for WS-E or a question for WS-E plus Mode.

>For example, the SubscriptionEnd message.

thanks
-Doug


Wu Chou, IEEE Fellow, Ph.D. | Director |Dialogue System Research | AVAYA | 233 Mt. Airy Road| Rm. 2D48 | Basking Ridge, NJ 07920 | Voice/Fax: 908-696-5198 / 908-696-5401 | wuchou@avaya.com<blocked::mailto:wuchou@avaya.com>


Take care:

    Dr. David Snelling < David . Snelling . UK . Fujitsu . com >
    Fujitsu Laboratories of Europe Limited
    Hayes Park Central
    Hayes End Road
    Hayes, Middlesex  UB4 8FE
    Reg. No. 4153469

    +44-7590-293439 (Mobile)






______________________________________________________________________

Fujitsu Laboratories of Europe Limited
Hayes Park Central, Hayes End Road, Hayes, Middlesex, UB4 8FE
Registered No. 4153469

This e-mail and any attachments are for the sole use of addressee(s) and
may contain information which is privileged and confidential. Unauthorised
use or copying for disclosure is strictly prohibited. The fact that this
e-mail has been scanned by Trendmicro Interscan and McAfee Groupshield does
not guarantee that it has not been intercepted or amended nor that it is
virus-free.