- From: Francis McCabe <fgm@fla.fujitsu.com>
- Date: Thu, 12 Feb 2004 15:24:17 -0800
- To: "Burdett, David" <david.burdett@commerceone.com>
- Cc: "'Steve Ross-Talbot'" <steve@enigmatec.net>, WS Choreography <public-ws-chor@w3.org>
Pub-sub is an interesting case as it is possible to demonstrate that it represents a corner case -- it is not possible to capture pub-sub with MEPs but is possible with choreography (assuming that semantics is part of choreography). Here is why: A given publication service will notify its clients of new events. The service may choose to package up all events for a given client into a single message. A service may give additional events depending on the semantics of the subscription request. A client may choose to register for multiple events, perhaps specified using a keyword notation (not important) -- due to the imprecise nature of this, a given subscription request may overlap with an older subscription request. The result is that it is not possible to identify a given event message as being part of a given subscription. However, all events have been requested in a subscription. I believe that without a semantic connection between event messages and subscription messages you cannot capture this scenario. MEPs are defined to be without semantics. Frank On Feb 12, 2004, at 10:22 AM, Burdett, David wrote: > Steve > > Giving this a bit of thought makes me think that the essence of > pub-sub is ... > > 1. The publisher sets up a service that accepts subscriptions requests > and changes for some other service that actually publishes > documents/messages. > > 2. The subscriber then requests subscription to a service by sending a > messge which then gets either accepted or rejected. > > 3. When the subscriber makes the request, they must include some kind > of "delivery address" that identifies where documents/messages etc, > generated by the publisher must be sent > > 4. If the request is accepted, then the publisher will return some > kind of "identifier" for the subscription that can later be used when > changing or cancelling the subscription > > 5. The publisher starts publishing documents. This is a one-way > message although it might be delivered reliably > 6. The publisher continues publishing documents until: a) the > subscription is cancelled, or b) the subscription runs-out, e.g. a > certain period of time has passed, a specific number of > messages/documents have been received, the subscriber hasn't paid. > > From a CDL perspective, the "delivery address" is what the Overview > Model calls a "Channel". This means that to use it, we need to have a > way of representing the Channel in XML and decie how it should be > included in the message. My thoughts would be the body. > > The rest sounds to me like a pretty regular Choreography Definition > with dependencies, e.g. you can't cancel a subscription unless you > managed to subscribe to it successfully. > > Another question is should such a Pub-Sub Choreography be standardized > as I am sure the need for Pub-Sub goes beyond WS Chor. For example you > could imagine a definition that allowed you to manage a subscription > to any web service then later cancel it. However you would need > standard XML docs to be used as Message Content for the Interactions > in the Pub Sub. > > Also where should such a spec be developed ... by WSDL, by WS Chor? > I'm not sure I know the answer to that one. > > David > > > > > -----Original Message----- > From: Steve Ross-Talbot [mailto:steve@enigmatec.net] > Sent: Thursday, February 12, 2004 5:41 AM > To: WS Choreography > Subject: Fwd: WSDL and pub/sub > > > > > How does this leave our stuff wrt the Barros usecase? > > Cheers > > Steve T > > > > Begin forwarded message: > > > From: "Jonathan Marsh" <jmarsh@microsoft.com> > > Date: 11 February 2004 20:44:41 GMT > > To: "Steve Ross-Talbot" <steve@enigmatec.net> > > Subject: RE: WSDL and pub/sub > > > > WSDL 2.0 part 3 [1] describes some message exchange patterns that > can > > be > > used as part of a pub/sub solution. Look at all the "out-*" patterns > > starting at section 3.4. A complete pub/sub solution is not > provided, > > as the address and mechanics of "sub"ing and providing the address > for > > the "pub" to be delivered are not standardized in WSDL (perhaps this > is > > an orchestration problem?). Note that the HTTP and SOAP bindings > don't > > support these message exchange patterns yet, but we have an issue > open > > on whether we should rectify this. > > > > Hope this helps. > > > > [1] > > http://dev.w3.org/cvsweb/~checkout~/2002/ws/desc/wsdl20/wsdl20- > > patterns. > > html#out-only > > > >> -----Original Message----- > >> From: Steve Ross-Talbot [mailto:steve@enigmatec.net] > >> Sent: Tuesday, February 10, 2004 10:07 AM > >> To: Jonathan Marsh > >> Subject: WSDL and pub/sub > >> > >> Jonathan, > >> > >> I seem to recollect that you indiciated that WSDL2.0 includes an MEP > > or > >> some such facility to represent pub/sub as a means of communication. > >> Could you verify this? And could you point me to the appropriate > >> description? This is something that the Choreography WG would very > > much > >> like to have so that a single message could be sent to multiple > >> sources without needing to bind to those sources. > >> > >> > >> Best Regards > >> > >> Steve Ross-Talbot > >> co-Chair W3C Web Services Choreography > >> > >> O: +44 207 397 8207 > >> C: +44 7855 268 848 > >> www.enigmatec.net
Received on Thursday, 12 February 2004 18:24:30 UTC