choreography protocol

I don't have much time to follow the latest developments of WSDL and MEP
might have well solved that problem. In WSDL 1.1, an operation can only
be a request with an optional response and any number of faults. So if
you wanted to describe something like 
request ->          / request processed signal <- 
	              / response <- / 
response processed signal ->

you would need at least two operations. If would be really great if WSDL
1.2 could support such protocol natively via MEPs for instance?

Actually making the protocol abstract is a really good idea, provided
that we could also provide default message format, but if people feel
that they have to invent new formats for these signals they can.

You may not want to have the process engine really being involved in
managing these signals. For instance if you implement a 2 signal
protocol like BPSS (message structure valid, message processed), the
message structure validation should be done by the service interface and
not below by the process engine for instance. The same could be true for
the "message processed" signal, you could have the application sending
it to the service interface via a call back. That would offload the
process engine. This is more questionable as you could say that the
application should really not know anything about the outside world, or
"message processed" could mean that 3 applications get to see the
message, in which case the process engine is certainly much better
position to compute and return that signal.

A choreography layer independent of the process engine is a better
architecture. This simplifies the process definition at least by a
factor of 2. The process definition should be really dealing with
business messages (request / response) not signals. This layer acts as
both an inbound/outbound gatekeeper to make sure that messages are
exchanges as specified by the choreography definition. 

In any case, the choreography should not assume that there is a BPEL
process engine inside, it should work with any service including the
ones with BPEL inside.

I don't really think that we need a separate spec. As I said, I think I
built the case for which a choreography protocol is almost mandatory for
any choreography spec that we could up with, it should be part of the
spec, if we decide it is in scope. This protocol layer should be common
to all services. 

Jean-Jacques 
 
 

>>-----Original Message-----
>>From: Ugo Corda [mailto:UCorda@SeeBeyond.com]
>>Sent: Freitag, 20. Juni 2003 13:15
>>To: Jean-Jacques Dubray; public-ws-chor@w3.org
>>Subject: RE: BPSS_f2f_june03.ppt
>>
>>JJ,
>>
>>Let me see if I understand your concept of choreography protocol by
asking
>>some concrete questions.
>>
>>I imagine you are proposing to define such a protocol at the abstract
>>level (i.e. part of the choreography abstract language we have been
>>talking about). So at the time we need to develop an executable
process we
>>would have to map that abstract protocol to the specific executable
>>language/protocols we plan to use.
>>
>>In the case of executable BPEL, would the abstract choreography
protocol
>>map to specific BPEL messages (i.e. your choreography protocol would
have
>>corresponding parts in the  WSDL descriptions of related messages)? Or
>>would it map to other BPEL features other than message contents?
>>
>>If it maps to message contents and the WSDL interfaces happen to be
bound
>>to SOAP, would that choreography protocol be mapped to a SOAP module
(i.e.
>>SOAP headers)? Do you envision a new spec that defines a SOAP-based
>>choreography protocol?
>>
>>Thank you,
>>Ugo
>>
>>> -----Original Message-----
>>> From: Jean-Jacques Dubray [mailto:jjd@eigner.com]
>>> Sent: Friday, June 20, 2003 4:30 AM
>>> To: Ugo Corda; 'Jean-Jacques Dubray'; public-ws-chor@w3.org
>>> Subject: RE: BPSS_f2f_june03.ppt
>>>
>>>
>>> Ugo:
>>>
>>> Basically a choreography protocol is needed to ensure that
>>> each peer in
>>> the choreography has the same view of the state in which the
>>> choreography instance is. Imagine a situation, where you send me a
PO
>>> and I am not supposed to respond until the goods are shipped
>>> and I will
>>> respond by sending you an invoice. So you send me a PO and
>>> the RM tells
>>> you "I got it" (just like a fax). The next thing that should
>>> happen then
>>> is you receive the goods and later on an invoice. If there was a
human
>>> behind a fax machine, and the order was garbled he could call
>>> and figure
>>> out the right information. In this case, the sender things the
>>> choreography is going ok. The responder on the contrary
>>> thinks that the
>>> collaboration is terminated on an error. This is why you need a
>>> protocol: to tell you that no exception occurred, each party has the
>>> same view of the state of the choreography.
>>>
>>> If you take a "highly-connected" system that has several hundred /
>>> thousands participants (not all participating in the same
choreography
>>> instance but rather having 2 by 2 conversations). You cannot
>>> expect that
>>> every message that will be exchanged in this setting would be of
high
>>> quality (the structure may be old or wrong, the content may be
>>> incoherent making the processing of the message impossible, i.e. a
>>> response cannot be created because the message could not get into
the
>>> system that was supposed to create the response).
>>>
>>> At this point, you can say I don't need/want a protocol. That
>>> means that
>>> when a choreography is designed, the designers must account for
these
>>> possible (yet common) errors. They will create specific
>>> messages to say
>>> "could not process your orders, it contains errors", and make these
>>> messages part of the choreography.
>>>
>>> On the other hand if you had a protocol, you would have a standard
set
>>> of exceptions (common to all message exchanges) and
>>> materialized with a
>>> set of standard messages. You could then express the
>>> choreography paths
>>> based on these error conditions (if success ... if failure ... if
>>> timeout ... if structure invalid ... if content invalid ...) with an
>>> implicit message exchange. The simplest set of exception for me are:
>>> message did not get there on time, message could not be processed by
>>> system/service of record.
>>>
>>> Again, all this has nothing to do with RM. The problem here
>>> is not that
>>> your message did not get to the recipient, it is rather that the
>>> recipient got a message that he could not process, hence
interrupting
>>> the choreography instance. A protocol would help you cover
>>> 80/90% of the
>>> common exceptions. Others can be dealt with at the design level.
>>>
>>> Cheers,
>>>
>>> Jean-Jacques Dubray____________________
>>> Chief Architect
>>> Eigner  Precision Lifecycle Management
>>> 200 Fifth Avenue
>>> Waltham, MA 02451
>>> 781-472-6317
>>> jjd@eigner.com
>>> www.eigner.com
>>>
>>>
>>>
>>> >>-----Original Message-----
>>> >>From: Ugo Corda [mailto:UCorda@SeeBeyond.com]
>>> >>Sent: Donnerstag, 19. Juni 2003 22:14
>>> >>To: Jean-Jacques Dubray; public-ws-chor@w3.org
>>> >>Subject: RE: BPSS_f2f_june03.ppt
>>> >>
>>> >>Jean-Jacques,
>>> >>
>>> >>I did not have a chance to listen to your presentation, so you
might
>>> have
>>> >>already explained this. In your slides you talk about a
>>> "choreography
>>> >>protocol", and I am not sure whether it is just regular
>>> messages plus
>>> an
>>> >>instance ID (as you mentioned in a previous message to the
>>> list) or it
>>> is
>>> >>more than that.
>>> >>
>>> >>Thank you,
>>> >>Ugo
>>> >>
>>> >>> -----Original Message-----
>>> >>> From: Martin Chapman [mailto:martin.chapman@oracle.com]
>>> >>> Sent: Thursday, June 19, 2003 12:18 PM
>>> >>> To: public-ws-chor@w3.org
>>> >>> Subject: FW: BPSS_f2f_june03.ppt
>>> >>>
>>> >>>
>>> >>>
>>> >>>
>>> >>> -----Original Message-----
>>> >>> From: Jean-Jacques Dubray [mailto:jjd@eigner.com]
>>> >>> Sent: Thursday, June 19, 2003 10:23 AM
>>> >>> To: 'Martin Chapman'; 'Steve Ross-Talbot';
>>> Daniel_Austin@grainger.com
>>> >>> Subject: BPSS_f2f_june03.ppt
>>> >>>
>>> >>>
>>> >>> Martin et al:
>>> >>>
>>> >>> This is my presentation for this afternoon. Please let me
>>> >>> know what time
>>> >>> and which number to call.
>>> >>>
>>> >>> Best regards,
>>> >>>
>>> >>> JJ-
>>> >>> 781-472-6317
>>> >>>
>>>
>>>

Received on Friday, 20 June 2003 14:38:41 UTC