RE: Requirements: Decision Points Requirement Proposals

It looks like BPEL chose to base its correlation mechanism on message contents (BPEL's Correlation Sets based on Message Properties) instead of abstract instance IDs. It would be interesting to hear from the BPEL people on this list the rationale for their choice (which I could not find in the BPEL spec itself).
 
Ugo

-----Original Message-----
From: Burdett, David [mailto:david.burdett@commerceone.com]
Sent: Monday, June 09, 2003 11:08 AM
To: 'Jean-Jacques Dubray'; Burdett, David; 'Yaron Y. Goland'; 'WS Chor Public'
Subject: RE: Requirements: Decision Points Requirement Proposals


JJ
 
I agree with you that correlation is not a topic we have discussed. I, like you prefer a choreography instance id as it simplifies the choreography definition and is indpendent of the detail of the message content. However it does mean that there can then be inconsistencies between the choreography instance id and some id inside the message content e.g. an order no.
 
David

-----Original Message-----
From: Jean-Jacques Dubray [mailto:jjd@eigner.com]
Sent: Monday, June 09, 2003 2:22 AM
To: 'Burdett, David'; 'Yaron Y. Goland'; 'WS Chor Public'
Subject: RE: Requirements: Decision Points Requirement Proposals



"The WS-Chor choreography definition MUST provide mechanisms by which the execution of one choreography definition is dependent on the execution of the instance of some other choreography definition". The use case for this is where you want to execute a choreography that determines the current state of processing of some earlier choreography. The "query" choreography can only validly be executed if there is some earlier instance of the a choreography that can be referenced.

[JJ] I would treat the "query/admin" problem as an independent problem like you seem to suggest in the following points. However, I can see the advantages of your proposal above in Multi-party-collaborations. This will save us from complex correlation schemes.

This is another area of requirements that I think we did not touch: correlation. There are two possible approaches: an explicit correlation mechanism based on the document contents, and a choreography protocol which envelope contains a choreography instance ID. I personally favor the choreography protocol as it works in all cases and simplifies the choreography definition.  

The following couple of requirements are ones that have been discussed much earlier on the list however I am not sure that we really want to do them, at least not initially, but I do think they are worth discussing ...

"The WS Choreography specification MUST provide standardized, reusable choreography definitions that allow one role to determine another roles state of processing of a choreography instance, no matter what choreography definition was being followed."

"The WS Choreography specification MUST provide standardized, reusable choreography definitions that allow one role to request another role to restart the processing of a "stalled" choreography instance, no matter what choreography definition was being followed." This could simply be a request to resend some earlier message that got lost.

The rationale for both of these is that querying the status of a choreography and re-starting a choreography will be common requirements for many (but not all) choreographies and therefore having a standard way of doing these functions will make choreographies easier to design and develop.

As stated earlier, more comments inline below. 

David 

-----Original Message----- 
From: Yaron Y. Goland [ mailto:ygoland@bea.com] 
Sent: Wednesday, June 04, 2003 4:28 PM 
To: WS Chor Public 
Subject: Requirements: Decision Points Requirement Proposals 

 

I propose the following requirements be added to the requirements document: 

The WS-Chor choreography description format MUST provide mechanisms to 
enable a choreography to specify that a process in a particular role MUST 
send zero, one or more messages from a statically defined set of messages in 
parallel, serial or any combination of the two. 
<DB> A couple of comments: 
1. I think a role that MUST send zero messages doesn't work as if the role MUST send zero messages, then why is it in the choreography. 

2. Why do you use the term "description format" instead of the simpler "definition' because, aren't the properties you seek a characteristic of the definition rather than the format of the definition.

3. The first sentence is circular as it says ... "The WS-Chor choreography description format MUST enable a choreography ..." without specifying what a choreography is. 

4. I think you mean when you say a "statically defined set of messages" that the actual messages definitions that can be sent are finite in number and from a proscribed list. There has been a lot of discussion on the idea of variability in the detailed message content which means that limiting a choreography to specific message formats will inhibit choreography reuse. Instead I thinkt that we should refer to "Message Types" or "Message Families" rather than "messages".

5. This requirement is also very similar to the next so my alternative is described below 
</DB> 

The WS-Chor choreography description format MUST be able to describe 
decision points where a process in a particular role MAY send zero, one or 
more messages from a statically defined set of messages in parallel, serial 
or any combination of the two. 
<DB>So how about the following requirement that combines the previous two and takes into account the comments I made ...

"The WS-Chor choreography definition MUST provide mechanisms that define the sequence in which one or more messages types are exchanged between two or more roles either in parallel, serially or any combination of the two, together with the conditions that cause those messages to be sent."</DB>

The WS-Chor choreography description format MUST be able to describe who is 
to receive a message by referencing their role. 
<DB>I would add the sender to this definition to give ... "The WS Choreography definition MUST be able to describe who the sender of a message is and who the receiver should be by referencing their role." The rationale for this is that what you do with a message may well depend on the role of the sender ... assuming that the same message can be sent by different roles.</DB>

The WS-Chor choreography description format MUST make it possible to specify 
a role's binding to an actual web service instance either statically, when a 
web service using that choreography is deployed, or dynamically at run time. 

The WS-Chor choreography description format MUST provide mechanisms to allow 
messages to be sent to a particular member of a set of web services in the 
same role. 
[Ed Note: What I'm very inelegantly trying to capture is the idea that if 
you are running an auction service and you just found out that one of the 
bidders isn't qualified to bid you need a way to say things like "I'm now 
going to send out an unsolicited 'get lost you dead beat' message to one web 
service that is in the role of bidder." This could then trigger a whole set 
of messages back and forth between the auction service and the dead beat 
bidder, the choreography needs some way to capture the fact that you are 
still talking to the same member of the role group.] 
<DB>This example introduces the idea of a role group, which I don't *think* we need. If we take this use case, then you can actually consider it as an internal "business process" problem, for example:

The auctioneer has a business process that consists of a set of separate individual identical choreographies between the auctioneer and the bidder where each choreography instance would take the following form ...

AUCTIONEER       BIDDER 
Bid Invite -------> 
Either ... 
Get Lost ---------> 
... or ... 
<--------------- Bid 
... etc ... 
The fact that there are several bidders involved is something that only the auctioneer needs to be concerned of. 

This means that this is really a business process (e.g. BPEL ) problem rather than a choreography problem especially as the auctioneer is in complete control of what goes on. For example, the auctioneer could treat all the interactions as being part of one choreography by using the same identifier for the correlation of all the messages irrespective of the bidder. 

Now there may be a use case where you really do the need the variability, but I can't think of one. On the other hand, if we can avoid this variability, then it will simplify the specification we need to write significantly.

</DB> 

The WS-Chor choreography description format MUST NOT require that the logic 
used by a sender in a decision point to decide how to act be exposed in the 
choreography. 
<DB>There's a corollary, I think, that says something like ... "The WS-Chor choreography definition MUST enable the results of decisions made by one role that affect the behavior of another role to be communicated to the other role." This is really about the transmission of relevant state information between roles.</DB>

The WS-Chor choreography description format MUST enable the annotation of 
all actions with human readable descriptions. 
<DB>I agree but would go further and replace the last phrase with "... with clear semantic definitions." Something might be human readable but that does not mean it explains the purpose well.</DB>

The WS-Chor choreography description format MUST provide an abstract 
mechanism where by the logic used to make a decision at a decision point can 
be expressed through reference to a WSBPEL abstract or executable process or 
similar machine readable logic. 
<DB>I don't have an alternative definition, but this pre-supposes a binding to WSBPEL that we might (or might not) want to make unless and until we collectively (i.e. WSBPEL and WSCHOR) work out what the goals and relationships of each activity will be.</DB>

The WS-Chor choreography description format base specification MUST NOT 
specify bindings for the abstract mechanism used to reference machine 
readable logic, rather extension specifications on top of the base 
specification MUST be used. 
<DB>As a general comment, we could do with developing definitions of various terms, e.g. "decision point", "base specification" which although quite intuitive, could be open to miss-interpretation.</DB>

I would appreciate it if objections to these requirements were stated in the 
form of alternate proposals. It's easy to say why something is wrong, it's a 
lot harder to spend the time to specify what is right. 

                Yaron