[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]
Subject: RE: [ebxml-bp] ActionItem 26 conitnued Nested Collaboration Multiparty, Roles in HaugenFletcher multiparty
dwm@ The jpeg sequence diagram and synchronization points seem OK to me. But is there a way to pull the bpss out of sdr files to view the bpss instance? My question was really how we "interpolate" RequestResponse BTs within a RequestResponse BT? This is a typical "coordination" problem, similar to having subtransactions or subcoherences. One approach is to have BPSS notation indicate nondeterminism on the AND - fork/join control flow, and let it go at that. This approach is hard to explain to implementers who want to use it to monitor flows for conformance, it seems to me. Another is to split the interpolated BT and sandwich in BTs as needed, marking the synchronization points as needed. This seems to give up on reusability, but would have the dependencies marked (we could add the xpath/documents within conditions as the problem suggests.) Other approaches would be to add to the notation to mark the dependencies. If we do the latter, the syntax should, in my opinion, accomodate approaches to distributed coordination (synchronization) ranging up to actual transaction commit protocols. These would be extensions to BPSS coming from other specifications (not in our charter scope). The very simple state synchronization signals of BPSS are not the final word in treating the problem of coordination, nor should the notation lock us into this. The "extra" syntax could be as an extension to the BTs. If so, these requirements should become inputs to Martin for his refactoring of the BT substitution family that he has been rapidly creating. Since OperationActivities would also be things that could be coordinated/synchronized, we should adopt an analogous treatment for them with respect to coordination/transaction services/messages. This may spill over into version 3.0 because despite having an OASIS committee draft from the BT group, there is as yet no standards body open coordination/synchronization spec (other than XA and earlier stuff). There are various WS- specs on the subject which may go to open review. There is also the WS-CAF group that has an ambition to produce specifications in this area. Because I think they may finish after the 2.0 draft, I think we will spill over to 3.0. But we should at least locate how the extension point will work (sort of) and mark the syntactical place(s) it will be shoehorned into the notation. All the above with a healthy sprinkling of IMO and as TC member. Dale Moberg -----Original Message----- From: David RR Webber [mailto:david@drrw.info] Sent: Sunday, February 22, 2004 3:30 PM To: Dale Moberg; ebXML BP Subject: Re: [ebxml-bp] ActionItem 26 conitnued Nested Collaboration Multiparty, Roles in HaugenFletcher multiparty DRRW@ Dale, This is also the same scenario that we did for the AutoTech demo last August for automotive use case. The JPEG diagram of this and VisualScript BPSS model is here: http://drrw.net/visualscripts/#ebXML Much easier to see a real diagram and participants to understand how the multiparty interactions and flows go. I'm planning to up date this diagram to include the forks and joins on the flow arrows (right now they are just blue lines with arrowheads showing the direction of the flow). If you want to change the use case - its really easy to do in VisualScript - its just a diagram with labels - and you can change the AIAG flavour to whatever industry and components makes sense. Thanks, DW. ----- Original Message ----- From: "Dale Moberg" <dmoberg@cyclonecommerce.com> To: "ebXML BP" <ebxml-bp@lists.oasis-open.org> Sent: Sunday, February 22, 2004 4:01 PM Subject: [ebxml-bp] ActionItem 26 conitnued Nested Collaboration Multiparty, Roles in HaugenFletcher multiparty Discussion: OASIS.ebBP.WI26-NestedCollaborationsAndRoles; Topic|Referring,Roles,Reusability; Point|HaugenFletcher MultipartyCollaboration use case dwm@ Haugen-Fletcher use case is essentially a retailer, distributer, manufacturer where the retailer orders something from the distributer, the dietributer orders from manufacturer and when distributer receives Response, there is a response to retailer. I will use an abbreviated way to sketch options, using MC for MultipartyCollaboration, BTA for BusinessTransactionActivity, BT for BusinessTransaction, and omitting a lot of syntax. ProbablyWrongOption #1: We just indicate a Fork over two BTs for a PO, POConfirm Transaction. Schematically, <BT name="POTransaction" ...> </BT> <!-- ununused in 1 --> <BT name="PONotification" ...> </BT> <BT name="POConfirmNotification" ...> </BT> <Role name="Retailer" ...> <Role name="Distributer" ...> <Role name="Manufacturer" ...> <!-- this is option 1 and may be wrong --> <MC> <Start ...> <Fork type="AND" .../> <BTA ... businessTransaction="POTransaction" ...> <Performs initiating="true" bindingBy="name" inScopeRole="Retailer" bindsToRole="Buyer" /> <Performs bindingBy="name" inScopeRole="Distributer" bindsToRole="Seller" / </BTA> <BTA ... businessTransaction="POTransaction" ...> <Performs initiating="true" bindingBy="name" inScopeRole="Distributer" bindsToRole="Buyer" /> <Performs initiating="true" bindingBy="name" inScopeRole="Manufacturer" bindsToRole="Seller" / </BTA> <Join name="J" waitForAll="true" ... /> <Success fromBusinessState=J"/> </MC> In the above, we nondeterministically indicate that both subtransactions need to occur. This may be the abstraction level we want for control flow, but it leaves a lot to implementers to guess at what would be a suitable implementation. There really isn't anything indicating that the second POTransaction has to complete before the first one can complete. By not giving this information to the lower layers, the layers might produce a deadlock, waiting to complete the first transaction before launching the second. The advantage of nondeterminism here is that BPSS doesn't have to worry about how to indicate a dependency at the BPSS abstraction level. Option # 1 is at least better than just listing BTAs in a sequence, which seems to me to be asking for a traversal "hang" when monitoring. Possibly better option # 2 Replace the MC part in option 1 by the following, eliminating the nondeterministic fork/join and interpolating a BT. <MC> <Start ...> <BTA ... businessTransaction="PONotification" ...> <Performs initiating="true" bindingBy="name" inScopeRole="Retailer" bindsToRole="Buyer" /> </BTA> <BTA ... businessTransaction="POTransaction" ...> <Performs initiating="true" bindingBy="name" inScopeRole="Distributer" bindsToRole="Buyer" /> <Performs initiating="true" bindingBy="name" inScopeRole="Manufacturer" bindsToRole="Seller" / </BTA> <BTA ... businessTransaction="POConfirmationNotification" ... name="BTA3"> <Performs initiating="true" bindingBy="name" inScopeRole="Distributer" bindsToRole="Seller" /> </BTA> <Success fromBusinessState="BTA3"/> </MC> In option 2, we interpolate a request-response BT between a one-way PO Request and a one-way PO Response. This interpolation is what is going on, but we introduce a number of new BTs and fail to reuse a buy-sell BT at different points in the control Option class #3: We add on some syntactical apparatus for indicating the dependency. For example, we nest a BTA element inside another to indicate that the included BTA must complete before the first BTA completes. Or, we add an attribute in the Performs binding, that points to the BTA that needs to complete (which is probably not a good idea because we would be limited to declaring just one BTA dependency-- what if we had N that needed to complete?) I am not going to throw out syntax before people indicate we need to explore things in option class 3. Maybe there is a good way to approach this use case other than 1 or 2 that I have just overlooked. CollaborationActivity options-- I did not see how to use these to our advantage in this case? Other approaches? At least I don't think Role association needs to be a problem with this use case. dwm@ In the following I assume we accept Martin Robert's proposals concerning specializations of BusinessTransaction, and also some form of JJ Dubray's proposal that the main containers (elements) for complex flow control include BinaryCollaboration, MultipartyCollaboration, and possibly CollaborationActivity. BinaryCollaborations and MultipartyCollaborations can basically have very similar content models in JJ's view, except that the number of "locally declared Roles" will differ. I will be discussing "references" made to the above items [the complex flow containers] and also references to the specialization siblings of BusinessTransaction made in BTAs. One of the difficulties that CPPA has had with aligning with BPSS is that CPPA needs to pick out of BPSS the BusinessTransactions that a given party will support, and also select from the global view of BPSS the Role perspective of one participant (party) (for a CPP). The problem is like finding a projection of 1-dimensional curves out of some n-dimensional model, and is a little tricky. What CPPA intends to do is, starting with some complex flow's entry point, trace a toplevel Role's trajectory through all the referenced complex flows to the baseline specializations of business transasction. What CPPA needs to do is to be able to track the Role in one flow container/businesstransaction down to the primitive collaboration patterns (Notification, RequestResponse, and so forth). Ideally explicit (and _uniform_) syntactical constructs, rather than implicit semantic conventions in the BPSS text, will mark out the "connections" through the flow constructs so that implementers can have a reasonably simple task to trace trajectories. BPSS, on the other hand, rightly demands that the containers it defines be reusable, and that different complex flows can refer to (and thereby reuse) the same syntactical units (possibly flow containers themselves but in the simple case, the specializations of BusinessTransaction). So reusability and reference rightly go together. Now the problem has so far been that if we put Role trajectory connections "in" a container, the container becomes far less reusable. Hence there has been a tension between the design requirement from CPPA for Role trajectories, and the design requirement from BPSS for reusability. I think there is a resolution for these apparently conflicting requirements that applies to all flow containers, and that is _uniform_. Here are the features of the proposed resolution: 1. There can be both "local" and "global" Role declarations. That is, Role elements can occur at the toplevel and can also occur within complex flow containers (current). The global Role elements are really BusinessPartyType declarations that can get associated with a toplevel flow starting container by being bound to its (local) Role values. For uniformity, all bindings use Performs elements, so the content models at various points will need to be expanded to allow sequences of Performs elements. [Adding global Roles to indicate BusinessPartyTypes is indicated by Anders recent model. It is actually an optional feature, but one I like.) 2. The connection between Roles currently visible and those in a referred to construct occurs where the reference is made using either IDs or names. I think this design is, more or less, the way we currently do it. But our current syntax is neither uniform nor does it generalize to the MC case. Let me take some time to expand this with some unusual, but possible use cases. a. Suppose we want to reuse a JJ-style MC flow involving a credit-approver, a seller, and a buyer. Suppose we want to reuse it from a BC with a buyer and seller Role defined locally. We want the localseller to bind to the MC seller, the local seller to the MC credit-approver, and the local buyer to the referenced buyer. Three Performs constructs within the BC's Transition element's content model would accomplish assoiating local Roles with referenced Roles. Reusability supported. If we need to indicate which is the initiating Role, an attribute on Performs could mark that. [That allows the reused MC container to not have a fixed initiating Role-- it always picks that up from the Referencing context.] b. Suppose we want to loop around a choice (XOR) over BTAs. Let the Referencing context have two local roles, A and B, and the Referenced BTA be in a context with two roles, counterProposer, proposalResponder. To alternate A with B for the Role of counterProposer, we would just alternate the Performs associations, viz., <Performs><Local>A</Local><Remote initiating="true">counterProposer</Remote></Performs><Performs><Local>B< /Local><Remote>proposalResponder</Remote></Performs> with <Performs><Local>B</Local><Remote initiating="true">counterProposer</Remote></Performs><Performs><Local>A< /Local><Remote>proposalResponder</Remote></Performs>. c. As these examples illustrate, the formal analogy here is that of actual parameters and formal parameters that are used to hook up a function invocation with a function declaration. The locally "visible" Roles are analogous to the actual parameters in the call, and the declared Role values in a flow container declaration are analogous to the formal parameters. IMO, it would be nice to have an attribute on Performs called "bindingBy" with two values, "name" or "ID" That way we would not need to have the repetitive attributes such as "fromRoleId" and "fromRoleName" all the time. Also, the whole "fromRole" "toRole" attribute apparatus would disappear in 2.0. These distinctions are inherently limited to a "BinaryFlow" perspective, and don't generalize well to Multiparty flow control. Also while initiating is an important property, I would like to see it placed within the content-model of Performs. d. Different initiator flag (to accomodate Martin's different "entry" mode use case) e. Distributor (Haugen/Fletcher style) cases. f. Yunker multiparty cases [I will try to do these other use cases if I get more time this weekend.] One implication of the above approach is that, for example, constructs containing a forward reference (such as, BusinessTransactionActivity, (OperationActivity if it is added), Transition, Start, CollaborationActivity) will have content-models expressing Role associations, which I have assumed to be provided by sequences of Performs, the uniform Role connection construct. The Success and Failure states also have references pointing back. I think we could probably use Performs here also. I think it is a separate small point that needs investigation though. How to mark the toplevel flow constructs (use a Start state?) is another pending item. A final issue is whether we should dispense with CollaborationActivity in Referencing BinaryCollaborations. Minimally, I think CollaborationActivities should allow me to feference either a MultipartyCollaboration or BinaryCollaboration. I am not certain what is allowed at present. @dwm @dwm @DRRW @dwm
[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]