[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]
Subject: Re: [wsbpel] Issue - 123 - Matching <reply> with <receive>
Hi, Paco,
I am for syntax simplicity of R1 also. At the same time, I want to
provide a clear semantics of that extra "usage" identifier means.
I have just replied an email to Maciej that may contain some answers to
your questions here.
I guess my latest "thinking-out-loud" viewpoint may actually fit your
expectation better. :-)
[ sorry for shifting my viewing angle :-) ]
Let me re-iterate a bit here:
If we look at the problem from a viewpoint from a relational tuple
viewpoint, the scope / lifetime of the usage identifier may be actually
a non-issue. Because, the partnerLink instance is always in the tuple
for matching. Therefore, my expected semantics can still be achieved
even when the usage identifier is scopeless. E.g. Maciej's usecase is
still legal and feasible. (That increase the modularity of our code).
<receive partnerLink="p1" operation="op1" usage="handleCust" />
<reply partnerLink="p1" operation="op1" usage="handleCust" />
<!-- p1.op1.handleCust in terminal state (request/response MEP) -->
<receive partnerLink="p1" operation="op1" usage="handleCust" />
<reply partnerLink="p1" operation="op1" usage="handleCust" />
Also, the "usage" identifier is not a resource. It is merely a piece of
string data or an identifier. It has no lifecyle cycle / scoping
implication.
Therefore, we can now put the scope issue of usage identifier aside.
We need to do is to make sure we got the right semantics of matching the
identification "tuples" of receive-and-reply pair. *** Particularly, the
"part (B)" that you, Satish and Yaron discussed about ... regarding to
whether to match optional CS between receive/reply. ***
Regarding to modularity, I still think this is a good goal to achieve.
But, that is a side topic, not a core part of Issue 123 anymore.
Regarding to "adding partner links to reduce ambiguity" part, I need to
admit I am stepping into a brand new territory, which is kind of
undefined by the spec and can be too hairy for users to understand.
That's why I still prefer using the "usage" identifier. I would put
this idea aside as well.
I hope my latest clarification make you feel more comfortable about my
suggested semantics.
Thanks!
Regards,
Alex Yiu
Francisco Curbera wrote:
>
>
>Alex,
>
>Thanks for the clarifications. It seems that the stronger argument you are
>making is the modularity of scopes from within a process, but since it is
>not clear what is exactly the goal of that modularity (reuse? support of a
>particular runtime architecture?) it is hard to assess the benefit. On the
>other hand there are now several identifiers that afaik are not scoped
>(link and activity names). Are you going to propose that we scope those as
>well?
>
>In also don't understand the point you make about adding partner links to
>reduce ambiguity. Are you saying that you will define two plinks in BPEL
>but map them to the same endpoint at runtime? Wouldn't that have the effect
>of obscuring the interaction protocol? That is, the fact that at runtime
>two messages are expected simultaneously from the same endpoint, their
>responses having to go back to that endpoint woud not be represented in the
>process definition.
>
>My main reason to insist on R1 is simplicity: no need to declare MEP
>instances, no need to handle visibility rules. Scope level declaration and
>inside out visibility rules are great to solve many problems, but are
>totally unjustified in this case. We are trying to solve a very simple
>problem, let's try keep the solution simple as well.
>
>Paco
>
>
>
>
> Alex Yiu
> <alex.yiu@oracle. To: Francisco Curbera/Watson/IBM@IBMUS
> com> cc: "Eckenfels. Bernd" <B.Eckenfels@seeburger.de>, Satish Thatte
> <satisht@Microsoft.com>, wsbpel@lists.oasis-open.org, ygoland@bea.com, Alex Yiu
> 08/05/2004 02:41 <alex.yiu@oracle.com>, Maciej Szefler <mbs@fivesight.com>
> AM Subject: Re: [wsbpel] Issue - 123 - Matching <reply> with <receive>
>
>
>
>
>
>
>Hi, Paco, (and others),
>
>I guess we can consolidate the text from "lifetime"/"namespace" into
>"scope", if you prefer. I was just reusing those terms from on Yuzo's
>original proposal.
>
>Let me answer some of your questions and re-iterate some of points
>expressed in the conf call today:
> I am NOT proposing the usage id is required. They are used only when
> we need to disambiguate multiple receive.
> This usage id will NOT affect the actual existing partnerLink
> behavior in anyway. It is a pure BPEL language feature, which has
> ZERO effects to the underlying WSDL layer. The usage id is just a
> supplementary qualifier for disambiguation.
> The key part of this proposal is try to bring the advantages of R1
> and R1A under one proposal. Semantic-wise, the scope of the usage
> identifier can have the following choices:
> (I) per BPEL process: Similar to "R1"
> (II) per partnerLink: Each partnerLink has its own namespace
> for usage identifier. That is my CURRENT proposed semantics.
> As of my current proposed semantics, "Scope"/"namespace" is per
> partnerLink. That means:
> The same identifier CANNOT be used for 2 receive/reply pairs on
> the same plink. That means the following is illegal.
> <scope>
> <partnerLink name="p1" ... />
> <receive partnerLink="p1" operation="op1" usage="u1" />
> <receive partnerLink="p1" operation="op1" usage="u1" />
> <reply partnerLink="p1" operation="op1" usage="u1" />
> <reply partnerLink="p1" operation="op1" usage="u1" />
> </scope>
> The usage identifier are not shared among partnerLink. It is
> legal to do the following:
> <receive partnerLink="p1" operation="op1" usage="handleCust"
> />
> <reply partnerLink="p1" operation="op1" usage="handleCust"
> />
> <receive partnerLink="p2" operation="op1" usage="handleCust"
> />
> <reply partnerLink="p2" operation="op1" usage="handleCust"
> />
> (No matter whether p1 and p2 are declared in the same scope or
> not)
> From the BPEL engine viewpoint, the first receive operation is
> now identified as the key tuple("p1"-"handCust", "op1", cs).
> "handCust" becomes a part of supplementary name of the
> partnerLInk.
> I totally agree with you that <receive> conflict / ambiguity should
> not happen that commonly. However, after passing Issue 75, I tend to
> think we want to make the <scope> construct as modular as possible.
> So, it will be easier for us to move / copy the <scope> around from
> one process to another process. A per-process namespace (I) for the
> "usage" identifier is not good enough for this modularity goal.
> The <receive> conflict / ambiguity can be avoided in a large extent
> by using a locally scope partnerLink without using the "usage"
> identifier already. Therefore, I suggest (II) semantics where the
> "usage" identifier fits as merely a supplement identifier to a
> partnerLink.
> I will send a separate email to illustrate how I see locally scoped
> partner can solve a majority of <receive> conflict / ambiguity
> problems already.
>
>Thanks!
>
>
>Regards,
>Alex Yiu
>
>
>
>[P.S.: A third alternative semantics: (III) coincide with the same scope
>which the used partnerLink is declared: that is a macro version of an
>implicit declaration of "messageExchange" element in "R1B". (I think that
>is what Maciej mentioned in the conf call. I said it is the current
>semantics of my proposal by mistake. Sorry about that.) ]
>
>
>
>Francisco Curbera wrote:
>
>
>
> Hi Alex,
>
> I like the simplicity approach that your R1B shares with R1. However,
> I
> don't really understand the following:
>
> Lifetime: you say the "lifetime" of an identifier is the plink.
> Does
> that mean that the same identifier can be used for 2 receive/reply
> pairs
> on the same plink - supposedly non conflicting ones? This seems to
> open
> the door for ambiguity or at least lack of clarity. Making ids
> unique
> per MEP instance is a much simpler approach. Maybe you really mean
> "scope"...
> "Namespace" (again I think this is actually "scope"): what is good
> about
> enabling the language so two MEP instances in different plinks can
> share
> the same usage id? The design looks particularly error prone,
> potentially misleading authors/consumers of process definitions to
> match
> receives and replies incorrectly because their usage ids match
> (but
> their plinks don't!). It certainly gives you flexibility but I
> don't
> know if the potential confusion and lack of readability makes it
> worth.
> I don't understand your last point about receive identification.
> Are you
> proposing that the usage id be required? Or that it have a runtime
> representation in messages targeted at that receive?
>
> I am really skeptical that dealing with name collisions is actually
> necessary in this case, seems to me we're trying to attack a problem
> that
> does not exist. The fact is that <receive> conflicts of this kind are
> probably going to be uncommon overall. Even when they happen I doubt
> that
> we'll see so many conflicts in a single process that you cannot deal
> with a
> single process-wide naming scope for MEP instance ids. The little
> benefit
> we can get out of the solutions proposed to deal with the id
> collission
> problem doesn't seem justify the additional complexity they all
> introduce.
>
> Paco
>
>
>
>
> Alex Yiu
>
> <alex.yiu@oracle. To:
> ygoland@bea.com
>
> com> cc: Satish
> Thatte <satisht@Microsoft.com>, Francisco Curbera/Watson/IBM@IBMUS,
>
> "Eckenfels. Bernd"
> <B.Eckenfels@seeburger.de>, wsbpel@lists.oasis-open.org, Alex Yiu
> 08/03/2004 11:02 <alex.yiu@oracle.com>
>
> PM Subject: Re: [wsbpel]
> Issue - 123 - Matching <reply> with <receive>
>
>
>
>
>
>
> Hi, all,
>
> After reading all the emails on this thread, I guess "R1" and "R1A"
> have
> its own pros and cons.
>
> I am think out loud again here to try to come up with a third
> proposal (
> "R1B") to address both simplicity and clarity of scope.
>
> (This is not my super firm position. The opinions polled split as
> 50-50 so
> far. This is just yet another "thinking-out-loud" attempt to merge
> two
> solutions into one):
>
> Proposal of "R1B":
> Syntax-wise, it is similar to "R1". Except I would suggest to
> rename
> to something like "usage" or "plinkUsage". E.g.:
> <receive partnerLink="plink1" usage="handleCustomer" ...>
> ...
> <receive partnerLink="plink1" usage="handlerShipper" ...>
> ...
> <reply partnerLink="plink1" usage="handlerCustomer" ...>
> ...
> <reply partnerLink="plink1" usage="handlerShipper" ...>
> The identifier used in the attribute does NOT need to be
> declared
> explicity in a scope beforehand. The identifier is declared
> implicitly in a sense. (This addresses the need of simplicity.)
> Semantics of the identifier used in the "usage" attribute:
> Lifetime: The lifetime of the "usage" identifier shares
> the the
> lifetime of the partnerLink used in the operation. This
> addresses both the need of clarity of lifetime and
> imposes a
> tighter control and make sure the lifetime of the
> identifier is
> sync with the lifetime of the partnerLink.
> Namespace: The namespace of the identifier is PER
> partnerLink.
> For examples: (all of the following are using the same
> operation and CS)
> <receive partnerLink="plink1" usage="handleCustomer"
> .../>
> <receive partnerLink="plink1" usage="handlerShipper"
> .../>
> These two receive operations do NOT collide.
> <receive partnerLink="plink1" usage="handleCustomer"
> .../>
> <receive partnerLink="plink2" usage="handlerCustomer"
> .../>
> These two receive operations do NOT collide.
> <receive partnerLink="plink1" usage="handleCustomer"
> .../>
> <receive partnerLink="plink1" usage="handlerCustomer"
> .../>
> These two receive operations do collide. This
> collision
> MUST be caught by static analysis.
> This addresses name collision prevention. And, it
> actually
> makes even easier to avoid any name collision.
> Unique key to identify a receive operation: We continue
> to have
> 3 components to uniquely identify a receive:
> [a] partnerLink (+ "usage" attribute, if used)
> [b] operation
> [c] any CS (if used)
>
> I hope you guys like my proposal.
> Thanks!
>
>
> If you guys still want to read further, here is my analysis on the
> current
> R1 and R1A proposal:
> R1:
> Advantage: simple syntax (just one attribute)
> Disadvantage: scope of the identifier used in that
> attribute is
> NOT clear (i.e. leading to "unclear MEP lifetime" and
> "harder
> name collision prevention" in Yuzo's previous email)
> R1A:
> Advantage: scope of identifier used in that attribute is
> clear.
> Disvantage:
> A bit too verbose
> An odd concept introduced:
> As of current spec stands, we have 3
> components to
> uniquely identify a receive: [a] partnerLink
> [b]
> operation [c] any CS (if used). With R1A, we
> now
> introduce yet one more entity
> ("messageExchange")
> to the formula. Its lifetime can be totally
> in a
> different scope compared with partnerLink or
> CS.
> We are making a 2 dimensional problem (two
> disjoint
> entities: partnerLink and CS) into 3
> dimensional
> one.
> Apart from this disambiguating usage, this
> messageExchange entity does not have other
> usages
> and meaning. Yet, it has the naming of this
> board
> semantics. It may tempt vendors and customers
> to
> overload and extend that term signficantly in
> a
> totally unpredictable way.
>
> Thanks!
>
>
>
> Regards,
> Alex Yiu
>
>
>
>
> To unsubscribe from this mailing list (and be removed from the roster
> of the OASIS TC), go to
> http://www.oasis-open.org/apps/org/workgroup/wsbpel/members/leave_workgroup.php
> .
>
>
>
>
>
>To unsubscribe from this mailing list (and be removed from the roster of the OASIS TC), go to http://www.oasis-open.org/apps/org/workgroup/wsbpel/members/leave_workgroup.php.
>
>
>
[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]