RE: [wsbpel] RE: completionHandler example

["Satish Thatte" <satisht@microsoft.com>]

I was too cryptic.  Cascading completion handlers are not a necessity.
I was simply pointing out that any time the scope you are completing is
intermediate, i.e., is not the named scope in named completion, writing
any completion logic sequentially following the occurrence of that scope
will not work.  Thus the "logic sequentially following the occurrence of
that scope" method is too fragile (what if someone adds a new completion
activity inside my scope?).

-----Original Message-----
From: Yaron Y. Goland [mailto:ygoland@bea.com] 
Sent: Thursday, September 30, 2004 2:16 PM
To: Satish Thatte
Cc: wsbpeltc
Subject: Re: [wsbpel] RE: completionHandler example

Doesn't this strike you as a bit complex? How many programmers are ever 
going to use a cascading completion set? Is the usage large enough that 
it justifies adding yet another type of handler to BPEL?

The use case for named completion is pretty obvious given that in BPEL 
you really can't do anything without declaring a scope so named 
completion is a necessity. But I'm having trouble seeing most or even a 
significant number of users implementing code that not only uses named 
completion but has a cascading set of completion handlers as well.

Could you help me understand the use cases that are likely to occur in 
real world code?

	Thanks,

		Yaron

Satish Thatte wrote:
> 
> 
> That doesn't work if you want to do cascading completions aka named
> completions.
> 
> -----Original Message-----
> From: Yaron Y. Goland [mailto:ygoland@bea.com]
> Sent: Thursday, September 30, 2004 11:17 AM
> To: Satish Thatte
> Cc: wsbpeltc
> Subject: Re: completionHandler example
> 
> I realize why putting the completion handler logic before the complete
> doesn't work but why doesn't putting it after the completion handler
> work?
> 
> scope
>     variables
>        variable name="MessageTrackerVariable"...
>     sequence
>        scope wrapper
>           flow
>              scope A_end_in_complete_to_wrapper
>                 ...
>              scope B_end_in_complete_to_wrapper
>                 ...
>              scope C_end_in_complete_to_wrapper
>                 ...
>        scope CompletionHandler
>           (Check MessageTrackerVariable and see if any cleanup is
> needed)
> 
> Because the exit point of a completed scope is always well defined
can't
> 
> the completion handler logic be placed after the completed scope?
> 
>         Yaron
> 
> 
> Satish Thatte wrote:
>  >
>  >
>  > Yaron,
>  > 
>  > During the F2F you asked for a completionHandler example.  Here is
my
>  > try with some explanations.  Please let me know if this makes sense
to
> you.
>  >
>  > 
>  > Satish
>  > 
>  >
>  > The completionHandler is the common process logic required at all
> points
>  > of premature completion within a scope.  The only difference
between
>  > copying it just before each occurrence of a <complete/> activity
thus
>  >
>  > 
>  >
>  > <sequence>
>  >
>  >     <completionHandler logic>
>  >
>  >     <complete/>
>  >
>  > </sequence>
>  >
>  > 
>  >
>  > and the proposed completionHandler feature is that the
> completionHandler
>  > logic executes *after* termination of all activities in the
> prematurely
>  > completed scope.  Thus it is a pure macro if the prematurely
completed
> 
>  > scope does not contain concurrent activities.  Many examples of the
> use
>  > of premature completion, e.g., completion of N out of M activities
in
> a
>  > flow, do involve concurrency.  Consider a case where M suppliers
had
>  > been contacted concurrently and the process was waiting for
> asynchronous
>  > responses (forgive the use of the word :-)) from them.  After at
least
> N
>  > of them send responses the scope completes.  Suppose the
conversation
>  > with each supplier is wrapped in a scope and a terminationHandler
is
>  > used to inform a supplier whose conversation was forcibly
terminated
>  > that their response is not needed.  But perhaps we did not or could
> not
>  > ensure that *exactly* N replies arrive because the race was too
tricky
> 
>  > to control.  Thus it is possible that more than N responses
arrived. 
>  > The completion handler could detect this situation, pick the N we
> really
>  > want and inform the rest that their responses have been rejected so
> that
>  > we exit the scope cleanly with exactly N accepted responses.
>  >
> 
> 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_workgr
oup.php. 
> 
>