I will receive one order which contain multiple details, based on the code
of detail, I need to start different process to handle the instruction
within detail parallel, then aggregate the result of each sub process and
product summary of the order to finish the job.

If I run through the loop to handle the detail, it cost time. Each detail
can be handled paralle. If I use Parallel shape, the number of branchs will
be dynamically determined, and it doesn't work. Is there anyone have ideas
or hints for this situation? I think the dynamical-allocate-then-aggregate
situation should have a design pattern to resolve. Thanks!

Michael

[ Post a follow-up to this message ]

One thing you can consider is using a "start" orchestration shape to handle
each line item, and pass each called orchestration a self-correlating port
reference for the called orchestration to call the main orch back - to
notify you of completions.  In a parallel branch within your main
orchestration, you can listen for completion notifications (in a loop,
knowing you can exit when # of completions = # of lines, or some timeout is
reached.)

I've used this technique for a scatter/gather pattern in the past with
success.

Scott Colestock
www.traceofthought.net


"Michael Peng" <michaelpeng@michaelpeng.idv.tw> wrote in message
news:eAtUyrHhFHA.2472@TK2MSFTNGP15.phx.gbl...
>I will receive one order which contain multiple details, based on the code
>of detail, I need to start different process to handle the instruction
>within detail parallel, then aggregate the result of each sub process and
>product summary of the order to finish the job.
>
> If I run through the loop to handle the detail, it cost time. Each detail
> can be handled paralle. If I use Parallel shape, the number of branchs
> will be dynamically determined, and it doesn't work. Is there anyone have
> ideas or hints for this situation? I think the
> dynamical-allocate-then-aggregate situation should have a design pattern
> to resolve. Thanks!
>
> Michael
>

[ Post a follow-up to this message ]

Thanks Pal!

"Scott Colestock" <scolestock@community.nospam> ¼¶¼g©ó¶l¥ó·s»D:u9olBsihFHA.3608@TK2MSFTNGP12
.phx.gbl...
> One thing you can consider is using a "start" orchestration shape to
> handle each line item, and pass each called orchestration a
> self-correlating port reference for the called orchestration to call the
> main orch back - to notify you of completions.  In a parallel branch
> within your main orchestration, you can listen for completion
> notifications (in a loop, knowing you can exit when # of completions = #
> of lines, or some timeout is reached.)
>
> I've used this technique for a scatter/gather pattern in the past with
> success.
>
> Scott Colestock
> www.traceofthought.net
>
>
> "Michael Peng" <michaelpeng@michaelpeng.idv.tw> wrote in message
> news:eAtUyrHhFHA.2472@TK2MSFTNGP15.phx.gbl... 
>
>

[ Post a follow-up to this message ]