On Thu, Aug 14, 2008 at 12:27 AM, Dwiputera Aries Fajar (CI/AFR-SG) <
[EMAIL PROTECTED]> wrote:

> Let say SERVICE A take about 3 days to complete its task...and when it is
> finished,
> How can this service know that it needs to send the reply (invoke) to my
> bpel process?


I would know because that's the application contract.  Nothing in BPEL
validates this; it's something that should be documented, checked and tested
at a higher level.


> Let say this SERVICE A is a marketplace, which is hit by thousands of web
> service...how can the service know that it will need to return this result
> (from thousands of results) to my bpel process?


Your request message should contain some information about the sender if it
intends to receives a reply.   This can be expressed in various forms, from
WS-Addressing headers to a customer number inside the message body.


> I know that inside WS-Addressing, they have some "return-address" field,
> which can be used as a reference to invoke or reply the result. But without
> WS-Addressing, how SERVICE A reply my invocation? Does it mean every web
> service needs to have some queue engine as its outer layer? Or is there any
> other mechanism which I'm not aware of?


There are various ways of doing it, but no standard outside of
WS-Addressing.  As another example, you could just include the service URL
in the request message.

Ode has its own protocol if you're interested:
http://ode.apache.org/implicit-correlations.html
http://ode.apache.org/stateful-exchange-protocol.html

alex

Reply via email to