Jorgen Thelin replied to
my earlier piece on Web
Services Proxies with some interesting thoughts. I'm somewhat ambivalent on this
whole issue. I partly agree with what he says, and I fully agree with his thoughts
that the industry must strive towards giving a clear view of these topics.
However, I also partially disagree with him. For example, while he
correctly points out that you can build an RPC-style proxy on top of an asynchronous
request/response interaction, I also believe there are some hidden caveats in that.
One particular that comes to mind is loosing on some of the potential reliability
gains you get from asynchronisity in the first place. Potentially, a system based
on async behavior can more easily deal with long wait times and specially some temporal
failures of some of the systems involved in the interaction with relatively little
development pain (and this is even more true on systems based on things like BPEL).
However, once you lay down sync-behavior it becomes much harder, and the failure semantics
just get more complicated (which might not be a problem for simple transactions, though).
I think this kinds of compromises should be much better explained, too.
On a related note (and please pardon my ignorance on this subject)
I also wonder whether the current incarnation of WSDL (even if it should
die like Clemens says) is good
enough for describing these kinds of interactions (such as asynchronous request/response),
or whether something like BPEL's roles in partnerLinks are a better (or worse) alternative
to describe them. Any thoughts?