I think a developer centric deployment model is fine as long as we
articulate how the exploded-archive should be built  ( i.e. all WSDL,
Schema, etc ... dependencies should be copied first and then the .bpel file
) and fail gracefully when artifacts can't be found. Is this what you had in
mind?

Lance

On 5/3/06, Alex Boisvert <[EMAIL PROTECTED]> wrote:


As a developer, I'm a big fan of the exploded-archive format  -- which
is effectively an oximoron for no deployment archive -- where individual
files are just moved or copied in a deployment area on the file system.

I think this approach reduces the need for deployment tooling, reduces
deployment times (faster test cycles) and reduces indirections, meaning
that I know exactly what's deployed instead of guessing what's been put
in the archive.

It would be nice if the engine supported this kind of straight-through
deployment in addition to the heavier archive deployment

alex

Matthieu Riou wrote:

> Hi,
>
> Lance mentioned before the need to agree on a deployment format for
> Ode, in short to answer to the question "what will I deploy in Ode?".
> Note that the question is 'what', not 'how' (yet). To start the debate
> I'll just drop some of the requirements I'd have, to see what you guys
> think.
>
> First, we could support two types of deployment unit, just like any
> J2EE app server does:
>
> 1. A raw format with no prior compilation, generation whatsoever. The
> magic happens on deployment.
> 2. A compiled, validated, ready to use format. Upon deployment this
> thing is just opened, read and we're ready to go.
>
> For this first format, I think we just need a self containing archive,
> including not only the process representation but also wsdl, xsl,
> deployment descriptor, whatever... End of requirements.
>
> The second format should be compiled and validated. So most (if not
> all) possible errors in the original bpel description, the wsld or
> anything else should be detected during the step that produces this
> deployment unit. The compiled version of the process should be as
> close as possible to the definition model used at runtime (something
> like a pre-processed object tree).
>
> This two format wouldn't be exclusive, it's just 2 steps in the
> lifetime of our deployed package. This 2 steps can happen all at one
> time (1 ->  2 -> running) or one after another (1 -> 2 // 2 ->
> running).
>
> Thoughts? Comments? Criticisms? Flames?
>
> Matthieu.



Reply via email to