marc fleury wrote:
>
> |Yep, that's exactly whay I was trying to get at. I feel that I'm lacking
> |in vision on the general "J2EE deployment" side of things, but I'm happy
> |to code something like this up if it's agreed that it's right direction.
>
> keep it simple.
>
> otherwise you end up with acyclic graphs with bidirectional (bipolar)
> dependencies on dog doodoo.
Well, yes, we wouldn't want that. Simple = good, dog doodoo dependencies
= bad.
> Define the MLet MBean (based on the new schema) and pass the "war, jar, jcx,
> xxx" argument as the extension seen by the deployer for a particular
> service.
>
> When such a service comes up he registers with the ApplicationManager,
> please hook up with Cristoph on the exact logic for when particular jars are
> redeployed within one application. I do believe his "large app with many
> dependent jars" problem needs a flexible "only redeploy this but not the
> whole application" solution. You have the knowledge of what CL is
> associated with what, that rests with the Application Management or even
> deeper (CL service anyone?).
It seems to me that there is potentially some confusion about what
constitutes an "application". A "J2EE Application" is fairly clearly
defined in the J2EE spec, being a collection of "J2EE modules", i.e. ejb
jars, client jars, wars and rars bundled into an ear (with deployment
descriptors etc).
As you've said, J2EE doesn't really say anything more about applications
- they're just collections of modules. What it sounds like is wanted is
some way of
(a) grouping J2EE applications together into one logical application
so that they can share classloaders
(b) specifying dependencies between J2EE applications so that we know
what needs redeploying.
In fact, this is almost exactly what Rickard said, so I guess I agree
with him.
I suggest that (b) is more important than a) because it facilitates the
management of large deployments - i.e. it is concerned with correctness,
whereas (a) is concerned with performance.
(b) really is a directed graph (I'd be more comfortable with it being
acyclic, but it would take some effort to enforce this), and (a) is a
tree, as Rickard said.
If this is an accurate understanding of what people require
(Chtistoph?), then I can throw something together, although I'm not able
to dedicate huge amounts of time at the moment.
> Essentialy you then call the different targets with the right jar, war,
> jcx...
What is this "jcx"? Do you mean rar, as in resource adapter?
Toby.
> go home you are done
>
> marc
--
Toby Allsopp
Research
Peace Software International Ltd
Ph +64-9-3730400
