Rickard,

> Since you point a URLClassLoader at it I would guess that you will never be
> able to remove it. Right?

Yep on Win2k! On my Linux box it works fine :-)

> > Once I know what type I have to deploy, I install it from the original
> > source. This doesnt seem smart, especially if we dont have much
> > bandwidth. But it was simpler to only pass the original source location,
> > which is needed to find libraries referenced by MANIFEST.MF/Class-Path:
> > entries in case of ejb or war packages.
> > I agree its not a clean solution, so maybe I should change this...
> 
> Ok. Yeah you should change to always deploy on the localCopy. This way it is
> always possible to put in a new version in /deploy. Right now it doesn't
> work very well (about half of my redeployments fail, consistently).

Ok will change that immediately...

> > This is a bug! ...in the AutoDeployer. The AutoDeployer calls the
> > J2eeDeployer too early (before the copy (or jar operation) on OS level
> > is completed).
> 
> No, that's not it. I made the wait longer, and it still appears. It is
> something else.

Hmmm... ???
I will try to investigate that...

> Also, if I restart the server, does the J2EE deployer redeploy ones that are
> not autodeployed? I.e. if I call deploy() through JMX and then restart, will
> it redeploy the package? If it doesn't, it should.

Uuhhh, this is a completely different story.
Since I remove all from the tmp/deploy directory on service startup, I
dont have anything left to redeploy (just read the deploymnet.cfg file
and call startApplication ())...

Ok, this is the Window hack but anyway I was thinking about that issue
too, especially on server crashes (will never happen ;-) it would be
fine to be able to recover the state of deployment (as I had this in an
earlier version). But how would that behave in conjunction with the
Autodeployer? J2eeDeployer starts, recoveres all deployments,
Autodeployer starts, redeploys all apps that are still available in the
observed directories.

Do you have an idea how to crack that (or anyone else)?


\Daniel

> 
> /Rickard

Reply via email to