Hello Toni,
On Sep 21, 2009, at 11:18 , Toni Menzel wrote:
Glad you asked, Angelo and I had further discussions on gchat and
agreed on
a way to go.
Here's the current status (lengthy version;)
Thanks for the warning! ;)
PART 1: General things
First of all, what makes ace assemblies so different from other Pax
Runner
setups:
a. Artifacts are flat file artifacts up until now (See
http://issues.apache.org/jira/browse/ACE-18 for some thoughts on this)
b. Artifacts are highly configured through config files who must be
at a
certain place on startup
We can (and should) overcome [a] easily. (see ideas on using Pax
Runner
Profiles in part 3)
Agreed, see below for comments on ACE-18.
[b] means:
we cannot just provide a single Pax Runner config file. We need to
copy
those configuration files to a certain position.
Like you say, a target consists of code (a set of bundles) and
configuration (now done with our configurator which reads
configuration files from a directory, but essentially anything that
imports Configuration Admin configs should do, we also have code to
use the XML format defined in the Auto Config section, which is used
together with the resource processor.
In the end we should be able to deploy all these targets using
deployment packages (containing bundles and configuration).
Have you thought about adding some kind of support for configurations
to Pax Runner (except the system properties)?
PART 2: Result of discussion so far
We need a number of pre-defined targets: a default target, a default
server,
a server-obr combo, etc.
For each of these, we would like a 'release' version containing a
framework
of our choice (latests felix), and does not require anything else at
runtime.
The 'dev' version is based on pax runner, can be configured to use any
framework you like, and contains possibly some additional bundles
(e.g. for
logging)
So, that would lead to something like six target xml's, resulting in
twelve
directories in the deploy/target directory.
Each of those targets will be produced by Pax Runner.
Release targets will have no pax runner reference and will be static
to
known (recommended?) target configuration set.
Benefit of using Pax Runner even in that scenario instead of the
current way
is (amonngst others): simple to switch to a different frameework and
version, same "language" used to define the assembly as in dev-
versions
(who will be just a pax runner config file).
Agreed.
PART 3: Ideas on using profiles
One other thing i see as well:
Pax Rummer has the notion of profiles / composites.
So, if we decide to publish ace artifacts to a snapshot repository
(maven)
somewhere, we can add ace profiles for each target here:
https://scm.ops4j.org/repos/ops4j/projects/pax/runner-repository/.
Can we also make this work when deploying to your own local Maven
repository (keeping everything you need local)?
This would make starting with ace very simple:
./pax-run.sh --profiles=org.apache.ace.gateway
and in exam:
profile("org.apache.ace.gateway")
Agreed, I would definitely like this if we can find some way to
provide the configuration along with the bundles.
I would also like something like:
./pax-run.sh --profiles=org.apache.ace.framework-with-ma -
Ddeploymentpackage=file:dp/ace-server.dp
For launching with a deployment package.
anyhow, this depends on how simple we can integrate the bridge to
maven (ant
will keep being the buildsystem for ace as far as i know).
See http://issues.apache.org/jira/browse/ACE-18
I think you already mentioned the biggest issue, the different
versioning scheme that Maven uses (having to use "snapshot" names for
anything that's not a release). I think we should be able to come up
with some kind of scheme for that.
I will provide a new patch for ACE-32 to meet the criterias in PART 2.
Ok, thanks!
Greetings, Marcel
