Good question.
I see it as a matter of roles. The jboss-all.xml file is full of
configurations that are basically about how the deployment functions
internally. That stuff is primarily the concern of the application
developer. The admin may want to make some adaptations to tailor to the
environment, but the contents of that file are basically the province of
the app developers.
The AS7-3694 request and HASingleton are primarily about under what
conditions the deployment is installed at all. That is something the
admin controls.
So, we could:
1) Go with jboss-all.xml + overlays and force the admin to mix his stuff
in with the app developer's stuff, keeping them in sync. (An overlay is
a complete override, not a merge.)
2) Add new xxx.xml files to control this stuff and use overlays to let
the admin insert his desired configs into the deployment. This would
basically just be a hack workaround because we don't want to do overlay
merging.
3) The management resource approach.
On 10/15/12 5:32 PM, Tomaž Cerar wrote:
Why are introducing yet another configuration mechanism for
deployments?
now we have
- jboss-deployment-structure.xml
- jboss-deployment-dependencies.xml
- jboss-all.xml
- jboss-xyz.xml - (too) many of them
that in combination with overlays this could provide same capability.
AFAIR primary idea behind jboss-all.xml was to unify configuration for
all subsystems (get rid of all other jboss-cmp.xml, jboss-ejb3.xml,
jboss-app.xml,....)
and overlays just add this extra capability of exposing/manipulating
this trough management.
This new approach has only one extra capability that is to enable
subsystem itself to push/write/change some configuration that is
deployment-wise.
From my point of view that is just wrong. Subsystems should not
run-time decide and change some deployment parameters that you can than
also manipulate trough mgmt.
Maybe I am not seeing some use-case but in general I don't like this
approach...
--
tomaz
On Mon, Oct 15, 2012 at 11:45 PM, Brian Stansberry
<brian.stansberry(a)redhat.com <mailto:brian.stansberry@redhat.com>> wrote:
I've been working $subject in order to help support Thomas Diesler's
request for AS7-3694[1]. The gist of this request is to add deployment
unit processing (DUP) configuration as children of the deployment
resource itself. Thomas' OSGi use case is one place where this would be
used. I expect HASingleton deployment will be another.
WIP is at [2]. I'm looking for feedback. :)
What I've done is based on what Thomas did at [3]. What I want to do is
move from the generic key/value pairs in that patch to a more formally
describable management API. Instead of:
<deployment name="foo.war"...>
<properties>
<property name="start.policy" value="DEFERRED"/>
<property>
</deployment>
It would be something analogous to how a profile configuration is done:
<deployment name="foo.war"...>
<deployment-subsystem xmlns="urn:jboss:domain:osgi:1.2">
<start-policy value="deferred"/>
</deployment>
</deployment>
The existing Extension API already has the hooks to support this.
Extensions can register xml parsers for children of the <deployment>
element and can register management resources to act as children of the
/deployment=foo.war resource as well. Several subsystems already take
advantage of the latter. Until now the former has been an unimplemented
API. The commit at [4] implements it.
A couple things giving me some concern:
1) The above xml:
<deployment-subsystem xmlns="urn:jboss:domain:osgi:1.2">
Nicer would be something like:
<deployers>
<subsystem xmlns="urn:jboss:domain:osgi:1.2">
I need to figure out if I can do some tricks with the parsing to allow
that to happen.
2) The structure of the resource tree. We already support resources like
this:
/deployment=foo.war/subsystem=web
Subsystems register resources like those to expose metrics. The commit
at [4] uses that same tree. When subsystems could now register child
resources to the deployment=* resource, they could include both runtime
stuff and configuration stuff.
I'm not sure that mixing the two is ideal, although it's what we do for
the regular subsystem resources in the profile. I'm vaguely concerned
that if someday the configuration that subsystems choose to expose via
this mechanism gets complex, the mixing of metrics with configuration in
the same tree will start to break down.
Comments are appreciated.
[1]
https://issues.jboss.org/browse/AS7-3694
[2]
https://github.com/bstansberry/jboss-as/commits/AS7-3694
[3]
https://github.com/jbossas/jboss-as/pull/3230
[4]
https://github.com/bstansberry/jboss-as/commit/6326003a104ac4ac825e8dda4c...
--
Brian Stansberry
Principal Software Engineer
JBoss by Red Hat
_______________________________________________
jboss-as7-dev mailing list
jboss-as7-dev(a)lists.jboss.org <mailto:jboss-as7-dev@lists.jboss.org>
https://lists.jboss.org/mailman/listinfo/jboss-as7-dev
--
Brian Stansberry
Principal Software Engineer
JBoss by Red Hat