7:27 p.m.
Moving a discussion to the dev list.
Paul, using parallel operations instead of storing metadata with the
deployment resource itself has a couple drawbacks:
1) On restart of the server the special handling information implicit in
the parallel deployment ops is lost and the deployment becomes a
non-singleton.
2) In a managed domain, the "special" operation can propagate to the
relevant servers in the domain at the time it is invoked, but subsequent
to that if another server is added to a server group that runs this
deployment, again the special handling information is lost.
The stuff Thomas is working on that I mentioned is at
https://github.com/jbossas/jboss-as/pull/2790.
On 9/12/12 12:17 AM, Paul Ferraro wrote:
I did spend some time thinking about this last week while staring out
at the Mediterranean Sea this past week.
Fundamentally, implementing singleton deployments simply requires wrapping the
corresponding Service<DeploymentUnit> in a SingletonService<DeploymentUnit>.
The effect is the deployment unit will be installed on all nodes in the cluster, but will
start on only 1 node at a time. The end result is more agile than what we had in EAP5,
since we're only starting/stopping existing services, not creating them from from
scratch.
From a domain model perspective, rather than tack on configuration meta data to the
deployment resource itself, I think it would be simpler/cleaner to provide a parallel set
of deploy/undeploy/redeploy operations. The deploy/redeploy operations will accept
additional optional parameters for cluster name and node preference (for a preferred node
election policy).
For standalone mode, it might still be nice to provide a separate file scanner using a
distinct location that creates singleton deployments. To clarify, this doesn't mean
making the deployment scanner itself a singleton service, but rather the operations
created by the scanner will install singleton deployment units instead of normal
deployment units.
Thoughts/comments?
Paul
----- Original Message -----
> From: "Brian Stansberry" <brian.stansberry(a)redhat.com>
> To: "Paul Ferraro" <pferraro(a)redhat.com>
> Sent: Wednesday, August 29, 2012 1:05:19 PM
> Subject: HASingleton related issues in the EAP 6.1 PRD
>
> Care to comment on $subject at
> https://docspace.corp.redhat.com/docs/DOC-105593 ?
>
> As I commented, I don't want to see an EAP 5-ish solution based on
> activating a scanner when a node becomes master. That would be a
> standalone-only hack. I think something along the lines of
> associating
> some configuration metadata with the deployment=* resource in the
> management model makes more sense, and then some clustering
> integration
> to "(de)activate" the deployment.
>
> If you agree, we should chat about this a bit on the jboss-as7-dev
> list
> before you go. The "associating some configuration metadata with the
> deployment=* resource" part intersects with some stuff Thomas Diesler
> was trying to do, and the "activate the deployment" part intersects
> with
> some stuff Stuart Douglas was working on, as well as some OSGi stuff.
>
> --
> Brian Stansberry
> Principal Software Engineer
> JBoss by Red Hat
>
--
Brian Stansberry
Principal Software Engineer
JBoss by Red Hat