It's annoying that replying to a posting on this list replies to the
sender, rather than the list. I'll bug helpdesk about it.
More below...
On Tue, 2008-12-09 at 11:06 -0600, Brian Stansberry wrote:
Paul Ferraro wrote:
> On Tue, 2008-12-09 at 09:42 -0600, Brian Stansberry wrote:
>> Paul Ferraro wrote:
>>> Forwarding to mod_cluster-dev list...
>>>
>>> -------- Forwarded Message --------
>>> From: Paul Ferraro <paul.ferraro(a)redhat.com>
>>> To: jean-frederic clere <jclere(a)redhat.com>
>>> Cc: Paul Ferraro <pferraro(a)redhat.com>, Brian Stansberry
<brian.stansberry(a)redhat.com>, Remy Maucherat <rmaucher(a)redhat.com>, Mladen
Turk <mturk(a)redhat.com>
>>> Subject: Re: releasing mod_cluster-1.0.0.Beta2
>>> Date: Mon, 08 Dec 2008 11:09:09 -0500
>>>
>>> On Mon, 2008-12-08 at 14:40 +0100, jean-frederic clere wrote:
>>>> Hi,
>>>>
>>>> I have started
http://www.jboss.org/mod_cluster/changelog.html, Paul
>>>> could you please add the JAVA part?
>>>>
>>> I'll do that now.
>>>
>>>> Any other open things preventing the release happening once that done?
>>>>
>>> Nope.
>>>
>> :-)
>>
>>>> What do you think is still missing for GA?
>>>>
>>> Here's my list (I'll get this all in jira today):
>>> * Investigate auto-registration of JBW listener/dependency
>> If this looks like significant work, I think moving to 1.1.0 would be fine.
>>
>
> This actually involves a few things:
Your list sounds like enough to say put it in a later release. :)
> * Adding the lifecycle listener dynamically
> * Adding a dependency on ModClusterService/HAModClusterService (notice
> the new names...) to the WebServer MC bean. This would mean that the
> user will have to manually remove "On Demand" from either the
> ModClusterService or HAModClusterService, since there will no longer be
> anything demanding its creation.
Hmm; points out a bit of conceptual ugliness. This is semi-OT, is my own
thinking, and is more a JBoss AS discussion than mod_cluster, so just
take it as food for thought.
IMHO, AS services shouldn't start until there is some end-user usage of
them. For example, why start JBoss Messaging until a queue or topic is
deployed? This is conceptually easy enough to do; the JBM beans are "On
Demand" and the queue/topic deployer adds a dependency on JBM, so
deploying queue/topic demands JBM.
Same logic applies to JBoss Web. In that case the infrastructure of a
deployer adding the dependency on JBoss Web to the war is already there.
But if you inject JBoss Web into ModClusterService and MCS isn't "On
Demand", now JBW will be started even if no war is deployed. A
workaround would be to make the war depend on MCS, but that's unnatural.
This is a bit non-real-world, since in most cases a war (jmx-console,
ROOT.war) will be deployed. But in a "developer" AS config designed for
fast startup times, maybe not.
Makes sense.
> * Moving jvmRoute to the mod_cluster config - and setting the
engine
> property dynamically.
>
A year or two ago in another context I mentioned setting the jvmRoute
dynamically. Remy said it wouldn't work or was a bad idea; don't recall
the details. Somewhere in the Design of Tomcat Integration forum.
I had a hunch that this would not be straight forward...
This begs the question - if the user will have to edit server.xml anyway
- why bother with dynamic lifecycle listener registration and the like?
>>> * Mechanism for excluding specific webapps from
mod_cluster httpd
>>> registration
>>> * Get existing integration tests runnable from maven/ant.
>>> * HA integration tests
>>> * Setup java build in Hudson
>>> * Setup automated testing in Hudson
>>>
>>>> Cheers
>>>>
>>>> Jean-Frederic
>>>
>>> _______________________________________________
>>> mod_cluster-dev mailing list
>>> mod_cluster-dev(a)lists.jboss.org
>>>
https://lists.jboss.org/mailman/listinfo/mod_cluster-dev
>>
>
> _______________________________________________
> mod_cluster-dev mailing list
> mod_cluster-dev(a)lists.jboss.org
>
https://lists.jboss.org/mailman/listinfo/mod_cluster-dev