Let me try to explain a bit more what I had in mind. Some of you may know I've been working on this stuff over the past weeks. When this thread came up, I thought some of you (as users of Karaf) might be interested in what I had worked on, so I actually pushed really hard to be able to open source bits (which is now available as Fabric). I think that would be helpful for some of us to be able to use those bits.
Now, my original idea was not really to embed Fabric in Karaf. Fabric has a broader goal than Karaf and has bits that are targeted for ServiceMix, Camel or ActiveMQ (see the Fabric extensions described at http://fabric.fusesource.org/documentation/user-guide.html#Fabric_Extensions). As such, it would not make sense to actually dump the code into Karaf. I sincerely think (but maybe I'm wrong), that the clustering thing we're talking about is a big project in itself, that it is more tied to OSGi than to Karaf itself (by that I mean, that it could be useful for people that use OSGi but not Karaf to some extent). I also think the notion of clustering is also really dependant on what you cluster. Clustering web applications has nothing to do with clustering activemq brokers. What does a Karaf clustering solution really mean ? Clustering activemq brokers is not simply deploying the same bundle in multiple nodes. For example, I see in https://cwiki.apache.org/confluence/display/KARAF/Karaf+clustering some thoughts about statefull / stateless: I think that in order to provide a generic way to cluster applications you need to control things. JEE defines a programming model using EJB, WebServices, etc... and everything can be controlled. At the end, you kinda have to enforce some things. Karaf is really designed to be a general purpose server side container. ServiceMix and Geronimo are using it already, maybe in the future other Apache projects may want to leverage it too. However, most of those I'm thinking about would not be interested in a clustering solution such as a generic one described at https://cwiki.apache.org/confluence/display/KARAF/Karaf+clustering. I think what we're discussing here is a different thing: some kind of environment based on OSGi which is highly distributed, with some kind of automatic provisioning. And I think if you want to actually have a reliable system, a provisioning agent need to be in full control of the provisioning. It seems to me very hard to imagine a reliable agent that can cope with any user interaction: you can't really support a user uninstalling a bundle, can you ? So you need to lock things down. And at the end, I think such a solution is not Karaf. It can reuse Karaf, but it's not Karaf. Now, for Fabric itself, I'm not sure if FuseFource will be willing to donate it to Apache. I can understand some of you (mostly the talend guys because they try to get into the same business as FuseSource) would rather start a new project than reusing Fabric. But for people that don't have a business closely related to Karaf, things are certainly a bit different. The ASL is not the only open source license in the world and being fanatic on anything is usually not good. I'm not aware of any ASL compatible OS, so I'm quite confident none of us is religious on those things. Each license has a purpose and one need to respect the will of the licensor. Last, if several people are willing to work on an OSGi based clustering (need a better term) and/or provisioning solution at the ASF, maybe a better way would be to go and start a project in the incubator. FWIW, the Ace project could definitely be interested in such an effort. I've seen a demo a few weeks ago of creating ec2 instances on the fly and provisioning them with Ace and that looked quite cool. On Sat, Apr 16, 2011 at 19:52, Ioannis Canellos <ioca...@gmail.com> wrote: > If I have misunderstood, then I apologize! Guillaume, could you please > sched some light to it? Do you intend to bring the source in, or you > intend to have fabric hosted to fuse source and used as a dependency, > like jansi? > > On Saturday, April 16, 2011, <j...@nanthrax.net> wrote: >> Hi Ioannis, >> >> Don't get me wrong: I assume that Guillaume talked about his implementation >> cause he can donate to Karaf. >> If cluster implementations can't be included in Karaf, there is no >> discussion ;) >> >> Regards >> JB >> -----Original Message----- >> From: Ioannis Canellos <ioca...@gmail.com> >> Date: Sat, 16 Apr 2011 11:37:59 >> To: <dev@karaf.apache.org> >> Reply-To: dev@karaf.apache.org >> Subject: Re: [PROPOSAL] Karaf clustering feature in Karaf 3.0.0 >> >>> >>> I have an absolute and total trust in Guillaume, especially about his >>> Apache fiber (keep in mind that Guillaume is a ASF member). If he proposed >>> his cluster implementation, he knows that we can use it in Karaf >>> >> >> We all do JB! I can't imagine there is anyone among us, that doesn't trust >> him and highly respect his work. >> >> My objection with Guillaume's proposal has nothing to do with Guillaume >> himself or Fuse (I respect Fuse and I've even proposed to my day work >> purchasing Fuse support :-)). >> >> >>> I don't care where the implementation comes from (Fuse, Talend, others), if >>> the implementation is a good one, and if the community accepts this >>> implementation. >> >> >> No ones cares about the origination of the solution. >> >> The only objection is that I believe that the default clustering solution >> should be part of karaf and not a third party dependency. >> If Guillaume proposed to brings his "source" in, none would object (at least >> I would step back). >> >> >> -- >> *Ioannis Canellos* >> * >> http://iocanel.blogspot.com >> >> Apache Karaf <http://karaf.apache.org/> Committer & PMC >> Apache ServiceMix <http://servicemix.apache.org/> Committer >> * >> >> > > -- > *Ioannis Canellos* > * > http://iocanel.blogspot.com > > Apache Karaf <http://karaf.apache.org/> Committer & PMC > Apache ServiceMix <http://servicemix.apache.org/> Committer > * > -- Cheers, Guillaume Nodet ------------------------ Blog: http://gnodet.blogspot.com/ ------------------------ Open Source SOA http://fusesource.com Connect at CamelOne May 24-26 The Open Source Integration Conference http://camelone.com/
