I've now renamed the old EndpointRegistry to be DomainRegistry [1] as has been discussed a while back on the ML.
I've also been adding methods to that to better support distributed runtimes. The main things added are: - methods about installed contributions and a new SPI class InstalledContribution [2] which holds information about the contributions attributes. Looking at that now perhaps InstalledContribution would be better renamed to something like ContrbutionDescription. The registry needs to know about the contributions so distributed runtimes can find out about available contributions to use, the attributes are needed so that those distributed runtimes don't have to actually load in the entire contribution themselves. The registry also now has a ContributionListener, thats needed so that distributed runtimes can find out when a contribution has been updated and clear out any local cache they might have of the old contribution. Besides for performance they need to cache the contribution so that when starting multiple composites in a runtime they all use the same contribution and classloader for that contribution (compliance JCI100010). - methods about running composites. These are needed so that runtimes can see composites are running elsewhere in the domain and so prevent a composite being started multiple times, and so that the domain composite can be created from anywhere based on whats currently running. - methods about what "members" are active in the domain and a way to send commands to remote members. These are used for starting and stopping composites on remote runtimes. The name "members" is just to avoid confusion with the word "node". This is where i've got to so far, I think there is refactoring that could be done but this is the current state. One thing thats not there yet which might be useful is something like a pluggable contribution repository. Currently the URL used when installing a contribution is what is shared in the registry, so "file:/C:/Tuscany/SVN/trunk/samples/getting-started/helloworld" is only going to work if all runtimes have access to that filesystem, if they don't then you need to use something they do all have access to, eg " https://repository.apache.org/content/groups/snapshots/org/apache/tuscany/sca/samples/helloworld/2.0-SNAPSHOT/helloworld-2.0-SNAPSHOT.jar";. If there was a pluggable repository (or repository manager?) then it could sort out ensuring things like file:/C:/... are installed and made available with an URL accessible domain wide. Another point is that although the above might sound specific to the DomainNode APIs virtually all the important logic is within the Tuscany runtime. So for example it would be relatively simple to have this also work with the NodeFactory/Node APIs by adding a NodeFactory.createNode method which takes contrbution URIs instead of URLs and have createNode just look up the contributions from the DomainRegistry. ...ant
