|Re: [ecf-dev] RPC for Remote P2 repositories|
Hi Jeff and Scott,I have a prototype of a remote p2 repository server running (well, my client consists of junit tests so far so it's not quite ready for production). I wrote this prototype using JSON-RPC, mainly because I was familiar with it, I knew it to introduce little overhead, and that all communication was based on http.
The difficulty that I found (and still find) most troublesome is the fact that many of the p2 objects cannot be serialized in a standardized fashion. I think it would be great if relevant objects implemented the java.io.Serializable interface or if some other generic serialization was made available. As it stands now, I have to provide custom marshalling for most of the objects. An approach that takes me too close to the actual implementation to feel comfortable.
Given the nature of the p2 resolver, I found that remoting the meta-data interface served a very limited purpose. The resolver must have access to "everything" anyway in order to do it's job so my approach so far has been more in the nature of a remotely controllable mirror that can serve up entire repositories. The artifact repository however, is easier.
I'm very much in favor of having the real implementation under the ECF and Equinox projects as I feel that this type of functionality should fit in one or the other entirely. I'm not sure how my current prototype overlaps with the Remote RCP Management but if what I'm doing can be merged into this project or my requirements can be met (me contributing of course), then that's the right way to go.
Regards, Thomas Hallgren Jeff McAffer wrote:
I am all for most of what you suggest. I do think that we should separate the issues of remote p2 and remote messaging. For example it is not clear that remote references and tc would me something we want to get into for p2. As such, ECG should be doing the comma infrastructre and then p2 should use that to make available it's function remotely. This this way the story is clear to consumers and we end up with a cleaner structure.Jeff On Jul 4, 2008, at 10:14 PM, Scott Lewis <slewis@xxxxxxxxxxxxx> wrote:Hi Jeff, Jeff McAffer wrote:Hey ThomasThe p2 repo API has not been particularly designed to support remoting. That is, I*Repository was not really intended to be used as a remote interface. Frankly we just did not think of it that way. Instead we imagined that people wanting to have a repo on another server would implement the actual repo using whatever server side technology appropriate (e.g., database, sql, ...) and then implement a "client-side" I*Repository. The client side would talk to the server using mechanisms appropriate to the configuration. There is no expectation that the server be Java etc etc. The client is also free to do any amount of caching etc that it wants to.In this way the users of I*Repository are completely isolated from the implementation details of the server and the repo implementors are free to optimize/implement as they choose.Does your usecase fit this model?Of course, it would also be interesting to consider changes to the p2 API to better facilitate remote calls. Please open enhancement requests if you see any such opportunities.I think it would be desirable to have a completely separate/new/simpler API for remote p2 access rather than attempt to modify the p2 API to facilitate transparent remote access.Why?1) There will likely be a number of differences in API behavior for things like failure/exception handling, timing (e.g. IProgressMonitor, etc), and synchronization 2) Consistent/proper handling of mechanisms like pass-by-reference/distributed garbage collection 3) Ability to use asynch messaging, futures, as well as call/return semantics for remote access...which is very helpful in remote situations, but unavailable via transparent remoting 4) Ability to use multiple/various network transport bindings as per ECF's remote services API 5) Ability to use network discovery mechanisms for remote access (i.e. discovery API)I would propose that ECF (with p2/Equinox/communities) should take this on as a sub-component within the Runtime project (where ECF is headed) and work together among interested parties. There is already some work begun in these directions within ECF...e.g. http://wiki.eclipse.org/Remote_Eclipse_RCP_Managementand obviously Thomas has already done some work on this as well. So given the basic goal here (remote provisioning of Equinox/RCP/Eclipse apps) it seems to me to be a good area for the Runtime project, Equinox p2, and ECF.Scott _______________________________________________ ecf-dev mailing list ecf-dev@xxxxxxxxxxx https://dev.eclipse.org/mailman/listinfo/ecf-dev_______________________________________________ ecf-dev mailing list ecf-dev@xxxxxxxxxxx https://dev.eclipse.org/mailman/listinfo/ecf-dev
Back to the top