On Thu, 2011-03-03 at 08:18 -0500, John R. Dunning wrote: > From: David Lutterkort <[email protected]> > Date: Wed, 02 Mar 2011 16:21:29 -0800 > > High level question: > > Are we trying to resolve > > 1. Does aeoulous/cloud-engine require that its backends support > stateless images? > > 2. Is the deltacloud api, itself, stateless? > > 3. Both 1 and 2. > > 4. Something else. > > > My understanding is that the answer to 1 has always been "Yes", and > that it continues to be. I'm agnostic about 2.
The answer to (1) is 'No' - with my proposal, Aeolus would simulate stateless instances on top of stateful images by cloning images itself. Aeolus would still present only stateless instances to the user, and the stateful/stateless gyrations would be contained within Condor. The big reason to do this in Aeolus rather than Deltacloud is that this simulation requires a lot of state, and cleanup if things go wrong. If we put this into Deltacloud, we'd have to have a task queue and similar functionality in no time (at that point we'd just as well pull Condor into Deltacloud) > > To revive this discussion, and put a firm proposal to resolve this > forward, here's what I think the best plan of action is: > > Deltacloud > ---------- > > Need to add the following capabilities to the API: > > * Indicate for each image whether it can be used for launching a > stateful or stateless instance [1] > * Make it possible to copy/clone an existing image to a new image > > > Are these futures? Hugh, is there now discussion of adding stateful > image support to CE for rev1? > > If CE doesn't need it, I wouldn't take it as a requirement for DC, > unless there's some unrelated-to-CE reason to think the api needs it. Doing it this way is the cleanest way to expose that functionality without blowing the complexity of DC up by an order of magnitude. David
