This all got started with an idea that Mike pitched earlier today and I'm not sure how much this has diverged from what he was originally thinking and it's still a rough concept but the more I think about this, the more I like the idea.
One of the big blockers that we have right now is support for disposable clients. Most of the exploration work that we've done so far has involved some sort of cloud system (openstack, open nebula etc.) but I'm wondering if we would be better served by avoiding cloud systems entirely. We've (or maybe just me) been assuming that the tasks have be executed directly on buildslaves like they currently are and any disposable client would have to be prepared and booted before we start task execution. What if we turned that on it's head a bit and put VM spawning into the task itself - spawn a vm local to the buildslave that is then responsible for the actual work in the task instead of doing all the work inside the buildslave process? This would allow for image and vm requirements to be described in the task without weird multiple parsings and could avoid most if not all of the problems/complexity that we've started seeing in getting buildbot to work with latent buildslaves. I've written up a brief article in the phab wiki: https://phab.qadevel.cloud.fedoraproject.org/w/planning/nocloud_disposable_clients/ Any thoughts on whether this is worth pursuing as an alternative to openstack et. al? If there's anything that isn't clear, please ask for clarification. Tim
pgpHvBkNBOwWw.pgp
Description: OpenPGP digital signature
_______________________________________________ qa-devel mailing list [email protected] https://admin.fedoraproject.org/mailman/listinfo/qa-devel
