I composed a tree with this change applied and tested it by bringing up a kube cluster using the ansible scripts at https://github.com/kubernetes/contrib/tree/master/ansible. It worked as expected.

To test the upgrade from f24's kube 1.2 case, I set up a kube cluster on f24, ran a simple nginx app, rebased to my f25+kube tree, and after reboot, the cluster was still working and the app returned.

This was basically what we agreed on in the meeting; patch looks sane. I am uncertain though about readding ceph-common - layering works for it (and presumably there is ceph containerization work).

Are we saying with this that we're not going to default to using layering at all?

It seemed simplest to just revert the change, but we can limit it further if you prefer. The issue with containers is that we don't yet have any official fedora containers for these components.

Layering does work with all but kubernetes-master, although it requires a reboot. If live layering (and the fcap support) is working by the time f26 comes out, then we'd have a much simpler layering or containers story for users.

I tested this today with https://github.com/kubernetes/contrib/tree/master/ansible - managed to get k8s up and running and use kompose to bring up an application. Can we get this merged?