On 06/06/2017 10:49 AM, Spyros Trigazis wrote: > Hello, > > I'm working on using system containers for our kubernetes configuration. > > What I found a little strange in a solution like [1], is that we don't specify > the version of the component. eg if i do atomic install --system > jasonbrooks/kubernetes-kubelet:rawhide > I don't control and I don't know the kubelet version. > > If we move to system containers, which containers we need to use? > Is ti recommended to maintain our own container images based on [2]?
We had a bit of a discussion on IRC about this but thought i would update this thread. With containerized kube you can feel free to build your own kube, pull from upstream kube or pull from the kube in the fedora registry: $ curl https://registry.fedoraproject.org/v2/_catalog 2>/dev/null | python -m json.tool | grep kube "f25/kubernetes-apiserver", "f25/kubernetes-controller-manager", "f25/kubernetes-kubelet", "f25/kubernetes-master", "f25/kubernetes-node", "f25/kubernetes-proxy", "f25/kubernetes-scheduler", These registry images currently don't have very meaningful tags associated with them. We are trying to automate the version population in the images so we can get reliable version updates when an rpm changes and the image is updated. See [1] Dusty [1] https://pagure.io/atomic-wg/issue/249
