kubeadm & weave: No networks found in /etc/cni/net.d

[discostur]

Hi,

i'm trying to setup k8s 1.8.x on fedora host 27 with the kubeadm system container:

setenforce 0
atomic install --system --system-package=no --name kubelet jasonbrooks/kubeadm
systemctl daemon-reload
systemd-tmpfiles --create /etc/tmpfiles.d/kubelet.conf
systemctl enable kubelet
kubeadm init --skip-preflight-checks

export kubever=$(kubectl version | base64 | tr -d '\n')
kubectl apply -f "https://cloud.weave.works/k8s/net?k8s-version=$kubever"

After that i'm seeing the following error messages:

Nov 20 11:47:43 fedora-atomic-host runc[3522]: W1120 10:47:43.656415    3532 cni.go:196] Unable to update cni config: No networks found in /etc/cni/net.d
Nov 20 11:47:43 fedora-atomic-host runc[3522]: E1120 10:47:43.656627    3532 kubelet.go:2095] Container runtime network not ready: NetworkReady=false reason:NetworkPluginNotReady message:docker: network plugin is not ready: cni config uninitialized
Nov 20 11:47:48 fedora-atomic-host runc[3522]: W1120 10:47:48.658418    3532 cni.go:196] Unable to update cni config: No networks found in /etc/cni/net.d
Nov 20 11:47:48 fedora-atomic-host runc[3522]: E1120 10:47:48.658588    3532 kubelet.go:2095] Container runtime network not ready: NetworkReady=false reason:NetworkPluginNotReady message:docker: network plugin is not ready: cni config uninitialized
Nov 20 11:47:48 fedora-atomic-host dockerd-current[1294]: 2017-11-20 10:47:48.904677 W | etcdserver: apply entries took too long [10.1646ms for 1 entries]
Nov 20 11:47:48 fedora-atomic-host dockerd-current[1294]: 2017-11-20 10:47:48.904714 W | etcdserver: avoid queries with large range/delete range! 

Kubectl shows me that the "po/kube-controller-manager-fedora-atomic-host" is pending in the "ContainerCreating" state.

Does anybody have an idea whats wrong?

[ashcrow]

/cc @jasonbrooks

[discostur]

@jasonbrooks Any news on this?

[jasonbrooks]

@discostur I've been trying to figure this out -- perhaps something has changed w/ 1.8

[jasonbrooks]

If I change the FROM line to registry.fedoraproject.org/fedora:27 I get a 1.7.x kube, and the container works.

It looks like for 1.8.x we might need kubernetes/kubernetes#55955

[jasonbrooks]

This container works on regular fedora, where /usr/libexec/kubernetes is writable. Kubeadm has this hardcoded, the PR linked above is supposed to fix that, and make it configurable.