Need to set --network-plugin=cni with docker mode

[elvizlai]

Describe the bug
when using docker, Pod Ip not managed by CNI plugin.

To Reproduce

1、wget https://github.com/containernetworking/plugins/releases/download/v0.7.5/cni-plugins-amd64-v0.7.5.tgz && tar zxvf cni-plugins-amd64-v0.7.5.tgz && mkdir -p /opt/cni/bin && mv * /opt/cni/bin/
2、install docker lastest version
3、using cmd `curl -sfL https://get.k3s.io | sh -s - --docker --no-flannel` to start k3s.
4、kubectl apply -f https://github.com/coreos/flannel/raw/master/Documentation/kube-flannel.yml
5、start ANY k8s example.

Expected behavior
Pod IP should be CNI related, not docker default 172.17.xxx

Additional context
Is there any way to indicate --network-plugin=cni ?

[mortenlj]

You can use the --kubelet-arg option to add an argument to kubelet:

k3s agent --docker --kubelet-arg network-plugin=cni

[deniseschannon]

@elvizlai Can you try with v0.5.0-rc3 to see if it's fixed for you?

[dnoland1]

Tested with v0.5.0-rc4. Ran steps above to download plugins, install docker, install k3s, and apply flannel plugin. Launched a test ubuntu pod and it appeared to get a pod IP (10.42.0.0/16):

root@k3s-node1:~# k3s kubectl get pods -o wide
NAME         READY   STATUS    RESTARTS   AGE   IP           NODE        NOMINATED NODE   READINESS GATES
ubuntu-pod   1/1     Running   0          31s   10.42.0.15   k3s-node1   <none>           <none>

[erikwilson]

Thanks for testing @dnoland1!