with docker-cri, k8s v1.24.1 uses docker0 as network for pod's

[jsachdev07]

Hi,
I have deployed k8s 1.24.1 and used docker-cri for the first time here as dockershim is removed.
The nodes went to ready state after cluster initialization without deploying any cni, earlier with dockershim till 1.23.x I used to deploy calico then only nodes go in ready state.
I noticed that docker0 is used as network for Pod's, all Pod's take an IP from docker0. does cri comes with an inbuilt cni ??
also I found that 2 pod's are taking same IP, let's suppose 2 pod's are running on diff worker nodes, they will take IP from respective node docker0 network, hence same IP's.
Also in such a case how can pod's talk that are running on different worker node ? as each pod takes IP from respective docker0 network
Can we by any chance switch to any other cni like calico or flannel or something else ??

[evol262]

cri-dockerd is a CRI, and nothing more. You can pick a desired CNI in k8s, and pass --network-plugin=cni to tell cri-dockerd to use it.

[korncola]

@jsachdev07 --network-plugin is removed from kubelet in 1.24, see:
https://kubernetes.io/docs/concepts/extend-kubernetes/compute-storage-net/network-plugins/
and
kubernetes/website#33640

better swtich to crio or containerd
https://kubernetes.io/docs/setup/production-environment/container-runtimes/

[evol262]

@korncola Your comment was in no way helpful.

--network-plugin is an argument to cri-dockerd, not kubelet in this case.

[jsachdev07]

@evol262 could you please let know how we can set network parameter for cri-dockerd
I was passing --network-plugin during kubeadm init which isn't working in 1.24.1

[evol262]

Adding --network-plugin=cni to the systemd unit file for cri-dockerd systemd unit file as part of the ExecStart stanza will take care of this.

It's a reasonable default with 1.24 shipped, which will be part of the next release.