We see on some Kubernetes cluster upgrading from 1.30 -> 1.31 following errors on cilium, coredns, kube-proxy, ... pods on Control Planes:

Warning  Failed     15s (x3 over 12s)  kubelet            Error: services have not yet been read at least once, cannot construct envvars

The pods will not start on the updated Control Plane, so we must do it with a little workaround to ensure a seamless upgrade.

Upgrade the Cluster without errors

First of all if you face that problem, no worries, you can easily rollback the kubeadm/kubelet and patch it then. You can also exchange the kubectl but its not needed.

Rollback of kubeadm (Debian based OS)

Redownload the old kubeadm/kubelet and restart with systemctl:

# Rollback to 1.30
wget -P /usr/local/bin https://dl.k8s.io/release/v1.30.X/bin/linux/amd64/kubeadm
wget -P /usr/local/sbin https://dl.k8s.io/release/v1.30.X/bin/linux/amd64/kubelet
systemctl restart kubelet

Upgrade the Cluster before update the binaries

Download the new kubeadm to your home directory and rename it to avoid confusion:

wget https://dl.k8s.io/release/v1.31.X/bin/linux/amd64/kubeadm
mv kubeadm kubeadm-v1.31.X
chmod +x kubeadm-v1.31.X
./kubeadm-v1.31.X   upgrade apply -y v1.31.X

If you have a cluster with more then one Control Plane update all Control Plane Nodes first before exchanging the binaries and reboot the Nodes.

And thats it !

If you still have problems with upgrading your Kubernetes clusters let us know. We can help you!