This is a walk through with commands needed to upgrade kubernetes from 1.11.4 to 1.13.0.
In general, start by checking this page for information related to the specific upgrade you are doing: https://kubernetes.io/docs/reference/setup-tools/kubeadm/kubeadm-upgrade/
Plan the upgrade:
sudo kubeadm upgrade plan
Note the output from the above command, specifically that you need to upgrade
kubeadm before you attempt the upgrade. After the upgrade, you also have to
upgrade kubelet on all nodes.
Upgrade kubeadm (ubuntu):
sudo apt-mark unhold kubeadm
sudo apt install kubeadm=1.13.0-00
sudo apt-mark hold kubeadm
Upgrade kubeadm (centos):
sudo yum install kubeadm-1.13.0-0 --disableexcludes=kubernetes
Upgrade control plane:
sudo kubeadm upgrade apply v1.13.0
Upgrade kubelet on the master
kubectl drain master --ignore-daemonsets --delete-local-data
# ubuntu
sudo apt-mark unhold kubelet
sudo apt install kubelet=1.13.0-00
sudo apt-mark hold kubelet
# centos
sudo yum install kubelet-1.13.0-0 --disableexcludes=kubernetes
When it is ready for action again, mark it as schedulable:
kubectl uncordon master
Drain the node:
kubectl drain <node-name> --ignore-daemonsets --delete-local-data
Upgrade kubelet and kubeadm:
# ubuntu
sudo apt-mark unhold kubelet kubeadm
sudo apt install kubelet=1.13.0-00 kubeadm=1.13.0-00
sudo apt-mark hold kubelet kubeadm
# centos
sudo yum install kubelet-1.13.0-0 kubeadm-1.13.0-0 --disableexcludes=kubernetes
Upgrade kubelet config:
sudo kubeadm upgrade node config --kubelet-version $(kubelet --version | cut -d ' ' -f 2)
sudo systemctl restart kubelet
When the node is ready for action again, mark it as schedulable:
kubectl uncordon <node-name>
Upgrading a cluster with HA control plane is quite similar to the above. See this page for precise instructions.
The main difference is that you should run
kubeadm upgrade node experimental-control-plane
on all additional control plane nodes. For the upgrade from 1.12 to 1.13 you also have to edit the kubeadm configmap manually.