Properly restarting nodes

dcarrascal75

Hi,
I'm doing my labs using two VirtualBox Ubuntu Servers on local.
It works as expected, but if I have to stop the training and shut down nodes, I cannot figure out how to properly restart them.
I think it would be great to include this part on the training.

I try to do like this:

sudo swapoff -a
sudo kubeadm reset
sudo kubeadm init --kubernetes-version 1.16.1 --pod-network-cidr 192.168.0.0/16

and then, as stated in last command output:

mkdir -p $HOME/.kube
sudo cp -i /etc/kubernetes/admin.conf $HOME/.kube/config
sudo chown $(id -u):$(id -g) $HOME/.kube/config

But in the best scenario I get new credentials so I am not able to relink the worker node...

Could somebody please provide a straight process for this?

Thank you and keep rocking !

David

chrispokorni

Hi David / @dcarrascal75,

Any reason why you chose to run a master reset and an init after restart?

The behavior you observe is expected when running a reset and init on master. In doing so, you wipe clean the configuration of the master and create new credentials for the cluster. Hence, the old worker is not able to join the cluster. An easy fix would be a reset and a join on the worker with the new token and hash values produced by the latest init of the master.

But why run reset at all? Assuming your nodes keep their IP addresses, and you permanently disabled swap on your nodes, a simple start of each node should bring up the k8s cluster as well.

Regards,
-Chris

serewicz

Hello,

If you are using Ubuntu 18.04 and the steps in our labs the kubelet process should be started by systemd. Basic flow is systemd -> kubelet -> kube-apiserver, etcd, kube-controller-manager. This would then start the rest of the pods as per what etcd has in the configuration file.

The boot process can take a bit. Run systemctl status kubelet to see if it is running. If it's not back after 2-3 minutes, use journalctl -u kubelet to see if there are other errors.

Let us know what you find,

Regards,

dcarrascal75

Hi @chrispokorni ,

Thank you for your response
Problem here is that if I shut down my master machine, when I restart it, the cluster is down. If I restart and execute

kubectl get nodes

I get an error (Unable to connect to the server) because kubernetes is down.
My question is, at this point, how to restart it?
I expected the server to restart on boot, but it is not doing it (unless I've missed some instruction when installing it)

Thanks again,

David

chrispokorni

Hi David / @dcarrascal75,

Since you are on VirtualBox, there may be a conflict between the node IP subnet and the Pod IP subnet. Calico uses 192.168.0.0/16 by default, and VBox uses a similar subnet. In such situations, issues may be encountered when rebooting nodes.

Also, if your master node does not retain the same IP address after reboot, then your kubectl is not able to talk to the master (based on kubeconfig) and the worker cannot find the master either.

Regards,
-Chris

dcarrascal75

Hi @chrispokorni , @serewicz,

I've finally found the issue thanks to journalctl -u kubelet command.
There was not a problem with IP, but is definitely something to keep in mind when working with VMs.

Problem was between the keyboard and the chair, as usual
I cloned my last master VM from an image where I have not still done the swapoff, so the kubelet is not able to start.

Thank you for your great help and support.
Amazing!

Greetings from Spain,

David