Allow scheduling of pods on Kubernetes master?

Question

I set up Kubernetes on CoreOS on bare metal using the generic install scripts. It's running the current stable release, 1298.6.0, with Kubernetes version 1.5.4.

We'd like to have a highly available master setup, but we don't have enough hardware at this time to dedicate three servers to serving only as Kubernetes masters, so I would like to be able to allow user pods to be scheduled on the Kubernetes master. I set --register-schedulable=true in /etc/systemd/system/kubelet.service but it still showed up as SchedulingDisabled.

I tried to add settings for including the node as a worker, including adding worker TLS certs to /etc/kubernetes/ssl, adding those settings to kubelet.service, adding an /etc/kubernetes/worker-kubeconfig.yaml that pointed to those certs, and added that information to the /etc/kubernetes/manifests/kube-proxy.yaml. I used my existing nodes as a template for what to add. This registered another node under the master's hostname and then both it and the original master node showed up as NotReady,SchedulingDisabled.

This question indicates that scheduling pods on the master node should be possible, but there is barely anything else that I can find on the subject.

Answer 1

Use the below command to untaint all masters

kubectl taint nodes --all node-role.kubernetes.io/master-

Answer 2

Another way to list all taints in nodes and untaint the tainted one.

root@lab-a:~# kubectl get nodes -o json | jq ".items[]|{name:.metadata.name, taints:.spec.taints}"
{
  "name": "lab-a",
  "taints": null
}
{
  "name": "lab-b",
  "taints": [
    {
      "effect": "NoSchedule",
      "key": "node-role.kubernetes.io/master"
    }
  ]
}

lab-a does not have any taint. so we untaint lab-b:

root@lab-a:~# k taint node lab-b node-role.kubernetes.io/master:NoSchedule-
node/lab-b untainted

Install jq in ubuntu by: apt-get install jq

Answer 3

If you are using Kubernetes 1.7 and above:

kubectl taint node mymasternode node-role.kubernetes.io/master:NoSchedule-

Answer 4

For anyone using kops on AWS. I wanted to enable scheduling of Pods on master.

$ kubectl get nodes -owide was giving me this output:

NAME                                          STATUS
...
...
ip-1**-**-**-***.********.compute.internal    Ready                      node
ip-1**-**-**-***.********.master.internal     Ready,SchedulingDisabled   master
                                                    ^^^^^^^^^^^^^^^^^^
ip-1**-**-**-***.********.compute.internal    Ready                      node
...
...

And $ kubectl describe nodes ip-1**-**-**-***.********.master.internal:

...
...
Taints:             <none>
Unschedulable:      true
...                 ^^^^
...

Patching the master with this command:

$ kubectl patch node MASTER_NAME -p "{\"spec\":{\"unschedulable\":false}}"

worked for me and scheduling of Pods is now enabled.

Ref: https://github.com/kubernetes/kops/issues/639#issuecomment-287015882

Answer 5

First, get the name of the master

kubectl get nodes

NAME     STATUS   ROLES    AGE   VERSION
yasin   Ready    master   11d   v1.13.4

as we can see there is one node with the name of yasin and the role is master. If we want to use it as worker we should run

kubectl taint nodes yasin node-role.kubernetes.io/master-

Answer 6

I don't know why the master node shows up as NotReady; it shouldn't. Try executing kubectl describe node mymasternode to find out.

The SchedulingDisabled is because the master node is tainted with dedicated=master:NoSchedule

Execute this command against all your masters to remove the taint:

kubectl taint nodes mymasternode dedicated-

To understand why that works read up on taints and tolerations.