I have a pod that is essentially a plugin for an apiserver, it's almost no workload pod which task is to externalize watches to another pubsub facility (serves like a bridge from one api to another) To reduce the latency and amount of real network connections I thought that it may make sense to always deploy its 1-replica deployment to same machine, that is running apiserver itself. It turns out that it's a master node. Pod almost does not take ram and CPU, pure streaming pod without any endpoints - bridge from k8s watches to something other. How can I do that?
If your intention is only to run a specific pod on the master node and not open up the master node, you should implement tolerations and nodeSelector. The sample below will always run busybox on the master node:
apiVersion: v1
kind: Pod
metadata:
name: busybox
labels:
run: busybox
spec:
restartPolicy: Never
nodeSelector:
<a unique label on your master node>: <the label value>
tolerations:
- key: node-role.kubernetes.io/master
operator: Exists
effect: NoSchedule
containers:
- name: busybox
image: busybox
imagePullPolicy: IfNotPresent
command: ["ash","-c","sleep 3600"]If you want deploy a pod on master node.
Just run:
kubectl taint nodes --all node-role.kubernetes.io/master-