Kube Scheduler

Kube Scheduler decides on which node the pod should be scheduled. It doesn’t create the pod on the required node. That is the job of kubelet service. It only decides which node the pod will go to.

It looks for pods that don’t have nodeName property set (usually every pod) and uses its scheduling algorithm to set the nodeName property. Once the nodeName property is set, the kubelet service running on that node requests the container runtime to create the pod.


apiVersion: v1
kind: Pod
metadata:
  labels:
    name: frontend
spec:
  containers:
	  - name: httpd
	    image: httpd:2.4-alpine
	nodeName: node02

If the cluster doesn’t have a scheduler, and the nodeName property is not set manually, the pod will remain in pending state.

When setting up the cluster from scratch, download the kube-scheduler binary and run it as a service.

If the cluster is set up using KubeAdmin, the kube-scheduler is automatically deployed as a static pod in the kube-system namespace on the master node. The config is present at /etc/kubernetes/manifests/kube-scheduler.yaml

Manual Scheduling

If the cluster doesn’t have kube-scheduler, we can manually schedule a pod to a node by setting the nodeName property. This can only be done when the pod is created. To schedule a pod on another node, recreate the pod.


apiVersion: v1
kind: Pod
metadata:
  labels:
    name: frontend
spec:
  containers:
	  - name: httpd
	    image: httpd:2.4-alpine
	nodeName: node02

Scheduling Plugins

The scheduler has 4 phases, each having a set of plugins that operate at that phase.

Scheduling Queue: pod are queued based on their priority

Filtering: nodes that don’t satisfy the pod requirements are discarded

Scoring: filtered nodes are scored based on various factors like the amount of resources present after scheduling the pod and whether or not the container image is already present on that node

Binding: the node with the highest score is selected