scheduler.md

Scheduler

Exercise 1: Disabling the scheduler

1. SSH to the master node.

2. Move the kube-scheduler manifest out of the /etc/kubernetes/manifests/ folder.

   sudo mv /etc/kubernetes/manifests/kube-scheduler.manifest ~
   

3. Wait until the kubelet shuts down the scheduler pod. This can be checked by listing all system pods.

   kubectl --namespace kube-system get pods
   

4. Deploy a pod normally.

5. Use the get pods command to list all pods. The one you've just deployed should be in a pending state with no node assigned to it.

6. Use the describe pod command to check which node is assigned to the pod.

7. Return the kube-scheduler manifest back to the /etc/kubernetes/manifests/ folder on the master node.

   sudo mv ~/kube-scheduler.manifest /etc/kubernetes/manifests/
   

8. Wait until the scheduler runs and make sure that a node is now assigned to the pod and the pod is running.

Exercise 2: Manually schedule a pod

1. While the default scheduler is disabled and a pod in the Pending state, try to manually assign a node to the container using API.
   • Use curl to sent a POST request to the /api/v1/namespaces/{namespace}/bindings endpoint.
   • The body of the request should have the following format {"apiVersion":"v1", "kind": "Binding", "metadata": {"name": "<pod-name>"}, "target": {"apiVersion": "v1", "kind": "Node", "name": "<node-name>"}}
   • Use the official Reference documentation and correct version of the API Reference to help.

   curl -X POST 127.0.0.1:8080/api/v1/namespaces/default/bindings -H "Content-Type:application/json" -d '{"apiVersion":"v1", "kind": "Binding", "metadata": {"name": "twocontainers"}, "target": {"apiVersion": "v1", "kind": "Node", "name": "master-us-west1-c-4lmf"}}'
   

   • Make sure you copy/move the kube-scheduler manifest back into /etc/kubernetes/manifests/