Resource Requirements & Limits

Resource Requests [Pod]

spec:
    containers:
    - name: myapp
        image: myapp:latest
        resources:
            requests:
                memory: "64Mi"
                cpu: "2"

Resource Limits [Pod]

spec:
    containers:
    - name: myapp
        image: myapp:latest
        resources:
            request:
                memory: "1Gi"
                cpu: "1"
            limits:
                memory: "2Gi"
                cpu: "2"

• Containers can't user more CPU than it's limit

• A container can use more memory than its limits

Limit Range

• Helps us to set a default limit or request.

spec:
    limits:
    - default:
        cpu: "500m"
      defaultRequest:
        cpu: "500m"
      max:
        cpu: "1"
      min:
        cpu: "100m"
      type: Container

Resource Quotas

• These are used to define and restrict how much membery & cpu can all pods utilize.

• Max resource limits are mentioned.

Lab Learning [Resources]

• We can't change pod resources during its execution hence on changes are saved in /temp/ directory.

• We can use kubectl replace --force -f </temp/location> it will terminate running pod and create new one.

Static Pods

• The pods created by kubelet without intervention from kube-apiserver are called as static pods.

• Manifests for pods are saved in /etc/kubernetes/manifests.

• We can't create deployments, replicasets, services using this.

Lab Learnings [Static Pods]

• Static pods are visible by get pods command.

• It will have node-name after it's name, that's how you find static pods

• eg: kube-apiserver-controlplane - controlplane is the node name.

• To delete pod, you have to delete manifest file in location.

• Similarly update it.

• To delete static pod in different node

  • ssh into node

  • go to /var/lib/kubelet/config.yaml

  • find staticPodPath

  • Go to the path and you'll find static pod manifest.

  • Delete it using rm command.

Manual Schedules

• The scheduler config can be found at /etc/kuberenetes/config/...yml

• If we have multiple mast nodes then only 1 scheduler configuration can run at a time.

• We can use leaderElection & leaderElect=true to define leader in multiple master scenario.

• We can specify what scheduler picks up the pod using schedulerName:<name-of-scheduler>

• kubectl logs <scheduler-name> -n ns

• kubectl get events

Stages of Scheduling

• Scheduling Queue

• Filtering

• Scoring

• Binding