How to use Kubernetes?

1. Via google gloud console

   Run cluster startup script

   Change cluster version to: --zone $ZONE --cluster-version "1.23.8-gke.1900" \

2. Use minikube for a local Kubernetes cluster

   • Pro: you don’t have to pay cloud costs.
   • Con: only one node
3. minikube is running in a docker container (check with docker ps)
4. minikube comes its own version of kubectl which is definitely compatible with minikube. So it’s safest to NOT run kubectl, but minikube kubectl for all commands.
5. Check that the running cluster is the minikube one via kubectl config view will contain (i.e. context should be pointing to cluster: minikube):

- context:
  cluster: minikube

• You can now run minikube dashboard for a nice dashboard of the cluster.
• Via Docker Desktop: Enable in Configurations

Output of kubectl config view will contain:

- context:
  cluster: docker-desktop

What?

Orchestrate many micro services.

View what’s happening

via kubectl

Run kubectl get pods (or kubectl get pods -A to view all pods - including internal ones, e.g. of minikube) every second:

watch -n 1 kubectl get pods

View endpoints, pods, nodes and services:

kubectl get endpoints,pods,nodes,svc -o wide

View everything:

kubectl get all

Get .yaml file + extra information of horizontal pod autoscaler:

kubectl get hpa -o yaml

via

Container

Kubernetes Yaml config file

apiVersion: v1
kind: Pod
metadata:
  name: my-pod
spec:
  containers:
    - name: nginx
      image: nginxdemos/hello:0.3
      ports:
        - containerPort: 80
      resources:
        requests:
          cpu: "100m"
          memory: "100Mi"
        limits:
          cpu: "250m"
          memory: "100Mi"

• CPU measured in millicores

  • If, for instance, we want to request 0.5 of a CPU, we should express it as 500m — where m represents millicores.
  • In the example above, we’ve specified the container’s requested CPU as 100m and the limit as 250m. This means the processing power that will be reserved for the container is 100m. Furthermore, if a process requires more than 100m, it can access the additional CPU resources that the scheduler will ensure are available on the node — up to the 250m limit.
• Memory resources are measured in bytes and can be expressed as fixed-point numbers, integers or power-of-two equivalents. We will use the most common type of expression — the power-of-two equivalent, Mi. This represents a Mebibyte, or 220 bytes.

see this link for more.

Multi container pods

apiVersion: v1
kind: Pod
metadata:
  name: my-multi-pod2
spec:
  containers:
    - name: container-a
      image: alpine:3.16.0
      command:
        - "/bin/sh"
      args:
        - "-c"
        - "while true; do echo $(date) > /tmp/buffer; sleep 1; done;"
      volumeMounts:
        - name: buffer
          mountPath: /tmp
    - name: container-b
      image: alpine:3.16.0
      command:
        - "/bin/sh"
      args:
        - "-c"
        - "while true; do cat /tmp/buffer; sleep 1; done;"
      volumeMounts:
        - name: buffer
          mountPath: /tmp
  volumes:
    - name: buffer
      emptyDir: {}

Jump into a running container (in a pod):

kubectl exec -it pod_name -- sh

kubectl exec -it pod_name container_name -- sh

• Files: ls -lisa
• Look at processes ps ef

Replica Sets and deployments

Selectors

• Each replica set has to have a selector which defines the set of pods.
• A label selector matches certain labels.

Rollout strategies

Services

Run nodes which contain pods.

Service types

• ClusterIP: Expose ports locally on localhost
• NodePort: Expose explicit external ports on nodes (on all node’s external IP addresses - view them via kubectl get nodes -o wide)
• LoadBalancer: Add external load balancer (not managed by Kubernetes)

Auto-Scaling

Horizontal pod autoscaling

Scale horizontally (aka horizontal pod autoscaling) with pods

apiVersion: autoscaling/v1
kind: HorizontalPodAutoscaler
metadata:
  name: my-autoscaler
spec:
  maxReplicas: 5
  minReplicas: 1
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: my-deployment
  targetCPUUtilizationPercentage: 5

• targetCPUUtilizationPercentage is chosen to be low (5%) so that scaling will be triggered fast.

Vertical node auto-scaling

• Scale vertically with nodes (instance groups)

Health checks

Probes

health-checks run by kubernetes to test whether everything is running well.

Types of probes: liveness probes, startup probes

Self-healing

Compare actual state to desired state -> act on differences

Concepts

Namespaces

kubectl get pods --namespace webshop

Useful

View spec

View spec of Kubernetes object (e.g. pod or deployment etc.):

kubectl get deployments.apps my-deployment -o yaml

Tools

helm the kubernetes package manager

• helm chart

  my-chart/
  templates/
    configmap.yaml
    deployment.yaml
    service.yaml
  Chart.yaml
  values.yaml   <-- default values file
  my-own-values.yaml

• values.yaml

  color: cyan
  replicas: 1

• Run all kubernetes objects defined in your helm chart via

  helm install my-release-defaults ./my-chart

Prometheus

Grafana

krew

krew# Plugin manager for kubectl.

*You can install it with brew. * kubectl-ctx

Allows to easily switch the context of kubernetes - switching between environments.

install

kubectl krew install ctx

usage

kubectl ctx kubectl-ns

Allows to easily display/switch kube namespaces.

install

kubectl krew install ns

usage

kubectl ns kubectl-view-secret

This plugin allows for easy secret decoding. Useful if you want to see what’s inside of a secret.

install

kubectl krew install view-secret

usage

kubectl view-secret