GKE Workload Optimization

admin

1 year ago

“GKE Workload Optimization”

Salah satu dari banyak manfaat menggunakan Google Cloud adalah model penagihannya yang menagih Anda hanya untuk sumber daya yang Anda gunakan. Dengan mengingat hal itu, Anda tidak hanya harus mengalokasikan sumber daya dalam jumlah yang wajar untuk aplikasi dan infrastruktur Anda, tetapi juga memanfaatkannya secara efisien. Dengan GKE, ada sejumlah alat dan strategi yang tersedia untuk Anda yang dapat mengurangi penggunaan berbagai sumber daya dan layanan sekaligus meningkatkan ketersediaan aplikasi Anda.

Praktikum

Provision lab environment

setting default zone

gcloud config set compute/zone us-central1-a

buat 3 node cluster

gcloud container clusters create test-cluster --num-nodes=3  --enable-ip-alias

buat manifest untuk gb-frontend pod

cat << EOF > gb_frontend_pod.yaml
apiVersion: v1
kind: Pod
metadata:
  labels:
    app: gb-frontend
  name: gb-frontend
spec:
    containers:
    - name: gb-frontend
      image: gcr.io/google-samples/gb-frontend:v5
      resources:
        requests:
          cpu: 100m
          memory: 256Mi
      ports:
      - containerPort: 80
EOF

create manifest

kubectl apply -f gb_frontend_pod.yaml

Task 1. Container-native load balancing through ingress

buat manifest cluster IP

cat << EOF > gb_frontend_cluster_ip.yaml
apiVersion: v1
kind: Service
metadata:
  name: gb-frontend-svc
  annotations:
    cloud.google.com/neg: '{"ingress": true}'
spec:
  type: ClusterIP
  selector:
    app: gb-frontend
  ports:
  - port: 80
    protocol: TCP
    targetPort: 80
EOF

apply ke cluster

kubectl apply -f gb_frontend_cluster_ip.yaml

Buat ingress untuk app

cat << EOF > gb_frontend_ingress.yaml
apiVersion: networking.k8s.io/v1
kind: Ingress
metadata:
  name: gb-frontend-ingress
spec:
  defaultBackend:
    service:
      name: gb-frontend-svc
      port:
        number: 80
EOF

kubectl apply -f gb_frontend_ingress.yaml

cek health status untuk backend service

BACKEND_SERVICE=$(gcloud compute backend-services list | grep NAME | cut -d ' ' -f2)

get health status dari service

gcloud compute backend-services get-health $BACKEND_SERVICE --global

retrieve

kubectl get ingress gb-frontend-ingress

Task 2. Load testing an application

download docker image

gsutil -m cp -r gs://spls/gsp769/locust-image .

build docker image

gcloud builds submit \
    --tag gcr.io/${GOOGLE_CLOUD_PROJECT}/locust-tasks:latest locust-image

cek list docker image

gcloud container images list

apply manifest

gsutil cp gs://spls/gsp769/locust_deploy_v2.yaml .
sed 's/${GOOGLE_CLOUD_PROJECT}/'$GOOGLE_CLOUD_PROJECT'/g' locust_deploy_v2.yaml | kubectl apply -f -

cek IP external

kubectl get service locust-main

cek web via -> [EXTERNAL_IP_ADDRESS]:8089

Masukan 200 user , dan 20 hatch rate
Click Start swarming
Buka Navigation menu > Kubernetes Engine
pilih workload
klik gb-frontend pod

Task 3. Readiness and liveness probes

buat demo

cat << EOF > liveness-demo.yaml
apiVersion: v1
kind: Pod
metadata:
  labels:
    demo: liveness-probe
  name: liveness-demo-pod
spec:
  containers:
  - name: liveness-demo-pod
    image: centos
    args:
    - /bin/sh
    - -c
    - touch /tmp/alive; sleep infinity
    livenessProbe:
      exec:
        command:
        - cat
        - /tmp/alive
      initialDelaySeconds: 5
      periodSeconds: 10
EOF

kubectl apply -f liveness-demo.yaml

cek list pod

kubectl describe pod liveness-demo-pod

delete file

kubectl exec liveness-demo-pod -- rm /tmp/alive

cek pod event

kubectl describe pod liveness-demo-pod

Setting up a readiness probe

buat single pod

cat << EOF > readiness-demo.yaml
apiVersion: v1
kind: Pod
metadata:
  labels:
    demo: readiness-probe
  name: readiness-demo-pod
spec:
  containers:
  - name: readiness-demo-pod
    image: nginx
    ports:
    - containerPort: 80
    readinessProbe:
      exec:
        command:
        - cat
        - /tmp/healthz
      initialDelaySeconds: 5
      periodSeconds: 5
---
apiVersion: v1
kind: Service
metadata:
  name: readiness-demo-svc
  labels:
    demo: readiness-probe
spec:
  type: LoadBalancer
  ports:
    - port: 80
      targetPort: 80
      protocol: TCP
  selector:
    demo: readiness-probe
EOF

kubectl apply -f readiness-demo.yaml

cek service

kubectl get service readiness-demo-svc

cek pod event

kubectl describe pod readiness-demo-pod

generate file

kubectl exec readiness-demo-pod -- touch /tmp/healthz
kubectl describe pod readiness-demo-pod | grep ^Conditions -A 5

Task 4. Pod disruption budgets

delete single app

kubectl delete pod gb-frontend

edit jadi 5 replica

cat << EOF > gb_frontend_deployment.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: gb-frontend
  labels:
    run: gb-frontend
spec:
  replicas: 5
  selector:
    matchLabels:
      run: gb-frontend
  template:
    metadata:
      labels:
        run: gb-frontend
    spec:
      containers:
        - name: gb-frontend
          image: gcr.io/google-samples/gb-frontend:v5
          resources:
            requests:
              cpu: 100m
              memory: 128Mi
          ports:
            - containerPort: 80
              protocol: TCP
EOF

kubectl apply -f gb_frontend_deployment.yaml

drain node

for node in $(kubectl get nodes -l cloud.google.com/gke-nodepool=default-pool -o=name); do
  kubectl drain --force --ignore-daemonsets --grace-period=10 "$node";
done

cek replica count

kubectl describe deployment gb-frontend | grep ^Replicas

uncordoning node

for node in $(kubectl get nodes -l cloud.google.com/gke-nodepool=default-pool -o=name); do
  kubectl uncordon "$node";
done

kubectl describe deployment gb-frontend | grep ^Replicas

buat pod disruption budget

kubectl create poddisruptionbudget gb-pdb --selector run=gb-frontend --min-available 4

drain lagi

for node in $(kubectl get nodes -l cloud.google.com/gke-nodepool=default-pool -o=name); do
  kubectl drain --timeout=30s --ignore-daemonsets --grace-period=10 "$node";
done

kubectl describe deployment gb-frontend | grep ^Replicas

Penutup

Sahabat Blog Learning & Doing demikianlah penjelasan mengenai GKE Workload Optimization. Semoga Bermanfaat . Sampai ketemu lagi di postingan berikut nya.