k8s-scaling-demo

Demo about K8s' scaling capabilities using GKE

Usage

Prior

• Set up GCP account (Gmail)
• Create project
• Set up billing
• Set up GKE API
• Go to GCP console
• Open Cloud Shell (small terminal icon upper-left)

Deployment

1. Spin up a GKE cluster:

   gcloud container clusters create k8s-demo-cluster --num-nodes 2 --zone europe-west1-b

2. Build the container and push to Google Cloud Registry:

   git clone https://github.com/GoogleCloudPlatform/nodejs-docs-samples.git

   cd nodejs-docs-samples/containerengine/hello-world/

   docker build -t gcr.io/$DEVSHELL_PROJECT_ID/hello-node:1.0 .

   gcloud docker -- push gcr.io/$DEVSHELL_PROJECT_ID/hello-node:1.0

3. Deploy a 'hello-world' app deployment:

   kubectl run hello-node --image=gcr.io/$DEVSHELL_PROJECT_ID/hello-node:1.0 --port=8080 --replicas 2

   Check status with kubectl get deployments and kubectl get pods.

4. Create a service and expose it externally through GCP's Load Balancer

   kubectl expose deployment hello-node --name=hello-node-svc --type=LoadBalancer --port=80 --target-port=8080

   Check status with kubectl get services and wait for the external ephimeral IP assignment.

   Verify success accesing the service through this external IP, either in your browser or with curl.

5. Update the deployment

   Increase the replica number. Go to 'Workloads' in GKE console and copy the yaml code corresponding to the 'hello-node' deployment into 'app-deployment.yaml', modifying the replica number.

   kubectl apply -f app-deployment.yaml

6. Create a horizontal pod autoscale resource

   With a mininum of 2 and a maximum of 10 replicas:

   kubectl autoscale deployment hello-node --min=2 --max=10 --cpu-percent=50

Load test

Install 'httperf' on your system, e. g. sudo apt-get install httperf. Consider using a VM instead of Cloud Shell for more throughput.

Create a sinthetic test load for the app and see the autoscaler in action:

httperf --server SERVICEIP --num-conns 100 --rate 10 --timeout 1

Play with some loads and see the autoscaler creating and destroying pods:

watch -n 1 kubectl get pods

Clean up

Clean up all resources from GCP's console: deployments, LB, cluster, firewall routes, etc.

Author

• Marcos Manuel Ortega
• [email protected]
• GDG Almería organizer
• GDG Cloud Español coorganizer
• 2/may/2018
• For AlmeriaJS meetup

License

See LICENSE file.