Google Kubernetes Engine

Deploy managed Kubernetes clusters on GCP.

Create Cluster

gcloud container clusters create my-cluster \
  --num-nodes=3 \
  --machine-type=e2-medium \
  --zone=us-central1-a \
  --enable-autoscaling \
  --min-nodes=1 \
  --max-nodes=5 \
  --workload-pool=${PROJECT_ID}.svc.id.goog

# Get credentials
gcloud container clusters get-credentials my-cluster --zone=us-central1-a

Node Pools

gcloud container node-pools create gpu-pool \
  --cluster=my-cluster \
  --zone=us-central1-a \
  --machine-type=n1-standard-4 \
  --accelerator=type=nvidia-tesla-k80,count=1 \
  --num-nodes=1

Workload Identity

gcloud iam service-accounts add-iam-policy-binding \
  --role=roles/iam.workloadIdentityUser \
  --member="serviceAccount:${PROJECT_ID}.svc.id.goog[NAMESPACE/KSA_NAME]" \
  GSA_NAME@${PROJECT_ID}.iam.gserviceaccount.com

Best Practices

• Use Workload Identity
• Enable VPC-native clusters
• Implement node auto-provisioning
• Use regional clusters for HA