Skip to content

kubernetes_deployment.md

Latest commit

 

History

History
136 lines (94 loc) · 1.94 KB

kubernetes_deployment.md

File metadata and controls

136 lines (94 loc) · 1.94 KB

Kubernetes Deployment Guide

Prerequisites

  • Kubernetes cluster (GKE, EKS, or local with minikube)
  • kubectl configured
  • Helm 3.x installed
  • Docker for building images

Quick Deploy

1. Build Docker Image

cd complete/
docker build -t fl-platform:latest .

2. Deploy with Helm

helm install fl-platform helm/fl-platform \
  --namespace fl-system \
  --create-namespace

3. Verify Deployment

kubectl get pods -n fl-system
kubectl get svc -n fl-system

Configuration

Custom Values

Create custom-values.yaml:

replicaCount:
  client: 10

config:
  numRounds: 50
  learningRate: 0.001
  
privacy:
  enabled: true
  epsilon: 3.0

Deploy with custom values:

helm upgrade fl-platform helm/fl-platform \
  -f custom-values.yaml \
  --namespace fl-system

Monitoring

Access Grafana

kubectl port-forward svc/grafana 3000:3000 -n fl-system

Visit http://localhost:3000 (admin/admin)

Access MLflow

kubectl port-forward svc/mlflow 5000:5000 -n fl-system

Visit http://localhost:5000

Access Prometheus

kubectl port-forward svc/prometheus 9090:9090 -n fl-system

Visit http://localhost:9090

Scaling

Manual Scaling

kubectl scale deployment fl-client --replicas=20 -n fl-system

Auto-scaling

HPA is configured by default:

  • Min replicas: 2
  • Max replicas: 20
  • Target CPU: 70%
  • Target Memory: 80%

Troubleshooting

Check Logs

# Server logs
kubectl logs -f deployment/fl-server -n fl-system

# Client logs
kubectl logs -f deployment/fl-client -n fl-system

# MLflow logs
kubectl logs -f statefulset/mlflow -n fl-system

Common Issues

Pods not starting:

kubectl describe pod <pod-name> -n fl-system

Out of resources:

kubectl top nodes
kubectl top pods -n fl-system

Cleanup

helm uninstall fl-platform -n fl-system
kubectl delete namespace fl-system