Vapora/kubernetes/kagent

Kagent Kubernetes Integration

Kubernetes manifests for deploying Google Kagent with VAPORA A2A protocol integration.

Directory Structure

kagent/
├── base/                 # Base configuration (environment-agnostic)
│   ├── namespace.yaml    # Kagent namespace
│   ├── rbac.yaml         # ServiceAccount, ClusterRole, ResourceQuota
│   ├── configmap.yaml    # Kagent configuration
│   ├── statefulset.yaml  # Kagent StatefulSet (3 replicas)
│   ├── service.yaml      # Kubernetes services
│   └── kustomization.yaml
├── overlays/
│   ├── dev/              # Development environment
│   │   ├── kustomization.yaml
│   │   └── statefulset-patch.yaml
│   └── prod/             # Production environment
│       ├── kustomization.yaml
│       └── statefulset-patch.yaml
└── README.md

Deployment

Prerequisites

• Kubernetes 1.24+
• kubectl configured
• Kustomize 4.0+
• VAPORA A2A server running

Deploy to Development

kubectl apply -k kubernetes/kagent/overlays/dev

Deploy to Production

kubectl apply -k kubernetes/kagent/overlays/prod

Features

StatefulSet Configuration

• Replicas: 3 (base), 1 (dev), 5 (prod)
• Image: google-kagent:latest
• Service Type: ClusterIP (Headless)
• Pod Anti-Affinity: Distributed across nodes

A2A Integration

Kagent is configured to discover and integrate with VAPORA A2A server:

a2a:
  enabled: true
  vapora_server: "http://vapora-a2a:8003"
  discover_interval: 30s (dev), 60s (prod)
  timeout: 30s

Ports

• 8080/http - REST API
• 50051/grpc - gRPC endpoint
• 9090/metrics - Prometheus metrics

Resource Limits

Development:

• CPU: 100m req, 500m limit
• Memory: 128Mi req, 512Mi limit

Production:

• CPU: 1000m req, 4000m limit
• Memory: 1Gi req, 4Gi limit

RBAC

Kagent service account has permissions to:

• List/watch pods, services
• Read configmaps
• Create/delete batch jobs
• Create events

Health Checks

• Liveness Probe: /health on port 8080 (30s initial, 10s interval)
• Readiness Probe: /ready on port 8080 (10s initial, 5s interval)

Environment Variables

• KAGENT_CONFIG - Config file path
• KAGENT_POD_NAME - Pod name (auto-filled)
• KAGENT_NAMESPACE - Pod namespace (auto-filled)
• A2A_SERVER_URL - VAPORA A2A server URL
• LOG_LEVEL - Logging level (debug/info/warn/error)

Storage

• ConfigMap Volume - Kagent configuration
• EmptyDir - Cache (1Gi) and tmp (500Mi)
• PersistentVolume - Data (10Gi per pod)

Networking

All services use internal cluster DNS:

• kagent.kagent.svc.cluster.local - Headless service
• kagent-api.kagent.svc.cluster.local - REST API
• kagent-grpc.kagent.svc.cluster.local - gRPC

Monitoring

Prometheus metrics exposed at /metrics:9090

Enable scraping with annotation:

prometheus.io/scrape: "true"
prometheus.io/port: "9090"

Troubleshooting

Check pod status

kubectl get pods -n kagent
kubectl describe pod -n kagent <pod-name>
kubectl logs -n kagent <pod-name>

Test A2A connectivity

kubectl exec -it -n kagent kagent-0 -- /bin/sh
curl http://vapora-a2a:8003/health

View events

kubectl get events -n kagent --sort-by='.lastTimestamp'

Cleanup

kubectl delete -k kubernetes/kagent/overlays/dev
kubectl delete -k kubernetes/kagent/overlays/prod