Eve library

Iris Kubernetes Deployment

Other

Revision 1

On this page

Kubernetes manifests for deploying the Iris AI Assistant Platform using Kustomize.

Directory Structure#

text
iris/
├── kustomization.yaml       # Root entrypoint, delegates to base/
├── base/                    # Base configuration for all environments
│   ├── kustomization.yaml   # Main deployable base
│   ├── rbac.yaml            # Service accounts and roles
│   ├── resource-quotas.yaml # Namespace resource limits
│   ├── limit-ranges.yaml    # Container resource defaults
│   ├── network-policies.yaml # Network segmentation
│   ├── priority-classes.yaml # Pod scheduling priorities
│   ├── namespaces/          # iris, iris-data namespaces
│   ├── services/            # Individual service deployments
│   └── autoscaling/         # HPA/VPA/custom/predictive scaling
└── overlays/                # Environment-specific configurations
    ├── development/       # Local/dev settings
    ├── staging/           # Staging environment
    └── production/        # Production environment

Services#

Service Description Port
iris-api API Gateway - main entry point 3000
iris-conversation Conversation lifecycle management 3000
iris-memory Hierarchical memory service 3000
iris-agent Agent execution and tool management 3000
iris-voice Speech-to-text and text-to-speech 3000
iris-worker Background jobs and maintenance 3000

Deployment#

Prerequisites#

  1. Kubernetes cluster (1.24+)
  2. kubectl configured
  3. Secrets supplied by the platform secret manager or replaced from the deployable placeholders in base/credentials-template.yaml

Create Required Secrets#

bash
kubectl create secret generic iris-secrets \
  --namespace iris \
  --from-literal=DATABASE_URL="postgresql://user:pass@postgres:5432/iris" \
  --from-literal=REDIS_URL="redis://redis:6379" \
  --from-literal=JWT_SECRET="your-jwt-secret" \
  --dry-run=client -o yaml | kubectl apply -f -

kubectl create secret generic iris-ai-secrets \
  --namespace iris \
  --from-literal=ANTHROPIC_API_KEY="sk-ant-..." \
  --from-literal=OPENAI_API_KEY="sk-..." \
  --from-literal=GOOGLE_AI_API_KEY="..." \
  --from-literal=ELEVENLABS_API_KEY="..." \
  --from-literal=CARTESIA_API_KEY="..." \
  --from-literal=DEEPGRAM_API_KEY="..." \
  --dry-run=client -o yaml | kubectl apply -f -

The base includes placeholder iris-secrets and iris-ai-secrets resources so Kustomize output is complete for development clusters. Staging and production must replace placeholder values before rollout.

Deploy Base#

bash
kubectl apply -k infra/k8s/iris

Deploy to Development Overlay#

bash
kubectl apply -k infra/k8s/iris/overlays/development

Deploy to Staging#

bash
kubectl apply -k infra/k8s/iris/overlays/staging

Deploy to Production#

bash
kubectl apply -k infra/k8s/iris/overlays/production

Configuration#

Environment Variables#

Common configuration is managed through ConfigMaps:

  • iris-common-config: LOG_LEVEL, NODE_ENV, telemetry settings
  • iris-redis-config: Redis connection details
  • iris-qdrant-config: Qdrant vector DB settings

Resource Limits#

Environment API Replicas Service Replicas Worker Replicas CPU Request Memory Request
Development 1 1 1 100m 256Mi
Staging 2 2 1 250m 512Mi
Production 3 2-3 2 250m 512Mi

Scaling#

HorizontalPodAutoscaler is configured for all services:

  • CPU target: 65-70% (service specific)
  • External custom metrics from Prometheus Adapter
  • Predictive metrics using 15-minute predict_linear forecasts
  • Min replicas vary by environment
  • Max replicas: up to 40 for bursty workloads

VerticalPodAutoscaler policies are also defined to optimize memory requests without conflicting with HPA replica decisions.

Networking#

Network Policies#

  • Default deny ingress to all pods
  • Allow ingress controller to API
  • Allow internal communication between Iris services
  • Allow Iris to access data namespace (PostgreSQL, Redis, Qdrant)
  • Allow Prometheus scraping

Service Mesh (Istio)#

The iris namespace has istio-injection: enabled for automatic sidecar injection. The base kustomization also patches Iris deployments with explicit Istio sidecar annotations and inbound port capture for port 3000.

Monitoring#

All services expose Prometheus metrics on port 3000 at /metrics:

yaml
annotations:
  prometheus.io/scrape: 'true'
  prometheus.io/port: '3000'

Troubleshooting#

View logs#

bash
kubectl logs -n iris deployment/iris-api
kubectl logs -n iris deployment/iris-conversation
kubectl logs -n iris deployment/iris-worker

Check pod status#

bash
kubectl get pods -n iris
kubectl describe pod -n iris <pod-name>

Test service connectivity#

bash
kubectl run -it --rm debug --image=curlimages/curl --restart=Never -- \
  curl http://iris-api.iris.svc.cluster.local:3000/health