Terraform configuration for the Iris AI Platform staging environment on AWS EKS.
Overview#
This configuration provisions:
- VPC: Multi-AZ network with public, private, and database subnets
- EKS: Kubernetes cluster with managed node groups
- RDS: PostgreSQL 16 with pgvector support
- ElastiCache: Redis 7.1 replication group
- S3: Buckets for conversations, memory, logs, and backups
- ECR: Container registries for Iris services
- IAM: IRSA roles for Kubernetes workloads
- Smoke Tests: Lambda-based health monitoring
The staging deployment manages the full Iris runtime set: iris-api,
iris-conversation, iris-memory, iris-agent, iris-voice, and
iris-worker.
Prerequisites#
- AWS CLI configured with appropriate credentials
- Terraform >= 1.10.0
- S3 bucket for Terraform state (
oshun-terraform-state) - DynamoDB table for state locking (
oshun-terraform-locks)
Usage#
Initialize#
bash
cd infra/terraform/iris/staging
terraform init
Plan#
bash
terraform plan -out=tfplan
Apply#
bash
terraform apply tfplan
Destroy#
bash
terraform destroy
Configuration#
The default configuration is optimized for staging:
| Component | Staging Configuration |
|---|---|
| EKS Nodes | SPOT instances (cost savings) |
| NAT Gateway | Single (cost savings) |
| RDS | db.t3.medium, Multi-AZ |
| ElastiCache | cache.t3.medium, 2 clusters |
| Node Scaling | 2-6 nodes |
Customization#
Override defaults using terraform.tfvars or command-line variables:
bash
terraform apply -var="general_node_capacity_type=ON_DEMAND"
Outputs#
After applying, key outputs include:
bash
# Get cluster configuration
terraform output kubectl_config
# Get database connection info
terraform output rds_endpoint
terraform output rds_secret_arn
# Get Redis connection info
terraform output elasticache_primary_endpoint
terraform output elasticache_secret_arn
# Get ECR repository URLs
terraform output ecr_repository_urls
Connecting to EKS#
bash
# Configure kubectl
eval $(terraform output -raw kubectl_config)
# Verify connection
kubectl get nodes
kubectl get pods -n iris
Smoke Tests#
Automated smoke tests run every 5 minutes via CloudWatch Events:
bash
# Invoke manually
aws lambda invoke --function-name iris-staging-smoke-test response.json
cat response.json
Architecture#
text
┌─────────────────────────────────────────────────────────────────┐
│ VPC (10.110.0.0/16) │
│ ┌─────────────────────────────────────────────────────────────┐│
│ │ Public Subnets ││
│ │ ┌─────────┐ ┌─────────┐ ┌─────────┐ ││
│ │ │ NAT GW │ │ ALB │ │ IGW │ ││
│ │ └─────────┘ └─────────┘ └─────────┘ ││
│ └─────────────────────────────────────────────────────────────┘│
│ ┌─────────────────────────────────────────────────────────────┐│
│ │ Private Subnets ││
│ │ ┌─────────────────────────────────────────────────────────┐││
│ │ │ EKS Node Group │││
│ │ │ ┌─────────┐ ┌─────────┐ ┌─────────┐ ┌─────────┐ │││
│ │ │ │iris-api │ │convers. │ │ memory │ │ agent │ │││
│ │ │ │ voice │ │ worker │ │││
│ │ │ └─────────┘ └─────────┘ └─────────┘ └─────────┘ │││
│ │ └─────────────────────────────────────────────────────────┘││
│ └─────────────────────────────────────────────────────────────┘│
│ ┌─────────────────────────────────────────────────────────────┐│
│ │ Database Subnets ││
│ │ ┌─────────────────────┐ ┌─────────────────────┐ ││
│ │ │ RDS PostgreSQL │ │ ElastiCache Redis │ ││
│ │ │ (Multi-AZ) │ │ (Multi-AZ) │ ││
│ │ └─────────────────────┘ └─────────────────────┘ ││
│ └─────────────────────────────────────────────────────────────┘│
└─────────────────────────────────────────────────────────────────┘
Security#
- All data encrypted at rest using KMS
- VPC endpoints for AWS services (ECR, S3, Secrets Manager)
- Network policies restrict pod communication
- IRSA for fine-grained IAM permissions
- Transit encryption for RDS and ElastiCache
Monitoring#
- CloudWatch Container Insights
- RDS Performance Insights
- VPC Flow Logs
- CloudWatch Alarms for:
- RDS CPU, memory, storage, connections
- ElastiCache CPU, memory, connections, evictions
- Smoke test failures
Cost Optimization (Staging)#
- SPOT instances for EKS nodes
- Single NAT Gateway
- Smaller RDS and ElastiCache instance types
- Shorter log retention periods
- No deletion protection
Deployment#
Automated deployments are handled by the GitHub Actions workflow:
.github/workflows/iris-staging-deploy.yml
Manual deployment:
bash
# Apply Terraform changes
terraform apply
# Deploy to Kubernetes
kubectl apply -k ../../../k8s/iris/overlays/staging
Troubleshooting#
EKS Access Denied#
bash
# Verify AWS identity
aws sts get-caller-identity
# Update kubeconfig
aws eks update-kubeconfig --name iris-staging --region us-east-1
Terraform State Lock#
bash
# Force unlock (use with caution)
terraform force-unlock <LOCK_ID>
Pod Failures#
bash
# Check pod status
kubectl describe pod -n iris <pod-name>
# View logs
kubectl logs -n iris <pod-name>
# Check events
kubectl get events -n iris --sort-by='.lastTimestamp'