Infrastructure as Code

Overview

MCP Server LangGraph uses Terraform as the Infrastructure as Code (IaC) tool to provision and manage cloud infrastructure across GCP and AWS. This modular approach enables consistent, repeatable, and version-controlled infrastructure deployments.

GCP Modules

6 production-ready Terraform modules for GKE

Multi-Environment

Dev, staging, and production configurations

State Management

GCS and S3 backends with locking

AWS Modules

EKS, RDS, ElastiCache modules (96/100 maturity)

Architecture Principles

Modularity

Reusable modules with clear interfaces

Each cloud resource is a self-contained module
Modules can be composed for different environments
No hard-coded values

Environment Parity

Consistent configs across dev/staging/prod

Same module versions
Parameter-driven differences (size, HA)
Promote confidence through consistency

State Isolation

Separate state per environment

Development: dev-tfstate bucket/prefix
Staging: staging-tfstate bucket/prefix
Production: prod-tfstate bucket/prefix

Security by Default

Zero-trust networking and IAM

Private clusters (no public IPs)
Workload Identity (GCP) / IRSA (AWS)
Encryption at rest and in transit

Module Catalog

GCP Infrastructure

GCP VPC (terraform/modules/gcp-vpc)

Purpose: VPC-native networking for GKE with Cloud NATResources:

VPC with custom subnets
Secondary IP ranges for pods and services
Cloud NAT with static IPs
Firewall rules (IAP, health checks)
Private Service Connection for Cloud SQL

Outputs: Network ID, subnet names, NAT IPsBest for: All GKE deployments

GKE Autopilot (terraform/modules/gke-autopilot)

Purpose: Fully managed Kubernetes clusterFeatures:

Pay-per-pod pricing (40-60% savings)
Auto-scaling, auto-repair, auto-upgrade
Workload Identity enabled
Binary Authorization ready
Security Posture Dashboard
Dataplane V2 (eBPF-based)

Outputs: Cluster name, endpoint, kubectl commandsBest for: Production workloads

Cloud SQL (terraform/modules/cloudsql)

Purpose: Managed PostgreSQL with HAFeatures:

Regional HA (99.95% SLA)
Point-in-time recovery
Read replicas
Query Insights
Automatic backups

Outputs: Connection name, private IP, credentialsBest for: Stateful applications

Memorystore (terraform/modules/memorystore)

Purpose: Managed Redis with HAFeatures:

STANDARD_HA tier (99.9% SLA)
Automatic failover
Persistence (RDB + AOF)
Cross-region replicas

Outputs: Host, port, auth stringBest for: Session storage, caching

Workload Identity (terraform/modules/gke-workload-identity)

Purpose: IAM for Kubernetes podsFeatures:

No service account keys
GCP service account per workload
Granular IAM bindings
Automatic credential injection

Outputs: Service account emails, K8s manifestsBest for: All GKE workloads

Backend Setup (terraform/backend-setup-gcp)

Purpose: GCS buckets for Terraform stateFeatures:

Versioning enabled
Object lifecycle policies
Access logging
Encryption at rest

Outputs: Bucket names, backend configBest for: Initial setup (run once)

AWS Infrastructure

AWS infrastructure modules for EKS are already at 96/100 maturity and production-ready. See existing documentation for details.

Available modules:

VPC with public/private subnets
EKS cluster with managed node groups
RDS PostgreSQL with Multi-AZ
ElastiCache Redis cluster
IRSA for pod IAM

Directory Structure

terraform/
├── backend-setup-gcp/          # State bucket bootstrap
│   ├── main.tf                 # GCS buckets with versioning
│   ├── variables.tf            # Project ID, region, bucket prefix
│   └── README.md               # Setup instructions
│
├── modules/                    # Reusable modules
│   ├── gcp-vpc/                # VPC networking
│   ├── gke-autopilot/          # GKE cluster
│   ├── cloudsql/               # PostgreSQL
│   ├── memorystore/            # Redis
│   └── gke-workload-identity/  # IAM bindings
│
└── environments/               # Environment configs
    ├── gcp-dev/                # Development
    │   ├── main.tf             # Module composition
    │   ├── variables.tf        # Input variables
    │   ├── terraform.tfvars    # Dev-specific values
    │   └── backend.tf          # State backend config
    │
    ├── gcp-staging/            # Staging (production-like)
    └── gcp-prod/               # Production (HA, multi-region)

State Management

Backend Configuration

GCP (GCS)
AWS (S3)

terraform {
  backend "gcs" {
    bucket  = "PROJECT-terraform-state"
    prefix  = "environments/production"
  }
}

Features:

Automatic state locking
Versioning enabled
Encrypted at rest
Access logs

Setup: See Backend Setup Guide

Environment Strategy

Development

Purpose: Testing and iterationConfiguration:

Zonal cluster (1 zone)
Smaller instances
No read replicas
BASIC Redis tier

Cost: ~$100/month

Staging

Purpose: Pre-production validationConfiguration:

Regional cluster (3 zones)
Production-like sizing
HA databases
Full monitoring

Cost: ~$310/month

Production

Purpose: Live workloadsConfiguration:

Regional cluster (3 zones)
HA for all components
Read replicas
Full observability
Disaster recovery

Cost: ~$970/month (Autopilot)

Deployment Workflow

Initialize Backend

cd terraform/backend-setup-gcp
terraform init
terraform apply

Creates GCS buckets for state storage.

Configure Environment

cd terraform/environments/gcp-prod

# Edit terraform.tfvars
vim terraform.tfvars

Set project ID, region, cluster name, etc.

Plan Infrastructure

terraform init
terraform plan -out=tfplan

Review planned changes.

Apply Changes

terraform apply tfplan

Provision infrastructure (~15-30 minutes for full stack).

Verify Deployment

# Get kubectl credentials
gcloud container clusters get-credentials CLUSTER_NAME \
  --region REGION \
  --project PROJECT_ID

# Verify nodes
kubectl get nodes

Best Practices

Module Design

DO: Use semantic versioning for module releases

DO: Validate all inputs with Terraform validation blocks

DO: Provide comprehensive outputs for module consumers

DO: Document all variables with descriptions and examples

DON’T: Hard-code values inside modules

DON’T: Mix resource types in a single module

DON’T: Use count/for_each for critical resources (hard to change)

State Management

DO: Use separate state per environment

DO: Enable state locking (GCS automatic, S3 with DynamoDB)

DO: Enable versioning on state buckets

DO: Restrict state bucket access (principle of least privilege)

DON’T: Commit state files to Git

DON’T: Share state across unrelated infrastructure

Security

DO: Use Workload Identity (GCP) or IRSA (AWS)

DO: Enable encryption at rest for all data stores

DO: Use private clusters (no public IPs on nodes)

DO: Rotate credentials regularly via Secret Manager/Secrets Manager

DON’T: Use service account keys or IAM access keys

DON’T: Store secrets in Terraform state (use external secret stores)

Cost Optimization

GKE Autopilot (40-60% savings)

Strategy: Pay-per-pod vs. paying for idle nodesImplementation:

Use GKE Autopilot instead of Standard
Right-size pod requests (use VPA)
Enable autoscaling

Savings: $200-400/month per environment

Committed Use Discounts (25-52%)

Strategy: Commit to 1-year or 3-year usageImplementation:

# Apply CUD to all compute resources
# Purchase via GCP Console: Billing → Commitments

Savings:

242/month (25%) with 1-year,

466/month (52%) with 3-year

Resource Right-Sizing

Strategy: Match resources to actual usageImplementation:

Monitor actual CPU/memory usage
Downgrade oversized Cloud SQL/Redis instances
Remove unused read replicas

Savings: $50-150/month

Development Auto-Shutdown

Strategy: Turn off dev environments after hoursImplementation:

# Cloud Scheduler: Scale to 0 at 6 PM, scale up at 6 AM
gcloud scheduler jobs create http scale-down-dev \
  --schedule="0 18 * * 1-5" \
  --uri="..." \
  --message-body='{"desiredNodeCount":0}'

Savings: $50-70/month per dev environment

Migration Paths

From Manual Infrastructure

Import Existing Resources

terraform import google_container_cluster.main projects/PROJECT/locations/REGION/clusters/CLUSTER_NAME

Generate Terraform Config

Use terraform plan to match existing state

Gradually Adopt IaC

Start with new resources, migrate existing over time

From Other IaC Tools

From Pulumi/CDK:

Export resource definitions
Convert to HCL syntax
Import state

From CloudFormation (AWS):

Use former2 to generate Terraform
Review and customize
Import stacks

Troubleshooting

State Lock Conflicts

Symptom: Error acquiring the state lockCause: Previous Terraform run didn’t release lock (crash, Ctrl+C)Solution:

# GCS (automatic after 1 minute)
# Wait or manually delete lock in GCS console

# S3 + DynamoDB
aws dynamodb delete-item \
  --table-name terraform-locks \
  --key '{"LockID": {"S": "BUCKET/PATH/terraform.tfstate"}}'

Backend Initialization Errors

Symptom: Backend configuration changedSolution:

terraform init -reconfigure

Module Version Conflicts

Symptom: Module ... does not matchSolution:

terraform init -upgrade

Permission Denied Errors

Symptom: Error 403: The caller does not have permissionSolution: Grant required IAM roles

# For GCP
gcloud projects add-iam-policy-binding PROJECT_ID \
  --member="user:YOUR_EMAIL" \
  --role="roles/editor"

GCP Terraform Modules

Complete guide to all 6 GCP modules

Backend Setup

Initialize Terraform state backends

Multi-Environment

Dev, staging, prod configurations

GKE Production

Deploy to production with Terraform + Kustomize

Next Steps

Set Up State Backend

Backend Setup Guide →

Review Module Documentation

Terraform GCP Modules →

Choose Environment Strategy

Multi-Environment Guide →

Deploy to Production

GKE Production Deployment →

Getting Started

Deployment Options

LangGraph Platform

Kubernetes - GKE

Kubernetes - EKS & AKS

Kubernetes - Best Practices

Monitoring & Observability

Advanced Deployment

Configuration

Operations

​Overview

GCP Modules

Multi-Environment

State Management

AWS Modules

​Architecture Principles

​Module Catalog

​GCP Infrastructure

​AWS Infrastructure

​Directory Structure

​State Management

​Backend Configuration

​Environment Strategy

Development

Staging

Production

​Deployment Workflow

​Best Practices

​Module Design

​State Management

​Security

​Cost Optimization

​Migration Paths

​From Manual Infrastructure

​From Other IaC Tools

​Troubleshooting

​Related Documentation

GCP Terraform Modules

Backend Setup

Multi-Environment

GKE Production

​Next Steps

Overview

Architecture Principles

Module Catalog

GCP Infrastructure

AWS Infrastructure

Directory Structure

State Management

Backend Configuration

Environment Strategy

Deployment Workflow

Best Practices

Module Design

State Management

Security

Cost Optimization

Migration Paths

From Manual Infrastructure

From Other IaC Tools

Troubleshooting

Related Documentation

Next Steps