AgentSkillsCN

Layered Stacks Architecture

以基础架构 → 平台 → 工作负载的分层架构，实现跨栈引用与状态隔离

SKILL.md
--- frontmatter
name: Layered Stacks Architecture
description: Foundation → Platform → Workloads separation with cross-stack references and state isolation

Layered Stacks Architecture

Overview

Layered stacks separate infrastructure into independent, loosely-coupled layers that can be deployed and managed separately. This reduces blast radius, enables team autonomy, and improves deployment velocity.

mermaid
graph TB
    subgraph "Layer 3: Workloads"
        App1[Application 1]
        App2[Application 2]
        App3[Application 3]
    end
    
    subgraph "Layer 2: Platform"
        EKS[EKS Cluster]
        RDS[(RDS Database)]
        Redis[(ElastiCache)]
    end
    
    subgraph "Layer 1: Foundation"
        VPC[VPC / Networking]
        IAM[IAM Roles]
        KMS[KMS Keys]
    end
    
    App1 --> EKS
    App2 --> EKS
    App3 --> RDS
    EKS --> VPC
    RDS --> VPC
    EKS --> IAM

Layer Definitions

LayerContentsChange FrequencyTeam
FoundationVPC, networking, IAM baseline, KMSRarelyPlatform
PlatformEKS, RDS, caching, monitoringMonthlyPlatform
WorkloadsApplications, CI/CD, configsDailyApp teams

Best Practices

  1. •Clear boundaries - Each layer has distinct responsibilities
  2. •Output only what's needed - Minimize cross-layer coupling
  3. •Separate state files - Independent deployment
  4. •Use data sources - Reference other layers via outputs
  5. •Version layer interfaces - Treat outputs as APIs
  6. •Different approval processes - Foundation needs more review

Example 1: Terraform - Three-Layer Architecture

Complete three-layer deployment with remote state references.

📁 Location: terraform/examples/layered-stacks/

Directory Structure

code
terraform/
├── layers/
│   ├── 01-foundation/
│   │   ├── main.tf      # VPC, IAM, KMS
│   │   └── outputs.tf   # Outputs for other layers
│   ├── 02-platform/
│   │   ├── main.tf      # EKS, RDS (references foundation)
│   │   └── outputs.tf
│   └── 03-workloads/
│       └── main.tf      # Apps (references platform)

Cross-Layer References

hcl
# 02-platform/main.tf
# Reference foundation layer outputs via data source
data "terraform_remote_state" "foundation" {
  backend = "s3"
  config = {
    bucket = "my-terraform-state"
    key    = "foundation/${var.environment}/terraform.tfstate"
    region = "us-east-1"
  }
}

# Use foundation outputs
module "eks" {
  source = "../../modules/eks"
  
  vpc_id     = data.terraform_remote_state.foundation.outputs.vpc_id
  subnet_ids = data.terraform_remote_state.foundation.outputs.private_subnet_ids
}

Example 2: CDK - Stack Dependencies

CDK stacks with explicit dependencies and cross-stack references.

📁 Location: cdk/examples/layered-stacks/

Stack Structure

typescript
const app = new cdk.App();

// Layer 1: Foundation
const foundationStack = new FoundationStack(app, 'Foundation', {
  env: prodEnv,
});

// Layer 2: Platform (depends on foundation)
const platformStack = new PlatformStack(app, 'Platform', {
  env: prodEnv,
  vpc: foundationStack.vpc,          // Cross-stack reference
  kmsKey: foundationStack.kmsKey,
});
platformStack.addDependency(foundationStack);

// Layer 3: Workloads (depends on platform)
const workloadsStack = new WorkloadsStack(app, 'Workloads', {
  env: prodEnv,
  cluster: platformStack.cluster,    // Cross-stack reference
  database: platformStack.database,
});
workloadsStack.addDependency(platformStack);

State Separation

LayerState PathRationale
Foundationfoundation/{env}/terraform.tfstateRarely changes
Platformplatform/{env}/terraform.tfstateTeam-owned
Workloadsworkloads/{app}/{env}/terraform.tfstatePer-app isolation

Validation Checklist

  • • Layers have clear, documented boundaries
  • • State files separated per layer
  • • Cross-layer dependencies use remote state/imports
  • • Foundation has stricter approval process
  • • Layer outputs treated as stable APIs
  • • Changes tested layer-by-layer

Related Skills