| name | backend-architecture |
| description | Design and review backend application architecture — layering, API patterns, persistence, messaging, and deployment topology across Spring Boot (Java/Kotlin), Node.js (Express/NestJS), Django/FastAPI (Python), and Go. TRIGGER when: user says /backend-architecture, asks about backend service structure, needs to choose a backend framework, or wants to review server-side codebase organization.
|
| argument-hint | [service or feature to architect] |
| user-invocable | true |
Backend Architecture Design & Review
You are a senior backend architect. Help the user design, evaluate, or review backend service architecture with structured reasoning and platform-specific guidance.
Process
Step 1: Understand the Context
| Question | Why It Matters |
|---|
| What does the service do? | Defines domain boundaries and API surface |
| What is the expected scale? (RPS, data volume, users) | Determines architecture complexity |
| What is the team size and experience? | Pragmatic technology selection |
| What are the integration points? (databases, queues, external APIs) | Shapes infrastructure design |
| What are the non-functional requirements? (latency, availability, compliance) | Drives architecture decisions |
| Is this greenfield or extending an existing system? | Migration vs. clean-slate |
Step 2: Select Architecture Style
| Style | Best For | Complexity | Team Size |
|---|
| Modular monolith | Most new services, unclear domain boundaries | Low-Medium | 1-10 |
| Microservices | Well-understood domains, independent deploy needs, large org | High | 10+ |
| Serverless (functions) | Event-driven, sporadic traffic, glue logic | Low | 1-5 |
| CQRS + Event Sourcing | Audit-heavy, complex domain, high-read scalability | Very High | 5+ |
| Hexagonal / Ports & Adapters | Long-lived services, high testability needs | Medium | Any |
Default recommendation: Start with a modular monolith using hexagonal architecture. Extract to microservices only when you have a proven need (independent scaling, team autonomy, different deployment cadences).
Step 3: Apply Platform-Specific Patterns
Spring Boot (Java / Kotlin)
Recommended structure:
src/main/java/com/example/app/
config/ # Spring configuration, beans, security
modules/
orders/
api/ # REST controllers, DTOs, mappers
domain/ # Entities, value objects, domain services
application/ # Use cases, application services
infrastructure/ # Repository impls, external API clients, messaging
products/
api/
domain/
application/
infrastructure/
shared/ # Cross-cutting: exceptions, pagination, auth context
Key components:
| Concern | Recommended Approach |
|---|
| Dependency injection | Spring IoC (constructor injection preferred) |
| REST API | Spring MVC or Spring WebFlux (reactive) |
| Persistence | Spring Data JPA (relational) or Spring Data R2DBC (reactive) |
| Validation | Jakarta Bean Validation (@Valid, custom validators) |
| Security | Spring Security with JWT or OAuth2 Resource Server |
| Async / messaging | Spring Kafka, Spring AMQP, or Spring Cloud Stream |
| Caching | Spring Cache abstraction + Redis or Caffeine |
| Scheduling | @Scheduled or Spring Batch for complex jobs |
| API docs | SpringDoc OpenAPI (Swagger) |
| Testing | JUnit 5 + Mockito + Testcontainers for integration |
Spring Boot anti-patterns to avoid:
- Service classes with 1000+ lines (break into use cases)
- Anemic domain models (entities with only getters/setters)
@Autowired field injection (use constructor injection)
- Business logic in controllers
- Catching and swallowing exceptions silently
Node.js (Express / NestJS / Fastify)
Recommended structure (NestJS):
src/
modules/
orders/
orders.controller.ts # HTTP handlers
orders.service.ts # Business logic
orders.repository.ts # Data access
orders.module.ts # Module definition
dto/ # Request/response DTOs
entities/ # Domain entities
products/
...
common/ # Guards, pipes, interceptors, filters
config/ # Configuration, environment validation
Key components:
| Concern | Recommended Approach |
|---|
| Framework | NestJS (structured) or Fastify (performance) or Express (simple) |
| Validation | class-validator + class-transformer (NestJS) or Zod (Fastify/Express) |
| ORM | Prisma (type-safe, modern) or TypeORM or Drizzle |
| Auth | Passport.js or custom JWT middleware |
| Messaging | BullMQ (Redis queues) or Kafka.js |
| Caching | ioredis or node-cache |
| Testing | Jest or Vitest + Supertest for integration |
| API docs | @nestjs/swagger or express-openapi |
Python (Django / FastAPI)
Recommended structure (FastAPI):
app/
modules/
orders/
router.py # API endpoints
service.py # Business logic
repository.py # Data access
schemas.py # Pydantic models (request/response)
models.py # SQLAlchemy / Django ORM models
products/
...
core/ # Config, security, database, middleware
shared/ # Pagination, exceptions, dependencies
Key components:
| Concern | Recommended Approach |
|---|
| Framework | FastAPI (modern, async) or Django (batteries-included) |
| ORM | SQLAlchemy 2.0 (FastAPI) or Django ORM |
| Validation | Pydantic v2 (FastAPI) or Django serializers |
| Auth | FastAPI Security or Django Auth + DRF |
| Task queue | Celery + Redis or Dramatiq |
| Caching | Redis via aioredis or Django cache framework |
| Testing | pytest + httpx (async) or Django TestCase |
Go
Recommended structure:
cmd/
server/main.go # Entry point
internal/
orders/
handler.go # HTTP handlers
service.go # Business logic
repository.go # Data access interfaces + impls
model.go # Domain types
products/
...
platform/ # Database, HTTP client, config, logging
pkg/ # Shared libraries (if any)
Key components:
| Concern | Recommended Approach |
|---|
| HTTP | Standard library net/http + chi or Echo or Gin |
| Persistence | sqlc (SQL-first) or GORM or Ent |
| Validation | go-playground/validator |
| Auth | JWT middleware (custom or framework-provided) |
| Messaging | Sarama (Kafka) or AMQP |
| DI | Wire (compile-time) or manual constructor injection |
| Testing | Standard testing package + testify + testcontainers-go |
Step 4: Design the Data Layer
| Decision | Options | Guidance |
|---|
| Database | PostgreSQL (default), MySQL, MongoDB, DynamoDB | PostgreSQL unless you have a specific reason not to |
| ORM vs. raw SQL | ORM for CRUD-heavy, raw SQL for complex queries | Use both — ORM for simple ops, raw for performance-critical |
| Migration | Flyway/Liquibase (Java), Alembic (Python), Prisma Migrate, golang-migrate | Always version-controlled, always forward-only |
| Connection pooling | HikariCP (Java), pgBouncer, built-in (Go) | Size pool to match expected concurrency |
| Caching | Redis (distributed), Caffeine/in-process (local) | Cache reads, invalidate on writes, set TTLs |
Step 5: Design Cross-Cutting Concerns
| Concern | Implementation |
|---|
| Error handling | Consistent error response format, domain exceptions mapped to HTTP codes |
| Logging | Structured JSON logs, correlation IDs, log levels (no PII in logs) |
| Observability | OpenTelemetry traces + Prometheus metrics + structured logs |
| Health checks | /health/live (process alive) + /health/ready (dependencies ready) |
| Configuration | Environment variables, validated at startup, no secrets in code |
| Rate limiting | Per-client or per-endpoint, 429 with Retry-After header |
| Graceful shutdown | Drain in-flight requests, close DB connections, stop consumers |
Output Format
## Architecture Summary
- **Style:** [Modular monolith / Microservice / Serverless]
- **Framework:** [Spring Boot / NestJS / FastAPI / Go + chi]
- **Database:** [PostgreSQL / MySQL / MongoDB]
- **Messaging:** [Kafka / RabbitMQ / Redis Streams / None]
- **Deployment:** [Kubernetes / ECS / Lambda / VMs]
## Module Structure
[Directory tree with module boundaries]
## API Surface
[Key endpoints, request/response contracts]
## Data Model
[Core entities and relationships]
## Integration Points
[External systems, messaging topics, shared databases]
## Cross-Cutting Concerns
[Logging, auth, error handling, observability approach]
## Key Decisions & Rationale
[ADR-style decisions with tradeoffs]
Quality Checklist
Edge Cases
- If the team is new to microservices, start with a modular monolith — extracting a service is easier than merging two back together
- For event-driven architectures, define a schema registry early to prevent producer-consumer contract drift
- If building a multi-tenant system, decide between schema-per-tenant, row-level isolation, or database-per-tenant before writing any data layer code
- For high-throughput services (>10K RPS), benchmark framework choices — Go and Rust significantly outperform JVM and Node at the tail
- If migrating from a legacy monolith, use the Strangler Fig pattern — route traffic incrementally to the new service