Python Backend Architecture

Production-ready patterns for Python backend systems following Clean Architecture, Domain-Driven Design, and modern distributed system patterns.

Quick Reference

Core Principles

Architecture Layers

┌─────────────────────────────────────────────┐
│           Presentation Layer                │  ← API routes, schemas, middleware
├─────────────────────────────────────────────┤
│           Application Layer                 │  ← Use cases, DTOs, interfaces
├─────────────────────────────────────────────┤
│             Domain Layer                    │  ← Entities, value objects, events
├─────────────────────────────────────────────┤
│          Infrastructure Layer               │  ← DB, cache, messaging, external APIs
└─────────────────────────────────────────────┘

Dependency Rule: Dependencies point inward. Domain has no external dependencies.

Key Principles

1. Dependency Injection – Depend on abstractions, not implementations
2. Single Responsibility – One reason to change per class/function
3. Explicit over Implicit – No hidden dependencies or magic
4. Async by Default – Use async/await for all I/O operations
5. Type Hints Everywhere – All public APIs fully typed
6. Immutability – Prefer immutable data structures (especially value objects)
7. Early Return – Reduce nesting with guard clauses

Architecture Decision Flowchart

flowchart TD
    A[New Feature] --> B{Complex domain logic?}
    B -->|Yes| C[Use Domain-Driven Design]
    B -->|No| D{Multiple data sources?}

    C --> E[Define Entities & Value Objects]
    E --> F[Create Repository Interfaces]
    F --> G[Implement Use Cases]

    D -->|Yes| H[Use Repository Pattern]
    D -->|No| I{Transaction across aggregates?}

    H --> G

    I -->|Yes| J[Use Unit of Work]
    I -->|No| K[Direct repository calls]

    J --> G
    K --> G

    G --> L{Cross-service operation?}
    L -->|Yes| M[Use Saga Pattern]
    L -->|No| N[Single transaction]

    M --> O[Publish Domain Events]
    N --> O

    O --> P{Needs real-time updates?}
    P -->|Yes| Q[Event-Driven Architecture]
    P -->|No| R[Request-Response]

Common Patterns Summary

Domain Layer

• Entities: Objects with identity and lifecycle. See domain-driven-design.md
• Value Objects: Immutable objects compared by value. See domain-driven-design.md
• Domain Events: Capture what happened. See domain-driven-design.md
• Aggregates: Consistency boundaries. See domain-driven-design.md

Application Layer

• Use Cases: Single operation orchestrators
• DTOs: Data transfer between layers
• Interfaces: Abstract ports for infrastructure

Infrastructure Layer

• Repositories: Data access abstraction. See architecture-patterns.md
• Unit of Work: Transaction management. See architecture-patterns.md
• Event Bus: Publish/subscribe messaging. See event-driven.md

Presentation Layer

• Routers: HTTP endpoint handlers. See api-design.md
• Schemas: Request/response validation. See api-design.md
• Dependencies: DI factory functions. See api-design.md

Error Handling Strategy

# Domain errors (business logic violations)
class DomainError(Exception): pass
class EntityNotFoundError(DomainError): pass
class BusinessRuleViolationError(DomainError): pass

# Infrastructure errors (external system failures)
class InfrastructureError(Exception): pass
class DatabaseConnectionError(InfrastructureError): pass
class ExternalAPIError(InfrastructureError): pass

# Map to HTTP in presentation layer
@router.get("/{id}")
async def get_item(id: UUID, use_case: GetItemUseCase = Depends(...)):
    try:
        return await use_case.execute(id)
    except EntityNotFoundError:
        raise HTTPException(status_code=404, detail="Not found")
    except BusinessRuleViolationError as e:
        raise HTTPException(status_code=400, detail=str(e))
    except InfrastructureError:
        raise HTTPException(status_code=503, detail="Service unavailable")

See code-style.md for full exception hierarchy.

Quick Start Patterns

Basic Repository + Use Case

# 1. Define interface (application layer)
class UserRepository(ABC):
    @abstractmethod
    async def get_by_id(self, id: UUID) -> User | None: pass

    @abstractmethod
    async def save(self, user: User) -> User: pass

# 2. Implement (infrastructure layer)
class PostgresUserRepository(UserRepository):
    def __init__(self, session: AsyncSession):
        self._session = session

    async def get_by_id(self, id: UUID) -> User | None:
        result = await self._session.execute(
            select(UserModel).where(UserModel.id == id)
        )
        model = result.scalar_one_or_none()
        return self._to_entity(model) if model else None

# 3. Use case (application layer)
@dataclass
class GetUserUseCase:
    user_repo: UserRepository

    async def execute(self, user_id: UUID) -> UserDTO:
        user = await self.user_repo.get_by_id(user_id)
        if not user:
            raise EntityNotFoundError("User", str(user_id))
        return UserDTO.from_entity(user)

# 4. Route (presentation layer)
@router.get("/{user_id}")
async def get_user(
    user_id: UUID,
    use_case: GetUserUseCase = Depends(get_user_use_case)
):
    return await use_case.execute(user_id)

Adding Caching

@dataclass
class GetUserUseCase:
    user_repo: UserRepository
    cache: CacheService

    async def execute(self, user_id: UUID) -> UserDTO:
        # Try cache
        cached = await self.cache.get(f"user:{user_id}")
        if cached:
            return UserDTO.parse_raw(cached)

        # Fetch from DB
        user = await self.user_repo.get_by_id(user_id)
        if not user:
            raise EntityNotFoundError("User", str(user_id))

        # Cache result
        dto = UserDTO.from_entity(user)
        await self.cache.set(f"user:{user_id}", dto.json(), ttl=300)
        return dto

See caching.md for advanced patterns.

Publishing Events

@dataclass
class CreateOrderUseCase:
    order_repo: OrderRepository
    event_bus: EventBus

    async def execute(self, command: CreateOrderCommand) -> OrderDTO:
        order = Order.create(user_id=command.user_id, items=command.items)
        saved = await self.order_repo.add(order)

        # Publish event after successful save
        await self.event_bus.publish(OrderCreatedEvent.from_entity(saved))

        return OrderDTO.from_entity(saved)

See event-driven.md for event handlers and saga patterns.

Production Readiness

Before deploying, verify:

• Observability: Logging, metrics, tracing configured. See observability.md
• Security: Auth, input validation, rate limiting. See security.md
• Testing: Unit, integration, E2E tests passing. See testing.md
• Resilience: Circuit breakers, retries, graceful shutdown. See microservices.md
• Deployment: Docker, K8s, health checks configured. See deployment.md