python-patterns

Python Backend Patterns Skill

Purpose

Best practices for Python backend development: layered architecture, async I/O, dependency injection, and clear separation between HTTP handling, business logic, and data access. The concrete examples below use FastAPI, SQLAlchemy, and Pydantic, but the patterns apply to any Python web framework, ORM, and validation library.

Auto-Invoke Triggers

• Creating backend routes / endpoints
• Working with ORM models
• Implementing async operations
• Creating request/response validation schemas

---

Layer Responsibilities

Layer	Responsibility
Endpoints	HTTP handling, request/response
Services	Business logic, orchestration
Repositories	Data access, queries
Models	Database schema
Schemas	Data validation, serialization

These layers are framework-agnostic — keep HTTP concerns, business rules, and data access in separate modules regardless of which framework/ORM you use.

---

Example: FastAPI + SQLAlchemy + Pydantic (illustrative)

Illustrative — this is one concrete stack shown as an example. Substitute your framework's equivalents (any ASGI/WSGI framework, ORM, and validation library). The layering and separation-of-concerns patterns above are the reusable part.

Project Structure

app/
└── backend/
    ├── app/
    │   ├── main.py              # FastAPI app initialization
    │   ├── config.py            # Settings (pydantic-settings)
    │   ├── api/v1/endpoints/    # Route handlers
    │   ├── core/                # Security, exceptions
    │   ├── models/              # SQLAlchemy models
    │   ├── schemas/             # Pydantic schemas
    │   ├── services/            # Business logic
    │   ├── repositories/        # Data access
    │   └── db/session.py        # Database session
    ├── tests/
    ├── alembic.ini
    └── requirements.txt

---

Async Database Patterns

Session Management

• Use async_sessionmaker for async sessions
• Use dependency injection for session
• Commit in dependency, rollback on exception
• Use expire_on_commit=False for response data