Fastapi

FastAPI — Python web APIs

Overview

FastAPI is an async Python web framework with automatic request validation (Pydantic v2) and OpenAPI/Swagger docs out of the box. Use it for REST/JSON backends and for serving ML models behind HTTP. Pairs with [[pydeseq2]]-style scientific code or any service layer; containerize with [[docker]], set up the project with [[modern-python]].

uv add fastapi "uvicorn[standard]"
uv run uvicorn app:app --reload          # dev server, interactive docs at /docs

Minimal app with validation

from fastapi import FastAPI, HTTPException, Query
from pydantic import BaseModel, Field

app = FastAPI(title="Example API")

class Item(BaseModel):
    name: str
    price: float = Field(gt=0)
    tags: list[str] = []

DB: dict[int, Item] = {}

@app.get("/items/{item_id}")
def get_item(item_id: int) -> Item:        # path param, type-coerced + validated
    if item_id not in DB:
        raise HTTPException(status_code=404, detail="not found")
    return DB[item_id]

@app.get("/items")
def list_items(limit: int = Query(10, le=100)) -> list[Item]:   # query param
    return list(DB.values())[:limit]

@app.post("/items/{item_id}", status_code=201)
def create_item(item_id: int, item: Item) -> Item:              # JSON body -> Pydantic
    DB[item_id] = item
    return item

Type hints drive everything: path/query params, JSON body parsing, response serialization, and the OpenAPI schema. Return a Pydantic model or set response_model= to control output.

Async & blocking work

import httpx

@app.get("/proxy")
async def proxy():
    async with httpx.AsyncClient() as client:
        r = await client.get("https://api.example.com/data")
    return r.json()

Use async def for I/O-bound endpoints (await async clients). For blocking CPU/IO work (a sync DB driver, a heavy model) use a plain def endpoint — FastAPI runs it in a threadpool so it won't block the event loop. Never call blocking code inside async def.

Dependency injection

from fastapi import Depends, Header

async def get_db():
    db = Session()
    try:
        yield db
    finally:
        db.close()

def require_token(x_token: str = Header()):
    if x_token != "secret":
        raise HTTPException(401, "bad token")

@app.get("/secure")
def secure(db=Depends(get_db), _=Depends(require_token)):
    return {"ok": True}

Depends resolves and caches per-request, supports yield for setup/teardown, and is the idiomatic way to share DB sessions, auth, and config.

Lifespan (startup/shutdown)

from contextlib import asynccontextmanager

@asynccontextmanager
async def lifespan(app: FastAPI):
    app.state.model = load_model()    # startup: load once
    yield
    app.state.model.close()           # shutdown

app = FastAPI(lifespan=lifespan)

Use the lifespan context manager (the modern replacement for the deprecated @app.on_event("startup")) to load models/connection pools once at boot.

Testing

from fastapi.testclient import TestClient
from app import app

client = TestClient(app)

def test_create_and_get():
    assert client.post("/items/1", json={"name": "x", "price": 2}).status_code == 201
    r = client.get("/items/1")
    assert r.status_code == 200 and r.json()["name"] == "x"

TestClient (sync, Starlette) needs no running server. For async tests use httpx.ASGITransport with httpx.AsyncClient. See [[pytest]] for fixtures/markers.

Gotchas

• Blocking inside async def stalls the whole server — use def endpoints for sync work.
• Pydantic v2: model_config, model_dump(), field_validator (v1 Config/.dict() are gone). Validation errors auto-return HTTP 422 with field detail.
• Production server: uvicorn for a single process; behind a process manager use gunicorn -k uvicorn.workers.UvicornWorker -w 4. Put it behind nginx/a load balancer.
• CORS: browsers need CORSMiddleware configured explicitly.
• Big request bodies / file uploads use UploadFile (streamed), not bytes.

Related

Test with [[pytest]]; containerize and serve via [[docker]]; project tooling in [[modern-python]]; CI in [[github-actions-ci]]; GPU/serverless model serving via [[modal]].