kennethreitz d9f8883e66 Harden background shutdown and Redis rate limiting; gate CI on lint/types (#610)
## Summary

Three targeted improvements surfaced by an audit of the framework — two
reliability fixes and one CI gap. Ships as v7.1.3.

### 1. Redis rate-limit atomicity (bug fix)
`RedisBackend.hit` / `AsyncRedisBackend.ahit` performed `INCR` then
`EXPIRE` as two separate round-trips. If the process died (or the
connection dropped) between them, the key never received a TTL —
permanently rate-limiting that client until manual eviction. Both now
use a single atomic Lua `EVAL` that increments and, on the first hit of
a window, sets the expiry.

### 2. Background task graceful shutdown
`api.background`'s `ThreadPoolExecutor` was never drained, so in-flight
fire-and-forget tasks were abandoned on shutdown. Added
`BackgroundQueue.shutdown()` and a registered shutdown event handler
that drains the pool off the event loop.

### 3. CI now enforces lint + types
`ruff` and `mypy` previously ran only in the manual release guard, so
lint/type regressions could land on `main`. New `Lint & Types` workflow
runs both on every push and PR. The tree passes both today.

## Tests
- New `tests/test_v713_improvements.py` (7 tests): background
drain/reject-after-shutdown, app-level shutdown drain, and Redis
atomic-eval behavior (including a fake that fails if `incr`/`expire` are
called directly).
- Updated the existing fake Redis in `tests/test_round4.py` to implement
`eval`.
- Full suite: 740 passed, ruff clean, mypy clean.

## Notes
Also corrected the stale `[Unreleased]` compare link in the changelog
(pointed at v7.1.1 despite v7.1.2 shipping).

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-01 12:22:31 -04:00
2026-06-30 13:39:18 -04:00
2026-06-30 13:44:33 -04:00
2018-10-11 13:57:55 -04:00
2026-06-30 12:29:49 -04:00
2026-06-29 14:34:28 -04:00
2024-10-24 02:27:32 +02:00
2026-06-30 12:29:49 -04:00
2026-06-30 13:21:14 -04:00

Responder

A familiar HTTP Service Framework for Python, powered by Starlette.

import responder

api = responder.API()

@api.route("/{greeting}")
async def greet_world(req, resp, *, greeting):
    resp.text = f"{greeting}, world!"

if __name__ == "__main__":
    api.run()
$ pip install responder

That's it. Supports Python 3.11+.

The Basics

  • resp.text sends back text. resp.html sends back HTML. resp.content sends back bytes.
  • resp.media sends back JSON (or YAML, with content negotiation).
  • resp.file("path.pdf") serves a file with automatic content-type detection.
  • File(...) uploads use streamed UploadFile objects; await file.save(path) writes them to disk.
  • req.headers is case-insensitive. req.params gives you query parameters.
  • Both sync and async views work — the async is optional.

Highlights

# Type-safe route parameters
@api.route("/users/{user_id:int}")
async def get_user(req, resp, *, user_id):
    resp.media = {"id": user_id}

# HTTP method filtering
@api.route("/items", methods=["POST"])
async def create_item(req, resp):
    data = await req.media()
    resp.media = {"created": data}

# Route-local hooks
def require_json(req, resp):
    if not req.is_json:
        resp.status_code = 415
        resp.media = {"error": "JSON required"}

@api.post("/events", before=require_json)
async def events(req, resp):
    resp.media = await req.media()

# Local dependencies
from responder import Depends

def current_user(req):
    return req.headers.get("X-User")

@api.get("/me")
def me(req, resp, *, user=Depends(current_user)):
    resp.media = {"user": user}

# Side-effect dependencies
@api.get("/ready", dependencies=[Depends(current_user)])
def ready(req, resp):
    resp.media = {"ready": True}

# Route-level auth with OpenAPI security
from responder.ext.auth import BearerAuth

auth = BearerAuth(tokens=["s3cret"])

@api.get("/private", auth=auth)
def private(req, resp, *, user):
    resp.media = {"user": user}

# Named policies keep route auth intent reusable
admin = api.policy("admin", auth.requires("admin"))

@api.get("/admin", auth=admin)
def admin(req, resp, *, user):
    resp.media = {"user": user}

# App-level auth with public route opt-out
secured_api = responder.API(auth=auth)

@secured_api.get("/health", auth=None)
def health(req, resp):
    resp.media = {"ok": True}

# Optional auth accepts anonymous requests but still rejects bad credentials
optional_auth = auth.optional()

@api.get("/maybe", auth=optional_auth)
def maybe(req, resp, *, user):
    resp.media = {"user": user}

# Class-based views
@api.route("/things/{id}")
class ThingResource:
    def on_get(self, req, resp, *, id):
        resp.media = {"id": id}
    def on_post(self, req, resp, *, id):
        resp.text = "created"

# Before-request hooks (auth, rate limiting, etc.)
@api.route(before_request=True)
def check_auth(req, resp):
    if not req.headers.get("Authorization"):
        resp.status_code = 401
        resp.media = {"error": "unauthorized"}

# Custom error handling
@api.exception_handler(ValueError)
async def handle_error(req, resp, exc):
    resp.status_code = 400
    resp.media = {"error": str(exc)}

# Problem-details enrichment
def problem_handler(payload, request, exc):
    payload["type"] = f"https://example.com/problems/{payload['status']}"
    payload["instance"] = request.url.path

api = responder.API(problem_handler=problem_handler, request_id=True)

# Lifespan events
from contextlib import asynccontextmanager

@asynccontextmanager
async def lifespan(app):
    print("starting up")
    yield
    print("shutting down")

api = responder.API(lifespan=lifespan)

# GraphQL
import graphene
api.graphql("/graphql", schema=graphene.Schema(query=Query))

# WebSockets
@api.route("/ws", websocket=True)
async def websocket(ws):
    await ws.accept()
    while True:
        name = await ws.receive_text()
        await ws.send_text(f"Hello {name}!")

# Mount WSGI/ASGI apps
from flask import Flask
flask_app = Flask(__name__)
api.mount("/flask", flask_app)

# Background tasks
@api.route("/work")
def do_work(req, resp):
    @api.background.task
    def process():
        import time; time.sleep(10)
    process()
    resp.media = {"status": "processing"}

Built-in OpenAPI docs, cookie-based sessions, gzip compression, static file serving, Jinja2 templates, and a production uvicorn server. Install responder[server] to add Granian for production ASGI serving.

Route convertors: str, int, float, uuid, path.

Framework errors use RFC 9457-style application/problem+json responses by default; pass problem_details=False to keep the legacy error format. Pass problem_handler= to enrich those payloads; request IDs are included when request_id=True or structured logging is enabled. OpenAPI documents the shared ProblemDetails schema, and generated clients expose it as APIError.problem. Handlers can send their own problem details with resp.problem(...), and use resp.created(...) / resp.no_content() for common REST responses.

For a polished, practical app, see examples/todo.py; for a tiny local-command example, see examples/fortunes.py. For the full contract fixture, see examples/atelier.py; the test suite imports it, validates its OpenAPI document, and drives a generated client against it.

Documentation

https://responder.kennethreitz.org

Languages
Python 99.7%
HTML 0.2%