Responder

The Python world certainly doesn't need more web frameworks. But, it does need more creativity, so I thought I'd spread some Hacktoberfest spirit around, bring some of my ideas to the table, and see what I could come up with.

An Example Web Service:

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()

That async declaration is optional.

This gets you a ASGI app, with a production static files server pre-installed, jinja2 templating (without additional imports), and a production webserver based on uvloop, serving up requests with gzip compression automatically.

Testimonials

"Pleasantly very taken with python-responder. @kennethreitz at his absolute best." —Rudraksh M.K.

"Buckle up!" —Tom Christie of APIStar and Django REST Framework

"I love that you are exploring new patterns. Go go go!" — Danny Greenfield, author of Two Scoops of Django

"Love what I have seen while it's in progress! Many features of Responder are from my wishlist for Flask, and it's even faster and even easier than Flask!" — Luna C.

"The most ambitious crossover event in history." —Pablo Cabezas, on Tom Christie joining the project

More Examples

Class-based views (and setting some headers and stuff):

@api.route("/{greeting}")
class GreetingResource:
    def on_request(req, resp, *, greeting):   # or on_get...
        resp.text = f"{greeting}, world!"
        resp.headers.update({'X-Life': '42'})
        resp.status_code = api.status_codes.HTTP_416

Render a template, with arguments:

@api.route("/{greeting}")
def greet_world(req, resp, *, greeting):
    resp.content = api.template("index.html", greeting=greeting)

The api instance is available as an object during template rendering.

Here, you can spawn off a background thread to run any function, out-of-request:

@api.route("/")
def hello(req, resp):

    @api.background.task
    def sleep(s=10):
        time.sleep(s)
        print("slept!")

    sleep()
    resp.content = "processing"

And even serve a GraphQL API:

import graphene

class Query(graphene.ObjectType):
    hello = graphene.String(name=graphene.String(default_value="stranger"))

    def resolve_hello(self, info, name):
        return "Hello " + name

api.add_route("/graph", graphene.Schema(query=Query))

We can then send a query to our service:

>>> requests = api.session()
>>> r = requests.get("http://;/graph", params={"query": "{ hello }"})
>>> r.json()
{'data': {'hello': 'Hello stranger'}}

Or, request YAML back:

>>> r = requests.get("http://;/graph", params={"query": "{ hello(name:\"john\") }"}, headers={"Accept": "application/x-yaml"})
>>> print(r.text)
data: {hello: Hello john}

Want HSTS?

api = responder.API(enable_hsts=True)

Boom.

Installing Responder

Install the latest release:

$ pipenv install responder
✨?✨

Or, install from the development branch:

$ pipenv install -e git+https://github.com/kennethreitz/responder.git#egg=responder

Only Python 3.6+ is supported.

Web Service Performance Characteristics

The objective of these benchmark tests is not testing deployment (like uwsgi vs gunicorn vs uvicorn etc) but instead test the performance of python-response against other popular Python web frameworks.

Methodology

The results below were gotten running the performance tests on a Lenovo W530, Intel(R) Core(TM) i7-3740QM CPU @ 2.70GHz, MEM: 32GB, Linux Mint 19. I used Python 3.7.0 with the WRK utility with params:
wrk -d20s -t10 -c200 (i.e. 10 threads and 200 connections).

  1. Simple "Hello World" benchmark

    python-responder v0.0.1 (Master branch)
    Requests/sec: 1368.23
    Transfer/sec: 163.01KB

    Django v2.1.2 (i18n == False)
    Requests/sec: 544.54
    Transfer/sec: 103.18KB

    Django v2.1.2 (i18n == True)
    Requests/sec: 535.12
    Transfer/sec: 101.38KB

    Django v2.1.2 (Minimal 1 file Django Application)
    https://gist.github.com/aitoehigie/ebcc1d3e460e66cd51e5501fa2636798
    Requests/sec: 701.53
    Transfer/sec: 99.34KB

    Flask v1.0.2
    Requests/sec: 896.24
    Transfer/sec: 144.41KB

The Basic Idea

The primary concept here is to bring the niceties that are brought forth from both Flask and Falcon and unify them into a single framework, along with some new ideas I have. I also wanted to take some of the API primitives that are instilled in the Requests library and put them into a web framework. So, you'll find a lot of parallels here with Requests.

  • Setting resp.text sends back unicode, while setting resp.content sends back bytes.
  • Setting resp.media sends back JSON/YAML (.text/.content override this).
  • Case-insensitive req.headers dict (from Requests directly).
  • resp.status_code, req.method, req.url, and other familiar friends.

Ideas

  • Flask-style route expression, with new capabilities -- all while using Python 3.6+'s new f-string syntax.
  • I love Falcon's "every request and response is passed into to each view and mutated" methodology, especially response.media, and have used it here. In addition to supporting JSON, I have decided to support YAML as well, as Kubernetes is slowly taking over the world, and it uses YAML for all the things. Content-negotiation and all that.
  • A built in testing client that uses the actual Requests you know and love.
  • The ability to mount other WSGI apps easily.
  • Automatic gzipped-responses.
  • In addition to Falcon's on_get, on_post, etc methods, Responder features an on_request method, which gets called on every type of request, much like Requests.
  • A production static file server is built-in.
  • Uvicorn built-in as a production web server. I would have chosen Gunicorn, but it doesn't run on Windows. Plus, Uvicorn serves well to protect against slowloris attacks, making nginx unnecessary in production.
  • GraphQL support, via Graphene. The goal here is to have any GraphQL query exposable at any route, magically.

Future Ideas

  • Cookie-based sessions are currently an afterthought, as this is an API framework, but websites are APIs too.
  • If frontend websites are supported, provide an official way to run webpack.

The Goal

The primary goal here is to learn, not to get adoption. Though, who knows how these things will pan out.

GitHub