If you've ever found yourself wanting a test http/s server that you can do the weirdest imaginable things with, overly is for you.
Let's start simple. Say we want to emulate the well known /get endpoint at httpbin.org, and stick it on a unit/pytest.
We use overly.Server as a decorator, passing it a series of steps to do when it gets a request. The decorator injects the server object in to the test, kinda like a fixture.
import requests
from overly import Server, delay, send_request_as_json, finish
@Server(("localhost", 25001), steps=[delay(2), send_request_as_json, finish])
def test_json_send(server):
r = requests.post(
server.http_test_url, json={"key_1": True, "key_2": "cheesestring"}
)
pprint(r.json())
# The response is:
#
# 'body': '{"key_1": true, "key_2": "cheesestring"}',
# 'files': [],
# 'forms': [],
# 'headers': [
# ['host', 'localhost:25001'],
# ['user-agent', 'python-requests/2.19.1'],
# ['accept-encoding', 'gzip, deflate'],
# ['accept', '*/*'],
# ['connection', 'keep-alive'],
# ['content-length', '40'],
# ['content-type', 'application/json']],
# 'http_version': '1.1',
# 'json': [
# {
# 'json': {'key_1': True, 'key_2': 'cheesestring'}
# }
# ],
# 'method': 'POST',
# 'path': '/',
# 'target': '/'}Simples! Now Let's Get Weird.
Want to see if your client will handle a redirect from https://example.com, to https://example.com/doowap, that returns a 404 page that's eight gigabytes, but the server only starts sending data 43 seconds later and kills the socket before finishing, just for kicks? We'll enforce the request order too so we don't accidentally write our tests to miss the redirection case, and skip the first redirect. Alrighty then:
from functools import partial
from overly import (
Server,
ssl_socket_wrapper,
HttpMethods,
send_303,
send_404,
delay,
just_kill
)
test_loc = ("localhost", 25001)
eight_gigs = 1 * (1024 ** 3) * 8)
Server(
test_loc,
max_requests=2,
ordered_steps=True,
ssl_socket_wrapper=ssl_socket_wrapper,
steps=[
[
(HttpMethods.GET, "/"),
partial(send_303, headers=("location", "/doowap")),
finish
],
[
(HttpMethods.GET, "/doowap"),
delay(43),
partial(send_404, data=b"x" * eight_gigs,
just_kill
]
]
).run()Bam. The worst 404 page of all time. The above can be used as a decorator, just like before. We can also just run the server with Server.run() for easy messing about.
As was mentioned, and as you can see, overly is overly configurabe :D All of your bases are covered!
The Server is concurrent, so you can test all of your weird async / threaded / voodoo clients, with keep-alive support.
Sure!
I'll break down the first example in detail.
The Server object can be used to wrap a function as a decorator. When the wrapped function is called, before it is run, overly starts the server in a thread, waits until the server declares it's listening socket is ready to go, and then runs whatever callable was passed to it. This also works with async calls, by passing a wrapper around your async func that invokes it correctly. An example of that is here.
The Server threads off new clients as they connect, pairing them with a ClientHandler instance. It's through the client handler that we communicate with the connected client, and fetch data about the request.
Server takes what I call "steps" as arguments. A step is any callable at all. There are many built in steps, and it's very straight forward to write a new one. Steps are invoked by the client handler. The client handler will pass its self to any step it invokes, and so all steps must accept at least one argument (the client handler.)
The first step we use is delay. delay is a simple function that just time.sleeps for t seconds.
Secondly, we have send_request_as_json. It a much more complex function, but given all it does it's still simple in theory. It takes a reference to the client handler as an argument, as described above. It accesses the underlying h11 request metadata object through client_hanlder.request, and the request body through client_handler.request_body.
It constructs a dict appropriate for jsoning as a response, and uses client_handler.http_send to send back h11.Response and h11.Data objects.
If you're unfamiliar with h11 that's about all you need to know, but you can read some more through the link there.
Finally we use finish to cleanly end the connection and inform both the client and our underlying h11.connection object we are finished with the request.