Quick and dirty mock service with Starlette

I had a challenge at work. The team needed to mock out a third party service in a testing environment. The service was slow and configuring it was painful. If we could mock it out, then the team could avoid those problems.

The challenge with mocking out the service is that part of the flow needs to invoke a webhook that will call back to my company's system to indicate that all work is done. Additionally, the webhook call must be after a delay because the service we're simulating takes a long time (i.e., more than 60 seconds).

The Problem: Run a mocked third party service with a delayed webhook callback.

The solution needs to:

• Respond to a POST request and return a 200 OK status.
• Call a webhook back to the system that will include extra identifiers to connect the webhook call to the original request.
• Be as simple as possible because this is not the core product.

Solution options

The Python ecosystem is full of strong options to address the first part of the solution. Django, Flask, Pyramid, Bottle, and any other web framework you can think of would handle that with ease.

The second part of the solution is harder. If I'm not careful, then being simple goes out the window, and I've destroyed the third objective.

Background task tools

As much as I love Django, I didn't think it was a good fit. In my thought process, calling the webhook later required some kind of background task. If you know Django, then you may immediately think of Celery, and that was my first thought too. The problem is that Celery requires a message broker like RabbitMQ. With that, the simplicity budget is totally shot.

An alternative to Celery is RQ, but it has similar problems and requires running Redis along with a separate worker process.

Could we solve this problem without using a separate background task tool?

Sans background task tools

Knowing that simplicity was an important goal, I really wanted to do this without a background task tool.

I considered two approaches:

1. Use threads.
2. Use cooperative concurrency (read: asynchronous programming)

I don't like threaded programming in Python. In fact, I don't like threaded programming. ¹ A threaded solution could totally work by executing the webhook call in a separate thread, but I was interested in other options.

Could I solve the problem with async programming? Async programming is much easier in recent versions of Python. The language now has built-in keywords like async and await that makes programming in that style friendlier. I've assessed a few async web frameworks, and I recalled a feature that I saw in the Starlette documentation: Background Tasks

Starlette is a web framework developed by the author of Django REST Framework (DRF), Tom Christie. DRF is such a solid project. Sharing the same creator bolstered my confidence that Starlette will be a well designed piece of software.

I decided to spike on the project to see if it was a viable path. A hour later, I was shocked to see that I had a final, working, solution.

The rest of this article will explore what I did to solve the problem using Starlette with a generic version that skips the boring details of the third party service.

Starlette

Since Starlette is an async framework, the Python tools needed to support a Starlette application are different from a standard WSGI app.

Instead of a synchronous web server like Gunicorn or uWSGI, an asynchronous web server is required. I selected Uvicorn, a web server similar to Gunicorn that uses uvloop to handle the asynchronous event loop.

To do the webhook callback, the emulator needs to make HTTP requests. The natural package to reach for is requests, but requests is a synchronous package that would block the event loop. I needed an asynchronous HTTP request library. For that requirement, I picked HTTPX. HTTPX is a very new library that shares the same API as requests and works with async programming.

If you want to follow along, first, install all the necessary dependencies. You'll need at least Python 3.6.

$ mkdir mock-service
$ cd mock-service
$ python3 -m venv venv
$ . venv/bin/activate
(venv) $ pip install starlette uvicorn httpx

For future readers, here are the versions that the pip install command installed into my local virtual environment at the time of writing this article.

$ pip freeze
certifi==2019.6.16
chardet==3.0.4
Click==7.0
h11==0.8.1
h2==3.1.1
hpack==3.0.0
hstspreload==2019.8.20
httptools==0.0.13
httpx==0.7.1
hyperframe==5.2.0
idna==2.8
rfc3986==1.3.2
starlette==0.12.8
uvicorn==0.8.6
uvloop==0.12.2
websockets==7.0

Let's start by listing the full example. I'll break down each part of the app, but sometimes it helps to get the full picture first. To try this out yourself, you can grab the example script.

import asyncio
import os
import uuid

import httpx
from starlette.applications import Starlette
from starlette.background import BackgroundTask
from starlette.responses import JSONResponse
import uvicorn

app = Starlette()
client = httpx.AsyncClient()
CALLBACK_URL = os.environ["CALLBACK_URL"]


@app.route("/api/endpoint", methods=["POST"])
async def fake_endpoint(request):
    identifier = str(uuid.uuid4())
    payload = {
        "identifier": identifier,
        "some_parameter": request.query_params.get("some_parameter"),
    }
    task = BackgroundTask(trigger_webhook, payload)
    return JSONResponse(
        {"identifier": identifier, "success": True}, background=task)


async def trigger_webhook(payload):
    await asyncio.sleep(5)
    params = {
        "success": True,
        "identifier": payload["identifier"],
        "some_parameter": payload["some_parameter"],
    }
    await client.get(CALLBACK_URL, params=params)


if __name__ == "__main__":
    uvicorn.run(app, host="0.0.0.0", port=8000)

Running the code

We can run the code end-to-end to simulate how this would plug into a real environment.

First, we need a webhook receiver. The Python built-in HTTP server is perfectly suited for this task. In a separate terminal, run:

$ python3 -m http.server 5000
Serving HTTP on 0.0.0.0 port 5000 (http://0.0.0.0:5000/) ...

Next, start the mock service.

(venv) $ CALLBACK_URL=http://0.0.0.0:5000 python3 mock_service.py
INFO: Started server process [47148]
INFO: Waiting for application startup.
INFO: Uvicorn running on http://0.0.0.0:8000 (Press CTRL+C to quit)

Notice that I've passed in CALLBACK_URL as an environment variable with the URL from our webhook receiver. The mock service will use this information to know where to send the webhook request.

With the core pieces in place, we need to trigger the mock service. I like HTTPie as a friendlier alternative to curl.

(venv) $ pip install httpie

Now we can get everything moving! Let's fire off a POST to the mock service.

(venv) $ http POST :8000/api/endpoint some_parameter==some_value
HTTP/1.1 200 OK
content-length: 68
content-type: application/json
date: Sat, 24 Aug 2019 15:54:07 GMT
server: uvicorn

{
    "identifier": "4a3a0ce2-ae0c-41d3-ba58-89a5a9579692",
    "success": true
}

From the mock service, you will see a log like:

INFO: ('127.0.0.1', 61647) - \
    "POST /api/endpoint?some_parameter=some_value HTTP/1.1" 200

Five seconds later, the fake webhook receiver should show:

127.0.0.1 - - [24/Aug/2019 11:54:12] \
    "GET /?success=true&identifier= \
    4a3a0ce2-ae0c-41d3-ba58-89a5a9579692&some_parameter=some_value \
    HTTP/1.1" 200 -

I've reformatted the lines so they will fit better in the article.

Awesome! The mock service did exactly what we wanted. The POST request got a successful response and, after a delay, the webhook got some of the same information.

Now that we've seen things work, let's break down the code.

Mock service explanation

We can look at this code in a few separate chunks.

import asyncio
import os
import uuid

import httpx
from starlette.applications import Starlette
from starlette.background import BackgroundTask
from starlette.responses import JSONResponse
import uvicorn

The import section isn't very exciting. I've divided the standard library imports from the third party imports. If we had more code in separate files, then I'd have a third section for local imports, but we don't need that because the entire emulator file fits in 39 lines.

app = Starlette()
client = httpx.AsyncClient()
CALLBACK_URL = os.environ["CALLBACK_URL"]

This section creates the module level globals that we are going to use. Sometimes globals are the best tool for a job even if they are often frowned upon. The CALLBACK_URL is the only setting for the emulator. Notice that I get the value from os.environ with the index syntax instead of the os.environ.get('CALLBACK_URL') syntax. The index style of fetching will ensure that we have the value since it's required to operate the emulator properly.

@app.route("/api/endpoint", methods=["POST"])
async def fake_endpoint(request):
    identifier = str(uuid.uuid4())
    payload = {
        "identifier": identifier,
        "some_parameter": request.query_params.get("some_parameter"),
    }
    task = BackgroundTask(trigger_webhook, payload)
    return JSONResponse(
        {"identifier": identifier, "success": True}, background=task)

This async function defines the core route. The style should be extremely familiar to Flask users. The body of the function does a handful of things:

1. Generates an identifier that a caller could use to associate this call with the webhook that will follow.
2. Extracts a value from the request's query string, some_parameter, and stores it for the background task.
3. Bundles data into a payload that will be available to the BackgroundTask which will execute the trigger_webhook function.
4. Responds with the identifier and sets the background task to run.

One weird aspect of this emulator is that it's handling POST requests, but there is no POST data processed. The real third party service included PDF files in the POST, but I wanted to ignore that aspect since it did not illustrate any extra value.

async def trigger_webhook(payload):
    await asyncio.sleep(5)
    params = {
        "success": True,
        "identifier": payload["identifier"],
        "some_parameter": payload["some_parameter"],
    }
    await client.get(CALLBACK_URL, params=params)

The webhook trigger sleeps for 5 seconds. It's important to use asyncio.sleep instead of time.sleep. The sleep function from time is a synchronous command and will block the event loop. By calling asyncio.sleep with await, the function yields execution back to the event loop until 5 seconds pass.

Once the delay is over, the HTTPX client calls the callback URL to return the identifier and parameter. This indicates to the caller that the emulator is done.

if __name__ == "__main__":
    uvicorn.run(app, host="0.0.0.0", port=8000)

The final bit of code connects the app to Uvicorn so that Uvicorn will start when the service is invoked with:

(venv) $ python3 mock_service.py

Starlette's promising future

Starlette's API made this emulator as quick to produce as I could possibly hope for. The framework is still new, but I think it has a very promising future in the Python ecosystem. It already has the best performance of any Python framework in the TechEmpower Benchmarks.

I hope you enjoyed looking at some async programming in Python. If you have any questions, please share on Twitter and reach me at @mblayman.

Thanks for reading!

This article first appeared on mattlayman.com.

---

1. I've been studying Rust on the side so maybe that language can change my mind about threaded programming, but the verdict is still out. ↩