Gunicorn

 

In this post we review usage of the python framework Guicorn.

Gunicorn provides a concurrent processing of sockets request, while using a polling mode that enables the same worker to handle multiple requests.

A Simple Flask Application

Gunicorn can be used to wrap the flask application. For example, let's assume we have the following flask application code:

main.py

from flask import Flask

app = Flask(__name__)

@app.route("/")
def index():
    return "Hello World"

if __name__ == "__main__":
    app.run(host='0.0.0.0', port=8080)

We deploy the dependencies:

pip install flask

and run the application:

python main.py

Let's stress test the flask application without gunicorn, first install the apache stress tool:

sudo apt install apache2-utils

and next run the stress with 10K requests and concurrency of 10 threads:

ab -n 10000 -c 10 http://localhost:8080/

The results are:

Wrap the Flask with Gunicorn

To wrap the flask application with gunicorn, we deploy:

pip install gunicorn

and run gunicorn:

gunicorn --workers=2 'main:app'

Notice that guicorn starts by default on the port 8000, so let's run the stress test:

ab -n 10000 -c 10 http://localhost:8000/

The results are:

And we've got ~10 times better results, which is pretty amazing.

Final Note

Using guicorn seems to make python run like a real multithreaded application, however we must be aware to the fact that gunicorn spawns workers processes, hence swe cannot share memory between the workers. 

Using guicorn is a nice solution if we do not have the resources to rewrite the application in GO, but notice that running gunicorn in kubernetes is a bad practice as the guideline is: 

one container == one process

and gunicorn does not follow it. Still, it is reasonable solution for non-critical services.