Gunicorn: Gunicorn doesn't allow requesting urls on the same server instance

[pistacchio]

Hi, I'm experiencing a strange problem. I have the following simple/sample Flask application:

# from __future__ import print_function
import requests

from flask import Flask

PORT = 6556
app = Flask(__name__)


@app.route('/')
def index():
    j = {'data': 1}

    print('INDEX CALLING API')
    response = requests.post('http://localhost:{}/fake_api'.format(PORT), json=j)
    print('INDEX CALLED API')

    return response.text

@app.route('/fake_api', methods=['POST'])
def fake_api():
    print('FAKE_API CALL RECEIVED')
    return 'Ok, it works'

if __name__ == '__main__':
    app.run(host='0.0.0.0', port=PORT, debug=True)

Basically, what I'm simulating is calling a url (http://localhost:6556/) and have the route call with requests another url of the same flask app (http://localhost:6556/fake_api). Being quite banal, it works fine if I simply start the app with python server.py.

The problem is that when I try to run the server with gunicorn like this

gunicorn -b 0.0.0.0:6556 server:app

the application hangs indefinitely on the request. In fact it prints INDEX CALLING API and then halts.

I've experienced the same problem with uWSGI. Any idea?

[jamadden]

The default type of worker for gunicorn is the sync worker, and the default number of workers is one. sync workers only support a single request at a time:

The most basic and the default worker type is a synchronous worker class that handles a single request at a time. This model is the simplest to reason about as any errors will affect at most a single request.

None of those defaults are being changed here. So you're trying to make a second request, but only one request is supported. Thus gunicorn blocks waiting for the first request to finish before handling the second---which can't happen because the first request is blocked waiting for the second request to finish. You'll need to use a different worker type or number of workers, or use a mechanism supplied by your framework to make an in-app request without spawning a new HTTP socket request (e.g., Pyramid's subrequest).

[pistacchio]

Ok, thanks. I got it working with:

# pip install gevent
gunicorn -b 0.0.0.0:6556 -k gevent server:app