falcon
falcon copied to clipboard
falconry
Expand deployment guide
Discuss deployments using some of the more popular web servers and cloud providers. This guide should only provide high-level guidance, and defer to the relevant project/vendor docs for details.
WSGI servers
- uWSGI
- Gunicorn
- httpd
Short discussion on reverse proxies, load balancers:
- HAProxy
- Nginx
- httpd
- lighttpd
- Appliances (e.g., F5)
- IaaS (e.g., ELB, Google Cloud LB)
Cloud providers:
- AWS
- Microsoft
- Digital Ocean
- Heroku
See also:
- http://bottlepy.org/docs/dev/deployment.html
- http://werkzeug.pocoo.org/docs/0.11/deployment/
- https://redmine.lighttpd.net/projects/lighttpd2/wiki/Howto_WSGI
Hi @anekix, just following up, were you still interested in taking this one?
OK, I'm assuming this one is fair game for anyone else who would like to pick it up. :)
I have some examples of working with uWSGI & NGINX. What would be most helpful? An example configuration file and maybe a wsgi.py file for how to attach an app to an uWSGI master?
@nZac I think it would be great to have a page added to the doc with sections for difference deployment scenarious and a TOC at the top. uwsgi + nginx would be a great first addition to that page.
For Gunicorn, you can reference the official Deploying Gunicorn docs which I found very helpful recently.
I think it would be good to make a recommendation on which workers are recommended. In my testing of a little hello world app, meinheld workers performed the best with Falcon.
falcon (req/sec)
--------- ------------------
eventlet 19.94k
gevent 19.95k
gunicorn 13.28k
meinheld 70.62k
uwsgi errors
I would really like to see some recommendations on uWSGI settings too as I would constantly see socket errors when smashing uWSGI with load.
Tested on the following instance:
AMI: CentOS 7 (x86_64) - with Updates HVM
Type: t2.2xlarge
vCPUs: 8
Memory (GiB): 32
Python: 3.6.5 (from IUS YUM repository)
With the following test command:
wrk --duration 10s --threads 1 --connections 200 http://wsgi:8000
And the following Gunicorn command:
gunicorn \
--workers 17 --worker-class=meinheld.gmeinheld.MeinheldWorker \
--bind 0.0.0.0:8000 \
hello_falcon:app
And finally here's the Falcon code:
import falcon
class HelloResource(object):
def on_get(self, req, resp):
resp.body = 'Hello World!'
app = falcon.API()
app.add_route('/', HelloResource())
To be honest, in all my testing thus far, I'm just not convinced of PyPy's performance once you add the WSGI server in, hence my issue #1309 so I'm keen to see your suggestions on deploying with PyPy too 😄
Keep up the great work! Fotis
Hope it's OK if I post another little tip I've learned here. If you are deploying with Gunicorn, it's important that you configure logging to write to Gunicorn's error logger or it won't log correctly during deployment.
To make this easy, I add the handler to the root logger like so:
# Configure the root handler log level and ensure that it is set to send logs
# to Gunicorn's error logs.
gunicorn_error_logger = logging.getLogger('gunicorn.error')
logging.root.handlers.extend(gunicorn_error_logger.handlers)
logging.root.setLevel(gunicorn_error_logger.level)
Perhaps the suggestion could then be that resources simply inherit this log level:
...
self.logger = logging.getLogger('app.endpoint')
self.logger.setLevel(logging.root.level)
...
Or of course, they can override it.
Hope this helps Fotis
I find the gunicorn + meinheld deployment very easy to set-up
@see https://github.com/the-benchmarker/web-frameworks/blob/master/python/falcon/Dockerfile#L12
