Limiter gives unexpected results when running with gunicorn

[vulcan25]

EDIT: solved, see next comment.

Flask-Limiter==1.0.1
limits==1.3

I have Flask Limiter setup on my /one endpoint with a limit as follows against the whole blueprint, as per the docs:

limiter.limit('4/hour')(bp)

Obviously with this config I would expect 4 200 response codes, followed by 429 until an hour has passed.

With Flask running in dev mode, using werkzeug's run_simple:

from werkzeug.serving import run_simple
#[snip]
if __name__ == '__main__':
    run_simple('localhost', 9999, app, use_reloader=True, use_debugger=True, use_evalex=True)

I can issue the following command to hit the endpoint 20 times.

for var in `seq 20` ;
do curl -w '%{http_code}-' http://localhost:9999/one -o /dev/null -s;
done

The following is the output of this command running 5 times, with me restarting the dev server each time (This is because I'm using Flask-Limiter's default config of in memory storage rather than redis, etc - restarting the dev sever forces the rate limits to reset).

200-200-200-200-429-429-429-429-429-429-429-429-429-429-429-429-429-429-429-429-%
200-200-200-200-429-429-429-429-429-429-429-429-429-429-429-429-429-429-429-429-%
200-200-200-200-429-429-429-429-429-429-429-429-429-429-429-429-429-429-429-429-%
200-200-200-200-429-429-429-429-429-429-429-429-429-429-429-429-429-429-429-429-%
200-200-200-200-429-429-429-429-429-429-429-429-429-429-429-429-429-429-429-429-%

This is what I expected. Four 200 errors and the rest 429.

However, when I launch the same application with gunicorn (again restarted per test)

gunicorn --bind 0.0.0.0:9999 --workers 4 run:app   --log-level debug        

The results are as follows.

200-200-200-200-200-429-200-200-200-200-200-200-200-200-429-429-200-429-429-429-%
200-200-200-200-200-200-200-200-200-429-429-429-429-429-429-429-200-429-200-200-%
200-200-200-200-200-200-200-200-200-200-200-200-200-200-429-429-200-429-429-429-%
200-200-200-200-200-200-200-429-200-200-200-200-200-200-200-429-429-200-429-429-%
200-200-200-200-200-200-200-200-200-200-429-200-200-429-200-429-429-429-429-429-%

Clearly the 4/hour rate limit is not being enforced, but whether I get a 400 or 200 seems to be quite random.

I ran another test without restarting gunicorn in between tests. With 100 requets as above, it eventually tends towards a 429:

200-200-200-200-200-200-200-200-429-200-429-200-429-200-200-200-200-200-429-429-%
429-429-429-429-429-429-429-200-429-429-429-429-429-429-429-429-429-429-429-429-%
[ snip - 3 more rows with all 429]... 

I tried hacking the Flask Limiter code to output more info to gunicorn's debug console.

I added a detail class to wrappers.Limit to output some information:

def detail(self):
    d = { 'limit':self.limit,
          '__scope': self.__scope,
          'per_method': self.per_method,
          'methods': self.methods,
          'error_message': self.error_message,
          'exempt_when': self.exempt_when
    }
    out = pprint.pformat(d) 
    return (out)

Then update extension.py adding a for loop to the __evaluate_limits function, which I understand to be the logic function.

def __evaluate_limits(self, endpoint, limits):
    for limit in limits: print (limit.detail())
    failed_limit = None
    ...

This at least confirms that the string '4/hour' is being interpreted correctly.

{'__scope': None,
 'error_message': None,
 'exempt_when': None,
 'limit': 4 per 1 hour,
 'methods': None,
 'per_method': False}

For better diagnostics, I suspect my function should grab information from somewhere else in the codebase, but I don't know where. I'm also not sure if this could be a gunicorn configuration value that is required; Some sort of internal caching causing this?

[vulcan25]

okay, so I figured this out. Running gunicorn with --workers 4 means each worker has it's own memory, hence the random results.

[sirjoe]

do you have some fix to get it working?

[VeNoMouS]

do you have some fix to get it working?

use memcached or redis?