"Too many open files" error, regardless of size parameter

[david-cliqz]

I am using grequests to test the performance of my service. Both the service and the machine from which I start the requests are inside Amazon EC2. It seems that I cannot start more requests than ulimit -n, even though I explicitly specify the size parameter for map() / imap()

My setup is roughly like this:

headers = {'Content-type': 'application/json', 'Accept': 'text/plain'}
params = [json.dumps({'url': urls[i]}) for i in urls]
reqs = [grequests.post(SERVICE_URL, data=params[i % len(params)], headers=headers)
            for i in xrange(no_requests)]
res = grequests.imap(reqs, size=200)
for r in res:
    ...
    r.close()

Doesn't the size parameter here tell grequests that it should issue no_requests, 200 at a time?

[rtdean]

ulimit -n shows the soft/hard limits for the number of open file descriptors a process can hold. Once you hit that limit, you can't open up any more FD's... this includes network sockets.

One of the things you could do is build a requests.Session object, and either use the default HTTPAdapter or mount a new one with a larger connection pool. Then, grequests/requests will end up doing HTTP Pipelining, reusing the existing connections where possible.

Try something like:

import requests
import grequests

session = requests.Session()
headers = {'Content-type': 'application/json', 'Accept': 'text/plain'}
params = [json.dumps({'url': urls[i]}) for i in urls]
reqs = (grequests.post(SERVICE_URL, data=params[i % len(params)], headers=headers, session=session)
            for i in xrange(no_requests))
for r in grequests.imap(reqs, size=200):
    ...

if you call r.close() in your for loop, you end up closing the underlying connection, and forcing a new connection to be established. (If you aren't using a Session, this ends up happening anyways).

You can also mount a new HTTPAdapter with a larger connection pool:

import requests
from requests.adapters import HTTPAdapter
import grequests

session = requests.Session()
session.mount('http://', HTTPAdapter(pool_connections=50, pool_maxsize=50))
headers = {'Content-type': 'application/json', 'Accept': 'text/plain'}
params = [json.dumps({'url': urls[i]}) for i in urls]
reqs = (grequests.post(SERVICE_URL, data=params[i % len(params)], headers=headers, session=session)
            for i in xrange(no_requests))
for r in grequests.imap(reqs, size=200):
    ...

[29x10]

@rtdean fixed my problem

[cyberstip]

For posterity, just adding a session to the request seems to fix the issue (as @rtdean mentioned):

session = requests.Session()
reqs = (grequests.get(url, session=session) for url in urls)
results = grequests.map(reqs, size=50)

[bbbco]

It would be great if you could add this info to a FAQ section in the README

[skasturi]

We just faced this issue on our servers bringing down the whole service. It would be good to have it in the README. Thank you!

[amitsquare]

Please do mention using Session Object in README .As till you complete system shutdown , user won't be searching the issue & end up here on this issue thread .

[jontesek]

@amitsquare @skasturi @bbbco Feel free to create a PR for this :).

[spyoungtech]

To provide some further context, this is (or at least was) a problem with HTTPS libraries in general... See psf/requests#239 urllib3/urllib3#291

As far as I can tell, this is not a direct issue with grequests, so I'm going to close the issue. As far as updating the README... Personally, I'm not sure it's worthwhile adding to the currently concise README. I see no particular reason to document it than any other issue not specifically within grequest's domain that affects HTTP libraries in general. Though, if someone feels this is particularly useful for grequests users, please feel free to submit a PR.