A high performance, concurrent http client library for python with gevent
Python C Makefile
Latest commit f50dbdc Aug 16, 2016 @rmohr rmohr committed on GitHub Merge pull request #82 from sbraz/manifest
Exclude compiled files from source distribution



Build Status

A high performance, concurrent HTTP client library for python using gevent.

gevent.httplib support was removed in gevent 1.0, geventhttpclient now provides that missing functionality.

geventhttpclient use a fast http parser, written in C, originating from nginx, extracted and modified by Joyent.

geventhttpclient has been specifically designed for high concurrency, streaming and support HTTP 1.1 persistent connections. More generally it is designed for efficiently pulling from REST APIs and streaming APIs like Twitter's.

Safe SSL support is provided by default. geventhttpclient depends on the certifi CA Bundle. This is the same CA Bundle which ships with the Requests codebase, and is derived from Mozilla Firefox's canonical set.

Python 2.7 and 3.4+ are supported. Python 2.6 is no longer supported.

Use of SSL/TLS with python 2.7.9 is not recommended and may be broken.

A simple example:


from geventhttpclient import HTTPClient
from geventhttpclient.url import URL

url = URL('http://gevent.org/')

http = HTTPClient(url.host)

# issue a get request
response = http.get(url.request_uri)

# read status_code

# read response body
body = response.read()

# close connections

httplib compatibility and monkey patch

geventhttpclient.httplib module contains classes for drop in replacement of httplib connection and response objects. If you use httplib directly you can replace the httplib imports by geventhttpclient.httplib.

# from httplib import HTTPConnection
from geventhttpclient.httplib import HTTPConnection

If you use httplib2, urllib or urllib2; you can patch httplib to use the wrappers from geventhttpclient. For httplib2, make sure you patch before you import or the super calls will fail.

import geventhttpclient.httplib

import httplib2

High Concurrency

HTTPClient has connection pool built in and is greenlet safe by design. You can use the same instance among several greenlets.

#!/usr/bin/env python

import gevent.pool
import json

from geventhttpclient import HTTPClient
from geventhttpclient.url import URL

# go to http://developers.facebook.com/tools/explorer and copy the access token
TOKEN = '<go to http://developers.facebook.com/tools/explorer and copy the access token>'

url = URL('https://graph.facebook.com/me/friends')
url['access_token'] = TOKEN

# setting the concurrency to 10 allow to create 10 connections and
# reuse them.
http = HTTPClient.from_url(url, concurrency=10)

response = http.get(url.request_uri)
assert response.status_code == 200

# response comply to the read protocol. It passes the stream to
# the json parser as it's being read.
data = json.load(response)['data']

def print_friend_username(http, friend_id):
    friend_url = URL('/' + str(friend_id))
    friend_url['access_token'] = TOKEN
    # the greenlet will block until a connection is available
    response = http.get(friend_url.request_uri)
    assert response.status_code == 200
    friend = json.load(response)
    if friend.has_key('username'):
        print '%s: %s' % (friend['username'], friend['name'])
        print '%s has no username.' % friend['name']

# allow to run 20 greenlet at a time, this is more than concurrency
# of the http client but isn't a problem since the client has its own
# connection pool.
pool = gevent.pool.Pool(20)
for item in data:
    friend_id = item['id']
    pool.spawn(print_friend_username, http, friend_id)



geventhttpclient supports streaming. Response objects have a read(N) and readline() method that read the stream incrementally. See src/examples/twitter_streaming.py for pulling twitter stream API.

Here is an example on how to download a big file chunk by chunk to save memory:

#!/usr/bin/env python

from geventhttpclient import HTTPClient, URL

url = URL('')
http = HTTPClient.from_url(url)
response = http.get(url.query_string)
assert response.status_code == 200

CHUNK_SIZE = 1024 * 16 # 16KB
with open('/tmp/100.dat', 'w') as f:
    data = response.read(CHUNK_SIZE)
    while data:
        data = response.read(CHUNK_SIZE)


The benchmark does 1000 get requests against a local nginx server with a concurrency of 10. See benchmarks folder.

  • httplib2 with geventhttpclient monkey patch (benchmarks/httplib2_patched.py): ~2500 req/s
  • geventhttpclient.HTTPClient (benchmarks/httpclient.py): ~4000 req/s