geventhttpclient

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

gevent.httplib support for patching http.client was removed in gevent 1.0, geventhttpclient now provides that missing functionality.

geventhttpclient uses a fast http parser, written in C.

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.

Since version 2.3, geventhttpclient features a largely requests compatible interface. It covers basic HTTP usage including cookie management, form data encoding or decoding of compressed data, but otherwise isn't as feature rich as the original requests. For simple use-cases, it can serve as a drop-in replacement.

import geventhttpclient as requests
requests.get("https://github.com").text
requests.post("http://httpbingo.org/post", data="asdfasd").json()

from geventhttpclient import Session
s = Session()
s.get("http://httpbingo.org/headers").json()
s.get("https://github.com").content

This interface builds on top of the lower level HTTPClient.

from geventhttpclient import HTTPClient
from geventhttpclient.url import URL

url = URL("http://gevent.org/")
client = HTTPClient(url.host)
response = client.get(url.request_uri)
response.status_code
body = response.read()
client.close()

httplib compatibility and monkey patch

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

# from http.client 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
geventhttpclient.httplib.patch()

import httplib2

High Concurrency

HTTPClient has a connection pool built in and is greenlet safe by design. You can use the same instance among several greenlets. It is the low level building block of this library.

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 = "<MY_DEV_TOKEN>"

url = URL("https://graph.facebook.com/me/friends", params={"access_token": TOKEN})

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

response = client.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(client, friend_id):
    friend_url = URL(f"/{friend_id}", params={"access_token": TOKEN})
    # the greenlet will block until a connection is available
    response = client.get(friend_url.request_uri)
    assert response.status_code == 200
    friend = json.load(response)
    if "username" in friend:
        print(f"{friend['username']}: {friend['name']}")
    else:
        print(f"{friend['name']} has no username.")

# 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, client, friend_id)

pool.join()
client.close()

Streaming

geventhttpclient supports streaming. Response objects have a read(n) and readline() method that read the stream incrementally. See 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:

from geventhttpclient import HTTPClient, URL

url = URL("http://127.0.0.1:80/100.dat")
client = HTTPClient.from_url(url)
response = client.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:
        f.write(data)
        data = response.read(CHUNK_SIZE)

Benchmarks

The benchmark runs 10000 GET requests against a local nginx server in the default configuration with a concurrency of 10. See benchmarks folder. The requests per second for a couple of popular clients is given in the table below. Please read benchmarks/README.md for more details. Also note, HTTPX is better be used with asyncio, not gevent.

HTTP Client	RPS
GeventHTTPClient	7268.9
Httplib2 (patched)	2323.9
Urllib3	2242.5
Requests	1046.1
Httpx	770.3

Linux(x86_64), Python 3.11.6 @ Intel i7-7560U

License

This package is distributed under the MIT license. Previous versions of geventhttpclient used http_parser.c, which in turn was based on http/ngx_http_parse.c from NGINX, copyright Igor Sysoev, Joyent, Inc., and other Node contributors. For more information, see http://github.com/joyent/http-parser