FastCGI server implementation using gevent coroutine-based networking library. No need to monkeypatch and slow down your favourite FastCGI server in order to make it "green".
Provides simple request handler API to allow for custom request handlers. Comes with two WSGI request hadler implementations -- one using standard wsgiref.handlers.BasicCGIHandler and another using original request hadler.
Full support for FastCGI protocol connection multiplexing feature.
Can fork multiple processes to better utilize multi-core CPUs.
Includes adapters for Django and frameworks that use PasteDeploy like Pylons / Pyramid and TurboGears to simplify depolyment.
To install gevent-fastcgi using pip run the following command:
$ pip install gevent-fastcgi
If you prefer easy_install here is how to use it:
$ easy_install gevent-fastcgi
This is how to use gevent-fastcgi in stand-alone mode:
from gevent_fastcgi.server import FastCGIServer
from gevent_fastcgi.wsgi import WSGIRequestHandler
def wsgi_app(environ, start_response):
start_response('200 OK', [('Content-type', 'text/plain')])
yield 'Hello World!'
request_handler = WSGIRequestHandler(wsgi_app)
server = FastCGIServer(('127.0.0.1', 4000), request_handler, num_workers=4)
server.serve_forever()
Using with PasteDeploy and friends
Gevent-fastcgi defines three paste.server_runner entry points. Each of them will run FastCGIServer with different request handler implementation:
- wsgi
- gevent_fastcgi.wsgi.WSGIRequestHandler will be used to handle requests. Application is expected to be a WSGI-application.
- wsgiref
- gevent_fastcgi.wsgi.WSGIRefRequestHandler which uses standard wsgiref.handlers will be used to handle requests. Application is expected to be a WSGI-application.
- fastcgi
- Application is expected to implement gevent_fastcgi.interfaces.IRequestHandler interface. It should use request.stdin to receive request body and request.stdout and/or request.stderr to send response back to Web-server.
Use it as following:
[server:main]
use = egg:gevent_fastcgi#wsgi
host = 127.0.0.1
port = 4000
# UNIX domain socket can be used by specifying path instead of host and port
# socket = /path/to/socket
# socket_mode = 0660
# The following values are used in reply to Web-server on `FCGI_GET_VALUES` request
#
# Maximum allowed simulteneous connections, i.e. the size of greenlet pool
# used for connection handlers.
max_conns = 1024
max_reqs = 1024
# Fork `num_workers` child processes after socket is bound.
# Must be equal or greate than 1. No children will be forked
# if set to 1 or not specified
num_workers = 8
# Call specified functions of gevent.monkey module before starting the server
gevent.monkey.patch_thread = yes
gevent.monkey.patch_time = no
gevent.monkey.patch_socket = on
gevent.monkey.patch_ssl = off
# or
gevent.monkey.patch_all = yes
Django adapter
Add gevent_fastcgi.adapters.django to INSTALLED_APPS of settings.py then run the following command (replace <address> with <host>:<port> or <unix-socket>):
$ python manage.py run_gevent_fastcgi <address>
Starting from version 0.1.16dev It is possible to use custom request handler with gevent_fastcgi.server.FastCGIServer. Such a handler should implement gevent_fastcgi.interfaces.IRequestHandler interface and basically is just a callable that accepts single positional argument request. gevent_fastcgi.wsgi module contains two implementations of IRequestHandler.
Request handler is run in separate greenlet. Request argument passed to request handler callable has the following attributes:
- environ
- Dictionary containing request environment. NOTE: contains whatever was sent by Web-server via FCGI_PARAM stream
- stdin
- File-like object that represents request body, possibly empty
- stdout
- File-like object that should be used by request handler to send response (including response headers)
- stderr
- File-like object that can be used to send error information back to Web-server
Following is sample of custom request handler implementation:
import os
from zope.interface import implements
from gevent import spawn, joinall
from gevent_subprocess import Popen, PIPE
from gevent_fastcgi.interfaces import IRequestHandler
# WARNING!!!
# CGIRequestHandler is for demonstration purposes only!!!
# IT MUST NOT BE USED IN PRODUCTION ENVIRONMENT!!!
class CGIRequestHandler(object):
implements(IRequestHandler)
def __init__(self, root, buf_size=1024):
self.root = os.path.abspath(root)
self.buf_size = buf_size
def __call__(self, request):
script_name = request.environ['SCRIPT_NAME']
if script_name.startswith('/'):
script_name = script_name[1:]
script_filename = os.path.join(self.root, script_name)
if script_filename.startswith(self.root) and
os.path.isfile(script_filename) and
os.access(script_filename, os.X_OK):
proc = Popen(script_filename, stdin=PIPE, stdout=PIPE, stderr=PIPE)
joinall((spawn(self.copy_stream, src, dest) for src, dest in [
(request.stdin, proc.stdin),
(proc.stdout, request.stdout),
(proc.stderr, request.stderr),
]))
else:
# report an error
request.stderr.write('Cannot locate or execute CGI-script %s' % script_filename)
# and send a reply
request.stdout.write('\r\n'.join((
'Status: 404 Not Found',
'Content-Type: text/plain',
'',
'No resource can be found for URI %s' % request.environ['REQUEST_URI'],
)))
def copy_stream(self, src, dest):
buf_size = self.buf_size
read = src.read
write = dest.write
while True:
buf = read(buf_size)
if not buf:
break
write(buf)
if __name__ == '__main__':
from gevent_fastcgi.server import FastCGIServer
address = ('127.0.0.1', 8000)
handler = CGIRequestHandler('/var/www/cgi-bin')
server = FastCGIServer(address, handler)
server.serve_forever()
This project could not be where it is now without help of the following great people:
Thank you guys for all your help!