Home

PooledProcessMixIn For Python

Fast Concurrent Pool of preforked-processes and threads Mix-in for python's socket server Replace the usual ThreadingMixIn and ForkingMixIn

This is a pure-python module that provides asynchronous mix-in similar to standard ThreadingMixIn and ForkingMixIn but provides better performance by utilizing a pool of processes forked at initialization time each process allocate a pool of given number of threads

Features

• by utilizing multiple processes, it will overcome Python's GIL problem and will distribute load over different cores.
• it does not use queues (which wast time Pickling python objects), instead it uses a single semaphore and event
• utilizing semaphore, event and shared memory boost performance because they are very basic IPC
• each process and thread in the pool is allocated once at initialization time not when getting requests which means requests are processed directly without wasting time allocating threads or forking processes
• each process in the pool have a pool of threads (so having 4 processes each one got 64 threads which gives us 256 concurrent workers at the same time)

Installation and Dependencies

This module has no external dependencies other than the standard python library.

To get the latest development snapshot type the following command

git clone https://github.com/muayyad-alsadi/python-PooledProcessMixIn.git

Example

from PooledProcessMixIn import PooledProcessMixIn
from BaseHTTPServer import BaseHTTPRequestHandler, HTTPServer

# ... define MyTestHandler somewhere

class MyHTTPTest (PooledProcessMixIn, HTTPServer):
    def __init__(self):
        self._process_n=7  # if not set will default to number of CPU cores
        self._thread_n=64  # if not set will default to number of threads
        HTTPServer.__init__(self, ('127.0.0.1',8888), MyTestHandler)
        self._init_pool() # this is optional, will be called automatically
        print "listing on http://127.0.0.1:8888/"

Usage

You can set initial number of processes by setting self._process_n before calling self._init_pool()

You can set initial number of threads for each forked process by setting self._thread_n before calling self._init_pool()

You should call self._init_pool() AFTER super class __init__ but if you did not call it, it will be called automatically when we get first request

Benchmarking

When Benchmarked against ThreadingMixIn and ForkingMixIn, it gives double performance (was able to handle about 1,500 request per second while other mix-ins reached 800 requests/second )

siege -b -c 100 -t10s localhost:8888/test
      Date & Time,  Trans,  Elap Time,  Data Trans,  Resp Time,  Trans Rate,  Throughput,  Concurrent,    OKAY,   Failed
2012-08-02 12:51:47,  14663,       9.58,           0,       0.01,     1530.58,        0.00,       22.87,   14663,       0
2012-08-02 12:52:44,   7653,       9.58,           0,       0.04,      798.85,        0.00,       29.42,    7653,       5
2012-08-02 12:53:14,   7726,       9.47,           0,       0.05,      815.84,        0.00,       43.57,    7726,       0

Can I use it to run my Django applications ?

Yes, Django provides WSGI application which can be used like the wsgi-demo included in this packages

import sys, os
    
from PooledProcessMixIn import PooledProcessMixIn
from wsgiref.simple_server import make_server, WSGIServer
os.environ.setdefault("DJANGO_SETTINGS_MODULE", "mysite.settings")
    
d=os.path.dirname(__file__)
sys.path.insert(0, d)
sys.path.insert(0, os.path.join(d, ".."))
    
from django import VERSION as DJ_VERSION
if DJ_VERSION>=(1,4):
    from django.core.wsgi import get_wsgi_application
else:
    from django.core.handlers.wsgi import WSGIHandler as get_wsgi_application
    
class WSGIServerPool(PooledProcessMixIn, WSGIServer):
    pass
    
application = get_wsgi_application()
make_server('127.0.0.1', 8080, application, server_class=WSGIServerPool).serve_forever()