|GUNICORN_PASTER(1)||User Contributed Perl Documentation||GUNICORN_PASTER(1)|
gunicorn_paster - Event-based HTTP/WSGI server, Paste application entry-point
gunicorn_paster [OPTIONS] [SETTINGS_PATH]
- -c CONFIG, --config=CONFIG
- Config file. [none]
- -b BIND, --bind=BIND
- Address to listen on. Ex. 127.0.0.1:8000 or unix:/tmp/gunicorn.sock
- -w WORKERS, --workers=WORKERS
- Number of workers to spawn. 
- -a ARBITER, --arbiter=ARBITER
- gunicorn arbiter entry point or module [egg:gunicorn#main]
- -p PIDFILE, --pid=PIDFILE
- Set the background PID FILE
- -D, --daemon
- Run daemonized in the background.
- -m UMASK, --umask=UMASK
- Define umask of daemon process
- -u USER, --user=USER
- Change worker user
- -g GROUP, --group=GROUP
- Change worker group
- -n PROC_NAME, --name=PROC_NAME
- Process name
- Log level below which to silence messages. [info]
- Log to a file. - equals stdout. [-]
- d, --debug
- Debug mode. only 1 worker.
- Show program's version number and exit
- -h, --help
- show this help message and exit
Green Unicorn (gunicorn) is an HTTP/WSGI server designed to serve fast clients or sleepy applications. That is to say; behind a buffering front-end server such as nginx or lighttpd.
* Optional support for Eventlet and Gevent to provide asynchronous long-polling ("Comet") connections. * Process management: Gunicorn reaps and restarts workers that die. * Easy integration with Django and Paster compatible applications (Pylons, TurboGears 2, etc. * Load balancing via pre-fork and a shared socket * Graceful worker process restarts * Upgrading without losing connections * Decode chunked transfers on-the-fly, allowing upload progress notifications or stream-based protocols over HTTP
There are various kernel parameters that you might want to tune in order to deal with a large number of simultaneous connections. Generally these should only affect sites with a large number of concurrent requests and apply to any sort of network server you may be running. They're listed here for ease of reference.
The commands listed are tested under Mac OS X 10.6. Your flavor of Unix may use slightly different flags. Always reference the appropriate man pages if uncertain.
INCREASING THE FILE DESCRIPTOR LIMIT¶
One of the first settings that usually needs to be bumped is the maximum number of open file descriptors for a given process. For the confused out there, remember that Unices treat sockets as files.
$ sudo ulimit -n 1024
INCREASING THE LISTEN QUEUE SIZE¶
Listening sockets have an associated queue of incoming connections that are waiting to be accepted. If you happen to have a stampede of clients that fill up this queue new connections will eventually start getting dropped.
$ sudo sysctl -w kern.ipc.somaxconn="1024"
WIDENING THE EPHEMERAL PORT RANGE¶
After a socket is closed it eventually enters the TIME_WAIT state. This can become an issue after a prolonged burst of client activity. Eventually the ephemeral port range is used up which can cause new connections to stall while they wait for a valid port.
This setting is generally only required on machines that are being used to test a network server.