Gunicorn 0.5 introduced the ability to use a Python configuration file. Gunicorn will look for gunicorn.conf.py in the current working directory or what ever path is specified on the command line with the -c option.
arbiter = "egg:gunicorn" # The arbiter to use for worker management
backlog = 2048 # The listen queue size for the server socket
bind = "127.0.0.1:8000" # Or "unix:/tmp/gunicorn.sock"
daemon = False # Whether work in the background
debug = False # Some extra logging
keepalive = 2 # Time we wait for next connection (in seconds)
logfile = "-" # Name of the log file
loglevel = "info" # The level at which to log
pidfile = None # Path to a PID file
workers = 1 # Number of workers to initialize
umask = 0 # Umask to set when daemonizing
user = None # Change process owner to user
group = None # Change process group to group
proc_name = None # Change the process name
tmp_upload_dir = None # Set path used to store temporary uploads
worker_connections=1000 # Maximum number of simultaneous connections
after_fork=lambda server, worker: server.log.info(
"Worker spawned (pid: %s)" % worker.pid)
before_fork=lambda server, worker: True
before_exec=lambda server: server.log.info("Forked child, reexecuting")
The arbiter manages the worker processes that actually serve clients. It handles launching new workers and killing misbehaving workers among other things. By default the arbiter is egg:gunicorn#main. This arbiter only supports fast request handling requiring a buffering HTTP proxy.
If your application requires the ability to handle prolonged requests to provide long polling, comet, or calling an external web service you'll need to use an async arbiter. Gunicorn has two async arbiters built in using Eventlet or Gevent. You can also use the Evenlet arbiter with the Twisted helper.