Rainbow Saddle

.. image:: https://travis-ci.org/flupke/rainbow-saddle.svg :target: https://travis-ci.org/flupke/rainbow-saddle

.. note:: I no longer use rainbow-saddle, and thus cannot actively maintain it.

rainbow-saddle is a wrapper around Gunicorn <http://gunicorn.org/>_ to simplify code reloading without dropping requests.

Installation

Install from pypi::

$ sudo pip install rainbow-saddle

Or from source::

$ sudo ./setup.py install

Why?

Sometimes doing a kill -HUP <gunicorn PID> is not sufficient to reload your code. For example it doesn't work well if you host your code behind a symlink <https://github.com/benoitc/gunicorn/issues/394>, or if a .pth in your installation is updated to point to a different directory <https://github.com/benoitc/gunicorn/issues/402>.

The correct way to reload code in such situations is a bit complicated::

# Reexec a new master with new workers
/bin/kill -s USR2 `cat "$PID"`
# Graceful stop old workers
/bin/kill -s WINCH `cat "$PIDOLD"`
# Graceful stop old master 
/bin/kill -s QUIT `cat "$PIDOLD"`

It also has the downside of changing the "master" process PID, which confuses tools such as supervisord.

rainbow-saddle handles all of this for you, and never changes its PID. Reloading code becomes as simple as sending a SIGHUP again::

$ rainbow-saddle --pid /tmp/mysite.pid gunicorn_paster development.ini --log-level debug 
$ kill -HUP `cat /tmp/mysite.pid`