template for a gunicorn upstart job that can run several instances of a django application

gunicorn-upstart.conf.template

# %(mysite)s - run %(mysite)s instances (default is the main production instance)
#
# This runs gunicorn-django for %(mysite)s; to install:
#  * sudo ln -s <this file> /etc/init/%(mysite)s
#  * sudo initctl reload-configuration 
#
# it expects the following directory layout:
#
# /home/%(mysite)s/public_html
#   \-env -> virtualenv
#   | \-bin
#   |   \-gunicorn_django
#   |   \-activate
#   \-<instance>
#   | \-app -> application
#   | \-conf
#   | | \-gunicorn_<instance>.conf.py
#   | | \-<this file>
#   | \-log -> logs
#   | \-run -> pid files and the like
#   \-<instance>
#   ...
#
# to use:
#  * sudo start/stop/restart/reload/status %(mysite)s
#  * sudo start/stop/restart/reload/status %(mysite)s instance=testing

description "%(mysite)s gunicorn instance, default is the main instance"

start on runlevel [2345]
stop on runlevel [06]

expect fork

respawn
respawn limit 10 5

instance ${instance:-main}

env APPHOME="/home/%(mysite)s/public_html"

pre-start script
    INST=${instance:-main}
    GUNICORN="$APPHOME/env/bin/gunicorn_django"

    test -n "$GUNICORN" -a -x "$GUNICORN" || { stop; exit 0; }
    test -d "$APPHOME" || { stop; exit 0; }
end script

script
    INST=${instance:-main}
    CONF="$APPHOME/$INST/conf/gunicorn_$INST.conf.py"
    GUNICORN="$APPHOME/env/bin/gunicorn_django"

    PIDFILE="$APPHOME/$INST/run/$INST.pid"
    LOGFILE="$APPHOME/$INST/log/gunicorn.log"

    cd "$APPHOME/$INST/app"
    source "$APPHOME/env/bin/activate"
    exec $GUNICORN --pid="$PIDFILE" --name="$INST" --user=django --group=django --config="$CONF" --daemon --log-file="$LOGFILE" 2>>"$LOGFILE"
end script

post-stop script
    INST=${instance:-main}
    rm -rf "$APPHOME/$INST/run/$INST".{pid,sock}
end script

# vim: set et sw=4 ts=4 sts=4

This would be really handy. It seems that $GUNICORN doesn't get passed from the pre-start to the run script though. I tried using the "export" command and didn't have much luck with that either.

it was actually a bug (fixed now), they are executed in different shell processes

I gave up on upstart a long time ago and am running gunicorn under supervisord now, btw :)

I got upstart working finally - though I found that it was impossible to pass variables around the way I was thinking. Your instance-based variable storage is probably the way to go. I'll probably revisit my script when launching multiple instances on a server.
Thanks for the gist. It was inspirational :)

What's the purpose of the source "$APPHOME/" on line 58? It was causing my upstart script to fail, and I can't really think of a good reason for sourcing a directory.

hmm, I have no idea why I left out the rest of the line, it was supposed to source the activate script, changing now

I don't think "--daemon" should be set in upstart, because you don't want to detach it from the upstart service. The daemon option is used when you run gunicorn_django directly from the command line.

Things to mention:

• Upstart in ubuntu 12.04 uses /bin/sh in script section, so source is . in /bin/sh
• Do we really need expect fork and --daemon in gunicorn part?