A lot of people don’t like Upstart, complaining that it’s a pointless solution to something that was never a problem in the first place. I’m nowhere near clever enough to have an opinion on this, but I like rolling with the times.
This site (and a few others I look after) runs on Django which is my current favourite-CMS-of-the-month. Django runs as a bunch of FastCGI processes which are then fronted by a Lighttpd reverse-proxy. This is all well and good, but when the server reboots, there needs to be some nice way of telling the FastCGI processes to start up. I’d previously rolled my own rather buggy init.d script, but there was no telling whether it would actually work from one reboot to the next.
Trawling the net, I found someone had written a (seemingly) decent upstart FastCGI script over at Sam Bashton’s blog. Bastardising this, I created my own version:
description "Basic Django fcgi Upstart config" author "Sam Bashton" start on (net-device-up and local-filesystems and runlevel ) stop on runlevel  respawn chdir /home/growse/django-sites/growse_com exec su -s /bin/sh -c 'exec "$0" "$@"' www-data -- /home/growse/django-sites/growse_com/manage.py runfcgi method=prefork socket=/home/growse/django-sites/run/growse_com.socket pidfile=/home/growse/django-sites/run/growse_com.pid minspare=4 maxspare=8 maxchildren=128 maxrequests=65535 daemonize=false
Several reboots later, this seems to work perfectly.
I quite like the theory behind upstart. Create a conf file, drop some dependencies and definitions in it and just add it to /etc/init. Not altogether sure what the fuss is all about…