I’ve been working on some software to help create flexible cloud-based infrastructure. The most recent component that I’ve been working on required the creation of a daemon that could pick up some messages and whisk them off to a queue for later processing. Writing a successful daemon is not that hard, but the details around how to do it seemed a bit vague to me so I thought I would document what I did to help other people in the future.
Any program could be a daemon. All it has to do is follow a few rules:
- Fork twice (this allows it to really go to the background and become the session leader)
- Change its working directory to /
- Set the umask to 0
- Redirect stdin, stdout and stderr to a logfile or /dev/null
- Write its PID to a pidfile that its init script references (This is probably not a requirement of all daemons, but if you’re working with Debian GNU/Linux and want to use the skeleton init scripts without much work you should do this)
After I figured all of that out it was pretty easy. I ended up implementing my program in Python and used this page as a reference.
The only other trick that I discovered was that the init script needs to know the exact name of the executable as if it were seeing it in ps. In Debian, init scripts are basically a front end for killall and without the actual process name they can’t stop the process.
So all in all, not too bad. It’s great to be able to write software that fits with the rest of the system. A consistent computer makes it easy when multiple people are involved with administering it.