This never came up running Arch on my servers at home and on my laptop, but now that I am running thousands of virtual machines with miles of automation scripts murphy's law is rearing it's ugly head.

The problem I am running into is the fact that many Arch init scripts don't block until they are finished killing a process. So what happens is that restarts fail to start because the daemon has not always stopped before the start is called. Also, some automation systems (like puppet) don't always call restart, but sometimes call stop and then start.

My argument is that using a sleep call in an init script's restart call is non deterministic behavior and we can do better.

My proposal for a fix to this problem is to add a killproc function to the default functions, I have attached an idea for one, but think it is a subject worth discussing. Basically, it waits until the pid in question dies, but can stat a failure if the pid does not die, and can take an optional parameter to use a signal 9 if the normal signal 15 does not work.

This file and a little script to test it out are also available in my gihub:
https://github.com/thatch45/sandbox/tree/master/arch-rc