After some messing about, I've finished my dependency system for runit,
and am now using it to start up and shut down my machine. =)

I've hacked about on svwaitup and svwaitdown a little; the diff is
enclosed and is required to make the dependencies work. Summary of
changes:
* svwaitup has an option to retry (with a variable delay) if the
  supervisor is not running or the service is not requested to be up.
  This was needed as the dependencies are brought up in background
  processes for parallelism, so the parent has to wait.
* svwaitdown has an option not to send the down command to services
  (still defaults to doing so). This is needed for the inverse situation
  of the above. =)
You need the previous patch I posted as well which creates the tools
svisup and svisdown.

The actual dependency management is done by having links in
servicedir/updeps that point to services which must be up first. The
reverse dependencies can be automatically generated by the enclosed
script svdepcalc, though at the moment it's really stupid and doesn't do
any kind of checking for cycles..etc.

Once the dependency links are there, svup and svdown will bring services
up and down in a way which obeys all dependency rules. The way I use it:
* All services are in a single directory; no pseudo-runlevels or
  anything
* All services have a 'down' file to stop them starting.
* The /etc/runit/1 script runs svup in the background on a dummy service
  which depends on everything I want to run (apache..etc).
* runsvdir starts and brings up the supervisors, and as it does so, svup
  brings services up.

Once the machine is up you can use svup and svdown freely to start and
stop services in a way which follows the dependencies. Then, to shut
down:

* /etc/runit/3 runs 'for SRV in /service/*; do svdown $SRV; done' which
  brings services down in line with dependencies. At least, if they
  terminate properly. There is a timeout for services that don't. (they
  are not killed at this stage)
* /etc/runit/3 runs svwaitdown -kx with a short timeout to clean up any
  services which did not get terminated cleanly by the above
* Halt.

I make few promises about the actual correct functioning of these
scripts and strongly recommend that you have a good read of them (in
fact, it'd be nice if you did that anyway to give me some feedback).
They work for me on exactly one machine, and have been tested a bit on
some fake services as well, but that's not a *lot* of testing.. =)

Thanks in advance for any comments, and I hope that at least someone
finds this useful =)

Torne