My friend Mike and I were talking a while back about Unix init systems
and came to the conclusion that mk's dependency tracking could come in
handy. I decided to implement it a few days ago using plan9port and
thought that some of the folks here might be interested. Although, I
still haven't decided if it's really neat, or completely evil :-)

The code is available here, along with a fuller README:

  https://lug.rose-hulman.edu/svn/misc/trunk/mkinit/


It consists of a bash script (the rc version has issues..) that runs as
the actually init daemon, a mkfile that contains a list of rules for
starting and stopping services, and a some helper programs.

It's based on mk so any mk rules could be used. What I use is a few
rules to denote runlevels and then a bunch of rules for starting and
stopping services of the form:

  foo-start:VPservice -u: deps-start ..
  foo-stop:VPservice -d: rdeps-stop ..
  
service -[ud] is just a helper script that determines if the services is
already up or down in order to avoid re-starting/stopping it. Dependency
tracking is pretty strait forward, but foo-stop dependencies are
backwards because the dependencies for the rule denote the services that
must be stopped before foo can be stopped. What this ends up creating is
two separate DAGs, one for booting and one for halting.

I've been using it the past couple days and it seems to work pretty
nicely, but it's still largely untested, and there are a few bugs such
utmp being broken.. It's also a good bit faster than sysvinit due to
mk's parallelization, bootcharts below.

  http://andy753421.ath.cx/temp/mkinit.png
  http://andy753421.ath.cx/temp/sysvinit.png


Reply via email to