init observations

[Luke Macken]

I've been playing around with initng quite a bit lately and have had a
good amount of luck getting it running successfully on my rawhide laptop, a
VMWare test environment, and my FC4 desktop.  Had a few issues right off the
bat, but eventually worked most of them out (thanks to the help of
Daniel Malmgren).  I believe fedora related patches even hit their repos
as well.

Here's what I've seen so far...

initng - http://initng.thinktux.net
======
Pros
 o Dynamic service dependencies
 o Service monitoring
 o Automatic respawning of services
 o Parallelized service startup
 o Plugin support
 o Very active and helpful community
   - Extremely open to getting initng working by default in Fedora
 o Supports /etc/rc.* scripts via a plugin, but uses it's own format by
   default (also supports xml init scripts via plugin)
 o FAST AS HELL[0]

Cons
 o Gentoo look-and-feel (brings back old memories)
   - Having such an open-minded community, giving initng a more
     unified/professional feel would hopefully be accepted
 o No inherent D-BUS support
   - initng's plugin support would allow this to be accomplished (but I
     would bring it up to the developers; who knows, they might want it
     upstream?)

- - -

After talking with Harald about his SystemManager, apparently he hasn't
actually gotten it complete enough to successfully start a system with
it yet, but he kindly answered all of the questions I had.

SystemManager - :pserver:anonymous at rhlinux.redhat.com:/usr/local/CVS servicemanager
=============
Pros
 o /etc/rc replacement; so integration into our current SysVinit setup
   would be close to trivial
 o Will work with our current init scripts without modification
 o Automatic/On-demand respawing of services
 o Exporting of services as D-BUS objects
 o Dynamic service dependencies

Cons
 o Not complete/tested
 o Still supports the /etc/rc.d/rc$i.d framework (which is horrible by
   design)

- - -

I'm sure I missed many aspects of the init/bootup procedure in the above
observations, as this is far from a complete analysis.  Please jump in if
you have any additional observations/questions/comments/corrections.

So.. our first step ?

Getting initng into extras definitely won't hurt.  It will give everyone
a chance to play it with, and hopefully will help work most of the kinks out.
As far as SystemManager goes; it's definitely a project with lots of
potential, and if the demand is there, I'm ready to start hacking.

What does everyone else think ?


luke

[0]: I generated bootcharts for a default FC4 install with minimal
tweaking (removed a few unnecessary services) and FC4 with a
default initng install.  These are in no way supposed to be an accurate
measurement of the true speeds of either of these versions of init
(they are also both running different init scripts).

http://people.redhat.com/lmacken/initng-bootchart.png
http://people.redhat.com/lmacken/SysVinit-bootchart.png