Re: rc improvements (wanted?)

On Fri, Jul 18, 2008 at 10:19:22PM -0700, Doug Barton wrote:
> Bernd Walter wrote:
> >Speaking about small systems, where startup time is more a problem than
> >on 08/15 desktop and server systems.
> >What I would love to see is that scripts like moused, ypserv, lpt, etc
> >are not started if the services are disabled.
> 
> That wold be a neat trick, how do you propose we accomplish it? (no, 
> I'm not being snide.)

Sorry - I have no proposal - just a few ideas, of which I know none is
perfect.
I think it would be great if rcorder can honour the settings somehow,
but I'm aware that rc.conf is a shell script and that some files are
not depended on any kind of _enable variable.
But if we can teach rcorder at least to read the foo_enable variables
and drop the script that provide foo.
Nevertheless I don't asume it to be a simple task.
I'm not familar enough with all the details on how to implement it nor
do I request it.
All I wanted to do is spread an idea and hope that someone can do
something with it.

> There are 144 scripts in /etc/rc.d/ on HEAD right now. Out of those I 
> count roughly 40 that I actually need, so let's round off to 100 
> unnecessary scripts to make the math easy. On my system (a pretty fast 
> C2D) it takes roughly .3 seconds of wall clock time to run one script 
> that is not enabled. Now cut that roughly in half since each of those 
> scripts will not have to suck in /etc/rc.subr and /etc/rc.conf* when 
> run at boot time, and that's 15 seconds of boot time that I could save 
> on average, let's say +/- 5 seconds. That's worth giving some thought to.

Well - my problem about this is more problematic:
[217]arm9# time /etc/rc.d/moused start
0.000u 0.000s 0:01.73 44.5%     1196+13491k 1+0io 2pf+0w

Ok - you wrote below that rc.subr and Co are loaded only once, so
the time on booting should be better.
Also the system is not completely idle right now.
But anyway - it's a much longer time for slow embedded systems.

rcorder is run only once and fast enough for that:
[220]arm9# time rcorder /etc/rc.d/* /usr/local/etc/rc.d/*
[...]
/etc/rc.d/bluetooth
/etc/rc.d/bgfsck
0.000u 0.000s 0:01.04 54.8%     777+17180k 0+0io 0pf+0w

> One way you could do this is to have /etc/rc.d/active and 
> /etc/rc.d/inactive (and probably an /etc/rc.d/system for critical 
> stuff that most people shouldn't touch). Then you could have a 
> vipw-like system to allow users to edit rc.conf that would move the 
> scripts to the right directory. Of course, this would be fraught with 
> potential for problems. :)

That's why I'm not doing it manualy, although there are scripts that
I will never need to enable.

> Another thing that would work for systems that a more sophisticated 
> admin/user updates with mergemaster would be to write a pre-compare 
> script that removes all the scripts you know you don't need from the 
> temproot/etc/rc.d so that they don't get installed. Of course the 
> benefit of that would not be nearly as wide spread, but it would also 
> not result in so much foot-shooting.
> 
> >So far each script is started, sucks in routines plus rc.conf
> 
> We're actually a little smarter than that. :) rc.subr and 
> rc.conf[.local] are loaded once by rc, then each script that runs 
> inherits those values.

That's bad to hear, since that fruit is already taken then ;-)

-- 
B.Walter <bernd_at_bwct.de> http://www.bwct.de
Modbus/TCP Ethernet I/O Baugruppen, ARM basierte FreeBSD Rechner uvm.