Startup needs overhaul

Robert Fleming rob at unixadministrator.org
Wed Aug 17 21:38:50 CDT 2005


launchd is probably the best answer to the whole startup issue.  However 
there are certain things that I would like to point out. 

We are still relying on the startup scripts that mimic the redhat 
startup processes.  They look cute but are they really helpful.  No.. In 
reality they provide no relative information.  I'm not concerned with a 
process "starting" or whether it came up [ ok ].  Yes it looks cute but 
it's not really functional. 

example. 
Hotplugd.  I'm a desktop user.  I have not hotplug network nics on the 
computer.  Nothing pcmcia, and nothing usb..  So why is the network 
configuration script hanging around waiting for hotplug to figure out 
that I have neither of the two and startup?  Hotplug services should be 
configuration optional  those that are working with USB devices should 
have the option after the initial startup to configure them.  I'm not 
running anything on my box that would ever need it so why make a 
dependency out of it.  eth0 is found by the kernel and the network 
address should immediately get configure, routed, and resolve.  No waiting.

The real dependencies are things like disk sub systems, network devices, 
etc.  They are things that init 1 and init 2 should take care of but in 
a non-redhat way.  No notification just get it going. 

If there is notice that should be given for failed services it should be 
in logs, email or some other method.  If a novice needs to know that 
something failed put it in tray app with a bouncer that notifies them 
when they have actually logged in. 

The startup should be configured in such a way that processes are 
backgrounded, threaded and non dependant.  Just because the gui comes up 
doesn't mean startup has completed successfully and that all services 
are running. 

I want my linux distro to actually startup like an enterprise system 
does.  If there is a failure it should be syslogged, and emailed.  If we 
are going to truly try to manage failure then we should write a startup 
app that can scan the loggs created by a service like postfix and deal 
with it in the best most efficent manner. 





More information about the ubuntu-devel mailing list