Revision controlled configurations.

Henning Glawe glaweh at physik.fu-berlin.de
Tue Mar 9 10:08:40 CET 2004


On Mon, Mar 08, 2004 at 07:39:50PM +0100, Josef Wolf wrote:
> While looking how to enable FAI to use subversion as a revision control
> backend, I got somewhat confused by following command sequence from the
> get_fai_cvs() function:
> 
>         export FAI_CONFIG_AREA=$FAI
>         export FAI=/tmp/$(basename $FAI_CONFIG_AREA)
> 	[ "$debug" ] && echo "\$FAI now points to $FAI"
> 	if [ -d "$FAI_CONFIG_AREA" ] ; then
> 	   echo "Config found at $FAI_CONFIG_AREA: Copying"
> 	   cp -a $FAI_CONFIG_AREA $FAI 
> 	fi
> 
> This effectively intermixes a (clients) local config area with the revision
> controlled config area. What is this good for? IMHO, a revision controlled
> configuration should be exactly what it is intended to be: revision
> controlled. Intermixing it with non-revision-controlled configuration
> options seems to be a BadThing(TM) to me.
> 
> Could someone please help to clarify my confusion?
$FAI_CONFIG_AREA contains the _revision controlled_ configuration space
from the last fai run on the client. I copy $FAI to /var/lib/config from
${FAI}/scripts/.
if a system is updated, the code you mention copys it back to /tmp/, cvs
updates it there and logs all changes to $LOGDIR. this logfile is later
used with fcopy's "-P" and "-L" options to minimize configuration file
updates and the necessairy daemon restarts when using
fai-update-system (which is basically a fai installation, skipping
partitioning, formatting and base.tgz unpacking).

only after a completed run $FAI_CONFIG_AREA is replaced by the updated
version, so in-update reboots or other failures don't lead to
configuration inconsitencies.

-- 
c u
henning



More information about the linux-fai mailing list