Dimitrie O. Paun wrote:
On Tue, Nov 30, 2004 at 11:08:30PM -0500, Jeremy Katz wrote:
at the moment... but this functionality is already at least sort of
there. Look at running anaconda with --rootpath /path to install
packages under /path.
I'd like to see a way of completely decoupling the hard disk imaging in
anaconda. With the -rootpath option and no hard disk imaging, I don't
see the point of keeping Kickstart Configurator around. You'd use the
same tool to build a kickstart file that you use install an anaconda
based system with. It seems that, if the rest of the python based
system tools were built carefully, they could be used to easily extend
anaconda in a meaningful two for the price of one kind-of way. A new
config tool could be put in the dispatch.py file. Hooks for these tools
in Anaconda sound like a way enhancing support for the goal of a read
only root in Stateless Linux.
You know I'd almost like to see the package install completely decoupled
too. If I create a standard that this group of machines have this set
of packages installed because of their role as a "Thick client"
http://fedora.redhat.com/docs/stateless/index.html#sn-overview-clients,
then I could use the proposed anaconda to configure /etc for the chunk
of code in /usr. I would then run the installer once to build /usr, and
then run the installer multiple times to build various /etc options in
each computer's role.
I've just tried to use anaconda with the --rootpath option.
It's not bad, but I think it's still not usable as a regular
application to create these virtual roots. Here are a few of
my observations:
1. it still does hardware detection:
[root@dimi ~]# anaconda --graphical --rootpath=/home/dimi/dev/kogaion/virtualroot.anaconda --method=http://dimi/FC3/
* Display mode = g
* Method = http://dimi/FC3/
Probing for video card: ATI Radeon 9600
Probing for monitor type: Dell 1901FP (Digital)
Probing for mouse type: Skipped mouse probe.
2. it still asks for langauge, keyboard settings instead of
just using the current ones, like any regular X app
I would think this is ok. Anaconda still has to understand it's
operating environment.
3. I can not close the window. I have to literaly kill the
app with kill(3) or xkill.
4. It still tries to mess with my disks:
* moving (1) to step partitionobjinit
* pv is /dev/hda2 in vg VolGroup00, size is 80866181120
* vg VolGroup00, size is 80866181120
* lv is VolGroup00/LogVol00, size of 79255568384
* lv is VolGroup00/LogVol01, size of 1577058304
* running vgchange failed: 1280. disabling lvm
These may be geared to the Stateless Linux push. If the partitioning is
pushed off to the very last step of anaconda before the install takes
off, then there may be some other advantages. One, it may be easier to
support the decoupling of running the installer without imaging hard
disks. (I don't recall right now, if there was a --test option that
turned off imaging disks.) Two, if the package selection is performed
right before imaging hard disks, then some actual numbers could be fed
into partition.py verses these hard coded guesses:
checkSizes = [('/usr', 250), ('/tmp', 50), ('/var', 384),
('/home', 100), ('/boot', 75)]
5. I don't think asking about timezones is appropriate
in this context, it should use the current one.
Again I'd rather see the disk issue solved. I'd rather put up with this
because it completely separates anaconda from code variations for a
"running system" verses "anaconda for a real install". I would imagine
this would add to additional testing. This is the same reason 2. above
seems ok to me.
<snip>
I got a few patches from jrb when he was working
on the Stateless stuff to get it working better, so he might have a
better idea of where it currently stands (cc'd).
Cool, it would be nice if we can get anaconda to have a "regular app" mode
so we can use it out of the box to create images for diskless clients.
Or other creative uses. ;-)
Greg
