FAI

[Henning Sprang]

Hi Florent,

On Mon, 2005-07-25 at 11:02 +0200, Florent Vuillemin wrote:
> Hi Henning,
> 
> now that you've returned from Debconf, here is what I made about a
> multi-distribution FAI:
> 
> - doc: I've written several articles in a MediaWiki to explain the
> changes I made, I will now need to publish them: I can either install
> a MediaWiki on a personal website or add my articles to your wiki
> (whatever you want)

We will have the FAI wiki up and running in the course of the next few
days, this should be the place to put that documentation then.

A more interesting question would be how we will share our code to work
on that stuff.

Thomas, can we use the FAI alioth subversion for this? Is this already
usable?

> 
> - Nfsroot: as redhat/mandriva do not have a base system image like
> Debian, we have several solutions to create the NFSROOT directory:
> ---> Use SmartPM, which can install rpms ands debs using different
> kinds of repositories (but the soft is quite young and not
> well-documented)
> ---> Create a base system image for each ditro and distribute them on
> mirrors: this is how I proceed at the moment, at least to test the
> installation. Then each distribution can use its own package manager :
> apt for debian
> yum for redhat
> urpmi for mandriva
> 
> I wrote a redhat-specific make-fai-nfsroot script and updated
> make-fai-nfsroot.conf to include more distrib-specific variables. The
> idea is to have several make-fai-nfsroot.conf files in /etc/fai  and
> several NFSROOT directories (one per distribution to install). Then,
> when a client boots, it will mount one these NFSROOTs, depending on
> DHCPd/bootpd configuration.
> 

I would not go for an extra nfsroot per distribution, as I did it for
the Ubuntu FAI Port, when using FAI at home for installing ubuntu and
sarge from the same server - this seemed to me like too complex and
simply overkill. If you name a reason why we need that, O.K., but my
plan was, only add functionality to make-fai-nfsroot to create multiple
base<DISTRIBUTION>.tgz files, one for each distibution, and then decide
at install time, for example depending on the variable FAI_DISTRIBUTION
which base.tgz has to be unpacked for a specific host.

What I don't understand is, why do you want to distribute the base.tgz
files on mirrors, and not just create them with make-fai-nfsroot and put
it where it is now, just with a nametag for the distribution it is meant
for?

Can smartpm bootsrap a base system for rpm based distribution? That
would be an interesting feature.

If not, I would try to look at mach - "make a chroot"
http://thomas.apestaart.org/projects/mach/ and use it's code for
creating the base.tgz files. Probably mach would need some additions to
support more systems, like I see RHEL, what you and Axel want to do, is
not contained in mach by now.

> - Installation : I also made some changes to rcS_fai (/usr/sbin/fai)
> and subroutines (/usr/shar/fai/subroutines*) to make them compatible
> with RedHat

Ack. rcS fai should generally be the same for any distribution,
distribution specific parts should be run depending on the variable
FAI_DISTRIBUTION mentioned above.

> 
> - Packages : at the moment, I'm working on the rpms installation (how
> to parse the package_config files and use package groups with Yum).
> When this is ok, I will also adapt the post-config files and FAI
> should be able to install redhat systems (I'm testing with an RHEL3)

This sounds great. How is your progress there? do you need help? I can
only help with stuff that can be done with free software, as I can't
afford any RHEL license. Probably I could do the same work with centos.

> 
> - Configuration space : I plan to add classes for the distribution
> (ex: REDHAT, DEBIAN etc) which are the same as the value of the
> $DISTRIBUTION variable in nfsroot-xxx.conf (former
> make-fai-nfsroot.conf). I think there is no need to change the file
> structure of the configuration space, as we can change the class names
> to be distrib-specific (ex: REDHAT_DNSSERVER)

For the first step, that's completely O.K. - in the long term I believe
that FAI needs a better way to store metadata but in the name of the
files belonging to a class.

> 
> - CVS: at the moment, I'm working with cvs on my localhost, we might
> be interested in a public cvs server. We will also have to communicate
> on fai-devel about this project soon.

There we go, question about how to handle distributed versioning and
development already issued to Thomas above...

Sound like you already did some great work, thanks!

Henning 

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
Url : http://lists.uni-koeln.de/pipermail/linux-fai-devel/attachments/20050726/0c1d37d3/attachment.bin