Re: [gentoo-installer] how to setup deployment server

[Chris Gianelloni]

On Thu, 2006-01-12 at 04:39 -0600, Tom McDonald wrote:
> Sven Vermeulen wrote:
>
> >On Mon, Jan 09, 2006 at 04:24:13PM -1000, Gavin Seo wrote:
> >
> >
> >>Does anyone have any documentation/guides on how to setup a automated
> >>deployment server?  I currently want to install gentoo on 30 nodes, I have
> >>an existing machine running gentoo that has identical hardware to the 30
> >>nodes that need gentoo system installed. If anyone has experience in
> >>setting up a deployment server, it would be greatly appreciated if you
> >>would share your experience.
> >>
> >>
> >
> >I have some experience on a quite similar field. The entire network is using
> >the same DHCP server; when the system boots from the network (which is only
> >allowed with the appropriate credentials af course) a small Linux image is
> >downloaded and started.
> >
> >That image fires up the machine, connects to a PartImage server and places
> >an image on the file system. The image itself is not tied to a particular
> >host as all configuration specific to the system is obtained from LDAP or
> >otherwise.
> >
> >However, due to the lack of PartImage development speed, I am contemplating
> >on developing something more robust and flexible, so I am also interested in
> >hearing from others...
> >
> >Wkr,
> >      Sven Vermeulen
> >
> >
> >
> Check out systemImager on source forge.

I have setup such a system for a previous employer.  Unfortunately, I
have not gotten permission to distribute it as an open source project
outside of the company.  Anyway, it is really fairly simple..

First, I built a custom kernel, using genkernel.  I used the initramfs
overlay option to add my own files to the initramfs, along with my own
linurc.  Essentially, I only extended genkernel's linuxrc to make it
fire up a bunch of extra scripts at the end.  The scripts determine the
architecture type of the machine, then proceed to download the latest
version of a stage4 tarball from a web location specified in the kernel
command line (similar to kickstart).  This allows for multiple server
types per architecture.  The drives are partitioned and the file-system
is built, then the tarball is uncompressed onto the file-system.  The
system boots via DHCP and reports its IP address to DNS using DDNS.  It
also fires off a provisioning script for our DHCP server to setup a
reservation.  After that, the bootloader is installed on the system, and
upon completion, the system sends an email to a specified address, then
reboots.  Of course, we also use LDAP for authentication, and store most
of our information in either LDAP or on large filers running NFS, so
configuration isn't as much of an issue, though the tarballs are
pre-configured.

Since this doesn't use any form of imaging, it is actually quite quick,
and also allows for any type of file-system supported by the kernel to
be used.

-- 
Chris Gianelloni
Release Engineering - Strategic Lead
x86 Architecture Team
Games - Developer
Gentoo Linux