+automatically configure / install linux distributions (generic)

Signed-off-by: Nico Schottelius <nico@kr.ethz.ch>
This commit is contained in:
Nico Schottelius 2011-08-06 02:35:14 +02:00
parent 11518ecec2
commit 60a890c60a

View file

@ -0,0 +1,93 @@
[[!meta title="Generic automatic installation for different Linux distributions"]]
## Introduction / Motivation
If you've been a little bit longer active as a sysadmin, you'll have faced
different Linux distributions and got used to their pros and cons.
One thing though that I cannot (and do not) want to get used to, is that
every distribution contains its own way of having an automated installation.
May it be [kickstart](http://fedoraproject.org/wiki/Anaconda/Kickstart),
[preseed](http://wiki.debian.org/DebianInstaller/Preseed) or
[aif](http://www.archlinux.org/packages/extra/any/aif/) (or any other
I've not mentioned), all contain their own specific settings to do
the same again and again.
Having a lot of discussions with regarding this topic, there is one
approach that seems to kick them all, can be automated easily and
extended for other distributions (or other Unices) and probably
easily developed as well:
## Unix = collection of files + metadata
Unices and Linux distributions are just a number of files in some particular
filesystem plus some metadata like partitions (or slices), as
some other people at [OpenQRM](http://openqrm.com/?q=node/2)
have recognised this as well:
We asked ourself "what is linux ?", it is the kernel, an initrd,
some modules and a root-filesystem. Those are all "just" files
..... so we should treat them like files by putting and managing
them on modern storage-servers.
But instead of doing the whole infrastructure management stuff,
virtualisation (hear the cloud?) and monitoring environment, I'll
just focus on the part of the installation.
## Installation into/from a directory
Assume the following: Every Linux distribution can somehow be put
into a directory. Either by tar'ing an existing installation or
by using specific tools like
[debootstrap](http://wiki.debian.org/Debootstrap),
[febootstrap](http://people.redhat.com/~rjones/febootstrap/) or
[archbootstrap](https://wiki.archlinux.org/index.php/Archbootstrap).
This proces is pretty easy and can be used as a starting point for
a generic installation.
## Add some metadata
Now take this directory, and put it onto the harddisk. Installation
done. Well. almost:
* Todo before
* Create partitions
* Create filesystems
* Todo after
* Adjust /etc/fstab
* Add correct boot loader
* Configure network + ssh
After you've created the partitions and the filesystem
(which is distribution independent!), you can copy over the files
to the target diretory, in which the selected partitions are mounted.
For post processing, you need to adjust the fstab, so partitions
(and swap, do not forget computers with 4 or 8 MiB main memory!)
will be used at the right location. Adding a bootloader would be of
great help for the boot process (almost distribution independent),
though booting from the network
via PXE may even get around this.
Finally the network configuration (distribution specific!) must be
adjusted and ssh should be installed, so the system can be configured
remotely (for instace with [[software/cdist]]).
## Todo
So the next steps are pretty straight forward:
Write something that automatically
* configures partitions
* creates filesystems
* mounts the filesystems
* copies files from a know location
* adjust the operating system
* umounts everything and
* reboots
Doesn't sound like very much and so
I'll probably give it a try in the next weeks, so stay tuned if you're
interested in this topic.
[[!tag eth unix]]