disscussion about new format of disk_config files

[Jan Jansen]

On Wednesday 05 October 2005 16:34, Henning Sprang wrote:
> On Wed, 2005-10-05 at 16:25 +0200, Michal Svamberg wrote:
> > Hi,
> > actual disk_config not supported this features:
> >  - RAID (mdadm)
i think the best way to implement a file based cfg for creating raid devices 
is to use the std. 'mdadm.conf' syntaxis and to store them in a file 
'disk_config/<class>.RAID'. This makes sure that special features of mdadm 
like autoscaning, assembling by uuid etc. can be easily used when creating 
the raid by simply running mdadm with the corresponding cfg file. This way 
makes also different, class depending raid cfgs possible. Additionaly, i 
cannot imagine how a detailed raid cfg can fit into the current cfg's format 
style without producin' textual overhead and complicating or reducing 
functions of the raid cfg possibilties!
Other as setup_harddisks does at the moment, raid devices shouldn't be allways 
created, but only if the cfg of the current install matches the cfg of an 
existant raid so the install can just assemble it and to not destroying all 
data on it (what would be very if just not thought of before running an 
install). Also, a usefull and dynamic creation of all disks/partition with a 
common size is required to not waste disk space if different sizes of disks 
exist. The chunk parameter should be included when creating fs's (e.g.: 
mkfs.ext3 --stride=?). 
Perhaps a repair mode for replacing/adding disks to an existant raid.?
> >  - LVM (lvm2)
Same as above for assembling existant volumes to avoid data destruction.
Should also process Chunk/Stride Parameters from raid setup to enhance 
performance.
I've written a hook some time ago, that srces a cfg file of the following 
type:
vg_config data
disks md0
ndb /var/data/ndb 150M rw,errors=remount-ro ext3 ;-E stride=8
h0a /var/data/h0a 3.8G rw,errors=remount-ro ext3 ;-E stride=8
h0b /var/data/h0b 75G  rw,errors=remount-ro ext3 ;-E stride=8
h0c /var/data/h0c 25G  rw,errors=remount-ro ext3 ;-E stride=8
h0d /var/data/h0d 3.8G rw,errors=remount-ro ext3 ;-E stride=8
h0e /var/data/h0e 5G  rw,errors=remount-ro ext3 ;-E stride=8
h0f /var/data/h0f 3.8G rw,errors=remount-ro ext3 ;-E stride=8
h0g /var/data/h0g 40G  rw,errors=remount-ro ext3 ;-E stride=8
h0h /var/data/h0h 50M  rw,errors=remount-ro ext3 ;-E stride=8
h0i /var/data/h0i 50M  rw,errors=remount-ro ext3 ;-E stride=8
home /net/home 35G rw,errors=remount-ro ext3 ;-E stride=8

which creates physical vol. md0, the vg 'data' with it and creates the logical 
volumes below the 'disks' line with the given size and filesystem.

I think this is more compact and simple than splitting pv/vg/lv cfgs.
> >  - change geometry of disk
should be supported by giving a value in the cfgfile like "-150M" or an size 
different to that of an existing lv. So the disk should be reduced by 150M or 
modified to the new static size. (Reduce must check if no data-loss can 
occour).

I'm very happy that these features are planned to be implemented in fai, 
because i use these two features (LVM/RAID) very often in my 
installations. :))
-- 
-----------------------------------------------------------------
EKF Elektronik GmbH Industrial Computers & Information Technology
Jan Jansen           mailto:jan at ekf.de        http://www.ekf.de
Philipp Reis Str. 4                      fax: +49 (0)2381-6890-90
D-59065 Hamm (Germany)                  phone: +49 (0)2381-6890-0
-----------------------------------------------------------------