Chad Walstrom said on Fri, Dec 05, 2003 at 02:17:55AM -0600: > > I did what you are trying to do using systemimager, cvs, and cvsup. > > ... There are a few rough spots (mostly in that I don't have a fully > > automatic way to restart daemons that have been updated in the golden > > client, so I have to restart them by hand), but in general it works > > very well, and has saved me a lot of time. > > This is all very well and find if you're using a single architecture > with nearly identical hardware setups. Granted, with discover and > read-edid, things are a little easier. I am using all x86, but I've got very different hardware setups; 4 different types of RAID, many 1U boxen with different ethernet cards that are not RAID, some SMP some not, some SCSI, some IDE, etc. Pretty much all of the non-autodetectable hardware configuration is restricted to three places: the kernel, /etc/fstab, and lilo.conf/grub. (Okay, 4 places if you include the X server config). I have an SMP and a non-SMP kernel installed in the golden image, and I have a SCSI vs. non-SCSI initial setup (due to the different device filenames requiring different setup commandlines and fstab), but other than that, there's nothing special necessary. The machines that are SMP have lilo configured to start the SMP kernel instead of the UP one (since there's no autodetect for that at boot time), but that's not a big deal. The kernel's are just the Debian default ones, which have most things built as modules; I just load the disk driver modules that I know I need from the initrd (which means that I'm wasting a little kernel memory, but I could unload the unused ones after boot; I just don't). FAI is good, but it doesn't handle updating the systems once you have them installed. M
Attachment:
pgpYTLzumLnzi.pgp
Description: PGP signature