FAI for the impatient, my personal notes

Steffen Grunewald steffen.grunewald at aei.mpg.de
Tue Aug 10 16:04:56 CEST 2004


I found that the documentation still hides important parts from the
user, at least from a somewhat stupid user like me. In particular, the
section "For the impatient user" will disappoint impatient users who
refuse to read the whole manual, and want to see it work the first time
(Although I'm a cluster admin, I've only used RedHat's kickstart before,
so first I'd like to get some feeling for what's happening there...
before converting everything to Debian and FAI.)

My setup (on top of my desk, not in the cluster room) is that of a
little geek inside a large institute, so for my tests I cannot refer
to name servers or central DHCP services, nor do I have a NIS at hand.
(there will be a NIS for the cluster, and perhaps also a DNS)

There are just two machines: one running sarge (pretty much up-to-date),
called "faiserver", and an old mainboard with a WD card, a hard disk,
and a floppy connected to it: "demohost". (Since it is quite slow, and
the graphics card is very old, I don't have a XF86Config-4 for it, I'll
be happy with text mode.)
The client is equipped with a non-PXE network interface card (WD8003 to
start with), and etherboot will help it to do PXE nevertheless.

After some unsuccessful runs (which also included the use of a 3COM905
card) I came up with the following checklist:

- install the necessary software as described in the docs
	apt-get install fai fai-kernels \
		mknbi dhcp3-server tftpd-hpa \
		rsh-server wget syslinux
	cd /usr/local/bin
	wget -r -l1 -nd www.informatik.uni-koeln.de/fai/download/imggen
	chmod +x imggen
- very important: copy the various config files to their locations from
  /usr/share/doc/fai/examples/etc (the documentation only recommends to
  read them, but I found it essential to *use* them). Add demohost to
  /etc/hosts. (demohost is currently [2.6] missing from most files.)
  add "demohost" to /etc/netgroup, and change the "netgroup" line in
  /etc/nsswitch.conf to "files,nis" (note: this doesn't seem to work, so
  I also added "demohost root" to ~fai/.rhosts! otherwise, task_chboot
  would fail. This needs further investigation. Anyone NIS-less?)
  edit /etc/dhcp3/dhcpd.conf (leave the root-path entry!!!)
  edit /etc/fai/sources.list to reflect the local mirror location.
- edit /etc/fai/fai.conf and .../make-fai-nfsroot.conf: important lines
  are NFSROOT_ETC_HOSTS (matching /etc/hosts!) and FAI_DEBOOTSTRAP
  (matching sources.list)
- check /etc/exports. If your nfsroot and config space are on the same
  filesystem, better export only that (nfs-kernel-server got problems).
  	/etc/init.d/nfs-kernel-server reload
- run
	fai-setup 2>&1 | tee fai-setup.log
- check /etc/exports again, you might have to remove added lines!!!
	/etc/init.d/nfs-kernel-server reload
- copy examples/simple/* to your faiconfig space
- edit /etc/inetd.conf, the last argument (after -s) should be /boot/fai
  (not /var/lib/tftpboot). 
  	cd /boot/fai
	ln -s . tftpboot
  Don't ask me why this is necessary... I get "pxelinux.0 file not
  found" messages from tftp otherwise.
- kill in.tftpd if it's running. also
	killall -1 inetd (alias /etc/init.d/inetd reload)
- Create the PXE links
	fai-chboot -IFv demohost
	fai-chboot -o default
- Check whether DHCP server is running, if not, there's a config problem
  with it (typical here is an invalid network/netmask combo)
  	/etc/init.d/dhcp3-server restart
- Now insert the etherboot floppy into the drive, and start the client

I now get a lot of "NFS server not responding, still retrying --- OK"
messages, perhaps this is due to the 10MBit connection I'm using, I will
try to get a EEpro100 card instead...

Thanks for the great work, Thomas!


Addendum: Since I got lots of NFS timeouts, I eventually swapped the WD
card (I had tried another one, too) for a Netgear GA622T that was still
in my shack, also not PXE-enabled. Since I couldn't get it to boot using
etherboot 5.3.8's driver (UDP checksum errors) I decided to make a
floppy instead:
	make-fai-bootfloppy -Fv FAI_ACTION=install
(my first attempt missed the FAI-ACTION part, and failed miserably with
TASKERROR action 21)
and it's running happily and fast (368 seconds)! I'll dump those WDs... 
There's only one small problem being reported in /tmp/fai/error.log:

# cat error.log
updatebase.log:cp: cannot create regular file `/tmp/target/etc/resolv.conf': Read-only file system

- Huh?!? Probably I can ignore this one... since faimond only gave "0"
  return codes? OK, the system rebooted cleanly, only the ns83820 driver
  now is missing from the kernel. Something to be fixed locally (add
  line "ns83820" to /etc/modules, in config or before reboot). Great!!!

Thanks again,


Steffen Grunewald * * * Merlin cluster admin (http://pandora.aei.mpg.de)
Albert-Einstein-Institut (MPI Gravitationsphysik, http://www.aei.mpg.de)
       Science Park Golm, Am Mühlenberg 1, 14476 Potsdam, Germany
e-mail: steffen.grunewald(*)aei.mpg.de * +49-331-567-{fon:7233,fax:7298}

More information about the linux-fai mailing list