> We didn't make any changes to kickstart in SL3 or 4 from what RedHat
> released, and currently do not plan on making any changes for SL5.

Sorry - I has assumed it would be an upstream change if anything. Just
wondered if people could point me to what part of the install made the
backup of the scripts.
> Here is RedHat's online documentation
.. which is what I followed.

FYI it's VERY worthwhile adding the "noipv6" if you haven't already -
saves me a good 50 seconds before the installer comes up. (unless you're
using IPv6)
> I do have to say that I've seen an ifcfg-eth0 get moved to
> ifcfg-eth0.bak.  ...
> Maybe this bears more investigation.

OK - that'd be very helpful. Google was v unhelpful I'm afraid. I suspect
there'll be a snippet on a mailing list somewhere.

> When you do a kickstart install, are you installing them using dhcp?
Yes.

Our full procedure is fully documented at
  http://www.gridpp.ac.uk/wiki/Glasgow_Cluster_YPF_Adding_A_New_Host
  http://www.gridpp.ac.uk/wiki/Glasgow_Cluster_YPF_Install

but consists basically of:
* dhcpd always has "correct" final IP address of node in. (we don't use a
special ip range for the installer)* Set PXE config to boot node into kickstart
* PXE Boot node - gets normal IP via DHCP, starts kickstart again with
DHCP address* kickstart runs. As part of the post stage we untar machine specific
configs into the right place on nodes (things like ntp.conf, yum.repos.d/*
and the /etc/sysconfig/network-scripts/ifcfg-ethN files)* node signals it's installed completely (wgets a php script) which sets
PXE config for that node back to "localboot"* node reboots, hopefully off local disk as told by PXE
* runs firstboot which, if authorized, gets SSH/cfengine keys pushed over
from master node and cfengine takes over management of the node.
Although we install using DHCP we have the nodes on static IP addresses as
it reduces a single point of failure. I wonder if it could be some DHCP
specific hook that creates the .bak version of the file.
I'm afraid I'm on leave for a week, so won't have chance to test anything
till w/c 16 April.
It's not a show-stopper as we only have 2 test nodes on SL5 just now - the
others are 30x for the worker nodes / grid servers and slc4x for the disk
servers.

Many thanks

Andrew