kickstart + PXE + NFS = WTF

[lruppert]

I have been trying now for a bit over twelve hours to do a simple kickstart install of a CentOS 7 box. Unfortunately, every time it gets to the network configuration, dracut decides it wants to reconfigure the eth0 interface, which drops the networking. Of the countless network directives I have tried, this one doesn't decide it needs to redo the ifcfg file, but it still drops the networking:

Code: Select all

network  --device=bootif

Here is what it looks like:

Code: Select all

20:29:09,407 DEBUG anaconda: network: devices found ['eth0']
20:29:09,435 DEBUG anaconda: network: ensure single initramfs connections
20:29:09,645 DEBUG anaconda: network: ensure active ifcfg connection for eth0 (9c6fdc35-5d9e-4a50-8964-09abc2135226 -> 89b33576-7c2c-434c-92c0-eb9601afd96d): activating
20:29:09,648 DEBUG anaconda: network: single connection ensured for devices ['eth0']
20:29:09,792 DEBUG anaconda: network: apply kickstart

At which point the machine does nothing and no longer responds to ping or offers a responsive shell on console.

I'm sure I can't be the only guy trying to boot a KVM PXE client to a kickstart server over NFS. And yet all of the people who seem to be doing PXE kickstart don't seem to be having my problem. Reminds me a bit of the cartoon of the guy sawing off the tree branch he is sitting on.

I'm also not sure why the UUID would change for the device. It is DHCP configured and gets a static lease and has the same MAC and is plugged into the same imaginary PCI slot as when it boots.

So...is there any way to tell dracut and kickstart to stop trying to be clever and just use leave the network interface as it is?
Or if not, does anyone have a simple bare bones pxelinux config and kickstart for a simple no-frills kickstart install of CentOS 7 that works with KVM?