Red Hat

I just finished setting up a Kickstart Server for our RHEL 5 installs. It's configured to use the PXE boot and NFS for the installation. Everything works fine when I just build one machine at a time. The installation works from begining to end.

But when I try the install three machines at once I run into problems. All machines boot up just fine, they get a DHCP lease and it works up until Anaconda loads. A lot of times at that point on 2 of the 3 machines, it fails to either get the DHCP lease back at this point, even though it got it just fine for the PXE boot. Or the other half of the time it fails to mount the NFS directory even though the directory is fully exported. Then at that point it asks for manual intervention. If I go and have it re-request the lease at that point it works fine, and I can mount the NFS directory manually when prompted. It doesn't do this all of the time but it just seems when I am building more than one machine at the time these issues arise. When I do one at a time it works fine. Does anyone know what the issue could be? Or places where I can begin to look?

You may want to try and tell Kickstart which interface to use....

add the ksdevice to your boot config:
linux ks=nfs:kserver.company.com:/kickstart.file ksdevice=eth0

or what ever you normal line is...

Thanks for your help. I tried putting ksdevice=eth0 in my pxelinux.cfg default file but it didn't really have any affect. It seems like a random network problem because sometimes even if all the devices do start installing it can hang in the middle of package installation too. My network is an isolated network on its own switch with only build devices off of it. The network is 10.254.254.0.

What does your /etc/dhcpd.client config file look like? Might be that your DHCP range is to small.

How far into the install does it get?

I actually figured out what the problem was by looking through the loging on the switch. I'm using a Cisco Catalyst 2960G switch. What was happening is that when Anaconda was requesting a DHCP lease it was taking down the eth0 interface and bringing it back up. By the way the switch is configured it takes about 60-70 seconds for the port to be renabled after the interface went down. By then Anaconda was timing out the request for the DHCP lease. So I tested this theory by using a cheap 5 port switch, and it ended up working.

Great news!!!