Sometimes I’m forced to question the sanity of my cow orkers. If you run our setup program and choose the option to set the time and date, you are presented with a string like “062716452005.40” As near as I can figure, that’s DDMMHHmmYYYY.SS, or translated into English, day, month, hour, minute, year, period, seconds. Besides the utterly moronic order of the elements in the string, the input routine has absolutely no flexibility in what you can enter and no error checking. Get one character wrong or miss a column, and you’re going to get a date and time that are utterly unlike what you expected, and you won’t find out until you exit the setup program and type “date”.
Ok, picture a network with one “controlling computer”, which I’ll call “the CMS”, and a bunch of satellite computers which I’ll call “the CPs”. These satellite computers live in projection booths in a theatre and have digital projectors hooked up to them, but that’s not important. The problem I’m dealing with is upgrading the machines from version 3.3 of our software to version 3.5. The software upgrade also necessitates an upgrade from RedHat 7.3 to CentOS 3.4.
I’ve got the upgrading of the CMS sorted (I have a non-bootable DVD with an apt repository with CentOS 3.4 and our software, and a kickstart file that does the upgrade without touching the partition with our data on it).
The CPs have hostnames of cp1 to cpN, and IPs of 192.168.30.101 and up. cp0 (192.168.30.100) is reserved.
What I’m working on now is upgrading the CPs. What I’ve been doing is making the CMS a PXE boot server, and wiping the boot partition on the CPs one at a time, re-installing them as cp0 and then when it comes back up, ssh-ing in and restoring the backed up configuration, including the hostname and IP.
The problem with that is that it takes 20 minutes per CP, and the powers that be are complaining that it takes too long. They’d like something more parallel.
So I’ve been thinking of retrieving the MAC addresses of each CP before I upgrade. Then I do them all in parallel, and use the MAC address afterwards to figure out which one is which. I understand that I can use “arp -a” to retrieve the MAC addresses. I’m wondering if there is something I can do to DHCP to give out the correct 192.168.30.1xx address to the right machine, or whether I should have DHCP hand out addresses in some other range, and then use “arp -a” again to find which machine has which address and fix them one at a time?