An audience of two

Tuesday October 10, 2006

Wrestling with jumpstart

I spent way too long yesterday debugging trivial jumpstart issues. When I first booted the client it would download the boot image via TFTP, then hang. So naturally my first reaction was to snoop the connection to see what was happening. This is what I saw:

client -> server    TFTP Ack  block 419
server -> client    TFTP Data block 420 (352 bytes) (last block)
client -> server    TFTP Ack  block 420
client -> BROADCAST BIFF R port=111
server -> client    BIFF C port=32771
client -> server    WHO R port=32777
server -> client    WHO C port=32777
client -> server    SYSLOG R port=32777
server -> client    SYSLOG C port=32777 BAD.FMT:

I googled "SYSLOG" and "BAD.FMT:" together with all kinds of other keywords, but couldn't find an answer. The clue finally came when I noticed that in the jumpstart window it was trying to resolve "127.0.0.1" - aha! I found the following entries in /etc/hosts

127.0.0.1 localhost server
10.1.2.3 server server-alias server.fulldomain.com

This fixed the first problem and it booted OK. However it then appeared to be finding but ignoring the sysidcfg file:

Beginning system identification...
Searching for configuration file(s)...
Using sysid configuration file server:/boot_server/sysidcfg
Search complete.
Discovering additional network configuration...
Completing system identification...
Starting remote procedure call (RPC) services:
─ The Solaris Installation Program ─────────────────────────────────────────────

  The Solaris installation program is divided into a series of short sections
  where you'll be prompted to provide information for the installation. 

It would then enter the interactive mode. Now I knew that the format of the sysidcfg was good as I'd copied it from another jumpstart server that was working. Once again, Google showed lots of people having the same issue, but no solutions.

So what was different betwen the two servers? After more headscratching I realised that this new system wasn't in the naming service yet. I'd manually added it to /etc/hosts and /etc/ethers on the jumpstart server, but neither NIS nor DNS could resolve client correctly. Fortunately there was a DNS entry for the IP address of the form dummy_10_1_2_4 so I updated /etc/ethers on the server and re-ran addinstall-client with the new hostname - Bingo.