On Thu, 2005-12-08 at 15:29 -0500, Brian Long wrote: > On Thu, 2005-12-08 at 14:44 -0500, Jeremy Katz wrote: > > On Thu, 2005-12-08 at 14:31 -0500, Brian Long wrote: > > > Have you ever considered wrapping sshd into the install image so remote > > > debugging can be performed on systems that don't support VNC? > > > > How can a system "not support VNC"? And what sort of debugging are you > > trying to do? > > For example, kickstarting a Sun V20z via console=ttyS0 (serial redirect > over SSH). vnc should still work... vnc works for s390. > I'd like to know when the F3 and F4 VT's have errors and be > able to track those errors. Unfortunately, all the stuff output to F3 > and F4 are not saved in files in /tmp, so I cannot save them in %post > (like I save other files in /tmp). The information from the ttys is logged elsewhere... /tmp/anaconda.log and /tmp/syslog for tty3 and tty4 respectively. And that then gets copied to /var/log/anaconda.log and /var/log/anaconda.log.syslog on the installed system. > > sshd is very heavy-weight and would add a lot of complication to the > > statically linked, minimal requirement first stage (where it would have > > to be). Note that there is currently telnet support, but I'm not sure > > if it accomplishes what you want. > > Understood. The cpqlinux site documents setting up /dev/vsc nodes so > you can cat /dev/vsc3 and see what was displayed to tty3. This is the > type of functionality I would like to see when doing serial-based > installs on remotely managed systems where I cannot stand in front of a > monitor. ... and that's why things primarily go to a file and the tty output is basically secondary.[1] Also, in FC5 and later, there's the ability to use syslog for the logging from anaconda so that the output goes to your remote syslog server. > Right now, the install.log.syslog file saved in /root is pretty sparse. > Logging more data into that file when things fail would be useful. For > example, if I kickstart a machine with %packages --nobase and specify > selinux --disabled, Anaconda does not complain that /usr/sbin/lokkit > does not exist (part of system-config-security-level-tui). This means > selinux is still enabled. > > Ideally, Anaconda would either automatically pull in its own > dependencies if they're missing from %packages or it would complain > loudly (not just in the F4 window) :) Ideally, --nobase never would have had to be added because the reason base always got pulled in was to protect people from shooting themselves in the foot like this. Unfortunately, enough people complained that they couldn't get things small enough. So now you can shoot yourself in the foot all you want. Using --nobase or excluding packages is inherently dangerous and _will_ override things giving the benefit of the doubt on the user knowing what they're doing. Jeremy [1] Not that this necessarily helps debugging the loader since you don't have a shell, but that's not your particular concern :)
