root login

[eichin-krb@thok.org]

> The master or the slave?  You do have a slave KDC, of course...  ;-)

That helps if the *machine* dies, but not if the kdc process does --
if it's a KDC bug that caused the crash, and it's one triggered by a
normal client, send_to_kdc (well, the b5 equivalent) will merrily
resend the same deathgram to all of your kdcs.  "oops" Having init
restart them can help, if you're keeping a stash file around anyway.
Having an openssh-based back door (as long as you're keeping that
up-to-date too) has some value as well, if you *really* want to avoid
going to the machine in person...