[3072] in testers
Re: sun4 8.1.0: update_ws, mkserv
daemon@ATHENA.MIT.EDU (Albert Dvornik)
Fri Apr 25 16:26:06 1997
To: Jonathon Weiss <jweiss@MIT.EDU>, Greg Hudson <ghudson@MIT.EDU>
Cc: testers@MIT.EDU, mkserv@MIT.EDU
From: Albert Dvornik <bert@MIT.EDU>
In-Reply-To: Jonathon Weiss's message of Wed, 23 Apr 1997 18:16:10 EDT
Date: Fri, 25 Apr 1997 16:25:56 EDT
OK, let me try explaining this a little better. I believe the
scenario of my problem worked like this:
(1) bert makes an rlogin into si as root and runs /srvd/update_ws.
(2) The machine starts the update and tracks in the new OS and all
that fun stuff.
(3) The machine reboots itself, as a part of the update process.
All TCP connections to the machine drop on the floor, including
bert's login.
[Somewhere around here bert confuses himself a little by blowing away
his X server, but this is irrelevant to the rest of the story.]
(4) The machine runs "mkserv update", as a part of the update process.
(Mkserv probably runs with the console as its controlling tty/
stdin, which will become relevant soon.)
(5) Mkserv runs remote.add to update "remote". This decides to ask
questions, implying "[ -t 0 ]" was true.
(6) Upon noticing that the machine isn't running inetd, bert gets
machine room access, switches si's serial terminal to get output
from si, and presses return, hoping to get a prompt. In response,
si displays the *second* of the questions asked by remote.add.
bert is happy that the default for the first question was the
former answer. =)
Jonathon said:
> [...], or if for some reason -t 0 evaluates to true during the post
> reboot stage of a reboot. If it is the latter, we need to do
I believe it is the latter, since nothing else would explain the
problem.
> something or mkserv remote machines that try to autoupdate will
> probably lose, and there are alot of those.
Oh. Yeah, you're right; I haven't thought of that.
If you give me machine room access, I can try auto-updating si (by
removing the last few lines from /etc/version and setting AUTOUPDATE
true) and see whether it, in fact, loses.
--bert
