[7239] in Release_7.7_team
Re: auto-update exploded
daemon@ATHENA.MIT.EDU (Jonathan Reed)
Tue Jan 25 20:24:45 2011
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0 (Apple Message framework v1082)
From: Jonathan Reed <jdreed@MIT.EDU>
In-Reply-To: <C8DABDEB-75EC-42F8-A6AD-CBDEA13CD69A@mit.edu>
Date: Tue, 25 Jan 2011 20:24:37 -0500
Message-Id: <A5AF8882-6486-4DE0-A9E8-2AC84F355A28@mit.edu>
To: "release-team@MIT.EDU Release Team" <release-team@mit.edu>
Content-Transfer-Encoding: 8bit
On Jan 25, 2011, at 10:10 AM, Jonathan Reed wrote:
> The auto-update exploded because Ubuntu pushed out an update to at(1), which killed atd(8), which in turn killed the update process. On machines which have not rebooted, atd is dead and no updates will occur. On machines which have rebooted, atd will attempt to continue, but dpkg --configure will fail because it doesn't like the state of the at package, and insists on reinstallation.
>
> We're looking at our options at this point.
I have asked acis-team to visit the public machines over the next few days and fix them using the procedure described at the bottom of this mail. Since the machines are still usable, this is not urgent, but still very high priority. I will also compose a mail to release-announce with recovery procedures.
For private workstations, users logged in with administrative privileges will possibly get a dialog box from update-manager, telling them they need to do something to make APT happy again. I will also consider a 3down post.
We can discuss this more on Friday (which should be listed in Exchange).
-Jon
Recovery procedure:
- For private workstations, become root, and "sudo apt-get install -f" which should force a reinstall of "at", among other things.
After running that, "dpkg --configure -a" should exit cleanly.
- For cluster workstations:
-- Hold down the left shift key at boot time until the GRUB menu appears.
-- Select the "recovery mode" boot option.
-- At the "Recovery Menu", select "dpkg Repair broken packages".
-- Answer "Y" to every question.
-- When it completes, reboot. (The "resume" option in the recovery menu doesn't work correctly.)
