distribution upgrade via update-manager fails over remote GDM

[Thomas Novin]

Binary package hint: update-manager

Hello

I was about to do a distribution upgrade from 8.10 latest to 9.04 alpha yesterday. I started it via a remote GDM session (which seems to run over SSH). I got a warning that the upgrade might fail and that a second sshd was started on port 9004. Well, probably nothings going to happen I thought.

After downloading all packages and upgrading for maybe 5-10 minutes my GDM session froze. I was unable to contact the client that I was upgrading, didn't even respond to ping.

When I was able to check the remote client out physically I noticed that my session previously active was still running. I couldn't use anything though, nothing was possible to start. I tried to restart X but it didn't start again. I could see lots of segfaults in dmesg and ifconfig showed that there was no eth0 any more.

My suggestion: Make the warning a LOT more scary! It took me a while to repair the system and my upgrade probably isn't that good because it's done mostly via aptitude and not update-manager.

[Chris Jones (cmsj)]

It seems like it would be sensible to simply not allow upgrades to be done in that manner. It's far too likely to go wrong.

[Michele Mangili (mangilimic)]

Thanks for reporting this bug and any supporting documentation. Since this bug has enough information provided for a developer to begin work, I'm going to mark it as confirmed and let them handle it from here. Thanks for taking the time to make Ubuntu better!

[Michael Vogt (mvo)]

I agree with Chris here, I think remote gdm upgrades are too risky, I will add detection code for this.

[Steve Langasek (vorlon)]

Michael,

The symptoms described here - segfaults in dmesg and eth0 going away - don't sound like they're related to gdm. Is added detection code really the answer here?

[Michele Mangili (mangilimic)]

Steve,

I don't know exactly what happened when Thomas upgraded that computer. He hasn't reported a bug against those segfaults in dmesg and eth0 going away, but he is just asking for making the warning message for gdm upgrades more scary. I believe that somehow he could solve that problem.

[Michael Vogt (mvo)]

Pointers/hints how to setup a test system so that I can try to get the detection done are very welcome :)

[Launchpad Janitor (janitor)]

This bug was fixed in the package update-manager - 1:0.100

---------------
update-manager (1:0.100) jaunty; urgency=low

  [ Andy Whitcroft ]
  * DistUpgrade/cdromupgrade:
    - if cdromupgrade is run with a relative path we will fail to find the
      installer components and error out. Ensure that the path is absolute.
      (LP: #335360)

  [ Michael Vogt ]
  * fix crash when help is not avaialble (LP: #338098)
  * data/glade/UpdateManager.glade:
    - remove the "Keep your system up-to-date" text (design team,
      LP: #336800)
  * fix crash when no network is avaialble for changelog fetching
    (LP: #334002)
  * debian/control:
    - add conflict against update-manager-kde to update-manager-hildon
      (LP: #333464)
  * UpdateManager/Core/MyCache.py:
    - show proper urls for sources with epochs (LP: #328164), thanks
      to Richie
  * UpdateManager/ChangelogViewer.py:
    - support copy to clipboard for URLs (LP: #85644), thanks to
      Richie
  * UpdateManager/UpdateManager.py:
    - disable fixed-hight mode, it can cause incorrect height
      calculation (thanks to Richie), LP: #273184
  * DistUpgrade/DistUpgradeController.py:
    - do not allow gtk/kde upgrades over ssh session (LP: #322482)
  * merged aufs branch (disabled by default but useful for testing)

 -- Michael Vogt <email address hidden> Tue, 10 Mar 2009 20:25:55 +0100