System won't boot after failed upgrade from 23.10 to 24.05

[Sam Varshavchik]

Walt Mankowski writes:

> Hi.
>
> I tried to upgrade from 23.10 to 24.04 today and it failed. It ran out
> of disk space on /usr midway through the install, leaving things in a
> weird state. I was able to run `apt upgrade` and install a bunch of
> them, but several 100 more were still in a state where they were being
> held. I tried doing another `apt do-release-upgrade`, but by that
> point it thought it was already on 24.04 and so it returned
> immediately.
>
> At this point I tried rebooting to see if that would clear things up,
> That was a mistake. Not it's not booting up. If gets pretty far into
> the boot process then fails. The messages scroll off the screen and I
> can't see all of them.. The last one says
>
> Failed to start svscan.service - Daemontools service scanner.
> See 'systemctl status svscan.service' for details.
>
> Above that I see that dbus.service failed to start. And above that I
> see a bunch of failed dependencies for half a dozen SSSD responder
> sockets. I'm not able to scroll up beyond that.
>
> I do nightly backups so with any luck I haven't lost any data. Is
> there any way I can get back in there in single user mode and have it
> finish the upgrade to 24.04?

I think you're out of luck. There's an endless number of things that can got  
screwed up here. Basically at some point:

- either in a middle of installing new version of some set of packages

- or attempting to configure some set of packages

- or attempting to uninstall or purge older version of packages

-- no way to tell which is the case without hands-on debugging of the system  
in question: the whole thing went sideways. And when the package  
installation process encountered an error who knows what happened after the  
fact before the whole thing shut down, making the whole situation worse.

It also doesn't help, very much, that Debian/Ubuntu's do_release_upgrade  
apparently plows ahead and does its thing on an existing, live, running  
system. I always thought that this was like attempting to replace all four  
tires on a car while it's still driving on the highway. I always believed  
that it would be more reliable to reboot into single user mode, then  
everything gets updated while the most of the system is shut down, like  
<cough> other, non-Debian based distros go about an upgrading to a new  
release.

Because now, the new release likely has many system libraries that are not  
binary ABI compatible with the previous release, so now when things blow up  
like this it's fairly likely that there are incompatible, different versions  
of packages that need to be cleaned up, preventing the system starting,  
which sounds like you're seeing. What are those? Only a hands-on, forensic  
inspection of the ailing system will determine.

I suppose that someone with sufficient expertise in these things can go and  
boot off a live image, mount the existing partitions, sift through dpkg/apt  
logs to determine what exactly was being installed/uninstalled when things  
went to crap, fix it, then finish updating, and maybe get lucky and be able  
to get things working well enough for 'apt dist-upgrade' to fix everything  
else.

But, all a mere mortal will be able to do is just boot up the same live usb  
image, mount the the existing partitions, make a backup of their data, and  
then just reinstall from scratch.

P.S. Every time I run 'apt upgrade', before asking for confirmation apt  
tells me how much additional disk space, if any, the updated packages will  
require. Bummer that do_release_upgrade does not do the same?

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 228 bytes
Desc: not available
URL: <https://lists.ubuntu.com/archives/ubuntu-users/attachments/20240525/4b58521d/attachment.sig>