On Sun, 3 Sep 2023 12:54:17 +0200 Bastian Blank <wa...@debian.org> wrote:
> But a really good solution this is not.
Yeah, I feel that's a slight understatement. A few of my users are on
backported kernels because they have new hardware and we haven't found
time to migrate our software stack to bookworm yet. One of them dutifully
ran regular updates through gpk-update-viewer , rebooted and bam... system
doesn't boot to a graphical desktop anymore.
What happened is that we're relying on the commercial Nvidia driver
(because we're doing CAD), which failed to update in tandem with with the
new kernel image by lack of corresponding headers. I guess some DKMS error
message must have briefly flashed by, but went unnoticed.
I know, I know,
* using the commercial nvidia driver is not really supported
* backports are a semi-unofficial favor that we shouldn't rely on in
production
* one could easily fix the problem either from a text terminal or by
choosing the previous kernel in grub
But it nevertheless leaves a poor impression on users if their system
"fails to boot" (as it's perceived) after running a routine update. One
risks getting comments like: "why, again, aren't we running Windows (or OSX)?"
So, I'm wondering about a mechanism that holds back on updating the kernel
image until the corresponding updated headers are available.
