TBH, I'm not sure why the pipeline failed :/

As this only checks for the booted kernel (which can differ from the one currently running) how does this behave when switching to a new system with 5.8+ configured, but having booted an older kernel? Is there an error message printed by nspawn?

Or are the links just named differently?

Nice catch - in that case switching fails with:

machine # [    4.283293] container really-long-name[741]: Failed to set alternative interface name 've-really-long-name' to 've-really-l3QgY', ignoring: Operation not supported
machine # [    8.726324] container really-long-name[2492]: Cannot find device "ve-really-long-name"

The ethernet device itself does get created - it's just that its alternative name remains unspecified, so it's only reachable via ve-really-l3QgY, instead of both ve-really-l3QgY and ve-really-long-name.

Maybe our container-startup-script (the one that crashes with Cannot find device above) could check kernel's version too and print more fine-grained error message telling user to reboot the machine?

Hmm - I wasn't entirely sure, just thinking out loudly.

If this is shown inside the activation script, it already gives quite a lot of information.

Maybe if we combine it with a 21.03 release note entry explaining the change, and a "make sure you're running a kernel with support for alternative interface names (5.8+)" at the end, that should be enough to connect the dots?

Release note seems alright for me :-)

Changing kernel versions requires users to make conscious effort (i.e. to modify some part of their configuration), so I consider nixos-rebuild boot instead of nixos-rebuild switch a natural order of things (and only doing nixos-rebuild switch with different kernel version yields that assertion useless).

Maybe we could add a warning (like please don't forget to reboot) for when someone does nixos-rebuild switch that changes kernel version?

Ran into this issue and would be nice to have this. What remains to be done/reviewed?

tl;dr as far as I'm concerned, this feature is ready 🙂 (though I'm not sure who I could ping to get it merged)

As @flokli noticed, we've got a small shortcoming that when you upgrade from an older kernel to 5.8+, the kernel version assertion passes and thus Nix compiles a configuration which then fails at runtime during nixos-rebuild switch (because the system is still running on the older kernel); imho trying to somehow solve this is not worth the hassle, as upgrading the kernel is anyway a "big change" that almost always requires reboot (which will work correctly), but feel free to add your 5 cents.

The problem here is that even on latest master, linuxPackages.kernel points to 5.4.x, which is < 5.8+.

The new test already contains:

{
    boot.kernelPackages = pkgs.linuxPackages_latest;    
}

... plus it passes on my machine, so I wouldn't bet my money on an older kernel being loaded.

I've just rebased my branch and we'll see now - maybe it was just some spurious failure.

Aaargh, sorry, I'm dumb. It's already too long ago - I forgot long names are only enabled on new kernels.

I locally cherry-picked the commit into master and successfully ran the test.

Can you:

• Add the test to nixos/tests/all-tests.nix, so it appears in the nixosTests attrset
• put the comment removal in a separate commit

Thanks!