I think I've got it.
The kernel is a red herring, the cause is the new systemd default link policy shipped by the pve-manager package bumped yesterday.
E.g., if you adde the following to lines below the
[link]
section in
/usr/lib/systemd/network/98-proxmox-ve-default.link
and reboot the kernel should not matter again anymore.
Code:
NamePolicy=keep kernel database onboard slot path
AlternativeNamesPolicy=database onboard slot path
This puts also back some sense in why my kernel testing yesterday didn't observe this already, I only installed the new kernel manually, not pulling in the new default-link yet (that was bumped only later).
We'll look into handling the default better, i.e., go back to the
99-deault-link.d
snippet approach where all configs are merged, or take in above properties in ours – will need to recheck the discussion I had with a colleague (who favored the separate file a bit more).
Rebooting into the old kernel seems to make udev re-use the name that was previously assigned to that, thus this was seemingly fix an issue and make it look like the kernel is at fault, while it really wasn't (that's my working theory, will focus on the fix before checkling that more closely).