Trent W. Buck wrote: > I can reproduce this issue. > It has bitten me 3 or 4 times. > I think happens every time the ABI bumps (5.10-n → 5.10-n+1). > > For me, the timeline is this: > > 1. unattended-upgrades installs new kernel > 2. kernel postinst builds new initrd > 3. unattended-upgrades installs new headers > 4. kernel postinst new zfs.ko
PS: from unattended-upgrades-dpkg.log, I'm pretty sure the problem is "above" dpkg, but I'm not clear if it's in unattended-upgrades or in apt. If it's happening in the apt layer, I guess that is because linux-image-N-amd64 does not Depends on linux-headers-N-amd64. So apt thinks it can do 2 separate dpkg runs, and (sometimes?) in the wrong order? If that's what happens, it's unreasonable to add stronger dependencies, because then EVERYONE will be forced to install headers. Is it reasonable for unattended-upgrades to have a special-case safety net for this? Something like If unattended-upgradse is upgrading a kernel AND a headers, then ensure headers installs first.