For about two weeks now, the boot-and-services integration test randomly fails because it detects a failed [email protected] unit after boot (we enforce that in a default systemd install there are no failed units):

Nov 23 20:42:00 autopkgtest systemd[1]: Started ifup for ens2.
Nov 23 20:42:00 autopkgtest systemd[1]: [email protected]: Unit is bound to inactive unit sys-subsystem-net-devices-ens2.device. Stopping, too.
Nov 23 20:42:00 autopkgtest systemd[1]: Stopping ifup for ens2...
Nov 23 20:42:00 autopkgtest ifdown[712]: /sbin/ifdown: waiting for lock on /run/network/ifstate.ens2
[...]
Nov 23 20:42:03 autopkgtest systemd[1]: Stopped ifup for ens2.
Nov 23 20:42:03 autopkgtest systemd[1]: [email protected]: Unit entered failed state.
Nov 23 20:42:03 autopkgtest systemd[1]: [email protected]: Failed with result 'exit-code'.

(full log)

The unusual bit is "Unit is bound to inactive unit .. .device". I haven't seen this downstream (yet), and never before ~ 2 weeks ago, so at first sight this looks like either a regression with udev, or maybe exposing a problem with [email protected] -- this is triggered from a udev rule, so I don't believe that the problem is that udev isn't running yet.

I haven't succeeded in reproducing this interactively yet, so I'm opening this to collect information and have a proper place to refer to when encountering this in tests.