Comment 7 for bug 1978785

Thank you! I can reproduce this now too. I see exactly the same "snap changes" output.

What is interesting is that we have a test that tries to simulate this behavior in our code (TestUpdateKernelBaseSingleRebootKernelUndo). However something in this test is misbehaving.

The real failure error in the auto-connect task and reports a rollback across reboots for the kernel:
"""
Error today at 10:10 UTC today at 10:12 UTC Automatically connect eligible plugs and slots of snap "pc-kernel"
"""

However the test does not show this, it checks that the rollback across reboots is detect for the "core20" snap - something that is not observed on the real device and that is also counter-intuitive as the failure happens on the kernel. So something in the test setup behaves differently from the real device and that seems to be screwing up the test.

What is interesting is that the undo path on the real device shows:
"""
90 Automatically connect eligible plugs and slots of snap "pc-kernel"
  2022-06-22T13:25:35Z ERROR cannot finish pc-kernel installation, there was a rollback across reboot
---
105 Make current revision for snap "core20" unavailable
  2022-06-22T13:25:36Z INFO Requested system restart.
"""
i.e. the undo of the core20 triggers another reboot.