We discussed in the sprint two weeks ago how we could prevent this from happening again. Part of the problem here is that this only shows up in a multi-stage upgrade, because it's a misbehavior of systemd that was triggered on the prerm script.
So normal upgrade tests do not catch it. We need to test upgrading *out* of the given release instead. So install the *new* snapd, do changes to the system by installing snaps, reinstall the *new* snapd on top of the existing ones, and play with the system.
We'll look into implementing this.
That said, as usual regression tests catch regressions. They don't catch new and creative ways in which the system can break. So the true answer is that nothing can prevent this from happening again, unfortunately. We can just continue to work on making the risks ever smaller.
Hi Colin,
We discussed in the sprint two weeks ago how we could prevent this from happening again. Part of the problem here is that this only shows up in a multi-stage upgrade, because it's a misbehavior of systemd that was triggered on the prerm script.
So normal upgrade tests do not catch it. We need to test upgrading *out* of the given release instead. So install the *new* snapd, do changes to the system by installing snaps, reinstall the *new* snapd on top of the existing ones, and play with the system.
We'll look into implementing this.
That said, as usual regression tests catch regressions. They don't catch new and creative ways in which the system can break. So the true answer is that nothing can prevent this from happening again, unfortunately. We can just continue to work on making the risks ever smaller.