How to test/simulate a failed upgrade?

[jsquyres]

I'm trying to test scenarios that result in automatically rebooting into fallback.

I've tried doing this:

1. Do a successful install
2. Do a successful upgrade (sudo kairos-agent upgrade ...; sudo reboot)
3. During the "boot" phase after the reboot, cause a fake failure (in my case, killall -s=FPE systemd)

This does cause a reboot (per https://kairos.io/docs/upgrade/boot_assessment/#automatic-reboot-in-case-of-systemd-crash), but https://kairos.io/docs/upgrade/boot_assessment/#booting-to-fallback implies that this should automatically boot to fallback.

However, in my case, this doesn't seem to be happening. Instead, the system just reboots back into the same active image, and therefore during the "boot" phase, my killall ... runs again, and I end up in a reboot loop.

Am I misunderstanding the docs in terms of what cases cause Kairos to automatically reboot into fallback mode?

---

Additionally, I have a similar but different scenario:

1. Kairos installs ABC image and boots it up successfully
2. Then run sudo kairos-agent upgrade DEF, which completes successfully
3. Then successfully reboot into the new DEF image
4. Some application logic runs (e.g., during the network phase after the reboot) that determines "this DEF image is broken -- we should reboot back into the original ABC image"

I see that I can sudo kairos-agent bootentry --select fallback; sudo reboot, but that's a one-time reboot into the original ABC image. If I reboot again after that, for example, it'll reboot back into the DEF image.

How can I swizzle the active/passive images again so that I can persistently reboot into the original ABC image?

Thank you!

[Itxaka]

umm, maybe network is a bit too late in the process to trigger the reboot? I cant really remember how it was, but maybe if we reach the network stage it wont work with the sentinels and such.

We will need to restest this to be able to test this programatically to know the exact place where we can do this, but very recently there was a live issue with one of the upgrades images in which after upgrade it automatically went to fallback, so we kind of know that it triggers.