Support a boot-once + broken new boot -> rollback model

[cgwalters]

Some bootloaders have support for "boot once". Combined with a userspace sanity check (say e.g. a multi-user.target service fails), trigger an automatic rollback.

This would help enable automatic updates; see #44

[cgwalters]

It'd probably make sense to bake in auto-rollback functionality into rpm-ostreed itself. We don't yet have a /etc/rpm-ostreed.conf, but not hard to add either.

Figure out a good way to hook in activating a service if reaching multi-user.target fails. Possibly using OnFailure= somewhere? Or would this have to be a custom always-running service that watches the journal?

Something like:

[autorollback]
policies=systemd-default-target

in /etc/rpm-ostreed.conf?

[cgwalters]

See also ostreedev/ostree#380

[cgwalters]

FWIW we actually now do have a config file.

[jlebon]

Note that there is a new project called greenboot that tries to address this: https://github.com/LorbusChris/greenboot.

[jlebon]

See also https://pagure.io/fedora-iot/issue/12.

[klausenbusk]

See also: https://systemd.io/AUTOMATIC_BOOT_ASSESSMENT.html

[jlebon]

Going to close this in favour of coreos/fedora-coreos-tracker#47 since I think in the end this is something that has to be designed at a higher level.