Open cgwalters opened 6 months ago
Kernel differences could be handled with kexec, see #435
Edit: Oh! That's also your issue! Oops.
Implementing this is going to require some work on ostree-prepare-root.service
to be able to run outside of the initramfs and target an explicit root. I don't think it'd be really hard but e.g. one thing we need to do in the general case is set up the composefs mount for example. We also need the /var
and /sysroot
bind mounts set up in the new deployment root.
xref https://www.freedesktop.org/software/systemd/man/latest/systemd-soft-reboot.service.html
We could theoretically do this by default if we detect there's no kernel changes...but per the doc it will cause state leakage in corner cases around things like kernel sysctl.d, module loading blacklists, etc.
systemd (understandably) just made this an opt-in thing. I think to start, we should do the same...just add the basic enablement features of setting up the next root.
Then users that want could invoke e.g.:
And hmm... we should have a flag that detects the case where the kernel is skewed, and errors out by default. But in theory we should also allow this, but ensure we've e.g. bind mounted in the old kernel data.