Refactor to switch to snapshotter interface

[davidcassany]

This commit adopts snapshotter interface in install, reset and upgrade commands. The change implies changes to the respective specs, grub configuration and dracut modules.

This commit also changes the behavior of recovery system upgrades. Now recovery upgrades are an optional step of a system upgrade. Recovery image can't be upgraded without upgrading the active system.

Finally build-disk command is also changed to be better aligned with upgrade and install procedures. Expandable disks are an unprivileged build and non expandable ones require privileges as they relay on snapshotter.

Fixes #1874
Fixes #1921
Fixes #1922

[codecov-commenter]

Codecov Report

Attention: 272 lines in your changes are missing coverage. Please review.

Comparison is base (c2d9965) 73.91% compared to head (e0255d5) 72.36%.

Files	Patch %	Lines
pkg/action/upgrade.go	59.77%	60 Missing and 12 partials ⚠️
pkg/snapshotter/loopdevice.go	57.60%	30 Missing and 9 partials ⚠️
pkg/action/reset.go	55.00%	27 Missing and 9 partials ⚠️
pkg/action/install.go	80.00%	19 Missing and 3 partials ⚠️
pkg/action/build-disk.go	71.01%	14 Missing and 6 partials ⚠️
pkg/types/v1/config.go	69.09%	9 Missing and 8 partials ⚠️
pkg/mocks/utils.go	53.12%	10 Missing and 5 partials ⚠️
pkg/elemental/elemental.go	89.18%	7 Missing and 1 partial ⚠️
pkg/config/config.go	88.52%	5 Missing and 2 partials ⚠️
pkg/types/v1/snapshotter.go	70.00%	4 Missing and 2 partials ⚠️
... and 7 more

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1906      +/-   ##
==========================================
- Coverage   73.91%   72.36%   -1.55%     
==========================================
  Files          71       72       +1     
  Lines        7762     7947     +185     
==========================================
+ Hits         5737     5751      +14     
- Misses       1600     1724     +124     
- Partials      425      472      +47     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[kkaempf]

We should do a code walk-through to get this merged 🤔

[frelon]

Big PR, so it's hard to get to everything.. I left some comments but generally looks very good!

We could also reconsider setting constants.MaxSnaps = 2 as discussed yesterday.

Good work!

[fgiudici]

Skimmed through the changes after yesterday code walk-through, looks good... well, actually great job!
Hope to see this merged soon, so we can start testing/using it 🥳
Nice to see the snapshotter interface getting alive, code looks more readable too.

[davidcassany]

We could also reconsider setting constants.MaxSnaps = 2 as discussed yesterday.

Yes I have to apply these little changes yet. As soon as I manage to manually upgrade an elemental cluster from current staging to a Dev including this PR I'll mark this PR ready to be merged and arrange this little details we discussed yesterday.

[davidcassany]

Finally got green run on the CI here and managed to run a regular UI based upgrade from staging to this current head.

Dev env requires also rancher/elemental#1186 to pass tests

[frelon]

Why go back to such an old release?

[davidcassany]

To keep grub compatibility, this is compatible with v1.1.0 but not really with v1.2 or v1.3. In fact I think we should stick with v1.1.0 here, but not sure we ever published it to the registry. This was one of the things I wanted to review after merging this.

The issue is with the ${mode} and ${img} variables in grub, in v1.2 and v1.3 we assume the image is referred by the ${mode} and the dracut module computes the path. This is hard make it compatible with v1.1.0, v1.2.0 and the upcoming v2.0.x as this implies having several paths adjustments.

If we assume the ${img} and is the full path and ${mode} is just to discriminate recovery vs active args this turns to be easier in terms of backwards compatibility. We have to keep in mind in the grub domain further changes are around the corner with the btrfs snapshotter and that our backward compatibility constraint should stick to v1.1.0 version, as being the latest released elemental-toolkit version. Because of that I took a very conservative approach, because we are likely to need revising it in any case.