Implement the bootc provision plugin

[ckyrouac]

This creates a new provision plugin that is built on top of the existing TestCloud (virtual) plugin. It adds new parameters to pass a Containerfile or container image. The plugin will then build a container image (if necessary) then build a bootc disk image from the container image using bootc image builder. Currently, bootc requires podman to be run as root when building a disk image. This is typically handled by running a podman machine as root.

An additional parameter "add-deps" toggles building a derived container image with the tmt test requirements.

Resolves #3013

Pull Request Checklist

• implement the feature
• write the documentation
• extend the test coverage
• adjust plugin docstring
• modify the json schema
• include a release note

---

This is a work in progress. I'm opening this PR early to get feedback on the high level design. I will add tests, docs, etc. after we solidify the higher level design.

If you want to try running the code here is an example fmf plan:

provision:
  how: bootc
  containerimage: quay.io/centos-bootc/centos-bootc:stream9
  disk: 20
summary: Testing bootc plugin
execute:
  how: tmt
  script: |
    echo "ok"

[ckyrouac]

@cgwalters fyi

[cgwalters]

Mostly a skim (I don't know the tmt codebase either) but looks sane to me!

[happz]

• LGTM, I like the reuse of existing plugins, that's very nice.
• Would it also make sense to support the container plugin? IIUIC, the image produced by podman can be run both with podman run, or, converted to qcow2, as a VM. I could imagine both ways being available to use, maybe through some runtime-plugin: container|virtual switch. Is it a bad idea?
• I will have more low-level comments related to the actual implementation, but I won't bother you now till you receive the high-level review. Just ping when you're ready for the boring stuff :)

[martinhoyer]

I'm still trying to speedrun reading/learning all things bootc (and tmt provisioning plugins), but fwiw, looks cool to me.

@happz about the container plugin support - Fedora docs says:

for fully-fledged tests it is not recommended to run a bootable container via, for instance, podman-run. One reason among others is that the filesystem is writable when being executed as an OCI container while most of the filesystem is mounted read-only on a deployed bootc system. That means the running container behaves differently than a deployed system. Yet, if you desire to run some quick tests it is recommended to run the container in detached mode.

From what I understand, podman-bootc could be pretty cool to use, once available.

[happz]

I'm still trying to speedrun reading/learning all things bootc (and tmt provisioning plugins), but fwiw, looks cool to me.

@happz about the container plugin support - Fedora docs says:

for fully-fledged tests it is not recommended to run a bootable container via, for instance, podman-run. One reason among others is that the filesystem is writable when being executed as an OCI container while most of the filesystem is mounted read-only on a deployed bootc system. That means the running container behaves differently than a deployed system. Yet, if you desire to run some quick tests it is recommended to run the container in detached mode.

From what I understand, podman-bootc could be pretty cool to use, once available.

I agree, it's not a perfect 1:1 substitution, but, exactly: for quick tests or basic test development, it may give me results faster than VM. I for one work on binutils and C/C++ toolchain in general, and my area of focus is fairly simple - compile this, run objdump on that, grep for expected section names, this kind of stuff. Learning about a typo in my reproducer quickly is very valuable, and once I'm done, I can always use full VM. I do it today already: I develop tests with container, then I switch to beaker or 1minutetip to get a more real environment before committing the new test to git. So, podman-run, for my trivial component, would be very welcome, even with caveats like the one you mentioned :)

[martinhoyer]

I agree, it's not a perfect 1:1 substitution, but, exactly: for quick tests or basic test development, it may give me results faster than VM. I for one work on binutils and C/C++ toolchain in general, and my area of focus is fairly simple - compile this, run objdump on that, grep for expected section names, this kind of stuff. Learning about a typo in my reproducer quickly is very valuable, and once I'm done, I can always use full VM. I do it today already: I develop tests with container, then I switch to beaker or 1minutetip to get a more real environment before committing the new test to git. So, podman-run, for my trivial component, would be very welcome, even with caveats like the one you mentioned :)

Thank you for the insight in your development process. I hope I can see it in more detail one day.

[ckyrouac]

Would it also make sense to support the container plugin? IIUIC, the image produced by podman can be run both with podman run, or, converted to qcow2, as a VM. I could imagine both ways being available to use, maybe through some runtime-plugin: container|virtual switch. Is it a bad idea?

I think the existing container plugin will handle this case without any additional code. The bootc image is just another container that can be run like a typical image.

[lukaszachy]

--add-deps doesn't install require/recommend yet, right?

Either way, we have chicken&egg problem in the case 'dist-git-source' is used as the list of packages is known after provision :/ But IMO we can ignore that for now to have something working.

[ckyrouac]

@happz This is ready for a review. I added some docs, tests, and code to cleanup the container images.

[psss]

/packit build

[psss]

/packit build

[psss]

Ok the xfail mystery should be solved by 57762fb.

[psss]

/packit build

[psss]

/packit build

[psss]

/packit build