Disable the creation of VMIs that are incompatible with live migration

[jean-edouard]

What this PR does / why we need it:
Live migration of VMIs with SRIOV/GPU access is very likely to fail.

Which issue(s) this PR fixes (optional, in fixes #<issue number>(, fixes #<issue_number>, ...) format, will close the issue(s) when PR gets merged):
Fixes #

Special notes for your reviewer:

Release note:

NONE

[kubevirt-bot]

[APPROVALNOTIFIER] This PR is NOT APPROVED

This pull-request has been approved by:
To complete the pull request process, please assign jean-edouard
You can assign the PR to them by writing /assign @jean-edouard in a comment when ready.

The full list of commands accepted by this bot can be found here.

Needs approval from an approver in each of these files:

• OWNERS

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[vladikr]

I'm not sure... why do you think it will fail with SRIOV/GPU?

[stu-gott]

Would it be possible to exercise this in a functional test or two?

[stu-gott]

@vladikr I think the issue here is with dedicated hardware. If a physical device is provided directly to a VM, is it migratable?

[davidvossel]

I'm not sure... why do you think it will fail with SRIOV/GPU?

yeah, this is confusing to me as well. it could fail in some scenarios, but it's unclear what makes it likely to fail. there are situations where it's likely to pass I believe... but it's possible i'm missing something.

[jean-edouard]

This PR is obviously not the right thing to do, closing for now.

[kubevirt-bot]

@jean-edouard: The following tests failed, say /retest to rerun all failed tests:

Test name	Commit	Details	Rerun command
pull-kubevirt-e2e-k8s-cnao-1.17	8a5b4ed	link	/test pull-kubevirt-e2e-k8s-cnao-1.17
pull-kubevirt-e2e-kind-k8s-1.17.0-ipv6	8a5b4ed	link	/test pull-kubevirt-e2e-kind-k8s-1.17.0-ipv6
pull-kubevirt-e2e-k8s-1.17	8a5b4ed	link	/test pull-kubevirt-e2e-k8s-1.17
pull-kubevirt-e2e-k8s-1.18	8a5b4ed	link	/test pull-kubevirt-e2e-k8s-1.18
pull-kubevirt-e2e-windows2016	8a5b4ed	link	/test pull-kubevirt-e2e-windows2016
pull-kubevirt-e2e-k8s-1.16	8a5b4ed	link	/test pull-kubevirt-e2e-k8s-1.16
pull-kubevirt-generate	8a5b4ed	link	/test pull-kubevirt-generate

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository. I understand the commands that are listed here.

[fabiand]

Could somebody explain the exact details of why this got closed?

[jean-edouard]

This got closed because it is unclear which settings are incompatible with live migration, if any.

Even if a VMI uses SRIOV/vGPU, it's probably still migratable to a node that also supports those (and has a similar config...)
Still, it might be difficult to test and even harder to fix potential issues with such migrations, so maybe we do want something like this PR, but it has to be made clear that it's a deliberate design decision and not necessarily a technical limitation.

[vladikr]

The main issue here is copying of the hardware state (pure hardware operations that were offloaded to the device) during live migration, which is not possible without the hardware and the hypervisor support (I don't think this exist)
However, other projects don't block as this specifically doesn't seem to pose a high risk.
I don't know if such an issue exists with GPU.

One other issue with SR-IOV is that we will need to rebind the network to the VF on the destination side.
I think @EdDev is looking into this.