install: Update node label prefix

[kartikjoshi21]

Update node label from node-role.kubernetes.io to node.kubernetes.io

Fixes: #194

[fidencio]

@kartikjoshi21, thanks for the contribution and I'd like to have this one merged before the release.

May I ask you to add the info you provided in the issue as part of the commit message?
"""
For Kubernetes version greater than or equal to 1.16, node label keys in the 'kubernetes.io' or 'k8s.io' namespace must begin with an allowed prefix (kubelet.kubernetes.io, node.kubernetes.io).
"""

This allows folks taking a look at the git history to easily understand why a change was made without having to visit GitHub to do so.

[kartikjoshi21]

@kartikjoshi21, thanks for the contribution and I'd like to have this one merged before the release.

May I ask you to add the info you provided in the issue as part of the commit message? """ For Kubernetes version greater than or equal to 1.16, node label keys in the 'kubernetes.io' or 'k8s.io' namespace must begin with an allowed prefix (kubelet.kubernetes.io, node.kubernetes.io). """

This allows folks taking a look at the git history to easily understand why a change was made without having to visit GitHub to do so.

Thanks @fidencio, I Have updated the commit message.

[stevenhorsman]

/test

[stevenhorsman]

@kartikjoshi21 - the tests are getting an error:

error: taint "node.kubernetes.io/master:NoSchedule" not found

which I think is related to this change?

[wainersm]

This is how I understand the situation:

• the operator CI deploys the test cluster with kubeadm, which labels the node as node-role.kubernetes.io. I will need to change the kubeadm configs to use the new label;
• the operator CI for enclave-cc deploys the test cluster with kind, which already labels the node as node.kubernetes.io. Additionally the workflow script add the node-role.kubernetes.io label, that explains why the enclave-cc job didn't fail (?)

This PR fixes #194 that is a duplicated of #130 . On this last issue there is some discussion whether change or not the label. I am in favor of the change btw, but as we are about to release 0.5 and it has potential to break stuffs: @kartikjoshi21 mind if we post-pone this change to after the release? If yes then I will put the "do-not-merge" label on this PR.

[kartikjoshi21]

@wainersm I think we can postpone it, and CAA one which is related to this one as well. Meanwhile i have a workaround to label node to node-role.* which is already merged.

[katexochen]

@stevenhorsman @wainersm @bpradipt What is the current state of this issue? Does it need more discussion? Would be nice if we could include a fix in the next release.

[stevenhorsman]

@stevenhorsman @wainersm @bpradipt What is the current state of this issue? Does it need more discussion? Would be nice if we could include a fix in the next release.

Hey Paul, thanks for bumping this topic. IIRC there was a bit of risk to it so we postponed it as we were late in the release, but I think as we are early in the cycle if it's re-based and the tests pass then we can get it in and if there are issue still back it out without being too disruptive. @kartikjoshi21 - does that sound okay?

[stevenhorsman]

/test

[kartikjoshi21]

Test failing due to error: taint "node.kubernetes.io/master:NoSchedule" not found

[stevenhorsman]

Test failing due to error: taint "node.kubernetes.io/master:NoSchedule" not found

When I raised this previously Wainer said:

the operator CI deploys the test cluster with kubeadm, which labels the node as node-role.kubernetes.io. I will need to change the kubeadm configs to use the new label;

[katexochen]

@wainersm any idea on how we could the CI passing for this PR? Could we temporarily add both labels to the CI kubeamd so both this PR and tests on main pass?

[wainersm]

hi @katexochen , sorry long delay to reply your msgs...

@wainersm any idea on how we could the CI passing for this PR? Could we temporarily add both labels to the CI kubeamd so both this PR and tests on main pass?

Here is the explanation: https://kubernetes.io/docs/setup/production-environment/tools/kubeadm/create-cluster-kubeadm/#managed-node-labels

I think we still need to untaint node-role.kubernetes.io but also ensure that node.kubernetes.io exists on the worker node.

[stevenhorsman]

/test

[katexochen]

tests-e2e-ubuntu-20.04_snp-x86_64-containerd_kata-qemu-snp

11:30:14 Cloning into '/tmp/tmp.nWm0FY5wVU/src/github.com/kata-containers/tests'...
11:30:16 ERROR: no known tests for runtime class kata-qemu-snp

[stevenhorsman]

tests-e2e-ubuntu-20.04_snp-x86_64-containerd_kata-qemu-snp

11:30:14 Cloning into '/tmp/tmp.nWm0FY5wVU/src/github.com/kata-containers/tests'...
11:30:16 ERROR: no known tests for runtime class kata-qemu-snp

SNP isn't supported at the moment, so that's an expected failure.

[stevenhorsman]

LGTM. Thanks

[fidencio]

lgtm, thanks @kartikjoshi21 and @katexochen!