chore(ci)!: Update tested Kubernetes versions

[spencergilbert]

Signed-off-by: Spencer Gilbert spencer.gilbert@gmail.com

While Vector still runs on kubernetes 1.14.x, as far as I can tell it's no longer offered/supported
by any major cloud provider. Right now #6564 is failing CI as we try to use a kubectl command that isn't
supported in 1.14.x (rollout restart).

This PR does the following:

• Drops 1.14.10

I'd also suggest consider adjusting our documentation around required/supported kubernetes versions. We'll also need to update and include more versions in a followup

[jszwedko]

Should we update the docs with this change too? 👍 on supporting only the lowest available version on major cloud providers.

[spencergilbert]

Should we update the docs with this change too? 👍 on supporting only the lowest available version on major cloud providers.

I wasn't sure if we wanted to - the chart still works on 1.14, we're no longer testing - I think it's a good idea to tighten the scope of supported versions, but that's also a larger discussion and (possibly) impactful to users/customers

[StephenWakely]

I'd say we need to update the docs. If we aren't testing it, we aren't supporting it.

[spencergilbert]

I'd say we need to update the docs. If we aren't testing it, we aren't supporting it.

I'd defer to the group/leadership, but there's definitely some semantics here - Vector is still compatible with 1.14, we're not testing it actively, but we're also "best effort" not breaking existing compatibilities. Perhaps we could adopt language similar to minikube's:

minikube follows the Kubernetes Version and Version Skew Support Policy, so we guarantee support for the latest build for the last 3 minor Kubernetes releases. When practical, minikube aims to support older releases as well so that users can emulate legacy environments.

[jszwedko]

I'd say we need to update the docs. If we aren't testing it, we aren't supporting it.

I'm inclined to agree with this. If we aren't testing it, we aren't really supporting it.

My preference would be to support the lowest common denominator version for the major cloud providers.

I do think we should update the docs with this PR.

[spencergilbert]

I'd say we need to update the docs. If we aren't testing it, we aren't supporting it.

I'm inclined to agree with this. If we aren't testing it, we aren't really supporting it.

👍

My preference would be to support the lowest common denominator version for the major cloud providers.

I think can figure this out in a following issue/rfc

I do think we should update the docs with this PR.

I'll just update the docs to be >=1.15 in this PR... though, do we need to consider this a breaking change?

[jszwedko]

I'll just update the docs to be >=1.15 in this PR... though, do we need to consider this a breaking change?

Yeah, this is a good question. I guess technically it should be a breaking change. I'd like to have us define a policy for what is breaking and not. I'll make a note of this for next quarter.

[spencergilbert]

I'll update it to be a breaking change, and I'll fix these patch versions - I went a little too new for these patches, I think the failing ones are to be released next week 😅

[spencergilbert]

Tested the two failed jobs locally and they passed, so I'm guessing just flakey? Or maybe related to the runner env

[StephenWakely]

Tested the two failed jobs locally and they passed, so I'm guessing just flakey?

I have a hunch about that flakey test. It's not to do with this PR. By creating the affinity pod in the same namespace as the test pod, sometimes the test picks up the logs from that pod instead. I'll get a fix.

[spencergilbert]

Tested the two failed jobs locally and they passed, so I'm guessing just flakey?

I have a hunch about that flakey test. It's not to do with this PR. By creating the affinity pod in the same namespace as the test pod, sometimes the test picks up the logs from that pod instead. I'll get a fix.

Talked a bit over slack we have two sources of flakiness here, cri-o related and one in our tests @StephenWakely will fix