Update code to use k8s.io/api/discovery/v1 instead of v1beta1

[invidian]

Closes #980

[invidian]

I probably rushed too much with this PR, as discovery/v1 is only available since K8s 1.21 and v1beta1 will be removed only in 1.25 release. And I'm not sure if it's worth it figuring out support for both in parallel. Even currently used version of kind in e2e tests deploys K8s 1.20 by default, hence the test failure I think.

BTW, I don't know why linter fails though. Running it locally with the same version does not report any issues to me and CI does not print why it fails 😕

See also https://kubernetes.io/docs/reference/using-api/deprecation-guide/#endpointslice-v125

[fedepaol]

@invidian , I'd rather wait to have 1.20 out of the support loop, which should happen fairly soon.
At that point we can bump the version of kind and let this in.
There is also another change I think makes sense, which is starting to use the nodeName field as opposed to the opology["kubernetes.io/hostname"] one

[fedepaol]

@invidian I think it's now time to revamp this since 1.20 is out of support lifecycle, are you willing to work on this again?

[invidian]

I'll rebase and see how it goes then 👍

[fedepaol]

I'll rebase and see how it goes then +1

Awesome, thanks! I see kind already pins to 1.21, so there is no need to update it.

[rata]

@fedepaol is 1.20 EOL yet? A new 1.20.x release was just created: https://groups.google.com/a/kubernetes.io/g/dev/c/Y1q7XUp5gjM/m/K_JZcn-5AgAJ

[fedepaol]

@fedepaol is 1.20 EOL yet? A new 1.20.x release was just created: https://groups.google.com/a/kubernetes.io/g/dev/c/Y1q7XUp5gjM/m/K_JZcn-5AgAJ

My understanding was that k8s maintain the n-3 latest release. According to https://endoflife.date/kubernetes , 1.20 active support ended in December, bug fixing will end next month.
Also note that in case ep slice is not available, we back off to endpoints (so it's not that metallb would stop working on older clusters).

[fedepaol]

@fedepaol is 1.20 EOL yet? A new 1.20.x release was just created: https://groups.google.com/a/kubernetes.io/g/dev/c/Y1q7XUp5gjM/m/K_JZcn-5AgAJ

My understanding was that k8s maintain the n-3 latest release. According to https://endoflife.date/kubernetes , 1.20 active support ended in December, bug fixing will end next month. Also note that in case ep slice is not available, we back off to endpoints (so it's not that metallb would stop working on older clusters).

Also, https://kubernetes.io/releases/_print/

[rata]

Ohh, great then! :)

[invidian]

I'm not sure why linter fails on CI (it runs quite old version BTW). This PR does not touch mentioned files...

And building was failing due to dependencies requiring Go 1.17 now, which wasn't quite explicitly said, so I had to downgrade the K8s dependencies to v0.22.x until Go is updated to 1.17.

[fedepaol]

Apologies for following up so late, let's wait for CI and merge this. And thanks.

[invidian]

Thanks for finishing this PR. I was on holidays and I couldn't do it :)

[fedepaol]

np, I felt guilty enough not to merge it before :P