Add support to set loopback to up

[MikeZappa87]

The CNI maintainers intend on deprecating the loopback plugin. This approach where we set the lo interface to up is similar to what CRI-O does. We (@mikebrow ) decided to have a flag to disable/enable and eventually we will go with removing the flags and just always setting the lo interface to up.

[k8s-ci-robot]

Skipping CI for Draft Pull Request.
If you want CI signal for your change, please convert it to an actual PR.
You can still manually trigger a test run with /test all

[aojea]

yay 👏

[MikeZappa87]

yay 👏

I had this on my queue for so long, figured I would just get it done.

[aojea]

I'm not familiar with the conventions of containerd for new APIs and config fields or the introduction of deprecations, so I can not judge that, but technically this LGTM

eventually we will go with removing the flags and just always setting the lo interface to up.

I still have question though that I think is worth to test it, what happens if an user uses this new option but still has the loopback cni plugin, if both loopback are configured are there any bad interactions or is a noop when a pod is created and runs both of them?

[MikeZappa87]

I'm not familiar with the conventions of containerd for new APIs and config fields or the introduction of deprecations, so I can not judge that, but technically this LGTM

eventually we will go with removing the flags and just always setting the lo interface to up.

I still have question though that I think is worth to test it, what happens if an user uses this new option but still has the loopback cni plugin, if both loopback are configured are there any bad interactions or is a noop when a pod is created and runs both of them?

The community CNI plugin just sets the lo interface to up. Cilium and Calico for example set the lo to up as well. I believe only flannel doesn't set it to up. This should not be impactful. Hopefully the users follow the upgrade instructions as well. However since the field defaults to the legacy behavior we should not see any issues.

[mikebrow]

/test pull-containerd-k8s-e2e-ec2

checking if "cmd: failed to assign an IP address to container" was a flake..

[MikeZappa87]

/test pull-containerd-k8s-e2e-ec2

[henry118]

LGTM. Just a minor nit on spacing :)

[mikebrow]

see comment on migration test in integration...

A nice to have here would be a test bucket for the config switch enabled vs disabled.. maybe an exec into busybox that uses localhost.. checks the result.

[MikeZappa87]

I manually tested this with 'use_internal_loopback = true'

lo was up.

[MikeZappa87]

/test pull-containerd-k8s-e2e-ec2

[mikebrow]

LGTM