Match this log message with the one above.

Also, this code looks to be the same, we might want to factor it out to a func? (name it MaybeRestartOnInconsistentPodCIDR()?)

I thought about that, the reasons I didn't do it is because these are 7 lines of code with a lot of deps on proxier methods ... the refactor to a func is not going to save more lines and will make it harder to read IMHO ...

to be honest, I like your idea, but with current state of the code on both proxies, I don't know if it is better to keep them isolated or to try to consolidate them

An alternative to avoid exposing the details of the detector everywhere:

add method to localDetector

err := localDetector.IsValid(node)
if err != nil {
  klkog.ErrorS(nil, "proxier.localDetector is no longer valid (%v), restarting kube-proxy", err)
  ...FlushAndExit()
}

I've added a new method to return the Value() of the local traffic discriminator

I meant instead of exposing the Value() and putting the logic outside, we expose just the notion that the detector is still valid (which is what is being checked here).

so the proxier call site just has

if proxier.localDetector.IsImplemented() {
  if err := proxier.localDetector.IsValid(node); err != nil {
    klog.ErrorS(...)
    klog.FlushAndExit(...)
  }
}

yeah, but localDetectors can use CIDRs or interfaces names/prefixes, or whatever we want to implement in the detector, adding a new method IsValid(node *v1.Node) is only valid for one of the traffic detector modes.
Do we want to add such specific method to an Interface?

I'm suggesting removing Mode() and Value() and instead have a IsValid(node *v1.Node).

For detectors that are always Valid, this can always return true.

ah, that makes more sense

Wouldn't that make it panic on_add since there is no local node UID set?

check I changed the logic to panic on PodCIDR only and added test cases to cover the different states, if no PodCIDR exists in the handler it means that the handler wasn't initialized, os it uses the new value for that