Usage of IPs returned by `InstancesV2().InstanceMetadata()`, and interaction with `--node-ip` about cloud-provider

[olemarkus]

The first entry is the primary node IP, yes. And yes, that is the correct annotation.

[deitch]

Anyone have insight into these issues?

[olemarkus]

If kubelet is configured to use --node-ip, this will end up in an annotation being set on the Node object that kubelet creates.
CCM creates a list of eligible IPs that the Node object can use. If the IP in the annotation is also in the list of IPs from CCM, this IP becomes the "primary" IP of the Node object.

Also note that Node IPs are either internal or external. This is not synonymous with private and public. Internal means used for intra-cluster communication, and can (often) be public IPs. For example when IPv6 is used, some CCMs always classify these as Internal.

[deitch]

Thanks for hopping in @olemarkus

Yes, I definitely get the private/public vs internal/external distinctions, but it is good to have them clearly here on this issue.

So to summarize. Setting node-ip has no impact on the node IP itself. What it does do, is add an annotation. The CCM, upon receiving the v1.Node object, can check for and use that annotation. The actual IPs used are those provided by the CCM, no matter what node-ip said.

When you wrote "list of eligible IPs", how does kubernetes decide, from that list, "this is the one I will use for internal comms, which is the one listed when I do kubectl get node? I'm asking both with setting node-ip and without? It's obvious when there's just one internal IP, but if there is a list?

[olemarkus]

Assuming an external CCM is used, then no, setting --node-ip doesn't really provide any guarantees. It is ultimately CCM that decides what IPs to use. An external CCM may even not use this library and provide their own logic.

"List of eligible IPs" is whatever is returned by the NodeAddressesByProviderID() function that is a part of the CCM Instance interface. Usually it involves calls to the relevant cloud API.

[deitch]

"List of eligible IPs" is whatever is returned by the NodeAddressesByProviderID() function that is a part of the CCM Instance interface. Usually it involves calls to the relevant cloud API

Sure, totally got that. When it gets returned, how does kubernetes decide which IP to use? Is it first in the list?

[deitch]

Also, is it this annotation, which is alpha? Or some other?

[deitch]

When I pass --node-ip to the kubelet, does the kubelet actually validate that it is a valid address? Or does it pass it on to the CCM as is? If I passed in --node-ip="Ole Markus is Helpful", will the kubelet catch that and error it out? Or is it up to the CCM to do so?

[deitch]

Hmm, while we are at it, is there any way for a user to provide an external IP as well? Or is the --node-ip, and how it gets passed to CCM, expected to be internal (although the CLI flag and annotation do not say so, just "provided")?

[olemarkus]

It will certainly be validated. Kubelet also uses the argument for what IPs to listen to on the kubelet server. Using an external IP probably works. But keep in mind that the control plane and a few pods typically need to reach the kubelet API.

[k8s-triage-robot]

The Kubernetes project currently lacks enough contributors to adequately respond to all issues and PRs.

This bot triages issues and PRs according to the following rules:

• After 90d of inactivity, lifecycle/stale is applied
• After 30d of inactivity since lifecycle/stale was applied, lifecycle/rotten is applied
• After 30d of inactivity since lifecycle/rotten was applied, the issue is closed

You can:

• Mark this issue or PR as fresh with /remove-lifecycle stale
• Mark this issue or PR as rotten with /lifecycle rotten
• Close this issue or PR with /close
• Offer to help out with Issue Triage

Please send feedback to sig-contributor-experience at kubernetes/community.

/lifecycle stale

[k8s-triage-robot]

The Kubernetes project currently lacks enough active contributors to adequately respond to all issues and PRs.

This bot triages issues and PRs according to the following rules:

• After 90d of inactivity, lifecycle/stale is applied
• After 30d of inactivity since lifecycle/stale was applied, lifecycle/rotten is applied
• After 30d of inactivity since lifecycle/rotten was applied, the issue is closed

You can:

• Mark this issue or PR as fresh with /remove-lifecycle rotten
• Close this issue or PR with /close
• Offer to help out with Issue Triage

Please send feedback to sig-contributor-experience at kubernetes/community.

/lifecycle rotten

[k8s-triage-robot]

The Kubernetes project currently lacks enough active contributors to adequately respond to all issues and PRs.

This bot triages issues according to the following rules:

• After 90d of inactivity, lifecycle/stale is applied
• After 30d of inactivity since lifecycle/stale was applied, lifecycle/rotten is applied
• After 30d of inactivity since lifecycle/rotten was applied, the issue is closed

You can:

• Reopen this issue with /reopen
• Mark this issue as fresh with /remove-lifecycle rotten
• Offer to help out with Issue Triage

Please send feedback to sig-contributor-experience at kubernetes/community.

/close not-planned

[k8s-ci-robot]

@k8s-triage-robot: Closing this issue, marking it as "Not Planned".

In response to this:

The Kubernetes project currently lacks enough active contributors to adequately respond to all issues and PRs.

This bot triages issues according to the following rules:

• After 90d of inactivity, lifecycle/stale is applied
• After 30d of inactivity since lifecycle/stale was applied, lifecycle/rotten is applied
• After 30d of inactivity since lifecycle/rotten was applied, the issue is closed

You can:

• Reopen this issue with /reopen
• Mark this issue as fresh with /remove-lifecycle rotten
• Offer to help out with Issue Triage

Please send feedback to sig-contributor-experience at kubernetes/community.

/close not-planned

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.