[WIP] first pass at calico/ipv6 kind setup

[mattfenwick]

No description provided.

[mattfenwick]

This doesn't work: some of the pods come up with ipv4 addresses and crash:

[lwr20]

I think you're missing a bit of config:
https://docs.projectcalico.org/networking/ipv6#enable-ipv6-only

Edit the CNI config (calico-config ConfigMap in the manifest) to disable IPv4 assignments and enable IPv6 assignments.

    "ipam": {
        "type": "calico-ipam",
        "assign_ipv4": "false",
        "assign_ipv6": "true"
    },

Not sure the best way to edit the configmap. Maybe kustomize (since its built into kubectl)?

[lwr20]

You could just make this change in calico.yaml now

[mattfenwick]

👍 awesome! looks like this is working on my mac now 😄

saving a copy of the calico yaml seems like the easiest approach, is that a horrible idea?

really appreciate the help, 🙇

[lwr20]

saving a copy of the calico yaml seems like the easiest approach, is that a horrible idea?

It'll work, but it means you'll need to raise a PR every time Calico does a release to get the latest.

I've recently been using kustomize for stuff - it might be more elegant here to:

• create a dir to hold the yaml and kustomize config - let's call it kustom
• download calico.yaml into kustom/calico.yaml
• create a kustom/kustomize.yaml file
• use kubectl apply -k kustom to modify and apply it in one step

[mattfenwick]

Gotcha, thanks for the suggestion! Opened #109 to track this 😄

[mattfenwick]

TCP requests to services on IPV6/KinD clusters seem to be slower than UDP/SCTP requests, and also slower than IPV4 requests -- in the two Linux environments I ran this PR on.

This seems to be general KinD/IPV6 behavior -- unrelated to Calico. Follow up in #108 to figure out why.

In the meantime, this works but is slower! https://github.com/mattfenwick/cyclonus/runs/2371412592?check_suite_focus=true