Unable to install k8s on IPv6-only node (or provisioning a node from a IPv6-only server)

[Maic-Rittmeier]

I am using kube-hetzner module v2.17.1.

When defining a control plane nodepool with disabling ipv4 like this:

  control_plane_nodepools = [
    {
      name        = "control-plane-fsn1",
      server_type = "cx22",
      location    = "fsn1",
      labels = [],
      taints = [],
      count       = 1

      disable_ipv4 = true
    }
  ]

I got following error during apply:

module.kube-hetzner.null_resource.first_control_plane (remote-exec): + INSTALL_K3S_SKIP_START=true
module.kube-hetzner.null_resource.first_control_plane (remote-exec): + INSTALL_K3S_SKIP_SELINUX_RPM=true
module.kube-hetzner.null_resource.first_control_plane (remote-exec): + INSTALL_K3S_CHANNEL=v1.31
module.kube-hetzner.null_resource.first_control_plane (remote-exec): + INSTALL_K3S_EXEC='server '
module.kube-hetzner.null_resource.first_control_plane (remote-exec): + sh -
module.kube-hetzner.null_resource.first_control_plane (remote-exec): + curl -sfL https://get.k3s.io/
module.kube-hetzner.null_resource.first_control_plane (remote-exec): [INFO]  Finding release for channel v1.31
module.kube-hetzner.null_resource.first_control_plane: Still creating... [2m20s elapsed]
module.kube-hetzner.null_resource.first_control_plane (remote-exec): curl: (28) Failed to connect to github.com port 443 after 135630 ms: Could not connect to server
module.kube-hetzner.null_resource.first_control_plane (remote-exec): [INFO]  Using v1.31.8+k3s1 as release
module.kube-hetzner.null_resource.first_control_plane (remote-exec): [INFO]  Downloading hash https://github.com/k3s-io/k3s/releases/download/v1.31.8+k3s1/sha256sum-amd64.txt
module.kube-hetzner.null_resource.first_control_plane: Still creating... [4m40s elapsed]
module.kube-hetzner.null_resource.first_control_plane (remote-exec): [ERROR]  Download failed
╷
│ Error: remote-exec provisioner error
│ 
│   with module.kube-hetzner.null_resource.first_control_plane,
│   on .terraform/modules/kube-hetzner/init.tf line 94, in resource "null_resource" "first_control_plane":
│   94:   provisioner "remote-exec" {
│ 
│ error executing "/tmp/terraform_1094854073.sh": Process exited with status
│ 1
╵

This seems to be a problem that the host gitlab.com could not reached from the ipv6 only node.

k3s-control-plane-fsn1-pyg:~ # ping -6 github.com

ping: github.com: Address family for hostname not supported

Are there any workarounds to get this setup running?

PS: I need this ipv6 only setup, because my previous issue was the initial provisioning from here:

# Wait for MicroOS to reboot and be ready.
  provisioner "local-exec" {
    command = <<-EOT
      timeout 600 bash <<EOF
          until ssh ${local.ssh_args} -i /tmp/${random_string.identity_file.id} -o ConnectTimeout=2 -p ${var.ssh_port} root@${coalesce(self.ipv4_address, self.ipv6_address, try(one(self.network).ip, null))} true 2> /dev/null
          do
            echo "Waiting for MicroOS to become available..."
            sleep 3
          done
      EOF
    EOT
  }

This initial ssh connection was not possible from our gitlab-runner instance (also using ipv6 only). The issue here is, when setting disable_ipv4 = false, the self.pv4_address is always uses to connect to the node via ssh.

[tskimmett]

There are some other discussion threads about ipv6 only nodes. The primary blocker for this is github's lack of ipv6 support, which was one of the issues you encountered.

To the maintainers of this repo, do you think it would be worth mentioning in the readme that ipv6-only nodes do not really work? I know for myself I went down this rabbit hole since hetzner gives a discount for not using ipv4, but quickly found it not to be worth the 80 cents.

[cyb3rko]

A workaround is using an IPv6 proxy to GitHub:
https://danwin1210.de/github-ipv6-proxy.php

Maybe the maintainers decide if they want to offer that functionality for IPv6-only clusters.