platform: Hetzner API implementation

[apricote]

Add support for the Hetzner Platform to kola and ore.

Usage looks like:

HETZNER_IMAGE_ID=$(./bin/ore \
  hetzner \
  --hetzner-token $(pass hetzner/token) \
  --hetzner-location fsn1 \
  create-image \
  --name flatcar-nightly \
  --file http://bincache.flatcar-linux.net/images/amd64/4011.0.0+nightly-20240624-2100/flatcar_production_hetzner_image.bin.bz2)

./bin/kola \
  --platform hetzner \
  --hetzner-token $(pass hetzner/token) \
  --hetzner-server-type cpx11 \
  --hetzner-location fsn1 \
  --hetzner-image "${HETZNER_IMAGE_ID}"
  run '*'

[tormath1]

Thanks! It looks good for a first iteration. As said on Matrix, we can see to implement a firewall (or update the default one) to allow connection on port 2379 for example. In api.go or in a network.go file.

EDIT: Just seen your message on Matrix

Only ports 25 & 465 (SMTP) are blocked by default.

[tormath1]

LGTM with the last two nits. Feel free to rearrange your commits as you wish - ideally we could rebase and squash your commits into those commits: https://github.com/flatcar/mantle/pull/533/commits to get one initial commit for the API implementation, one initial commit for the Kola platform implementation and one for the ore part.
I don't have a strong opinion on this, so it's up to you. :)

Tested with Flannel and Kubeadm:

=== RUN   cl.flannel.vxlan
=== RUN   cl.flannel.udp
=== RUN   kubeadm.v1.29.2.calico.base
=== RUN   kubeadm.v1.29.2.calico.base/node_readiness
=== RUN   kubeadm.v1.29.2.calico.base/nginx_deployment
=== RUN   kubeadm.v1.29.2.calico.base/NFS_deployment
--- PASS: cl.flannel.vxlan (208.44s)
        flannel.go:121: ping from 50821554(10.254.14.0) to 50821552(10.254.47.0)
--- PASS: cl.flannel.udp (271.60s)
        cluster.go:125: Timeout occurred while waiting for network connectivity.
        flannel.go:121: ping from 50821692(10.254.9.0) to 50821694(10.254.54.1)
--- PASS: kubeadm.v1.29.2.calico.base (591.11s)
    --- PASS: kubeadm.v1.29.2.calico.base/node_readiness (17.55s)
    --- PASS: kubeadm.v1.29.2.calico.base/nginx_deployment (24.24s)
    --- PASS: kubeadm.v1.29.2.calico.base/NFS_deployment (28.03s)
            cluster.go:125: WARNING: Kubernetes configuration file is group-readable. This is insecure. Location: /home/core/.kube/config
            cluster.go:125: WARNING: Kubernetes configuration file is world-readable. This is insecure. Location: /home/core/.kube/config
            cluster.go:125: WARNING: Kubernetes configuration file is group-readable. This is insecure. Location: /home/core/.kube/config
            cluster.go:125: WARNING: Kubernetes configuration file is world-readable. This is insecure. Location: /home/core/.kube/config
            cluster.go:125: jq: error (at <stdin>:121): Cannot iterate over null (null)
            cluster.go:125: jq: error (at <stdin>:121): Cannot iterate over null (null)
            cluster.go:125: jq: error (at <stdin>:121): Cannot iterate over null (null)
PASS, output in _kola_temp/hetzner-2024-07-24-1113-35472

[apricote]

• Squashed all commits into 5 along the lines you suggested and had in [wip] platform: add Hetzner #533.
• Rebased on current flatcar-master and changed the PR target branch to it.
• Updated hcloud-go to latest version

All tests besides one work for me with this version. The failing test looks weird to me, in that it tries to parse cloud-init config.

--- FAIL: cl.cloudinit.basic (292.05s)
        cluster.go:125: cat: /foo: No such file or directory
        cluster.go:145: "cat /foo" failed: output , status Process exited with status 1

[tormath1]

• Squashed all commits into 5 along the lines you suggested and had in [wip] platform: add Hetzner #533.

  • Rebased on current flatcar-master and changed the PR target branch to it.

  • Updated hcloud-go to latest version

All tests besides one work for me with this version. The failing test looks weird to me, in that it tries to parse cloud-init config.

--- FAIL: cl.cloudinit.basic (292.05s)
        cluster.go:125: cat: /foo: No such file or directory
        cluster.go:145: "cat /foo" failed: output , status Process exited with status 1

Great, thanks a lot for the cleanup. For the failing test, I suspect it's because the coreos-cloudinit support is not implemented for Hetzner: https://github.com/flatcar/coreos-cloudinit/tree/flatcar-master/datasource/metadata

Flatcar does not ship the "official" cloudinit but a rewriting in Go that implements a subset of the official cloud-init. I think for now we can skip this test and we can still implement the Hetzner cloudinit support later.

diff --git a/kola/tests/misc/cloudinit.go b/kola/tests/misc/cloudinit.go
index 9ef74d541..0abce80a4 100644
--- a/kola/tests/misc/cloudinit.go
+++ b/kola/tests/misc/cloudinit.go
@@ -156,7 +156,8 @@ write_files:
   - path: "/foo"
     content: bar`),
 		Distros:          []string{"cl"},
-		ExcludePlatforms: []string{"qemu-unpriv"},
+		// Hetzner: we need to implement coreos-cloudinit support for Hetzner.
+		ExcludePlatforms: []string{"qemu-unpriv", "hetzner"},
 		// This should run on all clouds
 	})
 	register.Register(&register.Test{

The CI failure is "normal" (here: https://github.com/flatcar/mantle/actions/runs/10253573202/job/28366511620?pr=536) - it's because you pushed from a fork.

[apricote]

Disabled the cloudinit test in 1ce68f9.

[tormath1]

Thanks for this implementation! Looking forward to seeing Flatcar tests running on Hetzner