Support bare-metal servers with hcloud-cloud-controller-manager

[apricote]

/kind feature

Describe the solution you'd like

Right now you only support Robot (bare-metal) servers when using syself/hetzner-cloud-controller-manager.

hetznercloud/hcloud-cloud-controller-manager (HCCM) added support for Robot servers in December 2023. It would be awesome if you would also support this with CAPH.

Anything else you would like to add:

As far as I could tell, the main problem is that CAPH expects the provider ID for bare-metal servers to be formatted as hcloud://bm-$ROBOTID. HCCM uses the format hrobot://$ROBOTID.

• cluster-api-provider-hetzner/api/v1beta1/hetznerbaremetalmachine_types.go

  Lines 40 to 41 in c05f7aa

  	// BareMetalHostNamePrefix is a prefix for all hostNames of bare metal servers.
  	BareMetalHostNamePrefix = "bm-"

• cluster-api-provider-hetzner/pkg/scope/baremetalhost.go

  Lines 142 to 151 in c05f7aa

  	// Hostname returns the desired host name.
  	func (s *BareMetalHostScope) Hostname() (hostname string) {
  	if s.hasConstantHostname() {
  	hostname = fmt.Sprintf("%s%s-%v", infrav1.BareMetalHostNamePrefix, s.Cluster.Name, s.HetznerBareMetalHost.Spec.ServerID)
  	} else {
  	hostname = infrav1.BareMetalHostNamePrefix + s.HetznerBareMetalHost.Spec.ConsumerRef.Name
  	}
  	return hostname
  	}

• cluster-api-provider-hetzner/controllers/csr_controller.go

  Line 328 in c05f7aa

  providerID := "hcloud://bm-" + serverID

• cluster-api-provider-hetzner/pkg/services/baremetal/baremetal/baremetal.go

  Lines 895 to 897 in c05f7aa

  	func providerIDFromServerID(serverID int) string {
  	return fmt.Sprintf("%s%s%d", providerIDPrefix, infrav1.BareMetalHostNamePrefix, serverID)
  	}

Related to hetznercloud/hcloud-cloud-controller-manager#702

Environment:

• cluster-api-provider-hetzner version:
• Kubernetes version: (use kubectl version)
• OS (e.g. from /etc/os-release):

[janiskemper]

@apricote would you be able to contribute here?

[simonostendorf]

@apricote @janiskemper did you already thought about a migration path? I think the "old" variant with bm- in the hostname and the provider id should be completely replaced with no prefix in the hostname and the hrobot:// provider prefix, but that would require a migration of existing nodes and the bm- is frequently used inside the code.

[janiskemper]

@simonostendorf we at Syself have the wish to migrate to the "official" one at one point, but did not spend the time on developing or testing a clear migration path. I'd be happy if @apricote or someone else can contribute!

[simonostendorf]

@simonostendorf we at Syself have the wish to migrate to the "official" one at one point, but did not spend the time on developing or testing a clear migration path. I'd be happy if @apricote or someone else can contribute!

@janiskemper If you have a defined migration path I can try and add this to my todo list, but I cannot promise anything.

[janiskemper]

@guettli has spent some time on it, but I don't think that it is a "defined migration path"

[apricote]

My idea would be:

1. Update CAPH to permit hrobot:// IDs
2. Update syself/hetzner-cloud-controller-manager to read from both IDs
3. In either tool add an automatic hook to rewrite the IDs from hcloud://bm- to hrobot://

• Question would be if anyone else is using hcloud://bm- and needs to be compatible with the new prefix. Quick search on GitHub only shows HCCM & CAPH as well as some forks of both projects.

[janiskemper]

thanks @apricote for your input!

@batistein you also had some thoughts on the difficulties!

[guettli]

My idea would be:

1. Update CAPH to permit hrobot:// IDs
2. Update syself/hetzner-cloud-controller-manager to read from both IDs
3. In either tool add an automatic hook to rewrite the IDs from hcloud://bm- to hrobot://

@apricote

Step 1 and Step 3 look good.

However, I am unsure about Step 3. What do you mean by "hook"? I would not modify existing node objects.

What do you think about these phases?

• Phase 1: We update all places that read the ProviderID to accept both schemas. Then, we wait a few months.
• Phase 2: We update the part that creates the old ProviderID schema so that it now generates the new hrobot:// schema.
• Phase 3: Three years later, we remove the code that reads the old hcloud://bm- schema.

[simonostendorf]

I think Phase 1 and 2 can be performed together, so that new nodes automatically get the hrobot provider id and the old variant is only for old nodes.