feat(node): disable zone label for cloud servers

docs/guides/README.md

@@ -7,4 +7,5 @@ In this folder, you should find guides for you to accomplish specific tasks with
 - [Load Balancer](load-balancer/README.md)
 - [Robot](robot/README.md)
 - [Address Family](address-family.md)
+- [Zone Label](zone-label.md)
 - [Troubleshooting](troubleshooting.md)

docs/guides/zone-label.md

@@ -0,0 +1,44 @@
+# Zone Label
+
+When a node is initialized, the hcloud-cloud-controller-manager sets these topology labels on it:
+
+| Label                                                                       | Value                                | Example     |
+| --------------------------------------------------------------------------- | ------------------------------------ | ----------- |
+| `topology.kubernetes.io/region`, `failure-domain.beta.kubernetes.io/region` | Location of the Server               | `fsn1`      |
+| `topology.kubernetes.io/zone`, `failure-domain.beta.kubernetes.io/zone`     | Legacy datacenter name of the Server | `fsn1-dc14` |
+
+The zone label can be disabled by setting `HCLOUD_INSTANCES_ZONE_LABEL_ENABLED` to `false`. By default, the value is set to `true`.
+
+**We recommend disabling the zone label for new clusters.** Datacenters are deprecated in the Hetzner Cloud API, and the legacy datacenter names the label uses are only known for a fixed set of older locations. For any newer location, the label falls back to the location name and is therefore identical to the region label. The location is the failure domain you actually want to spread workloads over, and it is already available via `topology.kubernetes.io/region`. We plan to remove the zone label entirely in the next major version.
+
+The label remains enabled by default because disabling it on a running cluster changes node labels, which existing workloads may depend on. See [Existing clusters](#existing-clusters) below.
+
+## Configuration via Helm
+
+```yaml
+# values.yaml
+---
+env:
+  HCLOUD_INSTANCES_ZONE_LABEL_ENABLED:
+    value: "false"
+```
+
+## Existing clusters
+
+Topology labels are only applied while a node is being initialized. Changing `HCLOUD_INSTANCES_ZONE_LABEL_ENABLED` does not relabel nodes that are already part of the cluster, so after disabling it you end up with a mixed cluster: existing nodes keep their zone label, and nodes joining afterwards do not get one. Anything that selects on the zone label — `nodeSelector`, node affinities, `topologySpreadConstraints`, or the node affinity of already-provisioned `PersistentVolume`s — will treat these two groups differently.
+
+Before disabling the label, check whether anything in your cluster relies on it:
+
+```sh
+kubectl get pods,pv --all-namespaces -o yaml | grep -c 'topology.kubernetes.io/zone\|failure-domain.beta.kubernetes.io/zone'
+```
+
+To reach a consistent state afterwards, either recreate your nodes, or remove the labels from the existing ones:
+
+```sh
+kubectl label nodes --all topology.kubernetes.io/zone- failure-domain.beta.kubernetes.io/zone-
+```
+
+## Robot servers
+
+This setting only affects Hetzner Cloud Servers. [Robot](robot/README.md) servers are always labeled with their datacenter (e.g. `fsn1-dc14`).

hcloud/cloud_test.go

@@ -74,6 +74,7 @@ func newTestEnv() testEnv {
 
 	cfg := config.HCCMConfiguration{}
 	cfg.Instance.AddressFamily = config.AddressFamilyIPv4
+	cfg.Instance.ZoneLabelEnabled = true
 
 	return testEnv{
 		Server:      server,

hcloud/instances.go

@@ -356,16 +356,23 @@ func (s hcloudServer) IsShutdown() (bool, error) {
 }
 
 func (s hcloudServer) Metadata(networkID int64, _ *corev1.Node, cfg config.HCCMConfiguration) (*cloudprovider.InstanceMetadata, error) {
-	return &cloudprovider.InstanceMetadata{
+	metadata := &cloudprovider.InstanceMetadata{
 		ProviderID:    providerid.FromCloudServerID(s.ID),
 		InstanceType:  s.ServerType.Name,
 		NodeAddresses: hcloudNodeAddresses(networkID, s.Server, cfg),
-		Zone:          legacydatacenter.NameFromLocation(s.Location.Name),
 		Region:        s.Location.Name,
 		AdditionalLabels: map[string]string{
 			ProvidedBy: "cloud",
 		},
-	}, nil
+	}
+
+	// By default, we continue to configure a zone label. The user
+	// can configure this behavior. We can drop this entirely in a v2.
+	if cfg.Instance.ZoneLabelEnabled {
+		metadata.Zone = legacydatacenter.NameFromLocation(s.Location.Name)
+	}
+
+	return metadata, nil
 }
 
 type robotServer struct {

internal/config/config.go

@@ -29,9 +29,10 @@ const (
 	robotRateLimitWaitTime  = "ROBOT_RATE_LIMIT_WAIT_TIME"
 	robotForwardInternalIPs = "ROBOT_FORWARD_INTERNAL_IPS"
 
-	hcloudInstancesAddressFamily = "HCLOUD_INSTANCES_ADDRESS_FAMILY"
-	hcloudServerCacheMode        = "HCLOUD_SERVER_CACHE_MODE"
-	hcloudServerCacheMaxAge      = "HCLOUD_SERVER_CACHE_MAX_AGE"
+	hcloudInstancesZoneLabelEnabled = "HCLOUD_INSTANCES_ZONE_LABEL_ENABLED"
+	hcloudInstancesAddressFamily    = "HCLOUD_INSTANCES_ADDRESS_FAMILY"
+	hcloudServerCacheMode           = "HCLOUD_SERVER_CACHE_MODE"
+	hcloudServerCacheMaxAge         = "HCLOUD_SERVER_CACHE_MAX_AGE"
 
 	// Disable the "master/server is attached to the network" check against the metadata service.
 	hcloudNetworkDisableAttachedCheck = "HCLOUD_NETWORK_DISABLE_ATTACHED_CHECK"
@@ -73,7 +74,8 @@ const (
 const ServerCacheDefaultMaxAge time.Duration = 10 * time.Second
 
 type InstanceConfiguration struct {
-	AddressFamily AddressFamily
+	AddressFamily    AddressFamily
+	ZoneLabelEnabled bool
 }
 
 type ServerCacheConfiguration struct {
@@ -185,6 +187,11 @@ func Read() (HCCMConfiguration, error) {
 		cfg.Instance.AddressFamily = AddressFamilyIPv4
 	}
 
+	cfg.Instance.ZoneLabelEnabled, err = getEnvBool(hcloudInstancesZoneLabelEnabled, true)
+	if err != nil {
+		errs = append(errs, err)
+	}
+
 	// ---- Server Cache ----
 
 	cfg.ServerCache = ServerCacheConfiguration{