Add Support to manually distribute CP machines in different zones with AvailabilityZones field

Makefile

@@ -178,7 +178,7 @@ ENVTEST ?= $(LOCALBIN)/setup-envtest
 
 ## Tool Versions
 KUSTOMIZE_VERSION ?= v5.1.1
-CONTROLLER_TOOLS_VERSION ?= v0.16.1
+CONTROLLER_TOOLS_VERSION ?= v0.16.5
 
 .PHONY: kustomize
 kustomize: $(KUSTOMIZE) ## Download kustomize locally if necessary. If wrong version is installed, it will be removed before downloading.

api/v1alpha1/ionoscloudmachine_types.go

@@ -124,10 +124,20 @@ type IonosCloudMachineSpec struct {
 	NumCores int32 `json:"numCores,omitempty"`
 
 	// AvailabilityZone is the availability zone in which the VM should be provisioned.
-	//+kubebuilder:validation:Enum=AUTO;ZONE_1;ZONE_2
-	//+kubebuilder:default=AUTO
+	// AvailabilityZone is mutually exclusive with AvailabilityZones.
+	// If specified, AvailabilityZone will be used to provision the VM.
+	// +kubebuilder:validation:default=AUTO
+	// +kubebuilder:validation:Enum=AUTO;ZONE_1;ZONE_2
 	//+optional
-	AvailabilityZone AvailabilityZone `json:"availabilityZone,omitempty"`
+	AvailabilityZone *AvailabilityZone `json:"availabilityZone,omitempty"`
+
+	// AvailabilityZones is the list of availability zones where the VM should be provisioned.
+	// AvailabilityZones is mutually exclusive with AvailabilityZone.
+	// If specified, and the machine is a CP the VM will be created in one of the specified availability zones.
+	// +kube:validation:MinItems=1
+	// +kubebuilder:validation:items:Enum=ZONE_1;ZONE_2
+	//+optional
+	AvailabilityZones []AvailabilityZone `json:"availabilityZones,omitempty"`
 
 	// MemoryMB is the memory size for the VM in MB.
 	// Size must be specified in multiples of 256 MB with a minimum of 1024 MB
@@ -216,9 +226,9 @@ type Volume struct {
 	SizeGB int `json:"sizeGB,omitempty"`
 
 	// AvailabilityZone is the availability zone where the volume will be created.
-	//+kubebuilder:validation:Enum=AUTO;ZONE_1;ZONE_2;ZONE_3
-	//+kubebuilder:default=AUTO
-	//+optional
+	// +kubebuilder:validation:Enum=AUTO;ZONE_1;ZONE_2;ZONE_3
+	// +kubebuilder:default=AUTO
+	// +optional
 	AvailabilityZone AvailabilityZone `json:"availabilityZone,omitempty"`
 
 	// Image is the image to use for the VM.

api/v1alpha1/ionoscloudmachine_types_test.go

@@ -43,7 +43,7 @@ func defaultMachine() *IonosCloudMachine {
 			ProviderID:       ptr.To("ionos://ee090ff2-1eef-48ec-a246-a51a33aa4f3a"),
 			DatacenterID:     "ee090ff2-1eef-48ec-a246-a51a33aa4f3a",
 			NumCores:         1,
-			AvailabilityZone: AvailabilityZoneTwo,
+			AvailabilityZone: ptr.To(AvailabilityZoneTwo),
 			MemoryMB:         2048,
 			CPUFamily:        ptr.To("AMD_OPTERON"),
 			Disk: &Volume{
@@ -162,32 +162,26 @@ var _ = Describe("IonosCloudMachine Tests", func() {
 		})
 
 		Context("Availability zone", func() {
-			It("should default to AUTO", func() {
-				m := defaultMachine()
-				// because AvailabilityZone is a string, setting the value as "" is the same as not setting anything
-				m.Spec.AvailabilityZone = ""
-				Expect(k8sClient.Create(context.Background(), m)).To(Succeed())
-				Expect(m.Spec.AvailabilityZone).To(Equal(AvailabilityZoneAuto))
-			})
 			It("should fail if not part of the enum", func() {
 				m := defaultMachine()
-				m.Spec.AvailabilityZone = "this-should-not-work"
+				var unknown AvailabilityZone = "this-should-not-work"
+				m.Spec.AvailabilityZone = ptr.To(unknown)
 				Expect(k8sClient.Create(context.Background(), m)).ToNot(Succeed())
 			})
 			DescribeTable("should work for value",
 				func(zone AvailabilityZone) {
 					m := defaultMachine()
-					m.Spec.AvailabilityZone = zone
+					m.Spec.AvailabilityZone = &zone
 					Expect(k8sClient.Create(context.Background(), m)).To(Succeed())
-					Expect(m.Spec.AvailabilityZone).To(Equal(zone))
+					Expect(ptr.Deref(m.Spec.AvailabilityZone, "")).To(Equal(zone))
 				},
 				Entry("AUTO", AvailabilityZoneAuto),
 				Entry("ZONE_1", AvailabilityZoneOne),
 				Entry("ZONE_2", AvailabilityZoneTwo),
 			)
 			It("Should fail for ZONE_3", func() {
 				m := defaultMachine()
-				m.Spec.AvailabilityZone = AvailabilityZoneThree
+				m.Spec.AvailabilityZone = ptr.To(AvailabilityZoneThree)
 				Expect(k8sClient.Create(context.Background(), m)).ToNot(Succeed())
 			})
 		})

config/crd/bases/infrastructure.cluster.x-k8s.io_ionoscloudclusters.yaml

@@ -3,7 +3,7 @@ apiVersion: apiextensions.k8s.io/v1
 kind: CustomResourceDefinition
 metadata:
   annotations:
-    controller-gen.kubebuilder.io/version: v0.16.1
+    controller-gen.kubebuilder.io/version: v0.16.5
   name: ionoscloudclusters.infrastructure.cluster.x-k8s.io
 spec:
   group: infrastructure.cluster.x-k8s.io

config/crd/bases/infrastructure.cluster.x-k8s.io_ionoscloudclustertemplates.yaml

@@ -3,7 +3,7 @@ apiVersion: apiextensions.k8s.io/v1
 kind: CustomResourceDefinition
 metadata:
   annotations:
-    controller-gen.kubebuilder.io/version: v0.16.1
+    controller-gen.kubebuilder.io/version: v0.16.5
   name: ionoscloudclustertemplates.infrastructure.cluster.x-k8s.io
 spec:
   group: infrastructure.cluster.x-k8s.io

config/crd/bases/infrastructure.cluster.x-k8s.io_ionoscloudmachines.yaml

@@ -3,7 +3,7 @@ apiVersion: apiextensions.k8s.io/v1
 kind: CustomResourceDefinition
 metadata:
   annotations:
-    controller-gen.kubebuilder.io/version: v0.16.1
+    controller-gen.kubebuilder.io/version: v0.16.5
   name: ionoscloudmachines.infrastructure.cluster.x-k8s.io
 spec:
   group: infrastructure.cluster.x-k8s.io
@@ -149,14 +149,28 @@ spec:
                   type: object
                 type: array
               availabilityZone:
-                default: AUTO
-                description: AvailabilityZone is the availability zone in which the
-                  VM should be provisioned.
+                description: |-
+                  AvailabilityZone is the availability zone in which the VM should be provisioned.
+                  AvailabilityZone is mutually exclusive with AvailabilityZones.
+                  If specified, AvailabilityZone will be used to provision the VM.
                 enum:
                 - AUTO
                 - ZONE_1
                 - ZONE_2
                 type: string
+              availabilityZones:
+                description: |-
+                  AvailabilityZones is the list of availability zones where the VM should be provisioned.
+                  AvailabilityZones is mutually exclusive with AvailabilityZone.
+                  If specified, and the machine is a CP the VM will be created in one of the specified availability zones.
+                items:
+                  description: AvailabilityZone is the availability zone where different
+                    cloud resources are created in.
+                  enum:
+                  - ZONE_1
+                  - ZONE_2
+                  type: string
+                type: array
               cpuFamily:
                 description: |-
                   CPUFamily defines the CPU architecture, which will be used for this VM.

config/crd/bases/infrastructure.cluster.x-k8s.io_ionoscloudmachinetemplates.yaml

@@ -3,7 +3,7 @@ apiVersion: apiextensions.k8s.io/v1
 kind: CustomResourceDefinition
 metadata:
   annotations:
-    controller-gen.kubebuilder.io/version: v0.16.1
+    controller-gen.kubebuilder.io/version: v0.16.5
   name: ionoscloudmachinetemplates.infrastructure.cluster.x-k8s.io
 spec:
   group: infrastructure.cluster.x-k8s.io
@@ -169,14 +169,28 @@ spec:
                           type: object
                         type: array
                       availabilityZone:
-                        default: AUTO
-                        description: AvailabilityZone is the availability zone in
-                          which the VM should be provisioned.
+                        description: |-
+                          AvailabilityZone is the availability zone in which the VM should be provisioned.
+                          AvailabilityZone is mutually exclusive with AvailabilityZones.
+                          If specified, AvailabilityZone will be used to provision the VM.
                         enum:
                         - AUTO
                         - ZONE_1
                         - ZONE_2
                         type: string
+                      availabilityZones:
+                        description: |-
+                          AvailabilityZones is the list of availability zones where the VM should be provisioned.
+                          AvailabilityZones is mutually exclusive with AvailabilityZone.
+                          If specified, and the machine is a CP the VM will be created in one of the specified availability zones.
+                        items:
+                          description: AvailabilityZone is the availability zone where
+                            different cloud resources are created in.
+                          enum:
+                          - ZONE_1
+                          - ZONE_2
+                          type: string
+                        type: array
                       cpuFamily:
                         description: |-
                           CPUFamily defines the CPU architecture, which will be used for this VM.

internal/service/cloud/server.go

@@ -21,6 +21,7 @@ import (
 	"encoding/base64"
 	"errors"
 	"fmt"
+	"math/rand"
 	"net/http"
 	"path"
 	"strconv"
@@ -29,7 +30,10 @@ import (
 	sdk "github.com/ionos-cloud/sdk-go/v6"
 	corev1 "k8s.io/api/core/v1"
 	apierrors "k8s.io/apimachinery/pkg/api/errors"
+	clusterv1 "sigs.k8s.io/cluster-api/api/v1beta1"
+	"sigs.k8s.io/cluster-api/util"
 	"sigs.k8s.io/cluster-api/util/conditions"
+	"sigs.k8s.io/controller-runtime/pkg/client"
 
 	infrav1 "github.com/ionos-cloud/cluster-api-provider-ionoscloud/api/v1alpha1"
 	"github.com/ionos-cloud/cluster-api-provider-ionoscloud/internal/util/ptr"
@@ -334,6 +338,11 @@ func (s *Service) createServer(ctx context.Context, secret *corev1.Secret, ms *s
 		return fmt.Errorf("image lookup: %w", err)
 	}
 
+	err = reconcileAvailabilityZone(ctx, ms)
+	if err != nil {
+		return fmt.Errorf("failed to reconcile availability zone %w", err)
+	}
+
 	renderedData := s.renderUserData(ms, string(bootstrapData))
 	copySpec := ms.IonosMachine.Spec.DeepCopy()
 	entityParams := serverEntityParams{
@@ -482,3 +491,67 @@ func (*Service) serversURL(datacenterID string) string {
 func (*Service) volumeName(m *infrav1.IonosCloudMachine) string {
 	return "vol-" + m.Name
 }
+
+func reconcileAvailabilityZone(ctx context.Context, ms *scope.Machine) error {
+	// Always return early if the AvailabilityZone is already set.
+	if ms.IonosMachine.Spec.AvailabilityZone != nil {
+		return nil
+	}
+
+	// Set default availability zone, if none is set.
+	if ms.IonosMachine.Spec.AvailabilityZone == nil && ms.IonosMachine.Spec.AvailabilityZones == nil {
+		ms.IonosMachine.Spec.AvailabilityZone = ptr.To(infrav1.AvailabilityZoneAuto)
+		return nil
+	}
+
+	if ms.IonosMachine.Spec.AvailabilityZones == nil {
+		return errors.New("availability zones are not set")
+	}
+
+	// if control plane machine, we distribute the machines across the zones.
+	if util.IsControlPlaneMachine(ms.Machine) {
+		machines, err := ms.ListMachines(ctx, client.MatchingLabels{
+			clusterv1.ClusterNameLabel:         ms.ClusterScope.Cluster.GetName(),
+			clusterv1.MachineControlPlaneLabel: "",
+		})
+		if err != nil {
+			return fmt.Errorf("failed to list machines %w", err)
+		}
+
+		// Track zone usage
+		usedZones := make(map[infrav1.AvailabilityZone]int)
+		for _, machine := range machines {
+			if machine.Name == ms.IonosMachine.GetName() {
+				// Skip the current machine
+				continue
+			}
+			if machine.Spec.AvailabilityZone != nil {
+				usedZones[*machine.Spec.AvailabilityZone]++
+			}
+		}
+
+		// Find the next least used availability zone
+		var selectedZone infrav1.AvailabilityZone
+
+		if len(usedZones) == 0 {
+			// If no zones are currently used, start with the first zone
+			selectedZone = ms.IonosMachine.Spec.AvailabilityZones[0]
+		} else {
+			minUsage := int(^uint(0) >> 1) // max int value for finding minimum
+
+			for _, zone := range ms.IonosMachine.Spec.AvailabilityZones {
+				if usage, found := usedZones[zone]; !found || usage < minUsage {
+					selectedZone = zone
+					minUsage = usage
+				}
+			}
+		}
+
+		// Set the selected zone
+		ms.IonosMachine.Spec.AvailabilityZone = ptr.To(selectedZone)
+	} else {
+		// if worker machine, we randomly select a zone.
+		ms.IonosMachine.Spec.AvailabilityZone = ptr.To(ms.IonosMachine.Spec.AvailabilityZones[rand.Intn(len(ms.IonosMachine.Spec.AvailabilityZones))]) //nolint:gosec
+	}
+	return nil
+}

internal/service/cloud/server_test.go

@@ -17,6 +17,7 @@ limitations under the License.
 package cloud
 
 import (
+	"context"
 	"fmt"
 	"net/http"
 	"path"
@@ -28,6 +29,7 @@ import (
 	"github.com/stretchr/testify/suite"
 	corev1 "k8s.io/api/core/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	clusterv1 "sigs.k8s.io/cluster-api/api/v1beta1"
 
 	infrav1 "github.com/ionos-cloud/cluster-api-provider-ionoscloud/api/v1alpha1"
 	"github.com/ionos-cloud/cluster-api-provider-ionoscloud/internal/ionoscloud/clienttest"
@@ -446,6 +448,72 @@ func (s *serverSuite) TestGetServerWithoutProviderIDFoundInList() {
 	s.NotNil(server)
 }
 
+func (s *serverSuite) TestReconcileAvailabilityZoneDefault() {
+	// Given
+	s.machineScope.IonosMachine.Spec.AvailabilityZone = nil
+
+	err := reconcileAvailabilityZone(context.Background(), s.machineScope)
+	s.NoError(err)
+	s.Equal(infrav1.AvailabilityZoneAuto, ptr.Deref(s.machineScope.IonosMachine.Spec.AvailabilityZone, ""))
+}
+
+func (s *serverSuite) TestReconcileAvailabilityZoneDistribute() {
+	// First Machine
+	// Given
+	s.capiMachine.Labels[clusterv1.MachineControlPlaneLabel] = ""
+	s.infraMachine.Labels[clusterv1.MachineControlPlaneLabel] = ""
+	s.machineScope.IonosMachine.Spec.AvailabilityZone = nil
+	s.machineScope.IonosMachine.Spec.AvailabilityZones = []infrav1.AvailabilityZone{infrav1.AvailabilityZoneOne, infrav1.AvailabilityZoneTwo}
+
+	err := reconcileAvailabilityZone(context.Background(), s.machineScope)
+	s.NoError(err)
+	s.Equal(infrav1.AvailabilityZoneOne, ptr.Deref(s.machineScope.IonosMachine.Spec.AvailabilityZone, ""))
+
+	// Update the machine to distribute
+	err = s.k8sClient.Update(context.Background(), s.machineScope.IonosMachine)
+	s.NoError(err)
+
+	// Second Machine
+	machine := &infrav1.IonosCloudMachine{
+		ObjectMeta: metav1.ObjectMeta{
+			Namespace: metav1.NamespaceDefault,
+			Name:      "test-cp-2",
+			Labels: map[string]string{
+				clusterv1.ClusterNameLabel:         s.capiCluster.Name,
+				clusterv1.MachineControlPlaneLabel: "",
+			},
+		},
+		Spec: infrav1.IonosCloudMachineSpec{
+			ProviderID:        ptr.To("ionos://" + exampleServerID),
+			DatacenterID:      "ccf27092-34e8-499e-a2f5-2bdee9d34a12",
+			NumCores:          2,
+			AvailabilityZones: []infrav1.AvailabilityZone{infrav1.AvailabilityZoneOne, infrav1.AvailabilityZoneTwo},
+			MemoryMB:          4096,
+			CPUFamily:         ptr.To("AMD_OPTERON"),
+			Disk: &infrav1.Volume{
+				Name:             "test-cp-2-hdd",
+				DiskType:         infrav1.VolumeDiskTypeHDD,
+				SizeGB:           20,
+				AvailabilityZone: infrav1.AvailabilityZoneAuto,
+				Image: &infrav1.ImageSpec{
+					ID: "3e3e3e3e-3e3e-3e3e-3e3e-3e3e3e3e3e3e",
+				},
+			},
+			Type: infrav1.ServerTypeEnterprise,
+		},
+		Status: infrav1.IonosCloudMachineStatus{},
+	}
+
+	err = s.k8sClient.Create(context.Background(), machine)
+	s.NoError(err)
+
+	s.machineScope.IonosMachine = machine
+
+	err = reconcileAvailabilityZone(context.Background(), s.machineScope)
+	s.NoError(err)
+	s.Equal(infrav1.AvailabilityZoneTwo, ptr.Deref(s.machineScope.IonosMachine.Spec.AvailabilityZone, ""))
+}
+
 //nolint:unused
 func (*serverSuite) exampleServer() sdk.Server {
 	return sdk.Server{