feat: Add VMSizeProperties support to disable SMT/hyperthreading (rebase of #1419)

pkg/apis/crds/karpenter.azure.com_aksnodeclasses.yaml

@@ -768,6 +768,23 @@ spec:
                   rule: self.all(k, !k.contains('\\'))
                 - message: tags values must be less than 256 characters
                   rule: self.all(k, size(self[k]) <= 256)
+              vmSizeProperties:
+                description: |-
+                  vmSizeProperties configures VM size customization options such as vCPU-to-physical-core ratio.
+                  This allows disabling hyperthreading (SMT) by setting vCPUsPerCore to 1.
+                  For more details see https://learn.microsoft.com/en-us/azure/virtual-machines/vm-customization
+                properties:
+                  vCPUsPerCore:
+                    description: |-
+                      vCPUsPerCore specifies the vCPU to physical core ratio.
+                      Setting this to 1 disables hyperthreading (SMT).
+                      Setting this to 2 or omitting enables hyperthreading (default).
+                      Note: Can only be set at VM creation time, not on running VMs.
+                    format: int32
+                    maximum: 2
+                    minimum: 1
+                    type: integer
+                type: object
               vnetSubnetID:
                 description: |-
                   vnetSubnetID is the subnet used by nics provisioned with this nodeclass.
@@ -1668,6 +1685,23 @@ spec:
                   rule: self.all(k, !k.contains('\\'))
                 - message: tags values must be less than 256 characters
                   rule: self.all(k, size(self[k]) <= 256)
+              vmSizeProperties:
+                description: |-
+                  vmSizeProperties configures VM size customization options such as vCPU-to-physical-core ratio.
+                  This allows disabling hyperthreading (SMT) by setting vCPUsPerCore to 1.
+                  For more details see https://learn.microsoft.com/en-us/azure/virtual-machines/vm-customization
+                properties:
+                  vCPUsPerCore:
+                    description: |-
+                      vCPUsPerCore specifies the vCPU to physical core ratio.
+                      Setting this to 1 disables hyperthreading (SMT).
+                      Setting this to 2 or omitting enables hyperthreading (default).
+                      Note: Can only be set at VM creation time, not on running VMs.
+                    format: int32
+                    maximum: 2
+                    minimum: 1
+                    type: integer
+                type: object
               vnetSubnetID:
                 description: |-
                   vnetSubnetID is the subnet used by nics provisioned with this nodeclass.

pkg/apis/v1alpha2/aksnodeclass.go

@@ -116,6 +116,24 @@ type AKSNodeClassSpec struct {
 	// https://learn.microsoft.com/en-us/azure/aks/custom-node-configuration
 	// +optional
 	LinuxOSConfig *LinuxOSConfiguration `json:"linuxOSConfig,omitempty"`
+	// vmSizeProperties configures VM size customization options such as vCPU-to-physical-core ratio.
+	// This allows disabling hyperthreading (SMT) by setting vCPUsPerCore to 1.
+	// For more details see https://learn.microsoft.com/en-us/azure/virtual-machines/vm-customization
+	// +optional
+	VMSizeProperties *VMSizeProperties `json:"vmSizeProperties,omitempty"`
+}
+
+// VMSizeProperties specifies VM size customization options.
+// See: https://learn.microsoft.com/en-us/azure/virtual-machines/vm-customization
+type VMSizeProperties struct {
+	// vCPUsPerCore specifies the vCPU to physical core ratio.
+	// Setting this to 1 disables hyperthreading (SMT).
+	// Setting this to 2 or omitting enables hyperthreading (default).
+	// Note: Can only be set at VM creation time, not on running VMs.
+	// +kubebuilder:validation:Minimum=1
+	// +kubebuilder:validation:Maximum=2
+	// +optional
+	VCPUsPerCore *int32 `json:"vCPUsPerCore,omitempty"`
 }
 
 // TODO: Add link for the aka.ms/nap/aksnodeclass-enable-host-encryption docs

pkg/apis/v1alpha2/crd_validation_cel_test.go

@@ -1077,4 +1077,54 @@ var _ = Describe("CEL/Validation", func() {
 			Expect(env.Client.Create(ctx, nodeClass)).To(Succeed())
 		})
 	})
+
+	Context("VMSizeProperties", func() {
+		It("should allow vmSizeProperties with vCPUsPerCore set to 1 (SMT disabled)", func() {
+			nodeClass := &v1alpha2.AKSNodeClass{
+				ObjectMeta: metav1.ObjectMeta{Name: strings.ToLower(randomdata.SillyName())},
+				Spec: v1alpha2.AKSNodeClassSpec{
+					VMSizeProperties: &v1alpha2.VMSizeProperties{
+						VCPUsPerCore: lo.ToPtr(int32(1)),
+					},
+				},
+			}
+			Expect(env.Client.Create(ctx, nodeClass)).To(Succeed())
+		})
+
+		It("should allow vmSizeProperties with vCPUsPerCore set to 2 (SMT enabled)", func() {
+			nodeClass := &v1alpha2.AKSNodeClass{
+				ObjectMeta: metav1.ObjectMeta{Name: strings.ToLower(randomdata.SillyName())},
+				Spec: v1alpha2.AKSNodeClassSpec{
+					VMSizeProperties: &v1alpha2.VMSizeProperties{
+						VCPUsPerCore: lo.ToPtr(int32(2)),
+					},
+				},
+			}
+			Expect(env.Client.Create(ctx, nodeClass)).To(Succeed())
+		})
+
+		It("should reject vmSizeProperties with vCPUsPerCore set to 0", func() {
+			nodeClass := &v1alpha2.AKSNodeClass{
+				ObjectMeta: metav1.ObjectMeta{Name: strings.ToLower(randomdata.SillyName())},
+				Spec: v1alpha2.AKSNodeClassSpec{
+					VMSizeProperties: &v1alpha2.VMSizeProperties{
+						VCPUsPerCore: lo.ToPtr(int32(0)),
+					},
+				},
+			}
+			Expect(env.Client.Create(ctx, nodeClass)).ToNot(Succeed())
+		})
+
+		It("should reject vmSizeProperties with vCPUsPerCore set to 3", func() {
+			nodeClass := &v1alpha2.AKSNodeClass{
+				ObjectMeta: metav1.ObjectMeta{Name: strings.ToLower(randomdata.SillyName())},
+				Spec: v1alpha2.AKSNodeClassSpec{
+					VMSizeProperties: &v1alpha2.VMSizeProperties{
+						VCPUsPerCore: lo.ToPtr(int32(3)),
+					},
+				},
+			}
+			Expect(env.Client.Create(ctx, nodeClass)).ToNot(Succeed())
+		})
+	})
 })

pkg/apis/v1beta1/aksnodeclass.go

@@ -133,6 +133,24 @@ type AKSNodeClassSpec struct {
 	// https://learn.microsoft.com/en-us/azure/aks/custom-node-configuration
 	// +optional
 	LinuxOSConfig *LinuxOSConfiguration `json:"linuxOSConfig,omitempty"`
+	// vmSizeProperties configures VM size customization options such as vCPU-to-physical-core ratio.
+	// This allows disabling hyperthreading (SMT) by setting vCPUsPerCore to 1.
+	// For more details see https://learn.microsoft.com/en-us/azure/virtual-machines/vm-customization
+	// +optional
+	VMSizeProperties *VMSizeProperties `json:"vmSizeProperties,omitempty"`
+}
+
+// VMSizeProperties specifies VM size customization options.
+// See: https://learn.microsoft.com/en-us/azure/virtual-machines/vm-customization
+type VMSizeProperties struct {
+	// vCPUsPerCore specifies the vCPU to physical core ratio.
+	// Setting this to 1 disables hyperthreading (SMT).
+	// Setting this to 2 or omitting enables hyperthreading (default).
+	// Note: Can only be set at VM creation time, not on running VMs.
+	// +kubebuilder:validation:Minimum=1
+	// +kubebuilder:validation:Maximum=2
+	// +optional
+	VCPUsPerCore *int32 `json:"vCPUsPerCore,omitempty"`
 }
 
 // TODO: Add link for the aka.ms/nap/aksnodeclass-enable-host-encryption docs

pkg/apis/v1beta1/crd_validation_cel_test.go

@@ -1097,6 +1097,57 @@ var _ = Describe("CEL/Validation", func() {
 			Expect(env.Client.Create(ctx, nodeClass)).To(Succeed())
 		})
 	})
+
+	Context("VMSizeProperties", func() {
+		It("should allow vmSizeProperties with vCPUsPerCore set to 1 (SMT disabled)", func() {
+			nodeClass := &v1beta1.AKSNodeClass{
+				ObjectMeta: metav1.ObjectMeta{Name: strings.ToLower(randomdata.SillyName())},
+				Spec: v1beta1.AKSNodeClassSpec{
+					VMSizeProperties: &v1beta1.VMSizeProperties{
+						VCPUsPerCore: lo.ToPtr(int32(1)),
+					},
+				},
+			}
+			Expect(env.Client.Create(ctx, nodeClass)).To(Succeed())
+		})
+
+		It("should allow vmSizeProperties with vCPUsPerCore set to 2 (SMT enabled)", func() {
+			nodeClass := &v1beta1.AKSNodeClass{
+				ObjectMeta: metav1.ObjectMeta{Name: strings.ToLower(randomdata.SillyName())},
+				Spec: v1beta1.AKSNodeClassSpec{
+					VMSizeProperties: &v1beta1.VMSizeProperties{
+						VCPUsPerCore: lo.ToPtr(int32(2)),
+					},
+				},
+			}
+			Expect(env.Client.Create(ctx, nodeClass)).To(Succeed())
+		})
+
+		It("should reject vmSizeProperties with vCPUsPerCore set to 0", func() {
+			nodeClass := &v1beta1.AKSNodeClass{
+				ObjectMeta: metav1.ObjectMeta{Name: strings.ToLower(randomdata.SillyName())},
+				Spec: v1beta1.AKSNodeClassSpec{
+					VMSizeProperties: &v1beta1.VMSizeProperties{
+						VCPUsPerCore: lo.ToPtr(int32(0)),
+					},
+				},
+			}
+			Expect(env.Client.Create(ctx, nodeClass)).ToNot(Succeed())
+		})
+
+		It("should reject vmSizeProperties with vCPUsPerCore set to 3", func() {
+			nodeClass := &v1beta1.AKSNodeClass{
+				ObjectMeta: metav1.ObjectMeta{Name: strings.ToLower(randomdata.SillyName())},
+				Spec: v1beta1.AKSNodeClassSpec{
+					VMSizeProperties: &v1beta1.VMSizeProperties{
+						VCPUsPerCore: lo.ToPtr(int32(3)),
+					},
+				},
+			}
+			Expect(env.Client.Create(ctx, nodeClass)).ToNot(Succeed())
+		})
+	})
+
 	Context("LinuxOSConfig sysctl cross-field validation", func() {
 		It("should reject rmemDefault > rmemMax", func() {
 			nodeClass := &v1beta1.AKSNodeClass{

pkg/providers/instance/vminstance.go

@@ -565,7 +565,8 @@ func newVMObject(opts *createVMOptions) *armcompute.VirtualMachine {
 		Identity: ConvertToVirtualMachineIdentity(opts.NodeIdentities),
 		Properties: &armcompute.VirtualMachineProperties{
 			HardwareProfile: &armcompute.HardwareProfile{
-				VMSize: lo.ToPtr(armcompute.VirtualMachineSizeTypes(opts.InstanceType.Name)),
+				VMSize:           lo.ToPtr(armcompute.VirtualMachineSizeTypes(opts.InstanceType.Name)),
+				VMSizeProperties: convertVMSizeProperties(opts.NodeClass.Spec.VMSizeProperties, opts.ProvisionMode),
 			},
 
 			StorageProfile: &armcompute.StorageProfile{
@@ -678,6 +679,21 @@ func setVMPropertiesSecurityProfile(vmProperties *armcompute.VirtualMachinePrope
 	}
 }
 
+// convertVMSizeProperties converts AKSNodeClass VMSizeProperties to Azure SDK format.
+// Returns nil when vmSizeProperties are unspecified or when the provision mode does not use the VM API.
+func convertVMSizeProperties(props *v1beta1.VMSizeProperties, provisionMode string) *armcompute.VMSizeProperties {
+	if props == nil || isAKSMachineAPIMode(provisionMode) {
+		return nil
+	}
+	return &armcompute.VMSizeProperties{
+		VCPUsPerCore: props.VCPUsPerCore,
+	}
+}
+
+func isAKSMachineAPIMode(provisionMode string) bool {
+	return provisionMode == consts.ProvisionModeAKSMachineAPI || provisionMode == consts.ProvisionModeAKSMachineAPIHeaderBatch
+}
+
 type createResult struct {
 	Poller *runtime.Poller[armcompute.VirtualMachinesClientCreateOrUpdateResponse]
 	VM     *armcompute.VirtualMachine