Add support for multiple instance types in AWSManagedMachinePool

Author: moko-poi

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

@@ -933,6 +933,16 @@ spec:
               instanceType:
                 description: InstanceType specifies the AWS instance type
                 type: string
+              instanceTypes:
+                description: |-
+                  InstanceTypes specifies a list of AWS instance types for the node group.
+                  When specified, this allows using multiple instance types which enhances
+                  the availability of Spot instances.
+                  At most one of InstanceType or InstanceTypes may be specified.
+                  See https://docs.aws.amazon.com/eks/latest/userguide/managed-node-groups.html#managed-node-group-capacity-types
+                items:
+                  type: string
+                type: array
               labels:
                 additionalProperties:
                   type: string

exp/api/v1beta1/conversion.go

@@ -149,6 +149,14 @@ func (src *AWSManagedMachinePool) ConvertTo(dstRaw conversion.Hub) error {
 	dst.Spec.RolePath = restored.Spec.RolePath
 	dst.Spec.RolePermissionsBoundary = restored.Spec.RolePermissionsBoundary
 
+	// Restore InstanceType and InstanceTypes from the hub version
+	// v1beta1 doesn't have InstanceTypes, but v1beta2 does
+	// We preserve the exact state through the annotation-based conversion data
+	// This includes cases where both fields are set (even though webhook would reject it)
+	// to ensure fuzzy conversion tests pass
+	dst.Spec.InstanceType = restored.Spec.InstanceType
+	dst.Spec.InstanceTypes = restored.Spec.InstanceTypes
+
 	return nil
 }
 
@@ -160,12 +168,26 @@ func (r *AWSManagedMachinePool) ConvertFrom(srcRaw conversion.Hub) error {
 		return err
 	}
 
+	// Preserve v1beta2 data through annotation
+	// This ensures fuzzy conversion tests pass
 	return utilconversion.MarshalData(src, r)
 }
 
 // Convert_v1beta2_AWSManagedMachinePoolSpec_To_v1beta1_AWSManagedMachinePoolSpec is a conversion function.
 func Convert_v1beta2_AWSManagedMachinePoolSpec_To_v1beta1_AWSManagedMachinePoolSpec(in *expinfrav1.AWSManagedMachinePoolSpec, out *AWSManagedMachinePoolSpec, s apiconversion.Scope) error {
-	return autoConvert_v1beta2_AWSManagedMachinePoolSpec_To_v1beta1_AWSManagedMachinePoolSpec(in, out, s)
+	if err := autoConvert_v1beta2_AWSManagedMachinePoolSpec_To_v1beta1_AWSManagedMachinePoolSpec(in, out, s); err != nil {
+		return err
+	}
+
+	// Convert v1beta2 InstanceTypes or InstanceType to v1beta1 InstanceType
+	// Prefer InstanceTypes if set (use first element), otherwise use InstanceType
+	if len(in.InstanceTypes) > 0 {
+		out.InstanceType = &in.InstanceTypes[0]
+	} else if in.InstanceType != nil {
+		out.InstanceType = in.InstanceType
+	}
+
+	return nil
 }
 
 func Convert_v1beta2_AWSMachinePoolStatus_To_v1beta1_AWSMachinePoolStatus(in *expinfrav1.AWSMachinePoolStatus, out *AWSMachinePoolStatus, s apiconversion.Scope) error {

exp/api/v1beta1/zz_generated.conversion.go

@@ -180,6 +180,14 @@ type AWSManagedMachinePoolSpec struct {
 	// +optional
 	InstanceType *string `json:"instanceType,omitempty"`
 
+	// InstanceTypes specifies a list of AWS instance types for the node group.
+	// When specified, this allows using multiple instance types which enhances
+	// the availability of Spot instances.
+	// At most one of InstanceType or InstanceTypes may be specified.
+	// See https://docs.aws.amazon.com/eks/latest/userguide/managed-node-groups.html#managed-node-group-capacity-types
+	// +optional
+	InstanceTypes []string `json:"instanceTypes,omitempty"`
+
 	// Scaling specifies scaling for the ASG behind this pool
 	// +optional
 	Scaling *ManagedMachinePoolScaling `json:"scaling,omitempty"`

exp/api/v1beta2/awsmanagedmachinepool_types.go

@@ -124,6 +124,20 @@ func (r *AWSManagedMachinePool) validateRemoteAccess() field.ErrorList {
 	return allErrs
 }
 
+func (r *AWSManagedMachinePool) validateInstanceTypes() field.ErrorList {
+	var allErrs field.ErrorList
+
+	// InstanceType and InstanceTypes are mutually exclusive
+	if r.Spec.InstanceType != nil && len(r.Spec.InstanceTypes) > 0 {
+		allErrs = append(allErrs, field.Invalid(
+			field.NewPath("spec", "InstanceTypes"),
+			r.Spec.InstanceTypes,
+			"cannot specify both instanceType and instanceTypes. Use instanceTypes for multiple instance types or instanceType for a single instance type"))
+	}
+
+	return allErrs
+}
+
 func (r *AWSManagedMachinePool) validateLaunchTemplate() field.ErrorList {
 	var allErrs field.ErrorList
 	if r.Spec.AWSLaunchTemplate == nil {
@@ -133,6 +147,9 @@ func (r *AWSManagedMachinePool) validateLaunchTemplate() field.ErrorList {
 	if r.Spec.InstanceType != nil {
 		allErrs = append(allErrs, field.Invalid(field.NewPath("spec", "InstanceType"), r.Spec.InstanceType, "InstanceType cannot be specified when LaunchTemplate is specified"))
 	}
+	if len(r.Spec.InstanceTypes) > 0 {
+		allErrs = append(allErrs, field.Invalid(field.NewPath("spec", "InstanceTypes"), r.Spec.InstanceTypes, "InstanceTypes cannot be specified when LaunchTemplate is specified"))
+	}
 	if r.Spec.DiskSize != nil {
 		allErrs = append(allErrs, field.Invalid(field.NewPath("spec", "DiskSize"), r.Spec.DiskSize, "DiskSize cannot be specified when LaunchTemplate is specified"))
 	}
@@ -171,6 +188,9 @@ func (*awsManagedMachinePoolWebhook) ValidateCreate(_ context.Context, obj runti
 	if errs := r.validateNodegroupUpdateConfig(); len(errs) > 0 {
 		allErrs = append(allErrs, errs...)
 	}
+	if errs := r.validateInstanceTypes(); len(errs) > 0 {
+		allErrs = append(allErrs, errs...)
+	}
 	if errs := r.validateLaunchTemplate(); len(errs) > 0 {
 		allErrs = append(allErrs, errs...)
 	}
@@ -216,6 +236,9 @@ func (*awsManagedMachinePoolWebhook) ValidateUpdate(_ context.Context, oldObj, n
 	if errs := r.validateNodegroupUpdateConfig(); len(errs) > 0 {
 		allErrs = append(allErrs, errs...)
 	}
+	if errs := r.validateInstanceTypes(); len(errs) > 0 {
+		allErrs = append(allErrs, errs...)
+	}
 	if errs := r.validateLaunchTemplate(); len(errs) > 0 {
 		allErrs = append(allErrs, errs...)
 	}
@@ -265,6 +288,8 @@ func (r *AWSManagedMachinePool) validateImmutable(old *AWSManagedMachinePool) fi
 	appendErrorIfMutated(old.Spec.SubnetIDs, r.Spec.SubnetIDs, "subnetIDs")
 	appendErrorIfSetAndMutated(old.Spec.RoleName, r.Spec.RoleName, "roleName")
 	appendErrorIfMutated(old.Spec.DiskSize, r.Spec.DiskSize, "diskSize")
+	appendErrorIfMutated(old.Spec.InstanceType, r.Spec.InstanceType, "instanceType")
+	appendErrorIfMutated(old.Spec.InstanceTypes, r.Spec.InstanceTypes, "instanceTypes")
 	appendErrorIfMutated(old.Spec.AMIType, r.Spec.AMIType, "amiType")
 	appendErrorIfMutated(old.Spec.RemoteAccess, r.Spec.RemoteAccess, "remoteAccess")
 	appendErrorIfSetAndMutated(old.Spec.CapacityType, r.Spec.CapacityType, "capacityType")

exp/api/v1beta2/awsmanagedmachinepool_webhook.go

@@ -225,7 +225,11 @@ func (s *NodegroupService) createNodegroup(ctx context.Context) (*ekstypes.Nodeg
 	if managedPool.DiskSize != nil {
 		input.DiskSize = managedPool.DiskSize
 	}
-	if managedPool.InstanceType != nil {
+	// Support both InstanceTypes (preferred for multiple types) and InstanceType (for single type)
+	// Webhook validation ensures they are mutually exclusive
+	if len(managedPool.InstanceTypes) > 0 {
+		input.InstanceTypes = managedPool.InstanceTypes
+	} else if managedPool.InstanceType != nil {
 		input.InstanceTypes = []string{aws.ToString(managedPool.InstanceType)}
 	}
 	if len(managedPool.Taints) > 0 {

exp/api/v1beta2/zz_generated.deepcopy.go

@@ -0,0 +1,39 @@
+apiVersion: infrastructure.cluster.x-k8s.io/v1beta2
+kind: AWSManagedMachinePool
+metadata:
+  name: test-pool-multiple-instance-types
+  namespace: default
+spec:
+  eksNodegroupName: test-nodegroup
+  # Use instanceTypes (plural) to specify multiple instance types
+  # This enhances availability when using Spot instances
+  instanceTypes:
+    - t3.medium
+    - t3a.medium
+    - t2.medium
+  capacityType: spot
+  scaling:
+    minSize: 1
+    maxSize: 10
+  updateConfig:
+    maxUnavailable: 1
+  tags:
+    usage: test
+---
+apiVersion: infrastructure.cluster.x-k8s.io/v1beta2
+kind: AWSManagedMachinePool
+metadata:
+  name: test-pool-single-instance-type
+  namespace: default
+spec:
+  eksNodegroupName: test-nodegroup-single
+  # Use instanceType (singular) for a single instance type (backward compatible)
+  instanceType: t3.medium
+  capacityType: on-demand
+  scaling:
+    minSize: 1
+    maxSize: 5
+  updateConfig:
+    maxUnavailable: 1
+  tags:
+    usage: test