✨ Add remediation strategy support for MachineSet

Signed-off-by: Vince Prignano <vince@prigna.com>

tests

Signed-off-by: Vince Prignano <vince@prigna.com>

api/v1beta1/machinedeployment_types.go

@@ -162,6 +162,11 @@ type MachineDeploymentStrategy struct {
 	// MachineDeploymentStrategyType = RollingUpdate.
 	// +optional
 	RollingUpdate *MachineRollingUpdateDeployment `json:"rollingUpdate,omitempty"`
+
+	// Remediation controls the strategy of remediating unhealthy machines
+	// and how remediating operations should occur during the lifecycle of the dependant MachineSets.
+	// +optional
+	Remediation *RemediationStrategy `json:"remediation,omitempty"`
 }
 
 // ANCHOR_END: MachineDeploymentStrategy
@@ -211,6 +216,31 @@ type MachineRollingUpdateDeployment struct {
 
 // ANCHOR_END: MachineRollingUpdateDeployment
 
+// ANCHOR: RemediationStrategy
+
+// RemediationStrategy allows to define how the MachineSet can control scaling operations.
+type RemediationStrategy struct {
+	// MaxInFlight determines how many in flight remediations should happen at the same time.
+	//
+	// Remediation only happens on the MachineSet with the most current revision, while
+	// older MachineSets (usually present during rollout operations) aren't allowed to remediate.
+	//
+	// Note: In general (independent of remediations), unhealthy machines are always
+	// prioritized during scale down operations over healthy ones.
+	//
+	// MaxInFlight can be set to a fixed number or a percentage.
+	// Example: when this is set to 20%, the MachineSet controller deletes at most 20% of
+	// the desired replicas.
+	//
+	// If not set, remediation is limited to all machines (bounded by replicas)
+	// under the active MachineSet's management.
+	//
+	// +optional
+	MaxInFlight *intstr.IntOrString `json:"maxInFlight,omitempty"`
+}
+
+// ANCHOR_END: RemediationStrategy
+
 // ANCHOR: MachineDeploymentStatus
 
 // MachineDeploymentStatus defines the observed state of MachineDeployment.

exp/topology/desiredstate/desired_state_test.go

@@ -1329,6 +1329,9 @@ func TestComputeMachineDeployment(t *testing.T) {
 	var clusterClassMinReadySeconds int32 = 20
 	clusterClassStrategy := clusterv1.MachineDeploymentStrategy{
 		Type: clusterv1.OnDeleteMachineDeploymentStrategyType,
+		Remediation: &clusterv1.RemediationStrategy{
+			MaxInFlight: ptr.To(intstr.FromInt32(5)),
+		},
 	}
 	md1 := builder.MachineDeploymentClass("linux-worker").
 		WithLabels(labels).
@@ -1390,6 +1393,9 @@ func TestComputeMachineDeployment(t *testing.T) {
 	var topologyMinReadySeconds int32 = 10
 	topologyStrategy := clusterv1.MachineDeploymentStrategy{
 		Type: clusterv1.RollingUpdateMachineDeploymentStrategyType,
+		Remediation: &clusterv1.RemediationStrategy{
+			MaxInFlight: ptr.To(intstr.FromInt32(5)),
+		},
 	}
 	mdTopology := clusterv1.MachineDeploymentTopology{
 		Metadata: clusterv1.ObjectMeta{

internal/apis/core/v1alpha3/conversion.go

@@ -187,14 +187,17 @@ func (src *MachineDeployment) ConvertTo(dstRaw conversion.Hub) error {
 		return err
 	}
 
-	if restored.Spec.Strategy != nil && restored.Spec.Strategy.RollingUpdate != nil {
+	if restored.Spec.Strategy != nil {
 		if dst.Spec.Strategy == nil {
 			dst.Spec.Strategy = &clusterv1.MachineDeploymentStrategy{}
 		}
-		if dst.Spec.Strategy.RollingUpdate == nil {
-			dst.Spec.Strategy.RollingUpdate = &clusterv1.MachineRollingUpdateDeployment{}
+		if restored.Spec.Strategy.RollingUpdate != nil {
+			if dst.Spec.Strategy.RollingUpdate == nil {
+				dst.Spec.Strategy.RollingUpdate = &clusterv1.MachineRollingUpdateDeployment{}
+			}
+			dst.Spec.Strategy.RollingUpdate.DeletePolicy = restored.Spec.Strategy.RollingUpdate.DeletePolicy
 		}
-		dst.Spec.Strategy.RollingUpdate.DeletePolicy = restored.Spec.Strategy.RollingUpdate.DeletePolicy
+		dst.Spec.Strategy.Remediation = restored.Spec.Strategy.Remediation
 	}
 
 	dst.Spec.Template.Spec.NodeDeletionTimeout = restored.Spec.Template.Spec.NodeDeletionTimeout
@@ -330,3 +333,11 @@ func Convert_v1alpha3_MachineStatus_To_v1beta1_MachineStatus(in *MachineStatus,
 	// Status.version has been removed in v1beta1, thus requiring custom conversion function. the information will be dropped.
 	return autoConvert_v1alpha3_MachineStatus_To_v1beta1_MachineStatus(in, out, s)
 }
+
+func Convert_v1beta1_MachineDeploymentStrategy_To_v1alpha3_MachineDeploymentStrategy(in *clusterv1.MachineDeploymentStrategy, out *MachineDeploymentStrategy, s apiconversion.Scope) error {
+	return autoConvert_v1beta1_MachineDeploymentStrategy_To_v1alpha3_MachineDeploymentStrategy(in, out, s)
+}
+
+func Convert_v1beta1_MachineSetSpec_To_v1alpha3_MachineSetSpec(in *clusterv1.MachineSetSpec, out *MachineSetSpec, s apiconversion.Scope) error {
+	return autoConvert_v1beta1_MachineSetSpec_To_v1alpha3_MachineSetSpec(in, out, s)
+}