✨ Add support for checking Machine conditions in MachineHealthCheck

api/core/v1beta1/conversion.go

@@ -73,6 +73,11 @@ func (src *Cluster) ConvertTo(dstRaw conversion.Hub) error {
 		return err
 	}
 
+	dst.Spec.Topology.ControlPlane.HealthCheck.Checks.UnhealthyMachineConditions = restored.Spec.Topology.ControlPlane.HealthCheck.Checks.UnhealthyMachineConditions
+	for i, md := range restored.Spec.Topology.Workers.MachineDeployments {
+		dst.Spec.Topology.Workers.MachineDeployments[i].HealthCheck.Checks.UnhealthyMachineConditions = md.HealthCheck.Checks.UnhealthyMachineConditions
+	}
+
 	// Recover intent for bool values converted to *bool.
 	clusterv1.Convert_bool_To_Pointer_bool(src.Spec.Paused, ok, restored.Spec.Paused, &dst.Spec.Paused)
 
@@ -145,6 +150,11 @@ func (src *ClusterClass) ConvertTo(dstRaw conversion.Hub) error {
 		return err
 	}
 
+	dst.Spec.ControlPlane.HealthCheck.Checks.UnhealthyMachineConditions = restored.Spec.ControlPlane.HealthCheck.Checks.UnhealthyMachineConditions
+	for i, md := range restored.Spec.Workers.MachineDeployments {
+		dst.Spec.Workers.MachineDeployments[i].HealthCheck.Checks.UnhealthyMachineConditions = md.HealthCheck.Checks.UnhealthyMachineConditions
+	}
+
 	// Recover intent for bool values converted to *bool.
 	for i, patch := range dst.Spec.Patches {
 		for j, definition := range patch.Definitions {
@@ -513,6 +523,8 @@ func (src *MachineHealthCheck) ConvertTo(dstRaw conversion.Hub) error {
 		return err
 	}
 
+	dst.Spec.Checks.UnhealthyMachineConditions = restored.Spec.Checks.UnhealthyMachineConditions
+
 	clusterv1.Convert_int32_To_Pointer_int32(src.Status.ExpectedMachines, ok, restored.Status.ExpectedMachines, &dst.Status.ExpectedMachines)
 	clusterv1.Convert_int32_To_Pointer_int32(src.Status.CurrentHealthy, ok, restored.Status.CurrentHealthy, &dst.Status.CurrentHealthy)
 	clusterv1.Convert_int32_To_Pointer_int32(src.Status.RemediationsAllowed, ok, restored.Status.RemediationsAllowed, &dst.Status.RemediationsAllowed)

api/core/v1beta2/cluster_types.go

@@ -725,6 +725,16 @@ type ControlPlaneTopologyHealthCheckChecks struct {
 	// +kubebuilder:validation:MinItems=1
 	// +kubebuilder:validation:MaxItems=100
 	UnhealthyNodeConditions []UnhealthyNodeCondition `json:"unhealthyNodeConditions,omitempty"`
+
+	// unhealthyMachineConditions contains a list of the machine conditions that determine
+	// whether a machine is considered unhealthy.  The conditions are combined in a
+	// logical OR, i.e. if any of the conditions is met, the machine is unhealthy.
+	//
+	// +optional
+	// +listType=atomic
+	// +kubebuilder:validation:MinItems=1
+	// +kubebuilder:validation:MaxItems=100
+	UnhealthyMachineConditions []UnhealthyMachineCondition `json:"unhealthyMachineConditions,omitempty"`
 }
 
 // ControlPlaneTopologyHealthCheckRemediation configures if and how remediations are triggered if a control plane Machine is unhealthy.
@@ -975,6 +985,16 @@ type MachineDeploymentTopologyHealthCheckChecks struct {
 	// +kubebuilder:validation:MinItems=1
 	// +kubebuilder:validation:MaxItems=100
 	UnhealthyNodeConditions []UnhealthyNodeCondition `json:"unhealthyNodeConditions,omitempty"`
+
+	// unhealthyMachineConditions contains a list of the machine conditions that determine
+	// whether a machine is considered unhealthy.  The conditions are combined in a
+	// logical OR, i.e. if any of the conditions is met, the machine is unhealthy.
+	//
+	// +optional
+	// +listType=atomic
+	// +kubebuilder:validation:MinItems=1
+	// +kubebuilder:validation:MaxItems=100
+	UnhealthyMachineConditions []UnhealthyMachineCondition `json:"unhealthyMachineConditions,omitempty"`
 }
 
 // MachineDeploymentTopologyHealthCheckRemediation configures if and how remediations are triggered if a MachineDeployment Machine is unhealthy.

api/core/v1beta2/clusterclass_types.go

@@ -281,6 +281,16 @@ type ControlPlaneClassHealthCheckChecks struct {
 	// +kubebuilder:validation:MinItems=1
 	// +kubebuilder:validation:MaxItems=100
 	UnhealthyNodeConditions []UnhealthyNodeCondition `json:"unhealthyNodeConditions,omitempty"`
+
+	// unhealthyMachineConditions contains a list of the machine conditions that determine
+	// whether a machine is considered unhealthy.  The conditions are combined in a
+	// logical OR, i.e. if any of the conditions is met, the machine is unhealthy.
+	//
+	// +optional
+	// +listType=atomic
+	// +kubebuilder:validation:MinItems=1
+	// +kubebuilder:validation:MaxItems=100
+	UnhealthyMachineConditions []UnhealthyMachineCondition `json:"unhealthyMachineConditions,omitempty"`
 }
 
 // ControlPlaneClassHealthCheckRemediation configures if and how remediations are triggered if a control plane Machine is unhealthy.
@@ -542,6 +552,16 @@ type MachineDeploymentClassHealthCheckChecks struct {
 	// +kubebuilder:validation:MinItems=1
 	// +kubebuilder:validation:MaxItems=100
 	UnhealthyNodeConditions []UnhealthyNodeCondition `json:"unhealthyNodeConditions,omitempty"`
+
+	// unhealthyMachineConditions contains a list of the machine conditions that determine
+	// whether a machine is considered unhealthy.  The conditions are combined in a
+	// logical OR, i.e. if any of the conditions is met, the machine is unhealthy.
+	//
+	// +optional
+	// +listType=atomic
+	// +kubebuilder:validation:MinItems=1
+	// +kubebuilder:validation:MaxItems=100
+	UnhealthyMachineConditions []UnhealthyMachineCondition `json:"unhealthyMachineConditions,omitempty"`
 }
 
 // MachineDeploymentClassHealthCheckRemediation configures if and how remediations are triggered if a MachineDeployment Machine is unhealthy.

api/core/v1beta2/machine_types.go

@@ -287,6 +287,10 @@ const (
 	// defined by a MachineHealthCheck object.
 	MachineHealthCheckUnhealthyNodeReason = "UnhealthyNode"
 
+	// MachineHealthCheckUnhealthyMachineReason surfaces when the machine does not pass the health checks
+	// defined by a MachineHealthCheck object.
+	MachineHealthCheckUnhealthyMachineReason = "UnhealthyMachine"
+
 	// MachineHealthCheckNodeStartupTimeoutReason surfaces when the node hosted on the machine does not appear within
 	// the timeout defined by a MachineHealthCheck object.
 	MachineHealthCheckNodeStartupTimeoutReason = "NodeStartupTimeout"

api/core/v1beta2/machinehealthcheck_types.go

@@ -111,6 +111,16 @@ type MachineHealthCheckChecks struct {
 	// +kubebuilder:validation:MinItems=1
 	// +kubebuilder:validation:MaxItems=100
 	UnhealthyNodeConditions []UnhealthyNodeCondition `json:"unhealthyNodeConditions,omitempty"`
+
+	// unhealthyMachineConditions contains a list of the machine conditions that determine
+	// whether a machine is considered unhealthy.  The conditions are combined in a
+	// logical OR, i.e. if any of the conditions is met, the machine is unhealthy.
+	//
+	// +optional
+	// +listType=atomic
+	// +kubebuilder:validation:MinItems=1
+	// +kubebuilder:validation:MaxItems=100
+	UnhealthyMachineConditions []UnhealthyMachineCondition `json:"unhealthyMachineConditions,omitempty"`
 }
 
 // MachineHealthCheckRemediation configures if and how remediations are triggered if a Machine is unhealthy.
@@ -227,7 +237,33 @@ type UnhealthyNodeCondition struct {
 
 	// timeoutSeconds is the duration that a node must be in a given status for,
 	// after which the node is considered unhealthy.
-	// For example, with a value of "1h", the node must match the status
+	// For example, with a value of "3600", the node must match the status
+	// for at least 1 hour before being considered unhealthy.
+	// +required
+	// +kubebuilder:validation:Minimum=0
+	TimeoutSeconds *int32 `json:"timeoutSeconds,omitempty"`
+}
+
+// UnhealthyMachineCondition represents a Machine condition type and value with a timeout
+// specified as a duration.  When the named condition has been in the given
+// status for at least the timeout value, a machine is considered unhealthy.
+type UnhealthyMachineCondition struct {
+	// type of Machine condition
+	// +kubebuilder:validation:Pattern=`^([a-z0-9]([-a-z0-9]*[a-z0-9])?(\.[a-z0-9]([-a-z0-9]*[a-z0-9])?)*/)?(([A-Za-z0-9][-A-Za-z0-9_.]*)?[A-Za-z0-9])$`
+	// +kubebuilder:validation:MinLength=1
+	// +kubebuilder:validation:MaxLength=316
+	// +kubebuilder:validation:XValidation:rule="self != 'Ready' && self != 'Available' && self != 'HealthCheckSucceeded' && self != 'OwnerRemediated' && self != 'ExternallyRemediated'",message="type must not be one of: Ready, Available, HealthCheckSucceeded, OwnerRemediated, ExternallyRemediated"
+	// +required
+	Type string `json:"type,omitempty"`
+
+	// status of the condition, one of True, False, Unknown.
+	// +required
+	// +kubebuilder:validation:Enum=True;False;Unknown
+	Status metav1.ConditionStatus `json:"status,omitempty"`
+
+	// timeoutSeconds is the duration that a machine must be in a given status for,
+	// after which the machine is considered unhealthy.
+	// For example, with a value of "3600", the machine must match the status
 	// for at least 1 hour before being considered unhealthy.
 	// +required
 	// +kubebuilder:validation:Minimum=0

api/core/v1beta2/v1beta1_condition_consts.go

@@ -157,6 +157,11 @@ const (
 
 	// UnhealthyNodeConditionV1Beta1Reason is the reason used when a machine's node has one of the MachineHealthCheck's unhealthy conditions.
 	UnhealthyNodeConditionV1Beta1Reason = "UnhealthyNode"
+
+	// UnhealthyMachineConditionV1Beta1Reason is the reason used when a machine has one of the MachineHealthCheck's unhealthy conditions.
+	// When both machine and node issues are detected, this reason takes precedence over node-related reasons
+	// (NodeNotFoundV1Beta1Reason, NodeStartupTimeoutV1Beta1Reason, UnhealthyNodeConditionV1Beta1Reason).
+	UnhealthyMachineConditionV1Beta1Reason = "UnhealthyMachine"
 )
 
 const (