Merge pull request #11290 from fabriziopandini/update-v1beta2-status-mhc-controller

✨ Update MHC with v1Beta2 status

Author: k8s-ci-robot

api/v1beta1/machine_types.go

@@ -200,22 +200,71 @@ const (
 	// Note: this could happen when creating the machine. However, this state should be treated as an error if it lasts indefinitely.
 	MachineNodeDoesNotExistV1Beta2Reason = ObjectDoesNotExistV1Beta2Reason
 
-	// MachineNodeDeletedV1Beta2Reason surfaces when  the node hosted on the machine has been deleted.
+	// MachineNodeDeletedV1Beta2Reason surfaces when the node hosted on the machine has been deleted.
 	// Note: controllers can't identify if the Node was deleted by the controller itself, e.g.
 	// during the deletion workflow, or by a users.
 	MachineNodeDeletedV1Beta2Reason = ObjectDeletedV1Beta2Reason
 )
 
-// Machine's HealthCheckSucceeded and OwnerRemediated conditions and corresponding reasons that will be used in v1Beta2 API version.
-// Note: HealthCheckSucceeded and OwnerRemediated condition are set by the MachineHealthCheck controller.
+// Machine's HealthCheckSucceeded condition and corresponding reasons that will be used in v1Beta2 API version.
+// Note: HealthCheckSucceeded condition is set by the MachineHealthCheck controller.
 const (
 	// MachineHealthCheckSucceededV1Beta2Condition is true if MHC instances targeting this machine report the Machine
 	// is healthy according to the definition of healthy present in the spec of the MachineHealthCheck object.
 	MachineHealthCheckSucceededV1Beta2Condition = "HealthCheckSucceeded"
 
+	// MachineHealthCheckSucceededV1Beta2Reason surfaces when a machine passes all the health checks defined by a MachineHealthCheck object.
+	MachineHealthCheckSucceededV1Beta2Reason = "HealthCheckSucceeded"
+
+	// MachineHealthCheckUnhealthyNodeV1Beta2Reason surfaces when the node hosted on the machine does not pass the health checks
+	// defined by a MachineHealthCheck object.
+	MachineHealthCheckUnhealthyNodeV1Beta2Reason = "UnhealthyNode"
+
+	// MachineHealthCheckNodeStartupTimeoutV1Beta2Reason surfaces when the node hosted on the machine does not appear within
+	// the timeout defined by a MachineHealthCheck object.
+	MachineHealthCheckNodeStartupTimeoutV1Beta2Reason = "NodeStartupTimeout"
+
+	// MachineHealthCheckNodeDeletedV1Beta2Reason surfaces when a MachineHealthCheck detect that the node hosted on the
+	// machine has been deleted while the Machine is still running.
+	MachineHealthCheckNodeDeletedV1Beta2Reason = "NodeDeleted"
+
+	// MachineHealthCheckHasRemediateAnnotationV1Beta2Reason surfaces a MachineHealthCheck detects a machine manually remediated
+	// via the remediate-machine annotation.
+	MachineHealthCheckHasRemediateAnnotationV1Beta2Reason = "HasRemediateAnnotation"
+)
+
+// Machine's OwnerRemediated conditions and corresponding reasons that will be used in v1Beta2 API version.
+// Note: OwnerRemediated condition is initially set by the MachineHealthCheck controller; then it is up to the Machine's
+// owner controller to update or delete this condition.
+const (
 	// MachineOwnerRemediatedV1Beta2Condition is only present if MHC instances targeting this machine
 	// determine that the controller owning this machine should perform remediation.
 	MachineOwnerRemediatedV1Beta2Condition = "OwnerRemediated"
+
+	// MachineOwnerRemediatedWaitingForRemediationV1Beta2Reason surfaces the machine is waiting for the owner controller
+	// to start remediation.
+	MachineOwnerRemediatedWaitingForRemediationV1Beta2Reason = "WaitingForRemediation"
+)
+
+// Machine's ExternallyRemediated conditions and corresponding reasons that will be used in v1Beta2 API version.
+// Note: ExternallyRemediated condition is initially set by the MachineHealthCheck controller; then it is up to the external
+// remediation controller to update or delete this condition.
+const (
+	// MachineExternallyRemediatedV1Beta2Condition is only present if MHC instances targeting this machine
+	// determine that an external controller should perform remediation.
+	MachineExternallyRemediatedV1Beta2Condition = "ExternallyRemediated"
+
+	// MachineExternallyRemediatedWaitingForRemediationV1Beta2Reason surfaces the machine is waiting for the
+	// external remediation controller to start remediation.
+	MachineExternallyRemediatedWaitingForRemediationV1Beta2Reason = "WaitingForRemediation"
+
+	// MachineExternallyRemediatedRemediationTemplateNotFoundV1Beta2Reason surfaces that the MachineHealthCheck cannot
+	// find the template for a external remediation request.
+	MachineExternallyRemediatedRemediationTemplateNotFoundV1Beta2Reason = "RemediationTemplateNotFound"
+
+	// MachineExternallyRemediatedRemediationRequestCreationFailedV1Beta2Reason surfaces that the MachineHealthCheck cannot
+	// create a request for the external remediation controller.
+	MachineExternallyRemediatedRemediationRequestCreationFailedV1Beta2Reason = "RemediationRequestCreationFailed"
 )
 
 // Machine's Deleting condition and corresponding reasons that will be used in v1Beta2 API version.

api/v1beta1/machinehealthcheck_types.go

@@ -24,6 +24,27 @@ import (
 	"k8s.io/apimachinery/pkg/util/intstr"
 )
 
+// MachineHealthCheck's RemediationAllowed condition and corresponding reasons that will be used in v1Beta2 API version.
+const (
+	// MachineHealthCheckRemediationAllowedV1Beta2Condition surfaces whether the MachineHealthCheck is
+	// allowed to remediate any Machines or whether it is blocked from remediating any further.
+	MachineHealthCheckRemediationAllowedV1Beta2Condition = "RemediationAllowed"
+
+	// MachineHealthCheckTooManyUnhealthyV1Beta2Reason is the reason used when too many Machines are unhealthy and
+	// the MachineHealthCheck is blocked from making any further remediation.
+	MachineHealthCheckTooManyUnhealthyV1Beta2Reason = "TooManyUnhealthy"
+
+	// MachineHealthCheckRemediationAllowedV1Beta2Reason is the reason used when the number of unhealthy machine
+	// is within the limits defined by the MachineHealthCheck, and thus remediation is allowed.
+	MachineHealthCheckRemediationAllowedV1Beta2Reason = "RemediationAllowed"
+)
+
+// MachineHealthCheck's Paused condition and corresponding reasons that will be used in v1Beta2 API version.
+const (
+	// MachineHealthCheckPausedV1Beta2Condition is true if this MachineHealthCheck or the Cluster it belongs to are paused.
+	MachineHealthCheckPausedV1Beta2Condition = PausedV1Beta2Condition
+)
+
 var (
 	// DefaultNodeStartupTimeout is the time allowed for a node to start up.
 	// Can be made longer as part of spec if required for particular provider.

api/v1beta1/v1beta2_condition_consts.go

@@ -225,16 +225,6 @@ const (
 	ClusterPausedV1Beta2Condition = PausedV1Beta2Condition
 )
 
-// Conditions that will be used for the MachineHealthCheck object in v1Beta2 API version.
-const (
-	// MachineHealthCheckRemediationAllowedV1Beta2Condition surfaces whether the MachineHealthCheck is
-	// allowed to remediate any Machines or whether it is blocked from remediating any further.
-	MachineHealthCheckRemediationAllowedV1Beta2Condition = "RemediationAllowed"
-
-	// MachineHealthCheckPausedV1Beta2Condition is true if this MachineHealthCheck or the Cluster it belongs to are paused.
-	MachineHealthCheckPausedV1Beta2Condition = PausedV1Beta2Condition
-)
-
 // Conditions that will be used for the ClusterClass object in v1Beta2 API version.
 const (
 	// ClusterClassVariablesReadyV1Beta2Condition is true if the ClusterClass variables, including both inline and external

internal/controllers/machinehealthcheck/machinehealthcheck_controller.go

@@ -50,6 +50,7 @@ import (
 	"sigs.k8s.io/cluster-api/util"
 	"sigs.k8s.io/cluster-api/util/annotations"
 	"sigs.k8s.io/cluster-api/util/conditions"
+	v1beta2conditions "sigs.k8s.io/cluster-api/util/conditions/v1beta2"
 	"sigs.k8s.io/cluster-api/util/patch"
 	"sigs.k8s.io/cluster-api/util/predicates"
 )
@@ -279,6 +280,13 @@ func (r *Reconciler) reconcile(ctx context.Context, logger logr.Logger, cluster
 			Message:  message,
 		})
 
+		v1beta2conditions.Set(m, metav1.Condition{
+			Type:    clusterv1.MachineHealthCheckRemediationAllowedV1Beta2Condition,
+			Status:  metav1.ConditionFalse,
+			Reason:  clusterv1.MachineHealthCheckTooManyUnhealthyV1Beta2Reason,
+			Message: message,
+		})
+
 		// If there are no unhealthy target, skip publishing the `RemediationRestricted` event to avoid misleading.
 		if len(unhealthy) != 0 {
 			r.recorder.Event(
@@ -321,6 +329,12 @@ func (r *Reconciler) reconcile(ctx context.Context, logger logr.Logger, cluster
 	m.Status.RemediationsAllowed = remediationCount
 	conditions.MarkTrue(m, clusterv1.RemediationAllowedCondition)
 
+	v1beta2conditions.Set(m, metav1.Condition{
+		Type:   clusterv1.MachineHealthCheckRemediationAllowedV1Beta2Condition,
+		Status: metav1.ConditionTrue,
+		Reason: clusterv1.MachineHealthCheckRemediationAllowedV1Beta2Reason,
+	})
+
 	errList := r.patchUnhealthyTargets(ctx, logger, unhealthy, cluster, m)
 	errList = append(errList, r.patchHealthyTargets(ctx, logger, healthy, m)...)
 
@@ -399,6 +413,13 @@ func (r *Reconciler) patchUnhealthyTargets(ctx context.Context, logger logr.Logg
 				from, err := external.Get(ctx, r.Client, m.Spec.RemediationTemplate, t.Machine.Namespace)
 				if err != nil {
 					conditions.MarkFalse(m, clusterv1.ExternalRemediationTemplateAvailableCondition, clusterv1.ExternalRemediationTemplateNotFoundReason, clusterv1.ConditionSeverityError, err.Error())
+
+					v1beta2conditions.Set(t.Machine, metav1.Condition{
+						Type:    clusterv1.MachineExternallyRemediatedV1Beta2Condition,
+						Status:  metav1.ConditionFalse,
+						Reason:  clusterv1.MachineExternallyRemediatedRemediationTemplateNotFoundV1Beta2Reason,
+						Message: fmt.Sprintf("error retrieving remediation template %s %s", m.Spec.RemediationTemplate.Kind, klog.KRef(t.Machine.Namespace, m.Spec.RemediationTemplate.Name)),
+					})
 					errList = append(errList, errors.Wrapf(err, "error retrieving remediation template %v %q for machine %q in namespace %q within cluster %q", m.Spec.RemediationTemplate.GroupVersionKind(), m.Spec.RemediationTemplate.Name, t.Machine.Name, t.Machine.Namespace, m.Spec.ClusterName))
 					return errList
 				}
@@ -428,16 +449,37 @@ func (r *Reconciler) patchUnhealthyTargets(ctx context.Context, logger logr.Logg
 				// Create the external clone.
 				if err := r.Client.Create(ctx, to); err != nil {
 					conditions.MarkFalse(m, clusterv1.ExternalRemediationRequestAvailableCondition, clusterv1.ExternalRemediationRequestCreationFailedReason, clusterv1.ConditionSeverityError, err.Error())
+
+					v1beta2conditions.Set(t.Machine, metav1.Condition{
+						Type:    clusterv1.MachineExternallyRemediatedV1Beta2Condition,
+						Status:  metav1.ConditionFalse,
+						Reason:  clusterv1.MachineExternallyRemediatedRemediationRequestCreationFailedV1Beta2Reason,
+						Message: "Please check controller logs for errors",
+					})
 					errList = append(errList, errors.Wrapf(err, "error creating remediation request for machine %q in namespace %q within cluster %q", t.Machine.Name, t.Machine.Namespace, t.Machine.Spec.ClusterName))
 					return errList
 				}
+
+				v1beta2conditions.Set(t.Machine, metav1.Condition{
+					Type:   clusterv1.MachineExternallyRemediatedV1Beta2Condition,
+					Status: metav1.ConditionFalse,
+					Reason: clusterv1.MachineExternallyRemediatedWaitingForRemediationV1Beta2Reason,
+				})
 			} else {
 				logger.Info("Target has failed health check, marking for remediation", "target", t.string(), "reason", condition.Reason, "message", condition.Message)
 				// NOTE: MHC is responsible for creating MachineOwnerRemediatedCondition if missing or to trigger another remediation if the previous one is completed;
 				// instead, if a remediation is in already progress, the remediation owner is responsible for completing the process and MHC should not overwrite the condition.
 				if !conditions.Has(t.Machine, clusterv1.MachineOwnerRemediatedCondition) || conditions.IsTrue(t.Machine, clusterv1.MachineOwnerRemediatedCondition) {
 					conditions.MarkFalse(t.Machine, clusterv1.MachineOwnerRemediatedCondition, clusterv1.WaitingForRemediationReason, clusterv1.ConditionSeverityWarning, "")
 				}
+
+				if ownerRemediatedCondition := v1beta2conditions.Get(t.Machine, clusterv1.MachineOwnerRemediatedV1Beta2Condition); ownerRemediatedCondition == nil || ownerRemediatedCondition.Status == metav1.ConditionTrue {
+					v1beta2conditions.Set(t.Machine, metav1.Condition{
+						Type:   clusterv1.MachineOwnerRemediatedV1Beta2Condition,
+						Status: metav1.ConditionFalse,
+						Reason: clusterv1.MachineOwnerRemediatedWaitingForRemediationV1Beta2Reason,
+					})
+				}
 			}
 		}