feat(k8sexporter/problemclient): allow deletion of deprecated condition types

cmd/options/options.go

@@ -88,6 +88,9 @@ type NodeProblemDetectorOptions struct {
 	MonitorConfigPaths types.ProblemDaemonConfigPathMap
 
 	// application options
+	// DeleteDeprecatedConditions is the flag determining whether to delete deprecated Conditions and Events from the node status.
+	DeleteDeprecatedConditions bool
+	DeprecatedConditionTypes   []string
 
 	// NodeName is the node name used to communicate with Kubernetes ApiServer.
 	NodeName string
@@ -131,6 +134,8 @@ func (npdo *NodeProblemDetectorOptions) AddFlags(fs *pflag.FlagSet) {
 		"127.0.0.1", "The address to bind the Prometheus scrape endpoint.")
 	fs.Float32Var(&npdo.QPS, "kube-api-qps", 500, "Maximum QPS to use while talking with Kubernetes API")
 	fs.IntVar(&npdo.Burst, "kube-api-burst", 500, "Maximum burst for throttle while talking with Kubernetes API")
+	fs.BoolVar(&npdo.DeleteDeprecatedConditions, "delete-deprecated-conditions", false, "Whether to delete deprecated Conditions and Events from the node status")
+	fs.StringSliceVar(&npdo.DeprecatedConditionTypes, "deprecated-condition-types", []string{}, "List of deprecated condition types to delete. This is ignored if --delete-deprecated-conditions is false.")
 	for _, exporterName := range exporters.GetExporterNames() {
 		exporterHandler := exporters.GetExporterHandlerOrDie(exporterName)
 		exporterHandler.Options.SetFlags(fs)

pkg/exporters/k8sexporter/k8s_exporter.go

@@ -63,6 +63,12 @@ func NewExporterOrDie(ctx context.Context, npdo *options.NodeProblemDetectorOpti
 	}
 
 	ke.startHTTPReporting(npdo)
+	if npdo.DeleteDeprecatedConditions {
+		err := ke.client.DeleteDeprecatedConditions(ctx, npdo.DeprecatedConditionTypes)
+		if err != nil {
+			klog.Errorf("Failed to delete deprecated conditions: %v", err)
+		}
+	}
 	ke.conditionManager.Start(ctx)
 
 	return &ke

pkg/exporters/k8sexporter/problemclient/fake_problem_client.go

@@ -60,6 +60,18 @@ func (f *FakeProblemClient) AssertConditions(expected []v1.NodeCondition) error
 	return nil
 }
 
+// DeleteDeprecatedConditions is a fake mimic of DeleteDeprecatedConditions, it only delete the internal condition cache.
+func (f *FakeProblemClient) DeleteDeprecatedConditions(ctx context.Context, types []string) error {
+	if err, ok := f.errors["DeleteDeprecatedConditions"]; ok {
+		return err
+	}
+	realTypes := generateConditionTypes(types)
+	for _, t := range realTypes {
+		delete(f.conditions, t)
+	}
+	return nil
+}
+
 // SetConditions is a fake mimic of SetConditions, it only update the internal condition cache.
 func (f *FakeProblemClient) SetConditions(ctx context.Context, conditions []v1.NodeCondition) error {
 	f.Lock()

pkg/exporters/k8sexporter/problemclient/problem_client.go

@@ -23,6 +23,7 @@ import (
 	"net/url"
 	"os"
 	"path/filepath"
+	"slices"
 
 	v1 "k8s.io/api/core/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
@@ -41,6 +42,8 @@ import (
 
 // Client is the interface of problem client
 type Client interface {
+	// DeleteDeprecatedConditions deletes the deprecated conditions of the node.
+	DeleteDeprecatedConditions(ctx context.Context, conditionTypeStrings []string) error
 	// GetConditions get all specific conditions of current node.
 	GetConditions(ctx context.Context, conditionTypes []v1.NodeConditionType) ([]*v1.NodeCondition, error)
 	// SetConditions set or update conditions of current node.
@@ -120,6 +123,44 @@ func (c *nodeProblemClient) SetConditions(ctx context.Context, newConditions []v
 	)
 }
 
+func (c *nodeProblemClient) DeleteDeprecatedConditions(ctx context.Context, conditionTypeStrings []string) error {
+	// get node object
+	klog.Infof("Deleting deprecated conditions %v (if present)...", conditionTypeStrings)
+	node, err := c.GetNode(ctx)
+	if err != nil {
+		return err
+	}
+
+	conditionTypes := generateConditionTypes(conditionTypeStrings)
+	// create a slice of the conditions we want to keep
+	newConditions := []v1.NodeCondition{}
+	for _, condition := range node.Status.Conditions {
+		if !slices.Contains(conditionTypes, condition.Type) {
+			newConditions = append(newConditions, condition)
+		} else {
+			klog.Infof("Deleting deprecated condition %s", condition.Type)
+		}
+	}
+
+	// update the node status if we need to,
+	// we only need to if the number of conditions has changed
+	if len(newConditions) < len(node.Status.Conditions) {
+		node.Status.Conditions = newConditions
+		return retry.OnError(retry.DefaultRetry,
+			func(error) bool {
+				return true
+			},
+			func() error {
+				_, err = c.client.Nodes().UpdateStatus(ctx, node, metav1.UpdateOptions{})
+				return err
+			},
+		)
+	} else {
+		klog.Infof("No deprecated conditions to delete")
+	}
+	return nil
+}
+
 func (c *nodeProblemClient) Eventf(eventType, source, reason, messageFmt string, args ...interface{}) {
 	recorder, found := c.recorders[source]
 	if !found {
@@ -163,3 +204,12 @@ func getNodeRef(namespace, nodeName string) *v1.ObjectReference {
 		Namespace: namespace,
 	}
 }
+
+func generateConditionTypes(conditionTypeStrings []string) []v1.NodeConditionType {
+	// convert the condition type strings to NodeConditionType
+	conditionTypes := []v1.NodeConditionType{}
+	for _, conditionTypeString := range conditionTypeStrings {
+		conditionTypes = append(conditionTypes, v1.NodeConditionType(conditionTypeString))
+	}
+	return conditionTypes
+}