descheduler: add a cache timeout argument for nodeAnomalyDetectors in…

… lowNodeLoad plugin. (koordinator-sh#1955)

Signed-off-by: bogo <bogo.ysh@gmail.com>
Co-authored-by: zengwang1 <zengwang1@xiaomi.com>

pkg/descheduler/apis/config/types_loadaware.go

@@ -80,6 +80,9 @@ type LowNodeLoadArgs struct {
 	// it is determined that the node is abnormal, and the Pods need to be migrated to reduce the load.
 	AnomalyCondition *LoadAnomalyCondition
 
+	// DetectorCacheTimeout indicates the cache expiration time of nodeAnomalyDetectors, the default is 5 minutes
+	DetectorCacheTimeout *metav1.Duration
+
 	// NodePools supports multiple different types of batch nodes to configure different strategies
 	NodePools []LowNodeLoadNodePool
 }

pkg/descheduler/apis/config/v1alpha2/defaults.go

@@ -45,6 +45,7 @@ const (
 	defaultMigrationEvictBurst         = 1
 	defaultSchedulerSupportReservation = "koord-scheduler"
 	defaultArbitrationInterval         = 500 * time.Millisecond
+	defaultDetectorCacheTimeout        = 5 * time.Minute
 )
 
 var (
@@ -263,6 +264,9 @@ func SetDefaults_LowNodeLoadArgs(obj *LowNodeLoadArgs) {
 	} else if obj.AnomalyCondition.ConsecutiveAbnormalities == 0 {
 		obj.AnomalyCondition.ConsecutiveAbnormalities = defaultLoadAnomalyCondition.ConsecutiveAbnormalities
 	}
+	if obj.DetectorCacheTimeout == nil {
+		obj.DetectorCacheTimeout = &metav1.Duration{Duration: defaultDetectorCacheTimeout}
+	}
 
 	if obj.NodeMetricExpirationSeconds == nil {
 		obj.NodeMetricExpirationSeconds = pointer.Int64(defaultNodeMetricExpirationSeconds)

pkg/descheduler/apis/config/v1alpha2/defaults_test.go

@@ -41,6 +41,23 @@ func TestSetDefaults_LowNodeLoadArgs(t *testing.T) {
 				NodeFit:                     pointer.Bool(false),
 				NodeMetricExpirationSeconds: pointer.Int64(defaultNodeMetricExpirationSeconds),
 				AnomalyCondition:            defaultLoadAnomalyCondition,
+				DetectorCacheTimeout:        &metav1.Duration{Duration: 5 * time.Minute},
+				ResourceWeights: map[corev1.ResourceName]int64{
+					corev1.ResourceCPU:    1,
+					corev1.ResourceMemory: 1,
+				},
+			},
+		},
+		{
+			name: "set detectorCacheTimeout",
+			args: &LowNodeLoadArgs{
+				DetectorCacheTimeout: &metav1.Duration{Duration: 10 * time.Minute},
+			},
+			expected: &LowNodeLoadArgs{
+				NodeFit:                     pointer.Bool(true),
+				NodeMetricExpirationSeconds: pointer.Int64(defaultNodeMetricExpirationSeconds),
+				AnomalyCondition:            defaultLoadAnomalyCondition,
+				DetectorCacheTimeout:        &metav1.Duration{Duration: 10 * time.Minute},
 				ResourceWeights: map[corev1.ResourceName]int64{
 					corev1.ResourceCPU:    1,
 					corev1.ResourceMemory: 1,
@@ -64,6 +81,7 @@ func TestSetDefaults_LowNodeLoadArgs(t *testing.T) {
 					ConsecutiveAbnormalities: defaultLoadAnomalyCondition.ConsecutiveAbnormalities,
 					ConsecutiveNormalities:   3,
 				},
+				DetectorCacheTimeout: &metav1.Duration{Duration: 5 * time.Minute},
 				ResourceWeights: map[corev1.ResourceName]int64{
 					corev1.ResourceCPU:    1,
 					corev1.ResourceMemory: 1,
@@ -87,6 +105,7 @@ func TestSetDefaults_LowNodeLoadArgs(t *testing.T) {
 				NodeFit:                     pointer.Bool(true),
 				NodeMetricExpirationSeconds: pointer.Int64(defaultNodeMetricExpirationSeconds),
 				AnomalyCondition:            defaultLoadAnomalyCondition,
+				DetectorCacheTimeout:        &metav1.Duration{Duration: 5 * time.Minute},
 				LowThresholds: ResourceThresholds{
 					corev1.ResourceCPU:    30,
 					corev1.ResourceMemory: 30,

pkg/descheduler/apis/config/v1alpha2/types_loadaware.go

@@ -79,6 +79,9 @@ type LowNodeLoadArgs struct {
 	// it is determined that the node is abnormal, and the Pods need to be migrated to reduce the load.
 	AnomalyCondition *LoadAnomalyCondition `json:"anomalyCondition,omitempty"`
 
+	// DetectorCacheTimeout indicates the cache expiration time of nodeAnomalyDetectors, the default is 5 minute
+	DetectorCacheTimeout *metav1.Duration `json:"detectorCacheTimeout,omitempty"`
+
 	// NodePools supports multiple different types of batch nodes to configure different strategies
 	NodePools []LowNodeLoadNodePool `json:"nodePools,omitempty"`
 }

pkg/descheduler/framework/plugins/loadaware/low_node_load.go

@@ -21,7 +21,6 @@ import (
 	"fmt"
 	"sort"
 	"strings"
-	"time"
 
 	gocache "github.com/patrickmn/go-cache"
 	corev1 "k8s.io/api/core/v1"
@@ -107,7 +106,7 @@ func NewLowNodeLoad(args runtime.Object, handle framework.Handle) (framework.Plu
 	koordSharedInformerFactory.Start(context.TODO().Done())
 	koordSharedInformerFactory.WaitForCacheSync(context.TODO().Done())
 
-	nodeAnomalyDetectors := gocache.New(5*time.Minute, 5*time.Minute)
+	nodeAnomalyDetectors := gocache.New(loadLoadUtilizationArgs.DetectorCacheTimeout.Duration, loadLoadUtilizationArgs.DetectorCacheTimeout.Duration)
 
 	return &LowNodeLoad{
 		handle:               handle,

pkg/descheduler/framework/plugins/loadaware/low_node_load_test.go

@@ -1030,7 +1030,8 @@ func TestLowNodeLoad(t *testing.T) {
 										},
 									},
 								},
-								EvictableNamespaces: tt.evictableNamespaces,
+								DetectorCacheTimeout: &metav1.Duration{Duration: 5 * time.Minute},
+								EvictableNamespaces:  tt.evictableNamespaces,
 							},
 						})
 					},

pkg/descheduler/framework/plugins/loadaware/utilization_util.go

@@ -360,7 +360,7 @@ func evictPods(
 			"nodePool", nodePoolName,
 		}
 		for k, v := range nodeInfo.usage {
-			keysAndValues = append(keysAndValues, k, v.String())
+			keysAndValues = append(keysAndValues, k.String(), v.String())
 		}
 		for resourceName, quantity := range totalAvailableUsages {
 			keysAndValues = append(keysAndValues, fmt.Sprintf("%s/totalAvailable", resourceName), quantity.String())