Merge pull request #1563 from huone1/optimize/selectClusters

[optimize]Consider available resources when selecting clusters

pkg/scheduler/core/generic_scheduler.go

@@ -141,8 +141,7 @@ func (g *genericScheduler) selectClusters(clustersScore framework.ClusterScoreLi
 	defer metrics.ScheduleStep(metrics.ScheduleStepSelect, time.Now())
 
 	groupClustersInfo := spreadconstraint.GroupClustersWithScore(clustersScore, placement, spec)
-
-	return spreadconstraint.SelectBestClusters(placement, groupClustersInfo)
+	return spreadconstraint.SelectBestClusters(placement, groupClustersInfo, spec.Replicas)
 }
 
 func (g *genericScheduler) assignReplicas(

pkg/scheduler/core/spreadconstraint/group_clusters.go

@@ -117,6 +117,7 @@ func (info *GroupClustersInfo) generateClustersInfo(clustersScore framework.Clus
 	clustersReplicas := calAvailableReplicas(clusters, rbSpec)
 	for i, clustersReplica := range clustersReplicas {
 		info.Clusters[i].AvailableReplicas = int64(clustersReplica.Replicas)
+		info.Clusters[i].AvailableReplicas += int64(getScheduledReplicas(rbSpec, clustersReplica.Name))
 	}
 
 	sortClusters(info.Clusters)
@@ -238,3 +239,19 @@ func sortClusters(infos []ClusterDetailInfo) {
 		return infos[i].Name < infos[j].Name
 	})
 }
+
+func getScheduledReplicas(rbSpec *workv1alpha2.ResourceBindingSpec, clusterName string) int32 {
+	if rbSpec == nil {
+		return 0
+	}
+
+	var replicas int32
+	for _, cluster := range rbSpec.Clusters {
+		if cluster.Name == clusterName {
+			replicas = cluster.Replicas
+			break
+		}
+	}
+
+	return replicas
+}

pkg/scheduler/core/spreadconstraint/select_clusters.go

@@ -3,53 +3,64 @@ package spreadconstraint
 import (
 	"fmt"
 
+	"k8s.io/klog/v2"
+
 	clusterv1alpha1 "github.com/karmada-io/karmada/pkg/apis/cluster/v1alpha1"
 	policyv1alpha1 "github.com/karmada-io/karmada/pkg/apis/policy/v1alpha1"
 )
 
 // SelectBestClusters selects the cluster set based the GroupClustersInfo and placement
-func SelectBestClusters(placement *policyv1alpha1.Placement, groupClustersInfo *GroupClustersInfo) ([]*clusterv1alpha1.Cluster, error) {
-	if len(placement.SpreadConstraints) != 0 {
-		return selectBestClustersBySpreadConstraints(placement.SpreadConstraints, groupClustersInfo)
+func SelectBestClusters(placement *policyv1alpha1.Placement, groupClustersInfo *GroupClustersInfo, needReplicas int32) ([]*clusterv1alpha1.Cluster, error) {
+	if len(placement.SpreadConstraints) == 0 || shouldIgnoreSpreadConstraint(placement) {
+		var clusters []*clusterv1alpha1.Cluster
+		for _, cluster := range groupClustersInfo.Clusters {
+			clusters = append(clusters, cluster.Cluster)
+		}
+		klog.V(4).Infof("select all clusters")
+		return clusters, nil
 	}
 
-	var clusters []*clusterv1alpha1.Cluster
-	for _, cluster := range groupClustersInfo.Clusters {
-		clusters = append(clusters, cluster.Cluster)
+	if shouldIgnoreAvailableResource(placement) {
+		needReplicas = InvalidReplicas
 	}
 
-	return clusters, nil
+	return selectBestClustersBySpreadConstraints(placement.SpreadConstraints, groupClustersInfo, needReplicas)
 }
 
 func selectBestClustersBySpreadConstraints(spreadConstraints []policyv1alpha1.SpreadConstraint,
-	groupClustersInfo *GroupClustersInfo) ([]*clusterv1alpha1.Cluster, error) {
+	groupClustersInfo *GroupClustersInfo, needReplicas int32) ([]*clusterv1alpha1.Cluster, error) {
 	if len(spreadConstraints) > 1 {
 		return nil, fmt.Errorf("just support single spread constraint")
 	}
 
 	spreadConstraint := spreadConstraints[0]
 	if spreadConstraint.SpreadByField == policyv1alpha1.SpreadByFieldCluster {
-		return selectBestClustersByCluster(spreadConstraint, groupClustersInfo)
+		return selectBestClustersByCluster(spreadConstraint, groupClustersInfo, needReplicas)
 	}
 
 	return nil, fmt.Errorf("just support cluster spread constraint")
 }
 
-func selectBestClustersByCluster(spreadConstraint policyv1alpha1.SpreadConstraint, groupClustersInfo *GroupClustersInfo) ([]*clusterv1alpha1.Cluster, error) {
-	totalClusterCnt := len(groupClustersInfo.Clusters)
-	if spreadConstraint.MinGroups > totalClusterCnt {
-		return nil, fmt.Errorf("the number of feasible clusters is less than spreadConstraint.MinGroups")
+func shouldIgnoreSpreadConstraint(placement *policyv1alpha1.Placement) bool {
+	strategy := placement.ReplicaScheduling
+
+	// If the replica division preference is 'static weighted', ignore the declaration specified by spread constraints.
+	if strategy != nil && strategy.ReplicaSchedulingType == policyv1alpha1.ReplicaSchedulingTypeDivided &&
+		strategy.ReplicaDivisionPreference == policyv1alpha1.ReplicaDivisionPreferenceWeighted &&
+		(strategy.WeightPreference != nil && len(strategy.WeightPreference.StaticWeightList) != 0 && strategy.WeightPreference.DynamicWeight == "") {
+		return true
 	}
 
-	needCnt := spreadConstraint.MaxGroups
-	if spreadConstraint.MaxGroups > totalClusterCnt {
-		needCnt = totalClusterCnt
-	}
-
-	var clusters []*clusterv1alpha1.Cluster
-	for i := 0; i < needCnt; i++ {
-		clusters = append(clusters, groupClustersInfo.Clusters[i].Cluster)
-	}
-
-	return clusters, nil
+	return false
+}
+
+func shouldIgnoreAvailableResource(placement *policyv1alpha1.Placement) bool {
+	strategy := placement.ReplicaScheduling
+
+	// If the replica division preference is 'Duplicated', ignore the information about cluster available resource.
+	if strategy == nil || strategy.ReplicaSchedulingType == policyv1alpha1.ReplicaSchedulingTypeDuplicated {
+		return true
+	}
+
+	return false
 }

pkg/scheduler/core/spreadconstraint/select_clusters_by_cluster.go

@@ -0,0 +1,97 @@
+package spreadconstraint
+
+import (
+	"fmt"
+
+	clusterv1alpha1 "github.com/karmada-io/karmada/pkg/apis/cluster/v1alpha1"
+	policyv1alpha1 "github.com/karmada-io/karmada/pkg/apis/policy/v1alpha1"
+)
+
+func selectBestClustersByCluster(spreadConstraint policyv1alpha1.SpreadConstraint, groupClustersInfo *GroupClustersInfo,
+	needReplicas int32) ([]*clusterv1alpha1.Cluster, error) {
+	totalClusterCnt := len(groupClustersInfo.Clusters)
+	if totalClusterCnt < spreadConstraint.MinGroups {
+		return nil, fmt.Errorf("the number of feasible clusters is less than spreadConstraint.MinGroups")
+	}
+
+	needCnt := spreadConstraint.MaxGroups
+	if totalClusterCnt < spreadConstraint.MaxGroups {
+		needCnt = totalClusterCnt
+	}
+
+	var clusterInfos []ClusterDetailInfo
+
+	if needReplicas == InvalidReplicas {
+		clusterInfos = groupClustersInfo.Clusters[:needCnt]
+	} else {
+		clusterInfos = selectClustersByAvailableResource(groupClustersInfo.Clusters, int32(needCnt), needReplicas)
+		if len(clusterInfos) == 0 {
+			return nil, fmt.Errorf("no enough resource when selecting %d clusters", needCnt)
+		}
+	}
+
+	var clusters []*clusterv1alpha1.Cluster
+	for i := range clusterInfos {
+		clusters = append(clusters, clusterInfos[i].Cluster)
+	}
+
+	return clusters, nil
+}
+
+// if needClusterCount = 2, needReplicas = 80, member1 and member3 will be selected finally.
+// because the total resource of member1 and member2 is less than needReplicas although their scores is highest
+// --------------------------------------------------
+// | clusterName      | member1 | member2 | member3 |
+// |-------------------------------------------------
+// | score            |   60    |    50   |    40   |
+// |------------------------------------------------|
+// |AvailableReplicas |   40    |    30   |    60   |
+// |------------------------------------------------|
+func selectClustersByAvailableResource(candidateClusters []ClusterDetailInfo, needClusterCount, needReplicas int32) []ClusterDetailInfo {
+	retClusters := candidateClusters[:needClusterCount]
+	restClusters := candidateClusters[needClusterCount:]
+
+	// the retClusters is sorted by cluster.Score descending. when the total AvailableReplicas of retClusters is less than needReplicas,
+	// use the cluster with the most AvailableReplicas in restClusters to instead the cluster with the lowest score,
+	// from the last cluster of the slice until checkAvailableResource returns true
+	var updateClusterID = len(retClusters) - 1
+	for !checkAvailableResource(retClusters, needReplicas) && updateClusterID >= 0 {
+		clusterID := GetClusterWithMaxAvailableResource(restClusters, retClusters[updateClusterID].AvailableReplicas)
+		if clusterID == InvalidClusterID {
+			updateClusterID--
+			continue
+		}
+
+		retClusters[updateClusterID], restClusters[clusterID] = restClusters[clusterID], retClusters[updateClusterID]
+		updateClusterID--
+	}
+
+	if updateClusterID < 0 {
+		return nil
+	}
+	return retClusters
+}
+
+func checkAvailableResource(clusters []ClusterDetailInfo, needReplicas int32) bool {
+	var total int64
+
+	for i := range clusters {
+		total += clusters[i].AvailableReplicas
+	}
+
+	return total >= int64(needReplicas)
+}
+
+// GetClusterWithMaxAvailableResource returns the cluster with maxAvailableReplicas
+func GetClusterWithMaxAvailableResource(candidateClusters []ClusterDetailInfo, originReplicas int64) int {
+	var maxAvailableReplicas = originReplicas
+	var clusterID = InvalidClusterID
+	for i := range candidateClusters {
+		if maxAvailableReplicas < candidateClusters[i].AvailableReplicas {
+			clusterID = i
+			maxAvailableReplicas = candidateClusters[i].AvailableReplicas
+		}
+	}
+
+	return clusterID
+}

pkg/scheduler/core/spreadconstraint/select_clusters_test.go

@@ -1,6 +1,7 @@
 package spreadconstraint
 
 import (
+	"fmt"
 	"reflect"
 	"testing"
 
@@ -27,41 +28,42 @@ func generateClusterInfo() []ClusterDetailInfo {
 		{
 			Name:              "member4",
 			Score:             60,
-			AvailableReplicas: 101,
+			AvailableReplicas: 50,
 			Cluster:           NewClusterWithTopology("member4", "P2", "R2", "Z2"),
 		},
 		{
 			Name:              "member2",
 			Score:             40,
-			AvailableReplicas: 101,
+			AvailableReplicas: 60,
 			Cluster:           NewClusterWithTopology("member2", "P1", "R1", "Z2"),
 		},
 		{
 			Name:              "member3",
 			Score:             30,
-			AvailableReplicas: 101,
+			AvailableReplicas: 80,
 			Cluster:           NewClusterWithTopology("member3", "P2", "R1", "Z1"),
 		},
 		{
 			Name:              "member1",
 			Score:             20,
-			AvailableReplicas: 101,
+			AvailableReplicas: 40,
 			Cluster:           NewClusterWithTopology("member1", "P1", "R1", "Z1"),
 		},
 	}
 }
 
 func TestSelectBestClusters(t *testing.T) {
-	clustetInfos := generateClusterInfo()
+	clusterInfos := generateClusterInfo()
 	type args struct {
 		placement         *policyv1alpha1.Placement
 		groupClustersInfo *GroupClustersInfo
+		needReplicas      int32
 	}
 	tests := []struct {
 		name    string
 		args    args
 		want    []*clusterv1alpha1.Cluster
-		wantErr bool
+		wantErr error
 	}{
 		{
 			name: "select clusters by cluster score",
@@ -76,20 +78,153 @@ func TestSelectBestClusters(t *testing.T) {
 					},
 				},
 				groupClustersInfo: &GroupClustersInfo{
-					Clusters: clustetInfos,
+					Clusters: clusterInfos,
 				},
+				needReplicas: 100,
 			},
 			want: []*clusterv1alpha1.Cluster{
-				clustetInfos[0].Cluster,
-				clustetInfos[1].Cluster,
+				clusterInfos[0].Cluster,
+				clusterInfos[1].Cluster,
 			},
 		},
+		{
+			name: "select clusters by cluster score and ignore available resources when scheduling strategy is duplicated",
+			args: args{
+				placement: &policyv1alpha1.Placement{
+					SpreadConstraints: []policyv1alpha1.SpreadConstraint{
+						{
+							SpreadByField: policyv1alpha1.SpreadByFieldCluster,
+							MaxGroups:     2,
+							MinGroups:     1,
+						},
+					},
+					ReplicaScheduling: &policyv1alpha1.ReplicaSchedulingStrategy{
+						ReplicaSchedulingType: policyv1alpha1.ReplicaSchedulingTypeDuplicated,
+					},
+				},
+				groupClustersInfo: &GroupClustersInfo{
+					Clusters: clusterInfos,
+				},
+				needReplicas: 120,
+			},
+			want: []*clusterv1alpha1.Cluster{
+				clusterInfos[0].Cluster,
+				clusterInfos[1].Cluster,
+			},
+		},
+		{
+			name: "select clusters by cluster score and ignore available resources when scheduling strategy is static weight",
+			args: args{
+				placement: &policyv1alpha1.Placement{
+					SpreadConstraints: []policyv1alpha1.SpreadConstraint{
+						{
+							SpreadByField: policyv1alpha1.SpreadByFieldCluster,
+							MaxGroups:     2,
+							MinGroups:     1,
+						},
+					},
+					ReplicaScheduling: &policyv1alpha1.ReplicaSchedulingStrategy{
+						ReplicaSchedulingType:     policyv1alpha1.ReplicaSchedulingTypeDivided,
+						ReplicaDivisionPreference: policyv1alpha1.ReplicaDivisionPreferenceWeighted,
+						WeightPreference: &policyv1alpha1.ClusterPreferences{
+							StaticWeightList: []policyv1alpha1.StaticClusterWeight{
+								{
+									TargetCluster: policyv1alpha1.ClusterAffinity{
+										ClusterNames: []string{"member1"},
+									},
+									Weight: 2,
+								},
+							},
+						},
+					},
+				},
+				groupClustersInfo: &GroupClustersInfo{
+					Clusters: clusterInfos,
+				},
+				needReplicas: 120,
+			},
+			want: []*clusterv1alpha1.Cluster{
+				clusterInfos[0].Cluster,
+				clusterInfos[1].Cluster,
+				clusterInfos[2].Cluster,
+				clusterInfos[3].Cluster,
+			},
+		},
+		{
+			name: "select clusters by cluster score and satisfy available resources",
+			args: args{
+				placement: &policyv1alpha1.Placement{
+					SpreadConstraints: []policyv1alpha1.SpreadConstraint{
+						{
+							SpreadByField: policyv1alpha1.SpreadByFieldCluster,
+							MaxGroups:     2,
+							MinGroups:     1,
+						},
+					},
+					ReplicaScheduling: &policyv1alpha1.ReplicaSchedulingStrategy{
+						ReplicaSchedulingType: policyv1alpha1.ReplicaSchedulingTypeDivided,
+					},
+				},
+				groupClustersInfo: &GroupClustersInfo{
+					Clusters: clusterInfos,
+				},
+				needReplicas: 120,
+			},
+			want: []*clusterv1alpha1.Cluster{
+				clusterInfos[0].Cluster,
+				clusterInfos[2].Cluster,
+			},
+		},
+		{
+			name: "select clusters by cluster score and insufficient resources",
+			args: args{
+				placement: &policyv1alpha1.Placement{
+					SpreadConstraints: []policyv1alpha1.SpreadConstraint{
+						{
+							SpreadByField: policyv1alpha1.SpreadByFieldCluster,
+							MaxGroups:     2,
+							MinGroups:     1,
+						},
+					},
+					ReplicaScheduling: &policyv1alpha1.ReplicaSchedulingStrategy{
+						ReplicaSchedulingType: policyv1alpha1.ReplicaSchedulingTypeDivided,
+					},
+				},
+				groupClustersInfo: &GroupClustersInfo{
+					Clusters: clusterInfos,
+				},
+				needReplicas: 200,
+			},
+			want:    nil,
+			wantErr: fmt.Errorf("no enough resource when selecting %d clusters", 2),
+		},
+		{
+			name: "select clusters by cluster score and exceeded the number of available clusters.",
+			args: args{
+				placement: &policyv1alpha1.Placement{
+					SpreadConstraints: []policyv1alpha1.SpreadConstraint{
+						{
+							SpreadByField: policyv1alpha1.SpreadByFieldCluster,
+							MaxGroups:     7,
+							MinGroups:     7,
+						},
+					},
+				},
+				groupClustersInfo: &GroupClustersInfo{
+					Clusters: clusterInfos,
+				},
+				needReplicas: 30,
+			},
+			want:    nil,
+			wantErr: fmt.Errorf("the number of feasible clusters is less than spreadConstraint.MinGroups"),
+		},
 	}
+
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			got, err := SelectBestClusters(tt.args.placement, tt.args.groupClustersInfo)
-			if (err != nil) != tt.wantErr {
-				t.Errorf("SelectBestClusters() error = %v, wantErr %v", err, tt.wantErr)
+			got, err := SelectBestClusters(tt.args.placement, tt.args.groupClustersInfo, tt.args.needReplicas)
+			if err != nil && err.Error() != tt.wantErr.Error() {
+				t.Errorf("SelectBestClusters() error = %v, wantErr = %v", err, tt.wantErr)
 				return
 			}
 			if !reflect.DeepEqual(got, tt.want) {

pkg/scheduler/core/spreadconstraint/util.go

@@ -14,6 +14,13 @@ import (
 	"github.com/karmada-io/karmada/pkg/util"
 )
 
+const (
+	// InvalidClusterID indicate a invalid cluster
+	InvalidClusterID = -1
+	// InvalidReplicas indicate that don't care about the available resource
+	InvalidReplicas = -1
+)
+
 func calAvailableReplicas(clusters []*clusterv1alpha1.Cluster, spec *workv1alpha2.ResourceBindingSpec) []workv1alpha2.TargetCluster {
 	availableClusters := make([]workv1alpha2.TargetCluster, len(clusters))