feat: refactoring thresholds and usage assessment

this commit refactors the thresholds and usage assessment for the node
utilization plugins. both high and low plugins are affected by this
change.

Author: ricardomaraschini

pkg/api/types.go

@@ -17,10 +17,13 @@ limitations under the License.
 package api
 
 import (
+	"math"
+
 	v1 "k8s.io/api/core/v1"
 	"k8s.io/apimachinery/pkg/api/resource"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/runtime"
+	"k8s.io/utils/ptr"
 )
 
 // +k8s:deepcopy-gen:interfaces=k8s.io/apimachinery/pkg/runtime.Object
@@ -75,6 +78,25 @@ type (
 	ResourceThresholds map[v1.ResourceName]Percentage
 )
 
+// Negative returns a new ResourceThresholds with all its values negated.
+func (r ResourceThresholds) Negative() ResourceThresholds {
+	reversed := make(ResourceThresholds, len(r))
+	for k, v := range r {
+		reversed[k] = -v
+	}
+	return reversed
+}
+
+// Round returns a new ResourceThresholds with all its values rounded according
+// to math.Round.
+func (r ResourceThresholds) Round() ResourceThresholds {
+	rounded := make(ResourceThresholds, len(r))
+	for k, v := range r {
+		rounded[k] = Percentage(math.Round(float64(v)))
+	}
+	return rounded
+}
+
 type PriorityThreshold struct {
 	Value *int32 `json:"value"`
 	Name  string `json:"name"`
@@ -114,3 +136,31 @@ type MetricsCollector struct {
 
 // ReferencedResourceList is an adaption of v1.ResourceList with resources as references
 type ReferencedResourceList = map[v1.ResourceName]*resource.Quantity
+
+// ReferencedResourceListForNodesCapacity returns a ReferencedResourceList for
+// the capacity of a list of nodes. If allocatable resources are present, they
+// are used instead of capacity.
+func ReferencedResourceListForNodesCapacity(node []*v1.Node) map[string]ReferencedResourceList {
+	capacities := map[string]ReferencedResourceList{}
+	for _, n := range node {
+		capacities[n.Name] = ReferencedResourceListForNodeCapacity(n)
+	}
+	return capacities
+}
+
+// ReferencedResourceListForNodeCapacity returns a ReferencedResourceList for
+// the capacity of a node. If allocatable resources are present, they are used
+// instead of capacity.
+func ReferencedResourceListForNodeCapacity(node *v1.Node) ReferencedResourceList {
+	capacity := node.Status.Capacity
+	if len(node.Status.Allocatable) > 0 {
+		capacity = node.Status.Allocatable
+	}
+
+	referenced := ReferencedResourceList{}
+	for name, quantity := range capacity {
+		referenced[name] = ptr.To(quantity)
+	}
+
+	return referenced
+}

pkg/api/types_test.go

@@ -0,0 +1,20 @@
+package api
+
+import (
+	"reflect"
+	"testing"
+)
+
+func TestResourceThresholdsRound(t *testing.T) {
+	limit := ResourceThresholds{
+		"cpu": 100 / 3,
+		"mem": 100 / 3,
+	}
+	expected := ResourceThresholds{
+		"cpu": 33,
+		"mem": 33,
+	}
+	if !reflect.DeepEqual(limit, expected) {
+		t.Errorf("Expected %v, got %v", expected, limit)
+	}
+}

pkg/framework/plugins/nodeutilization/highnodeutilization.go

@@ -102,11 +102,18 @@ func (h *HighNodeUtilization) Balance(ctx context.Context, nodes []*v1.Node) *fr
 	}
 
 	nodesMap, nodesUsageMap, podListMap := getNodeUsageSnapshot(nodes, h.usageClient)
-	nodeThresholdsMap := getStaticNodeThresholds(nodes, h.args.Thresholds, h.targetThresholds)
-	nodesUsageAsNodeThresholdsMap := nodeUsageToResourceThresholds(nodesUsageMap, nodesMap)
+
+	usage, thresholds := AssessNodesUsagesAndThresholds(
+		nodesUsageMap,
+		api.ReferencedResourceListForNodesCapacity(nodes),
+		h.args.Thresholds,
+		h.targetThresholds,
+		false,
+	)
+
 	nodeGroups := classifyNodeUsage(
-		nodesUsageAsNodeThresholdsMap,
-		nodeThresholdsMap,
+		usage,
+		thresholds,
 		[]classifierFnc{
 			// underutilized nodes
 			func(nodeName string, usage, threshold api.ResourceThresholds) bool {
@@ -128,16 +135,21 @@ func (h *HighNodeUtilization) Balance(ctx context.Context, nodes []*v1.Node) *fr
 	category := []string{"underutilized", "overutilized"}
 	for i := range nodeGroups {
 		for nodeName := range nodeGroups[i] {
-			klog.InfoS("Node is "+category[i], "node", klog.KObj(nodesMap[nodeName]), "usage", nodesUsageMap[nodeName], "usagePercentage", resourceUsagePercentages(nodesUsageMap[nodeName], nodesMap[nodeName], true))
+			klog.InfoS(
+				fmt.Sprintf("Node is %s", category[i]),
+				"node", klog.KObj(nodesMap[nodeName]),
+				"usage", nodesUsageMap[nodeName],
+				"usagePercentage", usage[nodeName].Round(),
+			)
 			nodeInfos[i] = append(nodeInfos[i], NodeInfo{
 				NodeUsage: NodeUsage{
 					node:    nodesMap[nodeName],
 					usage:   nodesUsageMap[nodeName], // get back the original node usage
 					allPods: podListMap[nodeName],
 				},
 				thresholds: NodeThresholds{
-					lowResourceThreshold:  resourceThresholdsToNodeUsage(nodeThresholdsMap[nodeName][0], nodesMap[nodeName]),
-					highResourceThreshold: resourceThresholdsToNodeUsage(nodeThresholdsMap[nodeName][1], nodesMap[nodeName]),
+					lowResourceThreshold:  resourceThresholdsToNodeUsage(thresholds[nodeName][0], nodesMap[nodeName]),
+					highResourceThreshold: resourceThresholdsToNodeUsage(thresholds[nodeName][1], nodesMap[nodeName]),
 				},
 			})
 		}

pkg/framework/plugins/nodeutilization/lownodeutilization.go

@@ -123,17 +123,18 @@ func (l *LowNodeUtilization) Balance(ctx context.Context, nodes []*v1.Node) *fra
 	}
 
 	nodesMap, nodesUsageMap, podListMap := getNodeUsageSnapshot(nodes, l.usageClient)
-	var nodeThresholdsMap map[string][]api.ResourceThresholds
-	if l.args.UseDeviationThresholds {
-		nodeThresholdsMap = getNodeThresholdsFromAverageNodeUsage(nodes, l.usageClient, l.args.Thresholds, l.args.TargetThresholds)
-	} else {
-		nodeThresholdsMap = getStaticNodeThresholds(nodes, l.args.Thresholds, l.args.TargetThresholds)
-	}
-	nodesUsageAsNodeThresholdsMap := nodeUsageToResourceThresholds(nodesUsageMap, nodesMap)
+
+	usage, thresholds := AssessNodesUsagesAndThresholds(
+		nodesUsageMap,
+		api.ReferencedResourceListForNodesCapacity(nodes),
+		l.args.Thresholds,
+		l.args.TargetThresholds,
+		l.args.UseDeviationThresholds,
+	)
 
 	nodeGroups := classifyNodeUsage(
-		nodesUsageAsNodeThresholdsMap,
-		nodeThresholdsMap,
+		usage,
+		thresholds,
 		[]classifierFnc{
 			// underutilization
 			func(nodeName string, usage, threshold api.ResourceThresholds) bool {
@@ -156,7 +157,12 @@ func (l *LowNodeUtilization) Balance(ctx context.Context, nodes []*v1.Node) *fra
 	listedNodes := map[string]struct{}{}
 	for i := range nodeGroups {
 		for nodeName := range nodeGroups[i] {
-			klog.InfoS("Node is "+category[i], "node", klog.KObj(nodesMap[nodeName]), "usage", nodesUsageMap[nodeName], "usagePercentage", resourceUsagePercentages(nodesUsageMap[nodeName], nodesMap[nodeName], true))
+			klog.InfoS(
+				fmt.Sprintf("Node is %x", category[i]),
+				"node", klog.KObj(nodesMap[nodeName]),
+				"usage", nodesUsageMap[nodeName],
+				"usagePercentage", usage[nodeName].Round(),
+			)
 			listedNodes[nodeName] = struct{}{}
 			nodeInfos[i] = append(nodeInfos[i], NodeInfo{
 				NodeUsage: NodeUsage{
@@ -165,15 +171,20 @@ func (l *LowNodeUtilization) Balance(ctx context.Context, nodes []*v1.Node) *fra
 					allPods: podListMap[nodeName],
 				},
 				thresholds: NodeThresholds{
-					lowResourceThreshold:  resourceThresholdsToNodeUsage(nodeThresholdsMap[nodeName][0], nodesMap[nodeName]),
-					highResourceThreshold: resourceThresholdsToNodeUsage(nodeThresholdsMap[nodeName][1], nodesMap[nodeName]),
+					lowResourceThreshold:  resourceThresholdsToNodeUsage(thresholds[nodeName][0], nodesMap[nodeName]),
+					highResourceThreshold: resourceThresholdsToNodeUsage(thresholds[nodeName][1], nodesMap[nodeName]),
 				},
 			})
 		}
 	}
 	for nodeName := range nodesMap {
 		if _, ok := listedNodes[nodeName]; !ok {
-			klog.InfoS("Node is appropriately utilized", "node", klog.KObj(nodesMap[nodeName]), "usage", nodesUsageMap[nodeName], "usagePercentage", resourceUsagePercentages(nodesUsageMap[nodeName], nodesMap[nodeName], true))
+			klog.InfoS(
+				"Node is appropriately utilized",
+				"node", klog.KObj(nodesMap[nodeName]),
+				"usage", nodesUsageMap[nodeName],
+				"usagePercentage", usage[nodeName].Round(),
+			)
 		}
 	}