koord-descheduler: Fix threshold initialization for prod. (#2130)

Signed-off-by: zwForrest <[email protected]>
koordinator-sh · Jul 16, 2024 · db93c61 · db93c61
1 parent 4c38309
commit db93c61
Show file tree

Hide file tree

Showing 2 changed files with 77 additions and 8 deletions.
diff --git a/pkg/descheduler/framework/plugins/loadaware/low_node_load_test.go b/pkg/descheduler/framework/plugins/loadaware/low_node_load_test.go
@@ -1238,6 +1238,68 @@ func TestLowNodeLoad(t *testing.T) {
 			expectedPodsEvicted: 5,
 			evictedPods:         []string{},
 		},
+		{
+			name:                   "node is always appropriately for node usage, prod pod rebalance successful",
+			useDeviationThresholds: true,
+			thresholds: ResourceThresholds{
+				corev1.ResourceCPU:  0,
+				corev1.ResourcePods: 0,
+			},
+			targetThresholds: ResourceThresholds{
+				corev1.ResourceCPU:  100,
+				corev1.ResourcePods: 100,
+			},
+			prodLowThresholds: ResourceThresholds{
+				corev1.ResourceCPU:  10,
+				corev1.ResourcePods: 10,
+			},
+			prodHighThresholds: ResourceThresholds{
+				corev1.ResourceCPU:  10,
+				corev1.ResourcePods: 10,
+			},
+			nodes: []*corev1.Node{
+				test.BuildTestNode(n1NodeName, 4000, 3000, 20, nil),
+				test.BuildTestNode(n2NodeName, 4000, 3000, 20, nil),
+				test.BuildTestNode(n3NodeName, 4000, 3000, 20, nil),
+			},
+			pods: []*corev1.Pod{
+				test.BuildTestPod("p1", 400, 0, n1NodeName, test.SetRSOwnerRef),
+				test.BuildTestPod("p2", 400, 0, n1NodeName, test.SetRSOwnerRef),
+				test.BuildTestPod("p3", 400, 0, n1NodeName, test.SetRSOwnerRef),
+				test.BuildTestPod("p4", 400, 0, n1NodeName, test.SetRSOwnerRef),
+				test.BuildTestPod("p5", 400, 0, n1NodeName, test.SetRSOwnerRef),
+				// These won't be evicted.
+				test.BuildTestPod("p6", 400, 0, n1NodeName, test.SetDSOwnerRef),
+				test.BuildTestPod("p7", 400, 0, n1NodeName, func(pod *corev1.Pod) {
+					// A pod with local storage.
+					test.SetNormalOwnerRef(pod)
+					pod.Spec.Volumes = []corev1.Volume{
+						{
+							Name: "sample",
+							VolumeSource: corev1.VolumeSource{
+								HostPath: &corev1.HostPathVolumeSource{Path: "somePath"},
+								EmptyDir: &corev1.EmptyDirVolumeSource{
+									SizeLimit: resource.NewQuantity(int64(10), resource.BinarySI),
+								},
+							},
+						},
+					}
+					// A Mirror Pod.
+					pod.Annotations = test.GetMirrorPodAnnotation()
+				}),
+				test.BuildTestPod("p8", 400, 0, n1NodeName, func(pod *corev1.Pod) {
+					// A Critical Pod.
+					pod.Namespace = "kube-system"
+					priority := utils.SystemCriticalPriority
+					pod.Spec.Priority = &priority
+				}),
+				test.BuildTestPod("p9", 400, 0, n2NodeName, test.SetRSOwnerRef),
+				test.BuildTestPod("p10", 400, 0, n3NodeName, test.SetRSOwnerRef),
+				test.BuildTestPod("p11", 400, 0, n3NodeName, test.SetRSOwnerRef),
+			},
+			expectedPodsEvicted: 4,
+			evictedPods:         []string{},
+		},
 	}
 
 	for _, tt := range testCases {

diff --git a/pkg/descheduler/framework/plugins/loadaware/utilization_util.go b/pkg/descheduler/framework/plugins/loadaware/utilization_util.go
@@ -90,6 +90,7 @@ func getNodeThresholds(
 	var averageResourceUsagePercent, prodAverageResourceUsagePercent ResourceThresholds
 	if useDeviationThresholds {
 		averageResourceUsagePercent, prodAverageResourceUsagePercent = calcAverageResourceUsagePercent(nodeUsages)
+		klog.V(4).InfoS("useDeviationThresholds", "node", averageResourceUsagePercent, "prod", prodAverageResourceUsagePercent)
 	}
 
 	nodeThresholdsMap := map[string]NodeThresholds{}
@@ -107,11 +108,14 @@ func getNodeThresholds(
 				if lowThreshold[resourceName] == MinResourcePercentage {
 					thresholds.lowResourceThreshold[resourceName] = &resourceCapacity
 					thresholds.highResourceThreshold[resourceName] = &resourceCapacity
-					thresholds.prodLowResourceThreshold[resourceName] = &resourceCapacity
-					thresholds.prodHighResourceThreshold[resourceName] = &resourceCapacity
 				} else {
 					thresholds.lowResourceThreshold[resourceName] = resourceThreshold(allocatable, resourceName, normalizePercentage(averageResourceUsagePercent[resourceName]-lowThreshold[resourceName]))
 					thresholds.highResourceThreshold[resourceName] = resourceThreshold(allocatable, resourceName, normalizePercentage(averageResourceUsagePercent[resourceName]+highThreshold[resourceName]))
+				}
+				if prodLowThreshold[resourceName] == MinResourcePercentage {
+					thresholds.prodLowResourceThreshold[resourceName] = &resourceCapacity
+					thresholds.prodHighResourceThreshold[resourceName] = &resourceCapacity
+				} else {
 					thresholds.prodLowResourceThreshold[resourceName] = resourceThreshold(allocatable, resourceName, normalizePercentage(prodAverageResourceUsagePercent[resourceName]-prodLowThreshold[resourceName]))
 					thresholds.prodHighResourceThreshold[resourceName] = resourceThreshold(allocatable, resourceName, normalizePercentage(prodAverageResourceUsagePercent[resourceName]+prodHighThreshold[resourceName]))
 				}
@@ -253,13 +257,15 @@ func classifyNodes(
 				lowNodes = append(lowNodes, nodeInfo)
 				nodeUsageExplain = "lower than node usage and it's appropriately for prod usage"
 			}
-			klog.InfoS("Node's utilization", "node", klog.KObj(nodeUsage.node), "result information", nodeUsageExplain, "node usage", nodeUsage.usage, "node usagePercentage", resourceUsagePercentages(nodeUsage, false),
-				"prod usage", nodeUsage.prodUsage, "prod usagePercentage", resourceUsagePercentages(nodeUsage, true))
+			klog.V(4).InfoS("Node's utilization", "node", klog.KObj(nodeUsage.node), "result information", nodeUsageExplain, "node usage", nodeUsage.usage, "node usagePercentage", resourceUsagePercentages(nodeUsage, false),
+				"node high threshold", nodeThresholds[nodeUsage.node.Name].highResourceThreshold, "node low threshold", nodeThresholds[nodeUsage.node.Name].lowResourceThreshold, "prod usage", nodeUsage.prodUsage,
+				"prod usagePercentage", resourceUsagePercentages(nodeUsage, true), "prod high threshold", nodeThresholds[nodeUsage.node.Name].prodHighResourceThreshold, "prod low threshold", nodeThresholds[nodeUsage.node.Name].prodLowResourceThreshold)
 		} else if highThresholdFilter(nodeUsage, nodeThresholds[nodeUsage.node.Name]) {
 			highNodes = append(highNodes, nodeInfo)
 			nodeUsageExplain = "higher than node usage"
-			klog.InfoS("Node's utilization", "node", klog.KObj(nodeUsage.node), "result information", nodeUsageExplain, "node usage", nodeUsage.usage, "node usagePercentage", resourceUsagePercentages(nodeUsage, false),
-				"prod usage", nodeUsage.prodUsage, "prod usagePercentage", resourceUsagePercentages(nodeUsage, true))
+			klog.V(4).InfoS("Node's utilization", "node", klog.KObj(nodeUsage.node), "result information", nodeUsageExplain, "node usage", nodeUsage.usage, "node usagePercentage", resourceUsagePercentages(nodeUsage, false),
+				"node high threshold", nodeThresholds[nodeUsage.node.Name].highResourceThreshold, "node low threshold", nodeThresholds[nodeUsage.node.Name].lowResourceThreshold, "prod usage", nodeUsage.prodUsage,
+				"prod usagePercentage", resourceUsagePercentages(nodeUsage, true), "prod high threshold", nodeThresholds[nodeUsage.node.Name].prodHighResourceThreshold, "prod low threshold", nodeThresholds[nodeUsage.node.Name].prodLowResourceThreshold)
 		} else {
 			if prodHighThresholdFilter(nodeUsage, nodeThresholds[nodeUsage.node.Name]) {
 				prodHighNodes = append(prodHighNodes, nodeInfo)
@@ -270,8 +276,9 @@ func classifyNodes(
 			} else {
 				nodeUsageExplain = "both appropriately for node && prod usage"
 			}
-			klog.InfoS("Node's utilization", "node", klog.KObj(nodeUsage.node), "result information", nodeUsageExplain, "node usage", nodeUsage.usage, "node usagePercentage", resourceUsagePercentages(nodeUsage, false),
-				"prod usage", nodeUsage.prodUsage, "prod usagePercentage", resourceUsagePercentages(nodeUsage, true))
+			klog.V(4).InfoS("Node's utilization", "node", klog.KObj(nodeUsage.node), "result information", nodeUsageExplain, "node usage", nodeUsage.usage, "node usagePercentage", resourceUsagePercentages(nodeUsage, false),
+				"node high threshold", nodeThresholds[nodeUsage.node.Name].highResourceThreshold, "node low threshold", nodeThresholds[nodeUsage.node.Name].lowResourceThreshold, "prod usage", nodeUsage.prodUsage,
+				"prod usagePercentage", resourceUsagePercentages(nodeUsage, true), "prod high threshold", nodeThresholds[nodeUsage.node.Name].prodHighResourceThreshold, "prod low threshold", nodeThresholds[nodeUsage.node.Name].prodLowResourceThreshold)
 		}
 	}