koord-descheduler: fix prod pod excessive eviction when node recover …

…to normal (#2225) Signed-off-by: JBbin <namechenjiabin@icloud.com>
koordinator-sh · Oct 16, 2024 · 34b17b8 · 34b17b8
1 parent 11701ae
commit 34b17b8
Show file tree

Hide file tree

Showing 2 changed files with 71 additions and 2 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
@@ -1300,6 +1300,64 @@ func TestLowNodeLoad(t *testing.T) {
 			expectedPodsEvicted: 3,
 			evictedPods:         []string{},
 		},
+		{
+			name: "stop prod pod eviction when the node recovers to normal.",
+			thresholds: ResourceThresholds{
+				corev1.ResourceCPU:  0,
+				corev1.ResourcePods: 0,
+			},
+			targetThresholds: ResourceThresholds{
+				corev1.ResourceCPU:  100,
+				corev1.ResourcePods: 100,
+			},
+			prodLowThresholds: ResourceThresholds{
+				corev1.ResourceCPU:  60,
+				corev1.ResourcePods: 60,
+			},
+			prodHighThresholds: ResourceThresholds{
+				corev1.ResourceCPU:  70,
+				corev1.ResourcePods: 70,
+			},
+			nodes: []*corev1.Node{
+				test.BuildTestNode(n1NodeName, 4000, 3000, 20, nil),
+				test.BuildTestNode(n2NodeName, 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),
+			},
+			expectedPodsEvicted: 1,
+			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
@@ -540,20 +540,31 @@ func evictPods(
 			if nodeUsage := nodeInfo.usage[resourceName]; nodeUsage != nil {
 				nodeUsage.Sub(quantity)
 			}
+			if prodUsage := nodeInfo.prodUsage[resourceName]; prod && prodUsage != nil {
+				prodUsage.Sub(quantity)
+			}
 		}
 
 		keysAndValues := []interface{}{
 			"node", nodeInfo.node.Name,
 			"nodePool", nodePoolName,
 		}
-		for k, v := range nodeInfo.usage {
+		usage := nodeInfo.usage
+		if prod {
+			usage = nodeInfo.prodUsage
+		}
+		for k, v := range usage {
 			keysAndValues = append(keysAndValues, k.String(), v.String())
 		}
 		for resourceName, quantity := range totalAvailableUsages {
 			keysAndValues = append(keysAndValues, fmt.Sprintf("%s/totalAvailable", resourceName), quantity.String())
 		}
 
-		klog.V(4).InfoS("Updated node usage", keysAndValues...)
+		if prod {
+			klog.V(4).InfoS("Updated node prodUsage", keysAndValues...)
+		} else {
+			klog.V(4).InfoS("Updated node usage", keysAndValues...)
+		}
 	}
 }