descheduler: Add detail logs for nonRemovable pods (#2000)

Signed-off-by: zwForrest <756495135@qq.com>

pkg/descheduler/controllers/migration/arbitrator/arbitrator.go

@@ -124,10 +124,12 @@ func (a *arbitratorImpl) Start(ctx context.Context) error {
 // Filter checks if a pod can be evicted
 func (a *arbitratorImpl) Filter(pod *corev1.Pod) bool {
 	if !a.filter.filterExistingPodMigrationJob(pod) {
+		klog.V(4).InfoS("Pod fails the following checks", "pod", klog.KObj(pod), "checks", "filterExistingPodMigrationJob")
 		return false
 	}
 
 	if !a.filter.reservationFilter(pod) {
+		klog.V(4).InfoS("Pod fails the following checks", "pod", klog.KObj(pod), "checks", "reservationFilter")
 		return false
 	}
 

pkg/descheduler/controllers/migration/arbitrator/filter.go

@@ -254,8 +254,8 @@ func (f *filter) filterMaxMigratingPerNode(pod *corev1.Pod) bool {
 	maxMigratingPerNode := int(*f.args.MaxMigratingPerNode)
 	exceeded := count >= maxMigratingPerNode
 	if exceeded {
-		klog.V(4).Infof("Pod %q fails to check maxMigratingPerNode because the Node %q has %d migrating Pods, exceeding the maxMigratingPerNode(%d)",
-			klog.KObj(pod), pod.Spec.NodeName, count, maxMigratingPerNode)
+		klog.V(4).InfoS("Pod fails the following checks", "pod", klog.KObj(pod),
+			"checks", "maxMigratingPerNode", "node", pod.Spec.NodeName, "count", count, "maxMigratingPerNode", maxMigratingPerNode)
 	}
 	return !exceeded
 }
@@ -285,8 +285,8 @@ func (f *filter) filterMaxMigratingPerNamespace(pod *corev1.Pod) bool {
 	maxMigratingPerNamespace := int(*f.args.MaxMigratingPerNamespace)
 	exceeded := count >= maxMigratingPerNamespace
 	if exceeded {
-		klog.V(4).Infof("Pod %q fails to check maxMigratingPerNamespace because the Namespace %q has %d migrating Pods, exceeding the maxMigratingPerNamespace(%d)",
-			klog.KObj(pod), pod.Namespace, count, maxMigratingPerNamespace)
+		klog.V(4).InfoS("Pod fails the following checks", "pod", klog.KObj(pod),
+			"checks", "maxMigratingPerNamespace", "namespace", pod.Namespace, "count", count, "maxMigratingPerNamespace", maxMigratingPerNamespace)
 	}
 	return !exceeded
 }
@@ -345,8 +345,9 @@ func (f *filter) filterMaxMigratingOrUnavailablePerWorkload(pod *corev1.Pod) boo
 	if len(migratingPods) > 0 {
 		exceeded := len(migratingPods) >= maxMigrating
 		if exceeded {
-			klog.V(4).Infof("The workload %s/%s/%s(%s) of Pod %q has %d migration jobs that exceed MaxMigratingPerWorkload %d",
-				ownerRef.Name, ownerRef.Kind, ownerRef.APIVersion, ownerRef.UID, klog.KObj(pod), len(migratingPods), maxMigrating)
+			klog.V(4).InfoS("Pod fails the following checks", "pod", klog.KObj(pod),
+				"checks", "maxMigratingPerWorkload", "owner", fmt.Sprintf("%s/%s/%s(%s)", ownerRef.Name, ownerRef.Kind, ownerRef.APIVersion, ownerRef.UID),
+				"migratingPods", len(migratingPods), "maxMigratingPerWorkload", maxMigrating)
 			return false
 		}
 	}
@@ -386,8 +387,9 @@ func (f *filter) filterExpectedReplicas(pod *corev1.Pod) bool {
 	if f.args.SkipCheckExpectedReplicas == nil || !*f.args.SkipCheckExpectedReplicas {
 		// TODO(joseph): There are f few special scenarios where should we allow eviction?
 		if expectedReplicas == 1 || int(expectedReplicas) == maxMigrating || int(expectedReplicas) == maxUnavailable {
-			klog.Warningf("maxMigrating(%d) or maxUnavailable(%d) equals to the replicas(%d) of the workload %s/%s/%s(%s) of Pod %q, or the replicas equals to 1, please increase the replicas or update the defense configurations",
-				maxMigrating, maxUnavailable, expectedReplicas, ownerRef.Name, ownerRef.Kind, ownerRef.APIVersion, ownerRef.UID, klog.KObj(pod))
+			klog.V(4).InfoS("Pod fails the following checks", "pod", klog.KObj(pod), "checks", "expectedReplicas",
+				"owner", fmt.Sprintf("%s/%s/%s(%s)", ownerRef.Name, ownerRef.Kind, ownerRef.APIVersion, ownerRef.UID),
+				"maxMigrating", maxMigrating, "maxUnavailable", maxUnavailable, "expectedReplicas", expectedReplicas)
 			return false
 		}
 	}
@@ -473,7 +475,8 @@ func (f *filter) filterLimitedObject(pod *corev1.Pod) bool {
 		defer f.limiterLock.Unlock()
 		if limiter := f.objectLimiters[ownerRef.UID]; limiter != nil {
 			if remainTokens := limiter.Tokens() - float64(1); remainTokens < 0 {
-				klog.Infof("Pod %q is filtered by workload %s/%s/%s is limited", klog.KObj(pod), ownerRef.Name, ownerRef.Kind, ownerRef.APIVersion)
+				klog.V(4).InfoS("Pod fails the following checks", "pod", klog.KObj(pod), "checks", "limitedObject",
+					"owner", fmt.Sprintf("%s/%s/%s", ownerRef.Name, ownerRef.Kind, ownerRef.APIVersion))
 				return false
 			}
 		}

pkg/descheduler/controllers/migration/util/util.go

@@ -22,6 +22,7 @@ import (
 	corev1 "k8s.io/api/core/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/util/intstr"
+	"k8s.io/klog/v2"
 
 	"github.com/koordinator-sh/koordinator/apis/extension"
 	sev1alpha1 "github.com/koordinator-sh/koordinator/apis/scheduling/v1alpha1"
@@ -115,5 +116,8 @@ func GetMaxMigrating(replicas int, intOrPercent *intstr.IntOrString) (int, error
 // FilterPodWithMaxEvictionCost rejects if pod's eviction cost is math.MaxInt32
 func FilterPodWithMaxEvictionCost(pod *corev1.Pod) bool {
 	cost, _ := extension.GetEvictionCost(pod.Annotations)
+	if cost == math.MaxInt32 {
+		klog.V(4).InfoS("Pod fails the following checks", "pod", klog.KObj(pod), "checks", "podWithMaxEvictionCost")
+	}
 	return !(cost == math.MaxInt32)
 }

pkg/descheduler/node/node.go

@@ -167,7 +167,7 @@ func PodFitsAnyNode(nodeIndexer podutil.GetPodsAssignedToNodeFunc, pod *corev1.P
 			klog.V(4).InfoS("Pod fits on node", "pod", klog.KObj(pod), "node", klog.KObj(node))
 			return true
 		} else {
-			klog.V(5).InfoS("Pod does not fit on node", "pod", klog.KObj(pod), "node", klog.KObj(node), "errors", utilerrors.NewAggregate(errors))
+			klog.V(4).InfoS("Pod does not fit on node", "pod", klog.KObj(pod), "node", klog.KObj(node), "errors", utilerrors.NewAggregate(errors))
 		}
 	}
 	return false

pkg/descheduler/pod/pods.go

@@ -24,6 +24,7 @@ import (
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/labels"
 	"k8s.io/apimachinery/pkg/util/sets"
+	"k8s.io/klog/v2"
 	qoshelper "k8s.io/kubernetes/pkg/apis/core/v1/helper/qos"
 
 	"github.com/koordinator-sh/koordinator/pkg/descheduler/framework"
@@ -100,12 +101,15 @@ func (o *Options) BuildFilterFunc() (FilterFunc, error) {
 			return false
 		}
 		if len(o.includedNamespaces) > 0 && !o.includedNamespaces.Has(pod.Namespace) {
+			klog.V(4).InfoS("Pod fails the following checks", "pod", klog.KObj(pod), "checks", "includedNamespaces")
 			return false
 		}
 		if len(o.excludedNamespaces) > 0 && o.excludedNamespaces.Has(pod.Namespace) {
+			klog.V(4).InfoS("Pod fails the following checks", "pod", klog.KObj(pod), "checks", "excludedNamespaces")
 			return false
 		}
 		if s != nil && !s.Matches(labels.Set(pod.GetLabels())) {
+			klog.V(4).InfoS("Pod fails the following checks", "pod", klog.KObj(pod), "checks", "labelSelector")
 			return false
 		}
 		return true