Merge pull request #8266 from elmiko/add-more-balance-logging

cluster-autoscaler: add logging for failed node balancing

Author: k8s-ci-robot

cluster-autoscaler/core/scaleup/orchestrator/orchestrator.go

@@ -469,6 +469,17 @@ func (o *ScaleUpOrchestrator) ComputeExpansionOption(
 	}
 
 	option.SimilarNodeGroups = o.ComputeSimilarNodeGroups(nodeGroup, nodeInfos, schedulablePodGroups, now)
+	if option.SimilarNodeGroups != nil {
+		// if similar node groups are found, log about them
+		similarNodeGroupIds := make([]string, 0)
+		for _, sng := range option.SimilarNodeGroups {
+			similarNodeGroupIds = append(similarNodeGroupIds, sng.Id())
+		}
+		klog.V(5).Infof("Found %d similar node groups: %v", len(option.SimilarNodeGroups), similarNodeGroupIds)
+	} else if o.autoscalingContext.BalanceSimilarNodeGroups {
+		// if no similar node groups are found and the flag is enabled, log about it
+		klog.V(5).Info("No similar node groups found")
+	}
 
 	estimateStart := time.Now()
 	expansionEstimator := o.estimatorBuilder(

cluster-autoscaler/processors/nodegroupset/compare_nodegroups.go

@@ -17,13 +17,15 @@ limitations under the License.
 package nodegroupset
 
 import (
+	"fmt"
 	"math"
 
 	apiv1 "k8s.io/api/core/v1"
 	"k8s.io/apimachinery/pkg/api/resource"
 	"k8s.io/autoscaler/cluster-autoscaler/config"
 	"k8s.io/autoscaler/cluster-autoscaler/simulator/framework"
 	"k8s.io/autoscaler/cluster-autoscaler/utils/scheduler"
+	klog "k8s.io/klog/v2"
 )
 
 // BasicIgnoredLabels define a set of basic labels that should be ignored when comparing the similarity
@@ -122,34 +124,44 @@ func IsCloudProviderNodeInfoSimilar(
 
 	for kind, qtyList := range capacity {
 		if len(qtyList) != 2 {
+			dissimilarNodesLog(n1.Node().Name, n2.Node().Name, fmt.Sprintf("missing capacity %q", kind))
 			return false
 		}
 		switch kind {
 		case apiv1.ResourceMemory:
 			if !resourceListWithinTolerance(qtyList, ratioOpts.MaxCapacityMemoryDifferenceRatio) {
+				dissimilarNodesLog(n1.Node().Name, n2.Node().Name, "memory not within tolerance")
 				return false
 			}
 		default:
 			// For other capacity types we require exact match.
 			// If this is ever changed, enforcing MaxCoresTotal limits
 			// as it is now may no longer work.
 			if qtyList[0].Cmp(qtyList[1]) != 0 {
+				dissimilarNodesLog(n1.Node().Name, n2.Node().Name, fmt.Sprintf("capacity resource %q does not match", kind))
 				return false
 			}
 		}
 	}
 
 	// For allocatable and free we allow resource quantities to be within a few % of each other
 	if !resourceMapsWithinTolerance(allocatable, ratioOpts.MaxAllocatableDifferenceRatio) {
+		dissimilarNodesLog(n1.Node().Name, n2.Node().Name, "allocatable resources not within tolerance")
 		return false
 	}
 	if !resourceMapsWithinTolerance(free, ratioOpts.MaxFreeDifferenceRatio) {
+		dissimilarNodesLog(n1.Node().Name, n2.Node().Name, "free resources not within tolerance")
 		return false
 	}
 
 	if !compareLabels(nodes, ignoredLabels) {
+		dissimilarNodesLog(n1.Node().Name, n2.Node().Name, "labels do not match")
 		return false
 	}
 
 	return true
 }
+
+func dissimilarNodesLog(node1, node2, message string) {
+	klog.V(5).Infof("nodes %q and %q are not similar, %s", node1, node2, message)
+}