chore: upgrade karpenter to 0.37.4

pkg/apis/crds/karpenter.sh_nodepools.yaml

@@ -617,6 +617,13 @@ spec:
                         memory leak protection, and disruption testing.
                       pattern: ^(([0-9]+(s|m|h))+)|(Never)$
                       type: string
+                    utilizationThreshold:
+                      description: |-
+                        UtilizationThreshold is defined as sum of requested resources divided by capacity
+                        below which a node can be considered for disruption.
+                      maximum: 100
+                      minimum: 1
+                      type: integer
                   type: object
                   x-kubernetes-validations:
                     - message: consolidateAfter cannot be combined with consolidationPolicy=WhenUnderutilized

pkg/apis/v1beta1/nodeclaim_status.go

@@ -22,12 +22,13 @@ import (
 )
 
 const (
-	ConditionTypeLaunched    = "Launched"
-	ConditionTypeRegistered  = "Registered"
-	ConditionTypeInitialized = "Initialized"
-	ConditionTypeEmpty       = "Empty"
-	ConditionTypeDrifted     = "Drifted"
-	ConditionTypeExpired     = "Expired"
+	ConditionTypeLaunched      = "Launched"
+	ConditionTypeRegistered    = "Registered"
+	ConditionTypeInitialized   = "Initialized"
+	ConditionTypeEmpty         = "Empty"
+	ConditionTypeUnderutilized = "Underutilized"
+	ConditionTypeDrifted       = "Drifted"
+	ConditionTypeExpired       = "Expired"
 )
 
 // NodeClaimStatus defines the observed state of NodeClaim

pkg/apis/v1beta1/nodepool.go

@@ -76,6 +76,12 @@ type Disruption struct {
 	// +kubebuilder:validation:Enum:={WhenEmpty,WhenUnderutilized}
 	// +optional
 	ConsolidationPolicy ConsolidationPolicy `json:"consolidationPolicy,omitempty"`
+	// UtilizationThreshold is defined as sum of requested resources divided by capacity
+	// below which a node can be considered for disruption.
+	// +kubebuilder:validation:Minimum:=1
+	// +kubebuilder:validation:Maximum:=100
+	// +optional
+	UtilizationThreshold *int `json:"utilizationThreshold,omitempty"`
 	// ExpireAfter is the duration the controller will wait
 	// before terminating a node, measured from when the node is created. This
 	// is useful to implement features like eventually consistent node upgrade,

pkg/apis/v1beta1/nodepool_validation.go

@@ -98,9 +98,6 @@ func (in *NodeClaimTemplate) validateRequirementsNodePoolKeyDoesNotExist() (errs
 
 //nolint:gocyclo
 func (in *Disruption) validate() (errs *apis.FieldError) {
-	if in.ConsolidateAfter != nil && in.ConsolidateAfter.Duration != nil && in.ConsolidationPolicy == ConsolidationPolicyWhenUnderutilized {
-		return errs.Also(apis.ErrGeneric("consolidateAfter cannot be combined with consolidationPolicy=WhenUnderutilized"))
-	}
 	if in.ConsolidateAfter == nil && in.ConsolidationPolicy == ConsolidationPolicyWhenEmpty {
 		return errs.Also(apis.ErrGeneric("consolidateAfter must be specified with consolidationPolicy=WhenEmpty"))
 	}

pkg/apis/v1beta1/nodepool_validation_cel_test.go

@@ -94,11 +94,6 @@ var _ = Describe("CEL/Validation", func() {
 			nodePool.Spec.Disruption.ConsolidationPolicy = ConsolidationPolicyWhenEmpty
 			Expect(env.Client.Create(ctx, nodePool)).To(Succeed())
 		})
-		It("should fail when setting consolidateAfter with consolidationPolicy=WhenUnderutilized", func() {
-			nodePool.Spec.Disruption.ConsolidateAfter = lo.ToPtr(MustParseNillableDuration("30s"))
-			nodePool.Spec.Disruption.ConsolidationPolicy = ConsolidationPolicyWhenUnderutilized
-			Expect(env.Client.Create(ctx, nodePool)).ToNot(Succeed())
-		})
 		It("should succeed when not setting consolidateAfter to 'Never' with consolidationPolicy=WhenUnderutilized", func() {
 			nodePool.Spec.Disruption.ConsolidateAfter = lo.ToPtr(MustParseNillableDuration("Never"))
 			nodePool.Spec.Disruption.ConsolidationPolicy = ConsolidationPolicyWhenUnderutilized

pkg/controllers/disruption/consolidation.go

@@ -101,6 +101,13 @@ func (c *consolidation) ShouldDisrupt(_ context.Context, cn *Candidate) bool {
 		c.recorder.Publish(disruptionevents.Unconsolidatable(cn.Node, cn.NodeClaim, fmt.Sprintf("NodePool %q has consolidation disabled", cn.nodePool.Name))...)
 		return false
 	}
+	// Only check when UtilizationThreshold is specified to make it compatible
+	if cn.nodePool.Spec.Disruption.UtilizationThreshold != nil {
+		if !cn.NodeClaim.StatusConditions().Get(v1beta1.ConditionTypeUnderutilized).IsTrue() ||
+			c.clock.Now().Before(cn.NodeClaim.StatusConditions().Get(v1beta1.ConditionTypeUnderutilized).LastTransitionTime.Add(*cn.nodePool.Spec.Disruption.ConsolidateAfter.Duration)) {
+			return false
+		}
+	}
 	return true
 }
 

pkg/controllers/nodeclaim/disruption/consolidation.go

@@ -0,0 +1,137 @@
+/*
+/*
+Copyright The Kubernetes Authors.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+package disruption
+
+import (
+	"context"
+	"fmt"
+
+	"github.com/prometheus/client_golang/prometheus"
+	v1 "k8s.io/api/core/v1"
+	"k8s.io/apimachinery/pkg/api/resource"
+	"k8s.io/utils/clock"
+	"knative.dev/pkg/logging"
+	"sigs.k8s.io/controller-runtime/pkg/client"
+	"sigs.k8s.io/controller-runtime/pkg/reconcile"
+
+	"sigs.k8s.io/karpenter/pkg/apis/v1beta1"
+	"sigs.k8s.io/karpenter/pkg/controllers/state"
+	"sigs.k8s.io/karpenter/pkg/metrics"
+	"sigs.k8s.io/karpenter/pkg/utils/node"
+	nodeclaimutil "sigs.k8s.io/karpenter/pkg/utils/nodeclaim"
+)
+
+// Consolidation is a nodeclaim sub-controller that adds or removes status conditions on nodeclaims when using WhenUnderutilized policy.
+type Consolidation struct {
+	kubeClient client.Client
+	cluster    *state.Cluster
+	clock      clock.Clock
+}
+
+//nolint:gocyclo
+func (e *Consolidation) Reconcile(ctx context.Context, nodePool *v1beta1.NodePool, nodeClaim *v1beta1.NodeClaim) (reconcile.Result, error) {
+	hasCondition := nodeClaim.StatusConditions().Get(v1beta1.ConditionTypeUnderutilized) != nil
+	if nodePool.Spec.Disruption.ConsolidationPolicy != v1beta1.ConsolidationPolicyWhenUnderutilized {
+		if hasCondition {
+			_ = nodeClaim.StatusConditions().Clear(v1beta1.ConditionTypeUnderutilized)
+		}
+		return reconcile.Result{}, nil
+	}
+	if initCond := nodeClaim.StatusConditions().Get(v1beta1.ConditionTypeInitialized); initCond == nil || initCond.IsFalse() {
+		if hasCondition {
+			_ = nodeClaim.StatusConditions().Clear(v1beta1.ConditionTypeUnderutilized)
+			logging.FromContext(ctx).Debugf("removing consolidated status condition, isn't initialized")
+		}
+		return reconcile.Result{}, nil
+	}
+	_, err := nodeclaimutil.NodeForNodeClaim(ctx, e.kubeClient, nodeClaim)
+	if err != nil {
+		if nodeclaimutil.IsDuplicateNodeError(err) || nodeclaimutil.IsNodeNotFoundError(err) {
+			_ = nodeClaim.StatusConditions().Clear(v1beta1.ConditionTypeUnderutilized)
+			if hasCondition {
+				logging.FromContext(ctx).Debugf("removing underutilized status condition, doesn't have a single node mapping")
+			}
+			return reconcile.Result{}, nil
+		}
+		return reconcile.Result{}, err
+	}
+
+	// Get the node to check utilization
+	n, err := nodeclaimutil.NodeForNodeClaim(ctx, e.kubeClient, nodeClaim)
+	if err != nil {
+		if nodeclaimutil.IsDuplicateNodeError(err) || nodeclaimutil.IsNodeNotFoundError(err) {
+			_ = nodeClaim.StatusConditions().Clear(v1beta1.ConditionTypeUnderutilized)
+			if hasCondition {
+				logging.FromContext(ctx).Debugf("removing underutilized status condition, doesn't have a single node mapping")
+			}
+			return reconcile.Result{}, nil
+		}
+		return reconcile.Result{}, err
+	}
+	pods, err := node.GetPods(ctx, e.kubeClient, n)
+	if err != nil {
+		return reconcile.Result{}, fmt.Errorf("retrieving node pods, %w", err)
+	}
+	// Check the node utilization if the utilizationThreshold is specified, the node can be disruptted only if the utilization is below the threshold.
+	threshold := nodePool.Spec.Disruption.UtilizationThreshold
+	if threshold != nil {
+		cpu, err := calculateUtilizationOfResource(n, v1.ResourceCPU, pods)
+		if err != nil {
+			return reconcile.Result{}, fmt.Errorf("failed to calculate CPU, %w", err)
+		}
+		memory, err := calculateUtilizationOfResource(n, v1.ResourceMemory, pods)
+		if err != nil {
+			return reconcile.Result{}, fmt.Errorf("failed to calculate memory, %w", err)
+		}
+		if cpu < float64(*threshold)/100 && memory < float64(*threshold)/100 {
+			if !hasCondition {
+				nodeClaim.StatusConditions().SetTrue(v1beta1.ConditionTypeUnderutilized)
+				logging.FromContext(ctx).Debugf("marking underutilizate")
+				metrics.NodeClaimsDisruptedCounter.With(prometheus.Labels{
+					metrics.TypeLabel:     metrics.ConsolidationReason,
+					metrics.NodePoolLabel: nodeClaim.Labels[v1beta1.NodePoolLabelKey],
+				}).Inc()
+			}
+		} else {
+			if hasCondition {
+				_ = nodeClaim.StatusConditions().Clear(v1beta1.ConditionTypeUnderutilized)
+				logging.FromContext(ctx).Debugf("removing underutilized status condition, utilization increased")
+			}
+		}
+	}
+	return reconcile.Result{}, nil
+}
+
+// CalculateUtilizationOfResource calculates utilization of a given resource for a node.
+func calculateUtilizationOfResource(node *v1.Node, resourceName v1.ResourceName, pods []*v1.Pod) (float64, error) {
+	allocatable, found := node.Status.Allocatable[resourceName]
+	if !found {
+		return 0, fmt.Errorf("failed to get %v from %s", resourceName, node.Name)
+	}
+	if allocatable.MilliValue() == 0 {
+		return 0, fmt.Errorf("%v is 0 at %s", resourceName, node.Name)
+	}
+	podsRequest := resource.MustParse("0")
+	for _, pod := range pods {
+		for _, container := range pod.Spec.Containers {
+			if resourceValue, found := container.Resources.Requests[resourceName]; found {
+				podsRequest.Add(resourceValue)
+			}
+		}
+	}
+	return float64(podsRequest.MilliValue()) / float64(allocatable.MilliValue()), nil
+}

pkg/controllers/nodeclaim/disruption/controller.go

@@ -51,9 +51,10 @@ type Controller struct {
 	kubeClient    client.Client
 	cloudProvider cloudprovider.CloudProvider
 
-	drift      *Drift
-	expiration *Expiration
-	emptiness  *Emptiness
+	drift         *Drift
+	expiration    *Expiration
+	emptiness     *Emptiness
+	consolidation *Consolidation
 }
 
 // NewController constructs a nodeclaim disruption controller
@@ -64,6 +65,7 @@ func NewController(clk clock.Clock, kubeClient client.Client, cluster *state.Clu
 		drift:         &Drift{cloudProvider: cloudProvider},
 		expiration:    &Expiration{kubeClient: kubeClient, clock: clk},
 		emptiness:     &Emptiness{kubeClient: kubeClient, cluster: cluster, clock: clk},
+		consolidation: &Consolidation{kubeClient: kubeClient, cluster: cluster, clock: clk},
 	}
 }
 
@@ -90,6 +92,7 @@ func (c *Controller) Reconcile(ctx context.Context, nodeClaim *v1beta1.NodeClaim
 		c.expiration,
 		c.drift,
 		c.emptiness,
+		c.consolidation,
 	}
 	for _, reconciler := range reconcilers {
 		res, err := reconciler.Reconcile(ctx, nodePool, nodeClaim)

pkg/controllers/provisioning/provisioner.go

@@ -311,26 +311,56 @@ func (p *Provisioner) Schedule(ctx context.Context) (scheduler.Results, error) {
 		return scheduler.Results{}, err
 	}
 	pods := append(pendingPods, deletingNodePods...)
+	// filter pods which are alredy handled in last 3 minute
+	targetPods := lo.FilterMap(pods, func(pod *v1.Pod, _ int) (*v1.Pod, bool) {
+		if p.isPodHandled(ctx, pod) {
+			return nil, false
+		}
+		return pod, true
+	})
 	// nothing to schedule, so just return success
-	if len(pods) == 0 {
+	if len(targetPods) == 0 {
 		return scheduler.Results{}, nil
 	}
-	s, err := p.NewScheduler(ctx, pods, nodes.Active())
+	s, err := p.NewScheduler(ctx, targetPods, nodes.Active())
 	if err != nil {
 		if errors.Is(err, ErrNodePoolsNotFound) {
 			log.FromContext(ctx).Info("no nodepools found")
 			return scheduler.Results{}, nil
 		}
 		return scheduler.Results{}, fmt.Errorf("creating scheduler, %w", err)
 	}
-	results := s.Solve(ctx, pods).TruncateInstanceTypes(scheduler.MaxInstanceTypes)
+	results := s.Solve(ctx, targetPods).TruncateInstanceTypes(scheduler.MaxInstanceTypes)
 	if len(results.NewNodeClaims) > 0 {
-		log.FromContext(ctx).WithValues("Pods", pretty.Slice(lo.Map(pods, func(p *v1.Pod, _ int) string { return klog.KRef(p.Namespace, p.Name).String() }), 5), "duration", time.Since(start)).Info("found provisionable pod(s)")
+		log.FromContext(ctx).WithValues("Pods", pretty.Slice(lo.Map(targetPods, func(p *v1.Pod, _ int) string { return klog.KRef(p.Namespace, p.Name).String() }), 5), "duration", time.Since(start)).Info("found provisionable pod(s)")
 	}
 	results.Record(ctx, p.recorder, p.cluster)
 	return results, nil
 }
 
+func (p *Provisioner) isPodHandled(ctx context.Context, pod *v1.Pod) bool {
+	var events v1.EventList
+	filter := client.MatchingFields{
+		"namespace":           pod.Namespace,
+		"involvedObject.kind": "Pod",
+		"involvedObject.name": pod.Name,
+		"reason":              "HandledByKarpenter",
+	}
+	if err := p.kubeClient.List(ctx, &events, filter); err == nil {
+		for _, event := range events.Items {
+			// ignore the pod if it's already handled in 3 minute
+			if time.Now().Before(event.LastTimestamp.Time.Add(3 * time.Minute)) {
+				log.FromContext(ctx).Info(fmt.Sprintf("pod %s/%s is handled", pod.Namespace, pod.Name))
+				return true
+			}
+		}
+	} else {
+		log.FromContext(ctx).Error(err, fmt.Sprintf("failed to get event for %s/%s", pod.Namespace, pod.Name))
+	}
+	p.recorder.Publish(scheduler.PodHandledEvent(pod))
+	return false
+}
+
 func (p *Provisioner) Create(ctx context.Context, n *scheduler.NodeClaim, opts ...functional.Option[LaunchOptions]) (string, error) {
 	ctx = log.IntoContext(ctx, log.FromContext(ctx).WithValues("NodePool", klog.KRef("", n.NodePoolName)))
 	options := functional.ResolveOptions(opts...)

pkg/controllers/provisioning/scheduling/events.go

@@ -59,3 +59,14 @@ func PodFailedToScheduleEvent(pod *v1.Pod, err error) events.Event {
 		DedupeTimeout:  5 * time.Minute,
 	}
 }
+
+func PodHandledEvent(pod *v1.Pod) events.Event {
+	return events.Event{
+		InvolvedObject: pod,
+		Type:           v1.EventTypeNormal,
+		Reason:         "HandledByKarpenter",
+		Message:        "Pod is handled by karpenter",
+		DedupeValues:   []string{string(pod.UID)},
+		DedupeTimeout:  5 * time.Minute,
+	}
+}

pkg/operator/operator.go

@@ -192,6 +192,18 @@ func NewOperator() (context.Context, *Operator) {
 	lo.Must0(mgr.GetFieldIndexer().IndexField(ctx, &v1.Node{}, "spec.providerID", func(o client.Object) []string {
 		return []string{o.(*v1.Node).Spec.ProviderID}
 	}), "failed to setup node provider id indexer")
+	lo.Must0(mgr.GetFieldIndexer().IndexField(ctx, &v1.Event{}, "involvedObject.kind", func(o client.Object) []string {
+		return []string{o.(*v1.Event).InvolvedObject.Kind}
+	}), "failed to setup event kind indexer")
+	lo.Must0(mgr.GetFieldIndexer().IndexField(ctx, &v1.Event{}, "involvedObject.name", func(o client.Object) []string {
+		return []string{o.(*v1.Event).InvolvedObject.Name}
+	}), "failed to setup event name indexer")
+	lo.Must0(mgr.GetFieldIndexer().IndexField(ctx, &v1.Event{}, "namespace", func(o client.Object) []string {
+		return []string{o.(*v1.Event).Namespace}
+	}), "failed to setup event namespace indexer")
+	lo.Must0(mgr.GetFieldIndexer().IndexField(ctx, &v1.Event{}, "reason", func(o client.Object) []string {
+		return []string{o.(*v1.Event).Reason}
+	}), "failed to setup event reason indexer")
 	lo.Must0(mgr.GetFieldIndexer().IndexField(ctx, &v1beta1.NodeClaim{}, "status.providerID", func(o client.Object) []string {
 		return []string{o.(*v1beta1.NodeClaim).Status.ProviderID}
 	}), "failed to setup nodeclaim provider id indexer")