Merge pull request #35275 from wojtek-t/cache_conditions

Automatic merge from submit-queue

Cache additional information in schedulercache.NodeInfo to speedup scheduler

Ref #35117

plugin/pkg/scheduler/algorithm/predicates/predicates.go

@@ -1119,12 +1119,7 @@ func (c *PodAffinityChecker) satisfiesPodsAffinityAntiAffinity(pod *v1.Pod, node
 }
 
 func PodToleratesNodeTaints(pod *v1.Pod, meta interface{}, nodeInfo *schedulercache.NodeInfo) (bool, []algorithm.PredicateFailureReason, error) {
-	node := nodeInfo.Node()
+	taints, err := nodeInfo.Taints()
-	if node == nil {
-		return false, nil, fmt.Errorf("node not found")
-	}
-
-	taints, err := v1.GetTaintsFromNodeAnnotations(node.Annotations)
 	if err != nil {
 		return false, nil, err
 	}
@@ -1174,11 +1169,6 @@ func isPodBestEffort(pod *v1.Pod) bool {
 // CheckNodeMemoryPressurePredicate checks if a pod can be scheduled on a node
 // reporting memory pressure condition.
 func CheckNodeMemoryPressurePredicate(pod *v1.Pod, meta interface{}, nodeInfo *schedulercache.NodeInfo) (bool, []algorithm.PredicateFailureReason, error) {
-	node := nodeInfo.Node()
-	if node == nil {
-		return false, nil, fmt.Errorf("node not found")
-	}
-
 	var podBestEffort bool
 	if predicateMeta, ok := meta.(*predicateMetadata); ok {
 		podBestEffort = predicateMeta.podBestEffort
@@ -1186,36 +1176,24 @@ func CheckNodeMemoryPressurePredicate(pod *v1.Pod, meta interface{}, nodeInfo *s
 		// We couldn't parse metadata - fallback to computing it.
 		podBestEffort = isPodBestEffort(pod)
 	}
-
 	// pod is not BestEffort pod
 	if !podBestEffort {
 		return true, nil, nil
 	}
 
-	// is node under pressure?
+	// is node under presure?
-	for _, cond := range node.Status.Conditions {
+	if nodeInfo.MemoryPressureCondition() == v1.ConditionTrue {
-		if cond.Type == v1.NodeMemoryPressure && cond.Status == v1.ConditionTrue {
+		return false, []algorithm.PredicateFailureReason{ErrNodeUnderMemoryPressure}, nil
-			return false, []algorithm.PredicateFailureReason{ErrNodeUnderMemoryPressure}, nil
-		}
 	}
-
 	return true, nil, nil
 }
 
 // CheckNodeDiskPressurePredicate checks if a pod can be scheduled on a node
 // reporting disk pressure condition.
 func CheckNodeDiskPressurePredicate(pod *v1.Pod, meta interface{}, nodeInfo *schedulercache.NodeInfo) (bool, []algorithm.PredicateFailureReason, error) {
-	node := nodeInfo.Node()
+	// is node under presure?
-	if node == nil {
+	if nodeInfo.DiskPressureCondition() == v1.ConditionTrue {
-		return false, nil, fmt.Errorf("node not found")
+		return false, []algorithm.PredicateFailureReason{ErrNodeUnderDiskPressure}, nil
 	}
-
-	// is node under pressure?
-	for _, cond := range node.Status.Conditions {
-		if cond.Type == v1.NodeDiskPressure && cond.Status == v1.ConditionTrue {
-			return false, []algorithm.PredicateFailureReason{ErrNodeUnderDiskPressure}, nil
-		}
-	}
-
 	return true, nil, nil
 }

plugin/pkg/scheduler/schedulercache/node_info.go

@@ -49,6 +49,14 @@ type NodeInfo struct {
 	// explicitly as int, to avoid conversions and improve performance.
 	allowedPodNumber int
 
+	// Cached tains of the node for faster lookup.
+	taints    []v1.Taint
+	taintsErr error
+
+	// Cached conditions of node for faster lookup.
+	memoryPressureCondition v1.ConditionStatus
+	diskPressureCondition   v1.ConditionStatus
+
 	// Whenever NodeInfo changes, generation is bumped.
 	// This is used to avoid cloning it if the object didn't change.
 	generation int64
@@ -122,6 +130,27 @@ func (n *NodeInfo) AllowedPodNumber() int {
 	return n.allowedPodNumber
 }
 
+func (n *NodeInfo) Taints() ([]v1.Taint, error) {
+	if n == nil {
+		return nil, nil
+	}
+	return n.taints, n.taintsErr
+}
+
+func (n *NodeInfo) MemoryPressureCondition() v1.ConditionStatus {
+	if n == nil {
+		return v1.ConditionUnknown
+	}
+	return n.memoryPressureCondition
+}
+
+func (n *NodeInfo) DiskPressureCondition() v1.ConditionStatus {
+	if n == nil {
+		return v1.ConditionUnknown
+	}
+	return n.diskPressureCondition
+}
+
 // RequestedResource returns aggregated resource request of pods on this node.
 func (n *NodeInfo) RequestedResource() Resource {
 	if n == nil {
@@ -148,12 +177,15 @@ func (n *NodeInfo) AllocatableResource() Resource {
 
 func (n *NodeInfo) Clone() *NodeInfo {
 	clone := &NodeInfo{
-		node:                n.node,
+		node:                    n.node,
-		requestedResource:   &(*n.requestedResource),
+		requestedResource:       &(*n.requestedResource),
-		nonzeroRequest:      &(*n.nonzeroRequest),
+		nonzeroRequest:          &(*n.nonzeroRequest),
-		allocatableResource: &(*n.allocatableResource),
+		allocatableResource:     &(*n.allocatableResource),
-		allowedPodNumber:    n.allowedPodNumber,
+		allowedPodNumber:        n.allowedPodNumber,
-		generation:          n.generation,
+		taintsErr:               n.taintsErr,
+		memoryPressureCondition: n.memoryPressureCondition,
+		diskPressureCondition:   n.diskPressureCondition,
+		generation:              n.generation,
 	}
 	if len(n.pods) > 0 {
 		clone.pods = append([]*v1.Pod(nil), n.pods...)
@@ -161,6 +193,9 @@ func (n *NodeInfo) Clone() *NodeInfo {
 	if len(n.podsWithAffinity) > 0 {
 		clone.podsWithAffinity = append([]*v1.Pod(nil), n.podsWithAffinity...)
 	}
+	if len(n.taints) > 0 {
+		clone.taints = append([]v1.Taint(nil), n.taints...)
+	}
 	return clone
 }
 
@@ -306,6 +341,18 @@ func (n *NodeInfo) SetNode(node *v1.Node) error {
 			}
 		}
 	}
+	n.taints, n.taintsErr = v1.GetTaintsFromNodeAnnotations(node.Annotations)
+	for i := range node.Status.Conditions {
+		cond := &node.Status.Conditions[i]
+		switch cond.Type {
+		case v1.NodeMemoryPressure:
+			n.memoryPressureCondition = cond.Status
+		case v1.NodeDiskPressure:
+			n.diskPressureCondition = cond.Status
+		default:
+			// We ignore other conditions.
+		}
+	}
 	n.generation++
 	return nil
 }
@@ -319,6 +366,9 @@ func (n *NodeInfo) RemoveNode(node *v1.Node) error {
 	n.node = nil
 	n.allocatableResource = &Resource{}
 	n.allowedPodNumber = 0
+	n.taints, n.taintsErr = nil, nil
+	n.memoryPressureCondition = v1.ConditionUnknown
+	n.diskPressureCondition = v1.ConditionUnknown
 	n.generation++
 	return nil
 }