EvenPodsSpread: PredicateMetadata initilization

- build a new `topologyPairsPodSpreadMap` into PredicateMetadata
- update ShallowCopy()
- unit tests

pkg/scheduler/algorithm/predicates/metadata.go

@@ -34,6 +34,9 @@ import (
 	schedutil "k8s.io/kubernetes/pkg/scheduler/util"
 )
 
+// MaxInt32 is the maximum value of int32
+const MaxInt32 = int32(^uint32(0) >> 1)
+
 // PredicateMetadata interface represents anything that can access a predicate metadata.
 type PredicateMetadata interface {
 	ShallowCopy() PredicateMetadata
@@ -66,6 +69,21 @@ type topologyPairsMaps struct {
 	podToTopologyPairs map[string]topologyPairSet
 }
 
+// topologyPairsPodSpreadMap combines []int32 and topologyPairsMaps to represent
+// (1) how existing pods match incoming pod on its spread constraints
+// (2) minimum match number of each hard spread constraint
+type topologyPairsPodSpreadMap struct {
+	minMatches []int32
+	*topologyPairsMaps
+}
+
+func newTopologyPairsPodSpreadMap() *topologyPairsPodSpreadMap {
+	return &topologyPairsPodSpreadMap{
+		// minMatches will be initilized with proper size later
+		topologyPairsMaps: newTopologyPairsMaps(),
+	}
+}
+
 // NOTE: When new fields are added/removed or logic is changed, please make sure that
 // RemovePod, AddPod, and ShallowCopy functions are updated to work with the new changes.
 type predicateMetadata struct {
@@ -91,6 +109,9 @@ type predicateMetadata struct {
 	// which should be accounted only by the extenders. This set is synthesized
 	// from scheduler extender configuration and does not change per pod.
 	ignoredExtendedResources sets.String
+	// Similar like map for pod (anti-)affinity, but impose additional min matches info
+	// to describe mininum match number on each topology spread constraint
+	topologyPairsPodSpreadMap *topologyPairsPodSpreadMap
 }
 
 // Ensure that predicateMetadata implements algorithm.PredicateMetadata.
@@ -137,17 +158,24 @@ func (pfactory *PredicateMetadataFactory) GetMetadata(pod *v1.Pod, nodeNameToInf
 	if pod == nil {
 		return nil
 	}
+	// existingPodSpreadConstraintsMap represents how existing pods matches "pod"
+	// on its spread constraints
+	existingPodSpreadConstraintsMap, err := getTPMapMatchingSpreadConstraints(pod, nodeNameToInfoMap)
+	if err != nil {
+		klog.Errorf("Error calculating spreadConstraintsMap: %v", err)
+		return nil
+	}
 	// existingPodAntiAffinityMap will be used later for efficient check on existing pods' anti-affinity
 	existingPodAntiAffinityMap, err := getTPMapMatchingExistingAntiAffinity(pod, nodeNameToInfoMap)
 	if err != nil {
-		klog.Errorf("[predicate meta data generation] error finding pods whose affinity terms are matched: %v", err)
+		klog.Errorf("Error calculating existingPodAntiAffinityMap: %v", err)
 		return nil
 	}
 	// incomingPodAffinityMap will be used later for efficient check on incoming pod's affinity
 	// incomingPodAntiAffinityMap will be used later for efficient check on incoming pod's anti-affinity
 	incomingPodAffinityMap, incomingPodAntiAffinityMap, err := getTPMapMatchingIncomingAffinityAntiAffinity(pod, nodeNameToInfoMap)
 	if err != nil {
-		klog.Errorf("[predicate meta data generation] error finding pods that match affinity terms: %v", err)
+		klog.Errorf("Error calculating incomingPod(Anti)AffinityMap: %v", err)
 		return nil
 	}
 	predicateMetadata := &predicateMetadata{
@@ -158,6 +186,7 @@ func (pfactory *PredicateMetadataFactory) GetMetadata(pod *v1.Pod, nodeNameToInf
 		topologyPairsPotentialAffinityPods:     incomingPodAffinityMap,
 		topologyPairsPotentialAntiAffinityPods: incomingPodAntiAffinityMap,
 		topologyPairsAntiAffinityPodsMap:       existingPodAntiAffinityMap,
+		topologyPairsPodSpreadMap:              existingPodSpreadConstraintsMap,
 	}
 	for predicateName, precomputeFunc := range predicateMetadataProducers {
 		klog.V(10).Infof("Precompute: %v", predicateName)
@@ -166,46 +195,224 @@ func (pfactory *PredicateMetadataFactory) GetMetadata(pod *v1.Pod, nodeNameToInf
 	return predicateMetadata
 }
 
+func getTPMapMatchingSpreadConstraints(pod *v1.Pod, nodeInfoMap map[string]*schedulernodeinfo.NodeInfo) (*topologyPairsPodSpreadMap, error) {
+	// we have feature gating in APIserver to strip the spec
+	// so don't need to re-check feature gate, just check length of constraints
+	constraints := getHardTopologySpreadConstraints(pod)
+	if len(constraints) == 0 {
+		return nil, nil
+	}
+
+	allNodeNames := make([]string, 0, len(nodeInfoMap))
+	for name := range nodeInfoMap {
+		allNodeNames = append(allNodeNames, name)
+	}
+
+	var lock sync.Mutex
+	var firstError error
+
+	topologyPairsPodSpreadMap := newTopologyPairsPodSpreadMap()
+
+	appendTopologyPairsMaps := func(toAppend *topologyPairsMaps) {
+		lock.Lock()
+		topologyPairsPodSpreadMap.appendMaps(toAppend)
+		lock.Unlock()
+	}
+	catchError := func(err error) {
+		lock.Lock()
+		if firstError == nil {
+			firstError = err
+		}
+		lock.Unlock()
+	}
+
+	ctx, cancel := context.WithCancel(context.Background())
+
+	processNode := func(i int) {
+		nodeInfo := nodeInfoMap[allNodeNames[i]]
+		node := nodeInfo.Node()
+		if node == nil {
+			catchError(fmt.Errorf("node %q not found", allNodeNames[i]))
+			cancel()
+			return
+		}
+		// Be design if NodeAffinity or NodeSelector is defined, spreading is
+		// applied to nodes that pass those filters.
+		if !podMatchesNodeSelectorAndAffinityTerms(pod, node) {
+			return
+		}
+		// ensure current node's labels contains all topologyKeys in 'constraints'
+		for _, constraint := range constraints {
+			if _, ok := node.Labels[constraint.TopologyKey]; !ok {
+				return
+			}
+		}
+
+		nodeTopologyMaps := newTopologyPairsMaps()
+		// nodeInfo.Pods() can be empty; or all pods don't fit
+		for _, existingPod := range nodeInfo.Pods() {
+			ok, err := podMatchesAllSpreadConstraints(existingPod, pod.Namespace, constraints)
+			if err != nil {
+				catchError(err)
+				cancel()
+				return
+			}
+			if ok {
+				for _, constraint := range constraints {
+					// constraint.TopologyKey is already guaranteed to be present
+					pair := topologyPair{key: constraint.TopologyKey, value: node.Labels[constraint.TopologyKey]}
+					nodeTopologyMaps.addTopologyPair(pair, existingPod)
+				}
+			}
+		}
+		// If needed, append topology pair without entry of pods.
+		// For example, on node-x, there is no pod matching spread constraints
+		// but node-x should be also considered as a match (with match number 0)
+		// i.e. <node: node-x>: {}
+		for _, constraint := range constraints {
+			// constraint.TopologyKey is already guaranteed to be present
+			pair := topologyPair{
+				key:   constraint.TopologyKey,
+				value: node.Labels[constraint.TopologyKey],
+			}
+			// addTopologyPairWithoutPods is a non-op if other pods match this pair
+			nodeTopologyMaps.addTopologyPairWithoutPods(pair)
+		}
+
+		appendTopologyPairsMaps(nodeTopologyMaps)
+	}
+	workqueue.ParallelizeUntil(ctx, 16, len(allNodeNames), processNode)
+
+	if firstError != nil {
+		return nil, firstError
+	}
+
+	// calculate min match for each topology pair
+	topologyPairsPodSpreadMap.minMatches = make([]int32, len(constraints))
+	tpKeyIdx := make(map[string]int)
+	for i, constraint := range constraints {
+		tpKeyIdx[constraint.TopologyKey] = i
+		topologyPairsPodSpreadMap.minMatches[i] = MaxInt32
+	}
+	for pair, podSet := range topologyPairsPodSpreadMap.topologyPairToPods {
+		idx := tpKeyIdx[pair.key]
+		// short circuit if we see 0 as min match of the topologyKey
+		if topologyPairsPodSpreadMap.minMatches[idx] == 0 {
+			continue
+		}
+		if l := int32(len(podSet)); l < topologyPairsPodSpreadMap.minMatches[idx] {
+			topologyPairsPodSpreadMap.minMatches[idx] = l
+		}
+	}
+	return topologyPairsPodSpreadMap, nil
+}
+
+func getHardTopologySpreadConstraints(pod *v1.Pod) (constraints []v1.TopologySpreadConstraint) {
+	if pod != nil {
+		for _, constraint := range pod.Spec.TopologySpreadConstraints {
+			if constraint.WhenUnsatisfiable == v1.DoNotSchedule {
+				constraints = append(constraints, constraint)
+			}
+		}
+	}
+	return
+}
+
+func podMatchesAllSpreadConstraints(pod *v1.Pod, ns string, constraints []v1.TopologySpreadConstraint) (bool, error) {
+	if len(constraints) == 0 || pod.Namespace != ns {
+		return false, nil
+	}
+	return podLabelsMatchesSpreadConstraints(pod.Labels, constraints)
+}
+
+// some corner cases:
+// 1. podLabels = nil, constraint.LabelSelector = nil => returns false
+// 2. podLabels = nil => returns false
+// 3. constraint.LabelSelector = nil => returns false
+func podLabelsMatchesSpreadConstraints(podLabels map[string]string, constraints []v1.TopologySpreadConstraint) (bool, error) {
+	if len(constraints) == 0 {
+		return false, nil
+	}
+	for _, constraint := range constraints {
+		selector, err := metav1.LabelSelectorAsSelector(constraint.LabelSelector)
+		if err != nil {
+			return false, err
+		}
+		if !selector.Matches(labels.Set(podLabels)) {
+			return false, nil
+		}
+	}
+	return true, nil
+}
+
 // returns a pointer to a new topologyPairsMaps
 func newTopologyPairsMaps() *topologyPairsMaps {
 	return &topologyPairsMaps{topologyPairToPods: make(map[topologyPair]podSet),
 		podToTopologyPairs: make(map[string]topologyPairSet)}
 }
 
-func (topologyPairsMaps *topologyPairsMaps) addTopologyPair(pair topologyPair, pod *v1.Pod) {
+func (m *topologyPairsMaps) addTopologyPair(pair topologyPair, pod *v1.Pod) {
 	podFullName := schedutil.GetPodFullName(pod)
-	if topologyPairsMaps.topologyPairToPods[pair] == nil {
+	if m.topologyPairToPods[pair] == nil {
-		topologyPairsMaps.topologyPairToPods[pair] = make(map[*v1.Pod]struct{})
+		m.topologyPairToPods[pair] = make(map[*v1.Pod]struct{})
 	}
-	topologyPairsMaps.topologyPairToPods[pair][pod] = struct{}{}
+	m.topologyPairToPods[pair][pod] = struct{}{}
-	if topologyPairsMaps.podToTopologyPairs[podFullName] == nil {
+	if m.podToTopologyPairs[podFullName] == nil {
-		topologyPairsMaps.podToTopologyPairs[podFullName] = make(map[topologyPair]struct{})
+		m.podToTopologyPairs[podFullName] = make(map[topologyPair]struct{})
 	}
-	topologyPairsMaps.podToTopologyPairs[podFullName][pair] = struct{}{}
+	m.podToTopologyPairs[podFullName][pair] = struct{}{}
 }
 
-func (topologyPairsMaps *topologyPairsMaps) removePod(deletedPod *v1.Pod) {
+// add a topology pair holder if needed
+func (m *topologyPairsMaps) addTopologyPairWithoutPods(pair topologyPair) {
+	if m.topologyPairToPods[pair] == nil {
+		m.topologyPairToPods[pair] = make(map[*v1.Pod]struct{})
+	}
+}
+
+func (m *topologyPairsMaps) removePod(deletedPod *v1.Pod) {
 	deletedPodFullName := schedutil.GetPodFullName(deletedPod)
-	for pair := range topologyPairsMaps.podToTopologyPairs[deletedPodFullName] {
+	for pair := range m.podToTopologyPairs[deletedPodFullName] {
-		delete(topologyPairsMaps.topologyPairToPods[pair], deletedPod)
+		delete(m.topologyPairToPods[pair], deletedPod)
-		if len(topologyPairsMaps.topologyPairToPods[pair]) == 0 {
+		if len(m.topologyPairToPods[pair]) == 0 {
-			delete(topologyPairsMaps.topologyPairToPods, pair)
+			delete(m.topologyPairToPods, pair)
 		}
 	}
-	delete(topologyPairsMaps.podToTopologyPairs, deletedPodFullName)
+	delete(m.podToTopologyPairs, deletedPodFullName)
 }
 
-func (topologyPairsMaps *topologyPairsMaps) appendMaps(toAppend *topologyPairsMaps) {
+func (m *topologyPairsMaps) appendMaps(toAppend *topologyPairsMaps) {
 	if toAppend == nil {
 		return
 	}
 	for pair := range toAppend.topologyPairToPods {
-		for pod := range toAppend.topologyPairToPods[pair] {
+		if podSet := toAppend.topologyPairToPods[pair]; len(podSet) == 0 {
-			topologyPairsMaps.addTopologyPair(pair, pod)
+			m.addTopologyPairWithoutPods(pair)
+		} else {
+			for pod := range podSet {
+				m.addTopologyPair(pair, pod)
+			}
 		}
 	}
 }
 
+func (m *topologyPairsMaps) clone() *topologyPairsMaps {
+	copy := newTopologyPairsMaps()
+	copy.appendMaps(m)
+	return copy
+}
+
+func (podSpreadMap *topologyPairsPodSpreadMap) clone() *topologyPairsPodSpreadMap {
+	// podSpreadMap could be nil when EvenPodsSpread feature is disabled
+	if podSpreadMap == nil {
+		return nil
+	}
+	copy := newTopologyPairsPodSpreadMap()
+	copy.minMatches = append([]int32(nil), podSpreadMap.minMatches...)
+	copy.topologyPairsMaps.appendMaps(podSpreadMap.topologyPairsMaps)
+	return copy
+}
+
 // RemovePod changes predicateMetadata assuming that the given `deletedPod` is
 // deleted from the system.
 func (meta *predicateMetadata) RemovePod(deletedPod *v1.Pod) error {
@@ -301,12 +508,10 @@ func (meta *predicateMetadata) ShallowCopy() PredicateMetadata {
 		ignoredExtendedResources: meta.ignoredExtendedResources,
 	}
 	newPredMeta.podPorts = append([]*v1.ContainerPort(nil), meta.podPorts...)
-	newPredMeta.topologyPairsPotentialAffinityPods = newTopologyPairsMaps()
+	newPredMeta.topologyPairsPotentialAffinityPods = meta.topologyPairsPotentialAffinityPods.clone()
-	newPredMeta.topologyPairsPotentialAffinityPods.appendMaps(meta.topologyPairsPotentialAffinityPods)
+	newPredMeta.topologyPairsPotentialAntiAffinityPods = meta.topologyPairsPotentialAntiAffinityPods.clone()
-	newPredMeta.topologyPairsPotentialAntiAffinityPods = newTopologyPairsMaps()
+	newPredMeta.topologyPairsAntiAffinityPodsMap = meta.topologyPairsAntiAffinityPodsMap.clone()
-	newPredMeta.topologyPairsPotentialAntiAffinityPods.appendMaps(meta.topologyPairsPotentialAntiAffinityPods)
+	newPredMeta.topologyPairsPodSpreadMap = meta.topologyPairsPodSpreadMap.clone()
-	newPredMeta.topologyPairsAntiAffinityPodsMap = newTopologyPairsMaps()
-	newPredMeta.topologyPairsAntiAffinityPodsMap.appendMaps(meta.topologyPairsAntiAffinityPodsMap)
 	newPredMeta.serviceAffinityMatchingPodServices = append([]*v1.Service(nil),
 		meta.serviceAffinityMatchingPodServices...)
 	newPredMeta.serviceAffinityMatchingPodList = append([]*v1.Pod(nil),

pkg/scheduler/algorithm/predicates/metadata_test.go

@@ -511,6 +511,39 @@ func TestPredicateMetadata_ShallowCopy(t *testing.T) {
 				},
 			},
 		},
+		topologyPairsPodSpreadMap: &topologyPairsPodSpreadMap{
+			minMatches: []int32{1},
+			topologyPairsMaps: &topologyPairsMaps{
+				topologyPairToPods: map[topologyPair]podSet{
+					{key: "name", value: "nodeA"}: {
+						&v1.Pod{ObjectMeta: metav1.ObjectMeta{Name: "p1", Labels: selector1},
+							Spec: v1.PodSpec{NodeName: "nodeA"},
+						}: struct{}{},
+					},
+					{key: "name", value: "nodeC"}: {
+						&v1.Pod{ObjectMeta: metav1.ObjectMeta{Name: "p2"},
+							Spec: v1.PodSpec{
+								NodeName: "nodeC",
+							},
+						}: struct{}{},
+						&v1.Pod{ObjectMeta: metav1.ObjectMeta{Name: "p6", Labels: selector1},
+							Spec: v1.PodSpec{NodeName: "nodeC"},
+						}: struct{}{},
+					},
+				},
+				podToTopologyPairs: map[string]topologyPairSet{
+					"p1_": {
+						topologyPair{key: "name", value: "nodeA"}: struct{}{},
+					},
+					"p2_": {
+						topologyPair{key: "name", value: "nodeC"}: struct{}{},
+					},
+					"p6_": {
+						topologyPair{key: "name", value: "nodeC"}: struct{}{},
+					},
+				},
+			},
+		},
 		serviceAffinityInUse: true,
 		serviceAffinityMatchingPodList: []*v1.Pod{
 			{ObjectMeta: metav1.ObjectMeta{Name: "pod1"}},
@@ -791,3 +824,409 @@ func TestGetTPMapMatchingIncomingAffinityAntiAffinity(t *testing.T) {
 		})
 	}
 }
+
+func TestPodLabelsMatchesSpreadConstraints(t *testing.T) {
+	tests := []struct {
+		name        string
+		podLabels   map[string]string
+		constraints []v1.TopologySpreadConstraint
+		want        bool
+		wantErr     bool
+	}{
+		{
+			name:      "normal match",
+			podLabels: map[string]string{"foo": "", "bar": ""},
+			constraints: []v1.TopologySpreadConstraint{
+				{
+					LabelSelector: &metav1.LabelSelector{
+						MatchExpressions: []metav1.LabelSelectorRequirement{
+							{
+								Key:      "foo",
+								Operator: metav1.LabelSelectorOpExists,
+							},
+						},
+					},
+				},
+			},
+			want: true,
+		},
+		{
+			name:      "normal mismatch",
+			podLabels: map[string]string{"foo": "", "baz": ""},
+			constraints: []v1.TopologySpreadConstraint{
+				{
+					LabelSelector: &metav1.LabelSelector{
+						MatchExpressions: []metav1.LabelSelectorRequirement{
+							{
+								Key:      "foo",
+								Operator: metav1.LabelSelectorOpExists,
+							},
+							{
+								Key:      "bar",
+								Operator: metav1.LabelSelectorOpExists,
+							},
+						},
+					},
+				},
+			},
+			want: false,
+		},
+		{
+			name: "podLabels is nil",
+			constraints: []v1.TopologySpreadConstraint{
+				{
+					LabelSelector: &metav1.LabelSelector{
+						MatchExpressions: []metav1.LabelSelectorRequirement{
+							{
+								Key:      "foo",
+								Operator: metav1.LabelSelectorOpExists,
+							},
+						},
+					},
+				},
+			},
+			want: false,
+		},
+		{
+			name: "constraint.LabelSelector is nil",
+			podLabels: map[string]string{
+				"foo": "",
+				"bar": "",
+			},
+			constraints: []v1.TopologySpreadConstraint{
+				{
+					MaxSkew: 1,
+				},
+			},
+			want: false,
+		},
+		{
+			name: "both podLabels and constraint.LabelSelector are nil",
+			constraints: []v1.TopologySpreadConstraint{
+				{
+					MaxSkew: 1,
+				},
+			},
+			want: false,
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			got, err := podLabelsMatchesSpreadConstraints(tt.podLabels, tt.constraints)
+			if (err != nil) != tt.wantErr {
+				t.Errorf("podLabelsMatchesSpreadConstraints() error = %v, wantErr %v", err, tt.wantErr)
+			}
+			if got != tt.want {
+				t.Errorf("podLabelsMatchesSpreadConstraints() = %v, want %v", got, tt.want)
+			}
+		})
+	}
+}
+
+func TestGetTPMapMatchingSpreadConstraints(t *testing.T) {
+	// we need to inject the exact pod pointers to want.topologyPairsMaps.topologyPairToPods
+	// otherwise, *pod (as key of a map) will always fail in reflect.DeepEqual()
+	tests := []struct {
+		name              string
+		pod               *v1.Pod
+		nodes             []*v1.Node
+		existingPods      []*v1.Pod
+		injectPodPointers map[topologyPair][]int
+		want              *topologyPairsPodSpreadMap
+	}{
+		{
+			name: "clean cluster with one spreadConstraint",
+			pod: makePod().name("p").label("foo", "").spreadConstraint(
+				1, "zone", hardSpread, makeLabelSelector().exists("foo").obj(),
+			).obj(),
+			nodes: []*v1.Node{
+				makeNode().name("node-a").label("zone", "zone1").label("node", "node-a").obj(),
+				makeNode().name("node-b").label("zone", "zone1").label("node", "node-b").obj(),
+				makeNode().name("node-x").label("zone", "zone2").label("node", "node-x").obj(),
+				makeNode().name("node-y").label("zone", "zone2").label("node", "node-y").obj(),
+			},
+			injectPodPointers: map[topologyPair][]int{
+				// denotes no existing pod is matched on this zone pair, but still needed to be
+				// calculated if incoming pod matches its own spread constraints
+				{key: "zone", value: "zone1"}: []int{},
+				{key: "zone", value: "zone2"}: []int{},
+			},
+			want: &topologyPairsPodSpreadMap{
+				minMatches: []int32{0},
+				topologyPairsMaps: &topologyPairsMaps{
+					podToTopologyPairs: make(map[string]topologyPairSet),
+				},
+			},
+		},
+		{
+			name: "normal case with one spreadConstraint",
+			pod: makePod().name("p").label("foo", "").spreadConstraint(
+				1, "zone", hardSpread, makeLabelSelector().exists("foo").obj(),
+			).obj(),
+			nodes: []*v1.Node{
+				makeNode().name("node-a").label("zone", "zone1").label("node", "node-a").obj(),
+				makeNode().name("node-b").label("zone", "zone1").label("node", "node-b").obj(),
+				makeNode().name("node-x").label("zone", "zone2").label("node", "node-x").obj(),
+				makeNode().name("node-y").label("zone", "zone2").label("node", "node-y").obj(),
+			},
+			existingPods: []*v1.Pod{
+				makePod().name("p-a1").node("node-a").label("foo", "").obj(),
+				makePod().name("p-a2").node("node-a").label("foo", "").obj(),
+				makePod().name("p-b1").node("node-b").label("foo", "").obj(),
+				makePod().name("p-y1").node("node-y").label("foo", "").obj(),
+				makePod().name("p-y2").node("node-y").label("foo", "").obj(),
+			},
+			injectPodPointers: map[topologyPair][]int{
+				// denotes existingPods[0,1,2]
+				{key: "zone", value: "zone1"}: []int{0, 1, 2},
+				// denotes existingPods[3,4]
+				{key: "zone", value: "zone2"}: []int{3, 4},
+			},
+			want: &topologyPairsPodSpreadMap{
+				minMatches: []int32{2},
+				topologyPairsMaps: &topologyPairsMaps{
+					podToTopologyPairs: map[string]topologyPairSet{
+						"p-a1_": newPairSet("zone", "zone1"),
+						"p-a2_": newPairSet("zone", "zone1"),
+						"p-b1_": newPairSet("zone", "zone1"),
+						"p-y1_": newPairSet("zone", "zone2"),
+						"p-y2_": newPairSet("zone", "zone2"),
+					},
+				},
+			},
+		},
+		{
+			name: "namespace mis-match doesn't count",
+			pod: makePod().name("p").label("foo", "").spreadConstraint(
+				1, "zone", hardSpread, makeLabelSelector().exists("foo").obj(),
+			).obj(),
+			nodes: []*v1.Node{
+				makeNode().name("node-a").label("zone", "zone1").label("node", "node-a").obj(),
+				makeNode().name("node-b").label("zone", "zone1").label("node", "node-b").obj(),
+				makeNode().name("node-x").label("zone", "zone2").label("node", "node-x").obj(),
+				makeNode().name("node-y").label("zone", "zone2").label("node", "node-y").obj(),
+			},
+			existingPods: []*v1.Pod{
+				makePod().name("p-a1").node("node-a").label("foo", "").obj(),
+				makePod().name("p-a2").namespace("ns1").node("node-a").label("foo", "").obj(),
+				makePod().name("p-b1").node("node-b").label("foo", "").obj(),
+				makePod().name("p-y1").namespace("ns2").node("node-y").label("foo", "").obj(),
+				makePod().name("p-y2").node("node-y").label("foo", "").obj(),
+			},
+			injectPodPointers: map[topologyPair][]int{
+				{key: "zone", value: "zone1"}: []int{0, 2},
+				{key: "zone", value: "zone2"}: []int{4},
+			},
+			want: &topologyPairsPodSpreadMap{
+				minMatches: []int32{1},
+				topologyPairsMaps: &topologyPairsMaps{
+					podToTopologyPairs: map[string]topologyPairSet{
+						"p-a1_": newPairSet("zone", "zone1"),
+						"p-b1_": newPairSet("zone", "zone1"),
+						"p-y2_": newPairSet("zone", "zone2"),
+					},
+				},
+			},
+		},
+		{
+			name: "normal case with two spreadConstraints",
+			pod: makePod().name("p").label("foo", "").
+				spreadConstraint(1, "zone", hardSpread, makeLabelSelector().exists("foo").obj()).
+				spreadConstraint(1, "node", hardSpread, makeLabelSelector().exists("foo").obj()).
+				obj(),
+			nodes: []*v1.Node{
+				makeNode().name("node-a").label("zone", "zone1").label("node", "node-a").obj(),
+				makeNode().name("node-b").label("zone", "zone1").label("node", "node-b").obj(),
+				makeNode().name("node-x").label("zone", "zone2").label("node", "node-x").obj(),
+				makeNode().name("node-y").label("zone", "zone2").label("node", "node-y").obj(),
+			},
+			existingPods: []*v1.Pod{
+				makePod().name("p-a1").node("node-a").label("foo", "").obj(),
+				makePod().name("p-a2").node("node-a").label("foo", "").obj(),
+				makePod().name("p-b1").node("node-b").label("foo", "").obj(),
+				makePod().name("p-y1").node("node-y").label("foo", "").obj(),
+				makePod().name("p-y2").node("node-y").label("foo", "").obj(),
+				makePod().name("p-y3").node("node-y").label("foo", "").obj(),
+				makePod().name("p-y4").node("node-y").label("foo", "").obj(),
+			},
+			injectPodPointers: map[topologyPair][]int{
+				{key: "zone", value: "zone1"}:  []int{0, 1, 2},
+				{key: "zone", value: "zone2"}:  []int{3, 4, 5, 6},
+				{key: "node", value: "node-a"}: []int{0, 1},
+				{key: "node", value: "node-b"}: []int{2},
+				{key: "node", value: "node-x"}: []int{},
+				{key: "node", value: "node-y"}: []int{3, 4, 5, 6},
+			},
+			want: &topologyPairsPodSpreadMap{
+				minMatches: []int32{3, 0},
+				topologyPairsMaps: &topologyPairsMaps{
+					podToTopologyPairs: map[string]topologyPairSet{
+						"p-a1_": newPairSet("zone", "zone1", "node", "node-a"),
+						"p-a2_": newPairSet("zone", "zone1", "node", "node-a"),
+						"p-b1_": newPairSet("zone", "zone1", "node", "node-b"),
+						"p-y1_": newPairSet("zone", "zone2", "node", "node-y"),
+						"p-y2_": newPairSet("zone", "zone2", "node", "node-y"),
+						"p-y3_": newPairSet("zone", "zone2", "node", "node-y"),
+						"p-y4_": newPairSet("zone", "zone2", "node", "node-y"),
+					},
+				},
+			},
+		},
+		{
+			name: "soft spreadConstraints should be bypassed",
+			pod: makePod().name("p").label("foo", "").
+				spreadConstraint(1, "zone", softSpread, makeLabelSelector().exists("foo").obj()).
+				spreadConstraint(1, "zone", hardSpread, makeLabelSelector().exists("foo").obj()).
+				spreadConstraint(1, "zone", softSpread, makeLabelSelector().exists("foo").obj()).
+				spreadConstraint(1, "node", hardSpread, makeLabelSelector().exists("foo").obj()).
+				obj(),
+			nodes: []*v1.Node{
+				makeNode().name("node-a").label("zone", "zone1").label("node", "node-a").obj(),
+				makeNode().name("node-b").label("zone", "zone1").label("node", "node-b").obj(),
+				makeNode().name("node-y").label("zone", "zone2").label("node", "node-y").obj(),
+			},
+			existingPods: []*v1.Pod{
+				makePod().name("p-a1").node("node-a").label("foo", "").obj(),
+				makePod().name("p-a2").node("node-a").label("foo", "").obj(),
+				makePod().name("p-b1").node("node-b").label("foo", "").obj(),
+				makePod().name("p-y1").node("node-y").label("foo", "").obj(),
+				makePod().name("p-y2").node("node-y").label("foo", "").obj(),
+				makePod().name("p-y3").node("node-y").label("foo", "").obj(),
+				makePod().name("p-y4").node("node-y").label("foo", "").obj(),
+			},
+			injectPodPointers: map[topologyPair][]int{
+				{key: "zone", value: "zone1"}:  []int{0, 1, 2},
+				{key: "zone", value: "zone2"}:  []int{3, 4, 5, 6},
+				{key: "node", value: "node-a"}: []int{0, 1},
+				{key: "node", value: "node-b"}: []int{2},
+				{key: "node", value: "node-y"}: []int{3, 4, 5, 6},
+			},
+			want: &topologyPairsPodSpreadMap{
+				minMatches: []int32{3, 1},
+				topologyPairsMaps: &topologyPairsMaps{
+					podToTopologyPairs: map[string]topologyPairSet{
+						"p-a1_": newPairSet("zone", "zone1", "node", "node-a"),
+						"p-a2_": newPairSet("zone", "zone1", "node", "node-a"),
+						"p-b1_": newPairSet("zone", "zone1", "node", "node-b"),
+						"p-y1_": newPairSet("zone", "zone2", "node", "node-y"),
+						"p-y2_": newPairSet("zone", "zone2", "node", "node-y"),
+						"p-y3_": newPairSet("zone", "zone2", "node", "node-y"),
+						"p-y4_": newPairSet("zone", "zone2", "node", "node-y"),
+					},
+				},
+			},
+		},
+		{
+			name: "different labelSelectors",
+			pod: makePod().name("p").label("foo", "").label("bar", "").
+				spreadConstraint(1, "zone", hardSpread, makeLabelSelector().exists("foo").obj()).
+				spreadConstraint(1, "node", hardSpread, makeLabelSelector().exists("bar").obj()).
+				obj(),
+			nodes: []*v1.Node{
+				makeNode().name("node-a").label("zone", "zone1").label("node", "node-a").obj(),
+				makeNode().name("node-b").label("zone", "zone1").label("node", "node-b").obj(),
+				makeNode().name("node-y").label("zone", "zone2").label("node", "node-y").obj(),
+			},
+			existingPods: []*v1.Pod{
+				makePod().name("p-a1").node("node-a").label("foo", "").obj(),
+				makePod().name("p-a2").node("node-a").label("foo", "").label("bar", "").obj(),
+				makePod().name("p-b1").node("node-b").label("foo", "").obj(),
+				makePod().name("p-y1").node("node-y").label("foo", "").obj(),
+				makePod().name("p-y2").node("node-y").label("foo", "").label("bar", "").obj(),
+				makePod().name("p-y3").node("node-y").label("foo", "").obj(),
+				makePod().name("p-y4").node("node-y").label("foo", "").label("bar", "").obj(),
+			},
+			injectPodPointers: map[topologyPair][]int{
+				{key: "zone", value: "zone1"}:  []int{1},
+				{key: "zone", value: "zone2"}:  []int{4, 6},
+				{key: "node", value: "node-a"}: []int{1},
+				{key: "node", value: "node-b"}: []int{},
+				{key: "node", value: "node-y"}: []int{4, 6},
+			},
+			want: &topologyPairsPodSpreadMap{
+				minMatches: []int32{1, 0},
+				topologyPairsMaps: &topologyPairsMaps{
+					podToTopologyPairs: map[string]topologyPairSet{
+						"p-a2_": newPairSet("zone", "zone1", "node", "node-a"),
+						"p-y2_": newPairSet("zone", "zone2", "node", "node-y"),
+						"p-y4_": newPairSet("zone", "zone2", "node", "node-y"),
+					},
+				},
+			},
+		},
+		{
+			name: "two spreadConstraints, and with podAffinity",
+			pod: makePod().name("p").label("foo", "").
+				nodeAffinityIn("node", []string{"node-a", "node-b", "node-y"}). // exclude node-x
+				spreadConstraint(1, "zone", hardSpread, makeLabelSelector().exists("foo").obj()).
+				spreadConstraint(1, "node", hardSpread, makeLabelSelector().exists("foo").obj()).
+				obj(),
+			nodes: []*v1.Node{
+				makeNode().name("node-a").label("zone", "zone1").label("node", "node-a").obj(),
+				makeNode().name("node-b").label("zone", "zone1").label("node", "node-b").obj(),
+				makeNode().name("node-x").label("zone", "zone2").label("node", "node-x").obj(),
+				makeNode().name("node-y").label("zone", "zone2").label("node", "node-y").obj(),
+			},
+			existingPods: []*v1.Pod{
+				makePod().name("p-a1").node("node-a").label("foo", "").obj(),
+				makePod().name("p-a2").node("node-a").label("foo", "").obj(),
+				makePod().name("p-b1").node("node-b").label("foo", "").obj(),
+				makePod().name("p-y1").node("node-y").label("foo", "").obj(),
+				makePod().name("p-y2").node("node-y").label("foo", "").obj(),
+				makePod().name("p-y3").node("node-y").label("foo", "").obj(),
+				makePod().name("p-y4").node("node-y").label("foo", "").obj(),
+			},
+			injectPodPointers: map[topologyPair][]int{
+				{key: "zone", value: "zone1"}:  []int{0, 1, 2},
+				{key: "zone", value: "zone2"}:  []int{3, 4, 5, 6},
+				{key: "node", value: "node-a"}: []int{0, 1},
+				{key: "node", value: "node-b"}: []int{2},
+				{key: "node", value: "node-y"}: []int{3, 4, 5, 6},
+			},
+			want: &topologyPairsPodSpreadMap{
+				minMatches: []int32{3, 1},
+				topologyPairsMaps: &topologyPairsMaps{
+					podToTopologyPairs: map[string]topologyPairSet{
+						"p-a1_": newPairSet("zone", "zone1", "node", "node-a"),
+						"p-a2_": newPairSet("zone", "zone1", "node", "node-a"),
+						"p-b1_": newPairSet("zone", "zone1", "node", "node-b"),
+						"p-y1_": newPairSet("zone", "zone2", "node", "node-y"),
+						"p-y2_": newPairSet("zone", "zone2", "node", "node-y"),
+						"p-y3_": newPairSet("zone", "zone2", "node", "node-y"),
+						"p-y4_": newPairSet("zone", "zone2", "node", "node-y"),
+					},
+				},
+			},
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			tt.want.topologyPairToPods = make(map[topologyPair]podSet)
+			for pair, indexes := range tt.injectPodPointers {
+				pSet := make(podSet)
+				for _, i := range indexes {
+					pSet[tt.existingPods[i]] = struct{}{}
+				}
+				tt.want.topologyPairToPods[pair] = pSet
+			}
+			nodeInfoMap := schedulernodeinfo.CreateNodeNameToInfoMap(tt.existingPods, tt.nodes)
+			if got, _ := getTPMapMatchingSpreadConstraints(tt.pod, nodeInfoMap); !reflect.DeepEqual(got, tt.want) {
+				t.Errorf("getTPMapMatchingSpreadConstraints() = %v, want %v", got, tt.want)
+			}
+		})
+	}
+}
+
+var (
+	hardSpread = v1.DoNotSchedule
+	softSpread = v1.ScheduleAnyway
+)
+
+func newPairSet(kv ...string) topologyPairSet {
+	result := make(topologyPairSet)
+	for i := 0; i < len(kv); i += 2 {
+		pair := topologyPair{key: kv[i], value: kv[i+1]}
+		result[pair] = struct{}{}
+	}
+	return result
+}

pkg/scheduler/algorithm/predicates/utils.go

@@ -19,8 +19,9 @@ package predicates
 import (
 	"strings"
 
-	v1 "k8s.io/api/core/v1"
+	"k8s.io/api/core/v1"
 	storagev1beta1 "k8s.io/api/storage/v1beta1"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/labels"
 	"k8s.io/apimachinery/pkg/util/sets"
 	utilfeature "k8s.io/apiserver/pkg/util/feature"
@@ -147,3 +148,169 @@ func isCSIMigrationOn(csiNode *storagev1beta1.CSINode, pluginName string) bool {
 
 	return mpaSet.Has(pluginName)
 }
+
+// utilities for building pod/node objects using a "chained" manner
+type nodeSelectorWrapper struct{ v1.NodeSelector }
+
+func makeNodeSelector() *nodeSelectorWrapper {
+	return &nodeSelectorWrapper{v1.NodeSelector{}}
+}
+
+// NOTE: each time we append a selectorTerm into `s`
+// and overall all selecterTerms are ORed
+func (s *nodeSelectorWrapper) in(key string, vals []string) *nodeSelectorWrapper {
+	expression := v1.NodeSelectorRequirement{
+		Key:      key,
+		Operator: v1.NodeSelectorOpIn,
+		Values:   vals,
+	}
+	selectorTerm := v1.NodeSelectorTerm{}
+	selectorTerm.MatchExpressions = append(selectorTerm.MatchExpressions, expression)
+	s.NodeSelectorTerms = append(s.NodeSelectorTerms, selectorTerm)
+	return s
+}
+
+func (s *nodeSelectorWrapper) obj() *v1.NodeSelector {
+	return &s.NodeSelector
+}
+
+type labelSelectorWrapper struct{ metav1.LabelSelector }
+
+func makeLabelSelector() *labelSelectorWrapper {
+	return &labelSelectorWrapper{metav1.LabelSelector{}}
+}
+
+func (s *labelSelectorWrapper) label(k, v string) *labelSelectorWrapper {
+	if s.MatchLabels == nil {
+		s.MatchLabels = make(map[string]string)
+	}
+	s.MatchLabels[k] = v
+	return s
+}
+
+func (s *labelSelectorWrapper) in(key string, vals []string) *labelSelectorWrapper {
+	expression := metav1.LabelSelectorRequirement{
+		Key:      key,
+		Operator: metav1.LabelSelectorOpIn,
+		Values:   vals,
+	}
+	s.MatchExpressions = append(s.MatchExpressions, expression)
+	return s
+}
+
+func (s *labelSelectorWrapper) notIn(key string, vals []string) *labelSelectorWrapper {
+	expression := metav1.LabelSelectorRequirement{
+		Key:      key,
+		Operator: metav1.LabelSelectorOpNotIn,
+		Values:   vals,
+	}
+	s.MatchExpressions = append(s.MatchExpressions, expression)
+	return s
+}
+
+func (s *labelSelectorWrapper) exists(k string) *labelSelectorWrapper {
+	expression := metav1.LabelSelectorRequirement{
+		Key:      k,
+		Operator: metav1.LabelSelectorOpExists,
+	}
+	s.MatchExpressions = append(s.MatchExpressions, expression)
+	return s
+}
+
+func (s *labelSelectorWrapper) notExist(k string) *labelSelectorWrapper {
+	expression := metav1.LabelSelectorRequirement{
+		Key:      k,
+		Operator: metav1.LabelSelectorOpDoesNotExist,
+	}
+	s.MatchExpressions = append(s.MatchExpressions, expression)
+	return s
+}
+
+func (s *labelSelectorWrapper) obj() *metav1.LabelSelector {
+	return &s.LabelSelector
+}
+
+type podWrapper struct{ v1.Pod }
+
+func makePod() *podWrapper {
+	return &podWrapper{v1.Pod{}}
+}
+
+func (p *podWrapper) obj() *v1.Pod {
+	return &p.Pod
+}
+
+func (p *podWrapper) name(s string) *podWrapper {
+	p.Name = s
+	return p
+}
+
+func (p *podWrapper) namespace(s string) *podWrapper {
+	p.Namespace = s
+	return p
+}
+
+func (p *podWrapper) node(s string) *podWrapper {
+	p.Spec.NodeName = s
+	return p
+}
+
+func (p *podWrapper) nodeSelector(m map[string]string) *podWrapper {
+	p.Spec.NodeSelector = m
+	return p
+}
+
+// particular represents HARD node affinity
+func (p *podWrapper) nodeAffinityIn(key string, vals []string) *podWrapper {
+	if p.Spec.Affinity == nil {
+		p.Spec.Affinity = &v1.Affinity{}
+	}
+	if p.Spec.Affinity.NodeAffinity == nil {
+		p.Spec.Affinity.NodeAffinity = &v1.NodeAffinity{}
+	}
+	nodeSelector := makeNodeSelector().in(key, vals).obj()
+	p.Spec.Affinity.NodeAffinity.RequiredDuringSchedulingIgnoredDuringExecution = nodeSelector
+	return p
+}
+
+func (p *podWrapper) spreadConstraint(maxSkew int, tpKey string, mode v1.UnsatisfiableConstraintResponse, selector *metav1.LabelSelector) *podWrapper {
+	c := v1.TopologySpreadConstraint{
+		MaxSkew:           int32(maxSkew),
+		TopologyKey:       tpKey,
+		WhenUnsatisfiable: mode,
+		LabelSelector:     selector,
+	}
+	p.Spec.TopologySpreadConstraints = append(p.Spec.TopologySpreadConstraints, c)
+	return p
+}
+
+func (p *podWrapper) label(k, v string) *podWrapper {
+	if p.Labels == nil {
+		p.Labels = make(map[string]string)
+	}
+	p.Labels[k] = v
+	return p
+}
+
+type nodeWrapper struct{ v1.Node }
+
+func makeNode() *nodeWrapper {
+	return &nodeWrapper{v1.Node{}}
+}
+
+func (n *nodeWrapper) obj() *v1.Node {
+	return &n.Node
+}
+
+func (n *nodeWrapper) name(s string) *nodeWrapper {
+	n.Name = s
+	return n
+}
+
+func (n *nodeWrapper) label(k, v string) *nodeWrapper {
+	if n.Labels == nil {
+		n.Labels = make(map[string]string)
+	}
+	n.Labels[k] = v
+	return n
+}