Merge pull request #58717 from resouer/extender-interface

Automatic merge from submit-queue (batch tested with PRs 60759, 60531, 60923, 60851, 58717). If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.

Implement preemption for extender with a verb and new interface

**What this PR does / why we need it**:

This is an alternative way of implementing #51656

**Which issue(s) this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close the issue(s) when PR gets merged)*:
Fixes #51656

**Special notes for your reviewer**:

We will also want to compare with #56296 to see which one is the best solution. See: https://github.com/kubernetes/kubernetes/pull/56296#discussion_r163381235

cc @ravigadde @bsalamat 

**Release note**:

```release-note
Implement preemption for extender with a verb and new interface
```

pkg/scheduler/algorithm/scheduler_interface.go

@@ -29,7 +29,9 @@ type SchedulerExtender interface {
 	// Filter based on extender-implemented predicate functions. The filtered list is
 	// expected to be a subset of the supplied list. failedNodesMap optionally contains
 	// the list of failed nodes and failure reasons.
-	Filter(pod *v1.Pod, nodes []*v1.Node, nodeNameToInfo map[string]*schedulercache.NodeInfo) (filteredNodes []*v1.Node, failedNodesMap schedulerapi.FailedNodesMap, err error)
+	Filter(pod *v1.Pod,
+		nodes []*v1.Node, nodeNameToInfo map[string]*schedulercache.NodeInfo,
+	) (filteredNodes []*v1.Node, failedNodesMap schedulerapi.FailedNodesMap, err error)
 
 	// Prioritize based on extender-implemented priority functions. The returned scores & weight
 	// are used to compute the weighted score for an extender. The weighted scores are added to
@@ -45,6 +47,23 @@ type SchedulerExtender interface {
 	// IsInterested returns true if at least one extended resource requested by
 	// this pod is managed by this extender.
 	IsInterested(pod *v1.Pod) bool
+
+	// ProcessPreemption returns nodes with their victim pods processed by extender based on
+	// given:
+	//   1. Pod to schedule
+	//   2. Candidate nodes and victim pods (nodeToVictims) generated by previous scheduling process.
+	//   3. nodeNameToInfo to restore v1.Node from node name if extender cache is enabled.
+	// The possible changes made by extender may include:
+	//   1. Subset of given candidate nodes after preemption phase of extender.
+	//   2. A different set of victim pod for every given candidate node after preemption phase of extender.
+	ProcessPreemption(
+		pod *v1.Pod,
+		nodeToVictims map[*v1.Node]*schedulerapi.Victims,
+		nodeNameToInfo map[string]*schedulercache.NodeInfo,
+	) (map[*v1.Node]*schedulerapi.Victims, error)
+
+	// SupportsPreemption returns if the scheduler extender support preemption or not.
+	SupportsPreemption() bool
 }
 
 // ScheduleAlgorithm is an interface implemented by things that know how to schedule pods

pkg/scheduler/api/types.go

@@ -168,6 +168,8 @@ type ExtenderConfig struct {
 	URLPrefix string
 	// Verb for the filter call, empty if not supported. This verb is appended to the URLPrefix when issuing the filter call to extender.
 	FilterVerb string
+	// Verb for the preempt call, empty if not supported. This verb is appended to the URLPrefix when issuing the preempt call to extender.
+	PreemptVerb string
 	// Verb for the prioritize call, empty if not supported. This verb is appended to the URLPrefix when issuing the prioritize call to extender.
 	PrioritizeVerb string
 	// The numeric multiplier for the node scores that the prioritize call generates.
@@ -200,11 +202,48 @@ type ExtenderConfig struct {
 	ManagedResources []ExtenderManagedResource
 }
 
+// ExtenderPreemptionResult represents the result returned by preemption phase of extender.
+type ExtenderPreemptionResult struct {
+	NodeNameToMetaVictims map[string]*MetaVictims
+}
+
+// ExtenderPreemptionArgs represents the arguments needed by the extender to preempt pods on nodes.
+type ExtenderPreemptionArgs struct {
+	// Pod being scheduled
+	Pod *v1.Pod
+	// Victims map generated by scheduler preemption phase
+	// Only set NodeNameToMetaVictims if ExtenderConfig.NodeCacheCapable == true. Otherwise, only set NodeNameToVictims.
+	NodeNameToVictims     map[string]*Victims
+	NodeNameToMetaVictims map[string]*MetaVictims
+}
+
+// Victims represents:
+//   pods:  a group of pods expected to be preempted.
+//   numPDBViolations: the count of violations of PodDisruptionBudget
+type Victims struct {
+	Pods             []*v1.Pod
+	NumPDBViolations int
+}
+
+// MetaPod represent identifier for a v1.Pod
+type MetaPod struct {
+	UID string
+}
+
+// MetaVictims represents:
+//   pods:  a group of pods expected to be preempted.
+//     Only Pod identifiers will be sent and user are expect to get v1.Pod in their own way.
+//   numPDBViolations: the count of violations of PodDisruptionBudget
+type MetaVictims struct {
+	Pods             []*MetaPod
+	NumPDBViolations int
+}
+
 // ExtenderArgs represents the arguments needed by the extender to filter/prioritize
 // nodes for a pod.
 type ExtenderArgs struct {
 	// Pod being scheduled
-	Pod v1.Pod
+	Pod *v1.Pod
 	// List of candidate nodes where the pod can be scheduled; to be populated
 	// only if ExtenderConfig.NodeCacheCapable == false
 	Nodes *v1.NodeList

pkg/scheduler/api/v1/types.go

@@ -142,6 +142,8 @@ type ExtenderConfig struct {
 	URLPrefix string `json:"urlPrefix"`
 	// Verb for the filter call, empty if not supported. This verb is appended to the URLPrefix when issuing the filter call to extender.
 	FilterVerb string `json:"filterVerb,omitempty"`
+	// Verb for the preempt call, empty if not supported. This verb is appended to the URLPrefix when issuing the preempt call to extender.
+	PreemptVerb string `json:"preemptVerb,omitempty"`
 	// Verb for the prioritize call, empty if not supported. This verb is appended to the URLPrefix when issuing the prioritize call to extender.
 	PrioritizeVerb string `json:"prioritizeVerb,omitempty"`
 	// The numeric multiplier for the node scores that the prioritize call generates.
@@ -178,7 +180,7 @@ type ExtenderConfig struct {
 // nodes for a pod.
 type ExtenderArgs struct {
 	// Pod being scheduled
-	Pod apiv1.Pod `json:"pod"`
+	Pod *apiv1.Pod `json:"pod"`
 	// List of candidate nodes where the pod can be scheduled; to be populated
 	// only if ExtenderConfig.NodeCacheCapable == false
 	Nodes *apiv1.NodeList `json:"nodes,omitempty"`
@@ -187,6 +189,43 @@ type ExtenderArgs struct {
 	NodeNames *[]string `json:"nodenames,omitempty"`
 }
 
+// ExtenderPreemptionResult represents the result returned by preemption phase of extender.
+type ExtenderPreemptionResult struct {
+	NodeNameToMetaVictims map[string]*MetaVictims `json:"nodeNameToMetaVictims,omitempty"`
+}
+
+// ExtenderPreemptionArgs represents the arguments needed by the extender to preempt pods on nodes.
+type ExtenderPreemptionArgs struct {
+	// Pod being scheduled
+	Pod *apiv1.Pod `json:"pod"`
+	// Victims map generated by scheduler preemption phase
+	// Only set NodeNameToMetaVictims if ExtenderConfig.NodeCacheCapable == true. Otherwise, only set NodeNameToVictims.
+	NodeNameToVictims     map[string]*Victims     `json:"nodeToVictims,omitempty"`
+	NodeNameToMetaVictims map[string]*MetaVictims `json:"nodeNameToMetaVictims,omitempty"`
+}
+
+// Victims represents:
+//   pods:  a group of pods expected to be preempted.
+//   numPDBViolations: the count of violations of PodDisruptionBudget
+type Victims struct {
+	Pods             []*apiv1.Pod `json:"pods"`
+	NumPDBViolations int          `json:"numPDBViolations"`
+}
+
+// MetaPod represent identifier for a v1.Pod
+type MetaPod struct {
+	UID string `json:"uid"`
+}
+
+// MetaVictims represents:
+//   pods:  a group of pods expected to be preempted.
+//     Only Pod identifiers will be sent and user are expect to get v1.Pod in their own way.
+//   numPDBViolations: the count of violations of PodDisruptionBudget
+type MetaVictims struct {
+	Pods             []*MetaPod `json:"pods"`
+	NumPDBViolations int        `json:"numPDBViolations"`
+}
+
 // FailedNodesMap represents the filtered out nodes, with node names and failure messages
 type FailedNodesMap map[string]string
 

pkg/scheduler/api/v1/zz_generated.deepcopy.go

@@ -29,7 +29,15 @@ import (
 // DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
 func (in *ExtenderArgs) DeepCopyInto(out *ExtenderArgs) {
 	*out = *in
-	in.Pod.DeepCopyInto(&out.Pod)
+	if in.Pod != nil {
+		in, out := &in.Pod, &out.Pod
+		if *in == nil {
+			*out = nil
+		} else {
+			*out = new(core_v1.Pod)
+			(*in).DeepCopyInto(*out)
+		}
+	}
 	if in.Nodes != nil {
 		in, out := &in.Nodes, &out.Nodes
 		if *in == nil {
@@ -188,6 +196,83 @@ func (in *ExtenderManagedResource) DeepCopy() *ExtenderManagedResource {
 	return out
 }
 
+// DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
+func (in *ExtenderPreemptionArgs) DeepCopyInto(out *ExtenderPreemptionArgs) {
+	*out = *in
+	if in.Pod != nil {
+		in, out := &in.Pod, &out.Pod
+		if *in == nil {
+			*out = nil
+		} else {
+			*out = new(core_v1.Pod)
+			(*in).DeepCopyInto(*out)
+		}
+	}
+	if in.NodeNameToVictims != nil {
+		in, out := &in.NodeNameToVictims, &out.NodeNameToVictims
+		*out = make(map[string]*Victims, len(*in))
+		for key, val := range *in {
+			if val == nil {
+				(*out)[key] = nil
+			} else {
+				(*out)[key] = new(Victims)
+				val.DeepCopyInto((*out)[key])
+			}
+		}
+	}
+	if in.NodeNameToMetaVictims != nil {
+		in, out := &in.NodeNameToMetaVictims, &out.NodeNameToMetaVictims
+		*out = make(map[string]*MetaVictims, len(*in))
+		for key, val := range *in {
+			if val == nil {
+				(*out)[key] = nil
+			} else {
+				(*out)[key] = new(MetaVictims)
+				val.DeepCopyInto((*out)[key])
+			}
+		}
+	}
+	return
+}
+
+// DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new ExtenderPreemptionArgs.
+func (in *ExtenderPreemptionArgs) DeepCopy() *ExtenderPreemptionArgs {
+	if in == nil {
+		return nil
+	}
+	out := new(ExtenderPreemptionArgs)
+	in.DeepCopyInto(out)
+	return out
+}
+
+// DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
+func (in *ExtenderPreemptionResult) DeepCopyInto(out *ExtenderPreemptionResult) {
+	*out = *in
+	if in.NodeNameToMetaVictims != nil {
+		in, out := &in.NodeNameToMetaVictims, &out.NodeNameToMetaVictims
+		*out = make(map[string]*MetaVictims, len(*in))
+		for key, val := range *in {
+			if val == nil {
+				(*out)[key] = nil
+			} else {
+				(*out)[key] = new(MetaVictims)
+				val.DeepCopyInto((*out)[key])
+			}
+		}
+	}
+	return
+}
+
+// DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new ExtenderPreemptionResult.
+func (in *ExtenderPreemptionResult) DeepCopy() *ExtenderPreemptionResult {
+	if in == nil {
+		return nil
+	}
+	out := new(ExtenderPreemptionResult)
+	in.DeepCopyInto(out)
+	return out
+}
+
 // DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
 func (in FailedNodesMap) DeepCopyInto(out *FailedNodesMap) {
 	{
@@ -283,6 +368,50 @@ func (in *LabelsPresence) DeepCopy() *LabelsPresence {
 	return out
 }
 
+// DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
+func (in *MetaPod) DeepCopyInto(out *MetaPod) {
+	*out = *in
+	return
+}
+
+// DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new MetaPod.
+func (in *MetaPod) DeepCopy() *MetaPod {
+	if in == nil {
+		return nil
+	}
+	out := new(MetaPod)
+	in.DeepCopyInto(out)
+	return out
+}
+
+// DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
+func (in *MetaVictims) DeepCopyInto(out *MetaVictims) {
+	*out = *in
+	if in.Pods != nil {
+		in, out := &in.Pods, &out.Pods
+		*out = make([]*MetaPod, len(*in))
+		for i := range *in {
+			if (*in)[i] == nil {
+				(*out)[i] = nil
+			} else {
+				(*out)[i] = new(MetaPod)
+				(*in)[i].DeepCopyInto((*out)[i])
+			}
+		}
+	}
+	return
+}
+
+// DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new MetaVictims.
+func (in *MetaVictims) DeepCopy() *MetaVictims {
+	if in == nil {
+		return nil
+	}
+	out := new(MetaVictims)
+	in.DeepCopyInto(out)
+	return out
+}
+
 // DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
 func (in *Policy) DeepCopyInto(out *Policy) {
 	*out = *in
@@ -483,3 +612,31 @@ func (in *ServiceAntiAffinity) DeepCopy() *ServiceAntiAffinity {
 	in.DeepCopyInto(out)
 	return out
 }
+
+// DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
+func (in *Victims) DeepCopyInto(out *Victims) {
+	*out = *in
+	if in.Pods != nil {
+		in, out := &in.Pods, &out.Pods
+		*out = make([]*core_v1.Pod, len(*in))
+		for i := range *in {
+			if (*in)[i] == nil {
+				(*out)[i] = nil
+			} else {
+				(*out)[i] = new(core_v1.Pod)
+				(*in)[i].DeepCopyInto((*out)[i])
+			}
+		}
+	}
+	return
+}
+
+// DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new Victims.
+func (in *Victims) DeepCopy() *Victims {
+	if in == nil {
+		return nil
+	}
+	out := new(Victims)
+	in.DeepCopyInto(out)
+	return out
+}

pkg/scheduler/api/validation/validation_test.go

@@ -61,6 +61,10 @@ func TestValidatePolicy(t *testing.T) {
 			policy:   api.Policy{ExtenderConfigs: []api.ExtenderConfig{{URLPrefix: "http://127.0.0.1:8081/extender", FilterVerb: "filter"}}},
 			expected: nil,
 		},
+		{
+			policy:   api.Policy{ExtenderConfigs: []api.ExtenderConfig{{URLPrefix: "http://127.0.0.1:8081/extender", PreemptVerb: "preempt"}}},
+			expected: nil,
+		},
 		{
 			policy: api.Policy{
 				ExtenderConfigs: []api.ExtenderConfig{

pkg/scheduler/api/zz_generated.deepcopy.go

@@ -29,7 +29,15 @@ import (
 // DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
 func (in *ExtenderArgs) DeepCopyInto(out *ExtenderArgs) {
 	*out = *in
-	in.Pod.DeepCopyInto(&out.Pod)
+	if in.Pod != nil {
+		in, out := &in.Pod, &out.Pod
+		if *in == nil {
+			*out = nil
+		} else {
+			*out = new(v1.Pod)
+			(*in).DeepCopyInto(*out)
+		}
+	}
 	if in.Nodes != nil {
 		in, out := &in.Nodes, &out.Nodes
 		if *in == nil {
@@ -188,6 +196,83 @@ func (in *ExtenderManagedResource) DeepCopy() *ExtenderManagedResource {
 	return out
 }
 
+// DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
+func (in *ExtenderPreemptionArgs) DeepCopyInto(out *ExtenderPreemptionArgs) {
+	*out = *in
+	if in.Pod != nil {
+		in, out := &in.Pod, &out.Pod
+		if *in == nil {
+			*out = nil
+		} else {
+			*out = new(v1.Pod)
+			(*in).DeepCopyInto(*out)
+		}
+	}
+	if in.NodeNameToVictims != nil {
+		in, out := &in.NodeNameToVictims, &out.NodeNameToVictims
+		*out = make(map[string]*Victims, len(*in))
+		for key, val := range *in {
+			if val == nil {
+				(*out)[key] = nil
+			} else {
+				(*out)[key] = new(Victims)
+				val.DeepCopyInto((*out)[key])
+			}
+		}
+	}
+	if in.NodeNameToMetaVictims != nil {
+		in, out := &in.NodeNameToMetaVictims, &out.NodeNameToMetaVictims
+		*out = make(map[string]*MetaVictims, len(*in))
+		for key, val := range *in {
+			if val == nil {
+				(*out)[key] = nil
+			} else {
+				(*out)[key] = new(MetaVictims)
+				val.DeepCopyInto((*out)[key])
+			}
+		}
+	}
+	return
+}
+
+// DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new ExtenderPreemptionArgs.
+func (in *ExtenderPreemptionArgs) DeepCopy() *ExtenderPreemptionArgs {
+	if in == nil {
+		return nil
+	}
+	out := new(ExtenderPreemptionArgs)
+	in.DeepCopyInto(out)
+	return out
+}
+
+// DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
+func (in *ExtenderPreemptionResult) DeepCopyInto(out *ExtenderPreemptionResult) {
+	*out = *in
+	if in.NodeNameToMetaVictims != nil {
+		in, out := &in.NodeNameToMetaVictims, &out.NodeNameToMetaVictims
+		*out = make(map[string]*MetaVictims, len(*in))
+		for key, val := range *in {
+			if val == nil {
+				(*out)[key] = nil
+			} else {
+				(*out)[key] = new(MetaVictims)
+				val.DeepCopyInto((*out)[key])
+			}
+		}
+	}
+	return
+}
+
+// DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new ExtenderPreemptionResult.
+func (in *ExtenderPreemptionResult) DeepCopy() *ExtenderPreemptionResult {
+	if in == nil {
+		return nil
+	}
+	out := new(ExtenderPreemptionResult)
+	in.DeepCopyInto(out)
+	return out
+}
+
 // DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
 func (in FailedNodesMap) DeepCopyInto(out *FailedNodesMap) {
 	{
@@ -283,6 +368,50 @@ func (in *LabelsPresence) DeepCopy() *LabelsPresence {
 	return out
 }
 
+// DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
+func (in *MetaPod) DeepCopyInto(out *MetaPod) {
+	*out = *in
+	return
+}
+
+// DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new MetaPod.
+func (in *MetaPod) DeepCopy() *MetaPod {
+	if in == nil {
+		return nil
+	}
+	out := new(MetaPod)
+	in.DeepCopyInto(out)
+	return out
+}
+
+// DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
+func (in *MetaVictims) DeepCopyInto(out *MetaVictims) {
+	*out = *in
+	if in.Pods != nil {
+		in, out := &in.Pods, &out.Pods
+		*out = make([]*MetaPod, len(*in))
+		for i := range *in {
+			if (*in)[i] == nil {
+				(*out)[i] = nil
+			} else {
+				(*out)[i] = new(MetaPod)
+				(*in)[i].DeepCopyInto((*out)[i])
+			}
+		}
+	}
+	return
+}
+
+// DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new MetaVictims.
+func (in *MetaVictims) DeepCopy() *MetaVictims {
+	if in == nil {
+		return nil
+	}
+	out := new(MetaVictims)
+	in.DeepCopyInto(out)
+	return out
+}
+
 // DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
 func (in *Policy) DeepCopyInto(out *Policy) {
 	*out = *in
@@ -483,3 +612,31 @@ func (in *ServiceAntiAffinity) DeepCopy() *ServiceAntiAffinity {
 	in.DeepCopyInto(out)
 	return out
 }
+
+// DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
+func (in *Victims) DeepCopyInto(out *Victims) {
+	*out = *in
+	if in.Pods != nil {
+		in, out := &in.Pods, &out.Pods
+		*out = make([]*v1.Pod, len(*in))
+		for i := range *in {
+			if (*in)[i] == nil {
+				(*out)[i] = nil
+			} else {
+				(*out)[i] = new(v1.Pod)
+				(*in)[i].DeepCopyInto((*out)[i])
+			}
+		}
+	}
+	return
+}
+
+// DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new Victims.
+func (in *Victims) DeepCopy() *Victims {
+	if in == nil {
+		return nil
+	}
+	out := new(Victims)
+	in.DeepCopyInto(out)
+	return out
+}

pkg/scheduler/core/BUILD

@@ -27,6 +27,7 @@ go_test(
         "//vendor/k8s.io/api/apps/v1beta1:go_default_library",
         "//vendor/k8s.io/api/core/v1:go_default_library",
         "//vendor/k8s.io/api/extensions/v1beta1:go_default_library",
+        "//vendor/k8s.io/api/policy/v1beta1:go_default_library",
         "//vendor/k8s.io/apimachinery/pkg/api/resource:go_default_library",
         "//vendor/k8s.io/apimachinery/pkg/apis/meta/v1:go_default_library",
         "//vendor/k8s.io/apimachinery/pkg/types:go_default_library",

pkg/scheduler/core/extender.go

@@ -41,6 +41,7 @@ const (
 // HTTPExtender implements the algorithm.SchedulerExtender interface.
 type HTTPExtender struct {
 	extenderURL      string
+	preemptVerb      string
 	filterVerb       string
 	prioritizeVerb   string
 	bindVerb         string
@@ -93,6 +94,7 @@ func NewHTTPExtender(config *schedulerapi.ExtenderConfig) (algorithm.SchedulerEx
 	}
 	return &HTTPExtender{
 		extenderURL:      config.URLPrefix,
+		preemptVerb:      config.PreemptVerb,
 		filterVerb:       config.FilterVerb,
 		prioritizeVerb:   config.PrioritizeVerb,
 		bindVerb:         config.BindVerb,
@@ -103,10 +105,139 @@ func NewHTTPExtender(config *schedulerapi.ExtenderConfig) (algorithm.SchedulerEx
 	}, nil
 }
 
+// SupportsPreemption returns if a extender support preemption.
+// A extender should have preempt verb defined and enabled its own node cache.
+func (h *HTTPExtender) SupportsPreemption() bool {
+	return len(h.preemptVerb) > 0
+}
+
+// ProcessPreemption returns filtered candidate nodes and victims after running preemption logic in extender.
+func (h *HTTPExtender) ProcessPreemption(
+	pod *v1.Pod,
+	nodeToVictims map[*v1.Node]*schedulerapi.Victims,
+	nodeNameToInfo map[string]*schedulercache.NodeInfo,
+) (map[*v1.Node]*schedulerapi.Victims, error) {
+	var (
+		result schedulerapi.ExtenderPreemptionResult
+		args   *schedulerapi.ExtenderPreemptionArgs
+	)
+
+	if !h.SupportsPreemption() {
+		return nil, fmt.Errorf("preempt verb is not defined for extender %v but run into ProcessPreemption", h.extenderURL)
+	}
+
+	if h.nodeCacheCapable {
+		// If extender has cached node info, pass NodeNameToMetaVictims in args.
+		nodeNameToMetaVictims := convertToNodeNameToMetaVictims(nodeToVictims)
+		args = &schedulerapi.ExtenderPreemptionArgs{
+			Pod: pod,
+			NodeNameToMetaVictims: nodeNameToMetaVictims,
+		}
+	} else {
+		nodeNameToVictims := convertToNodeNameToVictims(nodeToVictims)
+		args = &schedulerapi.ExtenderPreemptionArgs{
+			Pod:               pod,
+			NodeNameToVictims: nodeNameToVictims,
+		}
+	}
+
+	if err := h.send(h.preemptVerb, args, &result); err != nil {
+		return nil, err
+	}
+
+	// Extender will always return NodeNameToMetaVictims.
+	// So let's convert it to NodeToVictims by using NodeNameToInfo.
+	nodeToVictims, err := h.convertToNodeToVictims(result.NodeNameToMetaVictims, nodeNameToInfo)
+	if err != nil {
+		return nil, err
+	}
+	return nodeToVictims, nil
+}
+
+// convertToNodeToVictims converts "nodeNameToMetaVictims" from object identifiers,
+// such as UIDs and names, to object pointers.
+func (h *HTTPExtender) convertToNodeToVictims(
+	nodeNameToMetaVictims map[string]*schedulerapi.MetaVictims,
+	nodeNameToInfo map[string]*schedulercache.NodeInfo,
+) (map[*v1.Node]*schedulerapi.Victims, error) {
+	nodeToVictims := map[*v1.Node]*schedulerapi.Victims{}
+	for nodeName, metaVictims := range nodeNameToMetaVictims {
+		victims := &schedulerapi.Victims{
+			Pods: []*v1.Pod{},
+		}
+		for _, metaPod := range metaVictims.Pods {
+			pod, err := h.convertPodUIDToPod(metaPod, nodeName, nodeNameToInfo)
+			if err != nil {
+				return nil, err
+			}
+			victims.Pods = append(victims.Pods, pod)
+		}
+		nodeToVictims[nodeNameToInfo[nodeName].Node()] = victims
+	}
+	return nodeToVictims, nil
+}
+
+// convertPodUIDToPod returns v1.Pod object for given MetaPod and node name.
+// The v1.Pod object is restored by nodeInfo.Pods().
+// It should return error if there's cache inconsistency between default scheduler and extender
+// so that this pod or node is missing from nodeNameToInfo.
+func (h *HTTPExtender) convertPodUIDToPod(
+	metaPod *schedulerapi.MetaPod,
+	nodeName string,
+	nodeNameToInfo map[string]*schedulercache.NodeInfo) (*v1.Pod, error) {
+	var nodeInfo *schedulercache.NodeInfo
+	if nodeInfo, ok := nodeNameToInfo[nodeName]; ok {
+		for _, pod := range nodeInfo.Pods() {
+			if string(pod.UID) == metaPod.UID {
+				return pod, nil
+			}
+		}
+		return nil, fmt.Errorf("extender: %v claims to preempt pod (UID: %v) on node: %v, but the pod is not found on that node",
+			h.extenderURL, metaPod, nodeInfo.Node().Name)
+	}
+
+	return nil, fmt.Errorf("extender: %v claims to preempt on node: %v but the node is not found in nodeNameToInfo map",
+		h.extenderURL, nodeInfo.Node().Name)
+}
+
+// convertToNodeNameToMetaVictims converts from struct type to meta types.
+func convertToNodeNameToMetaVictims(
+	nodeToVictims map[*v1.Node]*schedulerapi.Victims,
+) map[string]*schedulerapi.MetaVictims {
+	nodeNameToVictims := map[string]*schedulerapi.MetaVictims{}
+	for node, victims := range nodeToVictims {
+		metaVictims := &schedulerapi.MetaVictims{
+			Pods: []*schedulerapi.MetaPod{},
+		}
+		for _, pod := range victims.Pods {
+			metaPod := &schedulerapi.MetaPod{
+				UID: string(pod.UID),
+			}
+			metaVictims.Pods = append(metaVictims.Pods, metaPod)
+		}
+		nodeNameToVictims[node.GetName()] = metaVictims
+	}
+	return nodeNameToVictims
+}
+
+// convertToNodeNameToVictims converts from node type to node name as key.
+func convertToNodeNameToVictims(
+	nodeToVictims map[*v1.Node]*schedulerapi.Victims,
+) map[string]*schedulerapi.Victims {
+	nodeNameToVictims := map[string]*schedulerapi.Victims{}
+	for node, victims := range nodeToVictims {
+		nodeNameToVictims[node.GetName()] = victims
+	}
+	return nodeNameToVictims
+}
+
 // Filter based on extender implemented predicate functions. The filtered list is
 // expected to be a subset of the supplied list. failedNodesMap optionally contains
 // the list of failed nodes and failure reasons.
-func (h *HTTPExtender) Filter(pod *v1.Pod, nodes []*v1.Node, nodeNameToInfo map[string]*schedulercache.NodeInfo) ([]*v1.Node, schedulerapi.FailedNodesMap, error) {
+func (h *HTTPExtender) Filter(
+	pod *v1.Pod,
+	nodes []*v1.Node, nodeNameToInfo map[string]*schedulercache.NodeInfo,
+) ([]*v1.Node, schedulerapi.FailedNodesMap, error) {
 	var (
 		result     schedulerapi.ExtenderFilterResult
 		nodeList   *v1.NodeList
@@ -133,7 +264,7 @@ func (h *HTTPExtender) Filter(pod *v1.Pod, nodes []*v1.Node, nodeNameToInfo map[
 	}
 
 	args = &schedulerapi.ExtenderArgs{
-		Pod:       *pod,
+		Pod:       pod,
 		Nodes:     nodeList,
 		NodeNames: nodeNames,
 	}
@@ -193,7 +324,7 @@ func (h *HTTPExtender) Prioritize(pod *v1.Pod, nodes []*v1.Node) (*schedulerapi.
 	}
 
 	args = &schedulerapi.ExtenderArgs{
-		Pod:       *pod,
+		Pod:       pod,
 		Nodes:     nodeList,
 		NodeNames: nodeNames,
 	}

pkg/scheduler/core/extender_test.go

@@ -22,12 +22,14 @@ import (
 	"time"
 
 	"k8s.io/api/core/v1"
+	policy "k8s.io/api/policy/v1beta1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/util/wait"
 	"k8s.io/kubernetes/pkg/scheduler/algorithm"
 	schedulerapi "k8s.io/kubernetes/pkg/scheduler/api"
 	"k8s.io/kubernetes/pkg/scheduler/schedulercache"
 	schedulertesting "k8s.io/kubernetes/pkg/scheduler/testing"
+	"k8s.io/kubernetes/pkg/scheduler/util"
 )
 
 type fitPredicate func(pod *v1.Pod, node *v1.Node) (bool, error)
@@ -111,22 +113,145 @@ type FakeExtender struct {
 	nodeCacheCapable bool
 	filteredNodes    []*v1.Node
 	unInterested     bool
+
+	// Cached node information for fake extender
+	cachedNodeNameToInfo map[string]*schedulercache.NodeInfo
+	cachedPDBs           []*policy.PodDisruptionBudget
+}
+
+func (f *FakeExtender) SupportsPreemption() bool {
+	// Assume preempt verb is always defined.
+	return true
+}
+
+func (f *FakeExtender) ProcessPreemption(
+	pod *v1.Pod,
+	nodeToVictims map[*v1.Node]*schedulerapi.Victims,
+	nodeNameToInfo map[string]*schedulercache.NodeInfo,
+) (map[*v1.Node]*schedulerapi.Victims, error) {
+	nodeToVictimsCopy := map[*v1.Node]*schedulerapi.Victims{}
+	// We don't want to change the original nodeToVictims
+	for k, v := range nodeToVictims {
+		// In real world implementation, extender's user should have his own way to get node object
+		// by name if needed (e.g. query kube-apiserver etc).
+		//
+		// For test purpose, we just use node from parameters directly.
+		nodeToVictimsCopy[k] = v
+	}
+
+	for node, victims := range nodeToVictimsCopy {
+		// Try to do preemption on extender side.
+		extenderVictimPods, extendernPDBViolations, fits := f.selectVictimsOnNodeByExtender(pod, node, nodeNameToInfo)
+		// If it's unfit after extender's preemption, this node is unresolvable by preemption overall,
+		// let's remove it from potential preemption nodes.
+		if !fits {
+			delete(nodeToVictimsCopy, node)
+		} else {
+			// Append new victims to original victims
+			nodeToVictimsCopy[node].Pods = append(victims.Pods, extenderVictimPods...)
+			nodeToVictimsCopy[node].NumPDBViolations = victims.NumPDBViolations + extendernPDBViolations
+		}
+	}
+	return nodeToVictimsCopy, nil
+}
+
+// selectVictimsOnNodeByExtender checks the given nodes->pods map with predicates on extender's side.
+// Returns:
+// 1. More victim pods (if any) amended by preemption phase of extender.
+// 2. Number of violating victim (used to calculate PDB).
+// 3. Fits or not after preemption phase on extender's side.
+func (f *FakeExtender) selectVictimsOnNodeByExtender(
+	pod *v1.Pod,
+	node *v1.Node,
+	nodeNameToInfo map[string]*schedulercache.NodeInfo,
+) ([]*v1.Pod, int, bool) {
+	// TODO(harry): add more test in generic_scheduler_test.go to verify this logic.
+	// If a extender support preemption but have no cached node info, let's run filter to make sure
+	// default scheduler's decision still stand with given pod and node.
+	if !f.nodeCacheCapable {
+		if fits, _ := f.runPredicate(pod, node); !fits {
+			return nil, 0, false
+		}
+		return []*v1.Pod{}, 0, true
+	}
+
+	// Otherwise, as a extender support preemption and have cached node info, we will assume cachedNodeNameToInfo is available
+	// and get cached node info by given node name.
+	nodeInfoCopy := f.cachedNodeNameToInfo[node.GetName()].Clone()
+
+	potentialVictims := util.SortableList{CompFunc: util.HigherPriorityPod}
+
+	removePod := func(rp *v1.Pod) {
+		nodeInfoCopy.RemovePod(rp)
+	}
+	addPod := func(ap *v1.Pod) {
+		nodeInfoCopy.AddPod(ap)
+	}
+	// As the first step, remove all the lower priority pods from the node and
+	// check if the given pod can be scheduled.
+	podPriority := util.GetPodPriority(pod)
+	for _, p := range nodeInfoCopy.Pods() {
+		if util.GetPodPriority(p) < podPriority {
+			potentialVictims.Items = append(potentialVictims.Items, p)
+			removePod(p)
+		}
+	}
+	potentialVictims.Sort()
+
+	// If the new pod does not fit after removing all the lower priority pods,
+	// we are almost done and this node is not suitable for preemption.
+	if fits, _ := f.runPredicate(pod, nodeInfoCopy.Node()); !fits {
+		return nil, 0, false
+	}
+
+	var victims []*v1.Pod
+
+	// TODO(harry): handle PDBs in the future.
+	numViolatingVictim := 0
+
+	reprievePod := func(p *v1.Pod) bool {
+		addPod(p)
+		fits, _ := f.runPredicate(pod, nodeInfoCopy.Node())
+		if !fits {
+			removePod(p)
+			victims = append(victims, p)
+		}
+		return fits
+	}
+
+	// For now, assume all potential victims to be non-violating.
+	// Now we try to reprieve non-violating victims.
+	for _, p := range potentialVictims.Items {
+		reprievePod(p.(*v1.Pod))
+	}
+
+	return victims, numViolatingVictim, true
+}
+
+// runPredicate run predicates of extender one by one for given pod and node.
+// Returns: fits or not.
+func (f *FakeExtender) runPredicate(pod *v1.Pod, node *v1.Node) (bool, error) {
+	fits := true
+	var err error
+	for _, predicate := range f.predicates {
+		fits, err = predicate(pod, node)
+		if err != nil {
+			return false, err
+		}
+		if !fits {
+			break
+		}
+	}
+	return fits, nil
 }
 
 func (f *FakeExtender) Filter(pod *v1.Pod, nodes []*v1.Node, nodeNameToInfo map[string]*schedulercache.NodeInfo) ([]*v1.Node, schedulerapi.FailedNodesMap, error) {
 	filtered := []*v1.Node{}
 	failedNodesMap := schedulerapi.FailedNodesMap{}
 	for _, node := range nodes {
-		fits := true
-		for _, predicate := range f.predicates {
-			fit, err := predicate(pod, node)
-			if err != nil {
-				return []*v1.Node{}, schedulerapi.FailedNodesMap{}, err
-			}
-			if !fit {
-				fits = false
-				break
-			}
+		fits, err := f.runPredicate(pod, node)
+		if err != nil {
+			return []*v1.Node{}, schedulerapi.FailedNodesMap{}, err
 		}
 		if fits {
 			filtered = append(filtered, node)
@@ -340,7 +465,7 @@ func TestGenericSchedulerWithExtenders(t *testing.T) {
 		}
 		cache := schedulercache.New(time.Duration(0), wait.NeverStop)
 		for _, name := range test.nodes {
-			cache.AddNode(&v1.Node{ObjectMeta: metav1.ObjectMeta{Name: name}})
+			cache.AddNode(createNode(name))
 		}
 		queue := NewSchedulingQueue()
 		scheduler := NewGenericScheduler(
@@ -362,3 +487,7 @@ func TestGenericSchedulerWithExtenders(t *testing.T) {
 		}
 	}
 }
+
+func createNode(name string) *v1.Node {
+	return &v1.Node{ObjectMeta: metav1.ObjectMeta{Name: name}}
+}

pkg/scheduler/core/generic_scheduler.go

@@ -54,13 +54,7 @@ type FitError struct {
 	FailedPredicates FailedPredicateMap
 }
 
-// Victims describes pod victims.
-type Victims struct {
-	pods             []*v1.Pod
-	numPDBViolations int
-}
-
-// ErrNoNodesAvailable defines an error of no nodes available.
+// ErrNoNodesAvailable is used to describe the error that no nodes available to schedule pods.
 var ErrNoNodesAvailable = fmt.Errorf("no nodes available to schedule pods")
 
 const (
@@ -234,34 +228,64 @@ func (g *genericScheduler) Preempt(pod *v1.Pod, nodeLister algorithm.NodeLister,
 	if err != nil {
 		return nil, nil, nil, err
 	}
-	nodeToVictims, err := selectNodesForPreemption(pod, g.cachedNodeInfoMap, potentialNodes, g.predicates, g.predicateMetaProducer, g.schedulingQueue, pdbs)
+	nodeToVictims, err := selectNodesForPreemption(pod, g.cachedNodeInfoMap, potentialNodes, g.predicates,
+		g.predicateMetaProducer, g.schedulingQueue, pdbs)
 	if err != nil {
 		return nil, nil, nil, err
 	}
-	for len(nodeToVictims) > 0 {
-		node := pickOneNodeForPreemption(nodeToVictims)
-		if node == nil {
-			return nil, nil, nil, err
-		}
-		passes, pErr := nodePassesExtendersForPreemption(pod, node.Name, nodeToVictims[node].pods, g.cachedNodeInfoMap, g.extenders)
-		if passes && pErr == nil {
-			// Lower priority pods nominated to run on this node, may no longer fit on
-			// this node. So, we should remove their nomination. Removing their
-			// nomination updates these pods and moves them to the active queue. It
-			// lets scheduler find another place for them.
-			nominatedPods := g.getLowerPriorityNominatedPods(pod, node.Name)
-			return node, nodeToVictims[node].pods, nominatedPods, err
-		}
-		if pErr != nil {
-			glog.Errorf("Error occurred while checking extenders for preemption on node %v: %v", node, pErr)
-		}
-		// Remove the node from the map and try to pick a different node.
-		delete(nodeToVictims, node)
+
+	// We will only check nodeToVictims with extenders that support preemption.
+	// Extenders which do not support preemption may later prevent preemptor from being scheduled on the nominated
+	// node. In that case, scheduler will find a different host for the preemptor in subsequent scheduling cycles.
+	nodeToVictims, err = g.processPreemptionWithExtenders(pod, nodeToVictims)
+	if err != nil {
+		return nil, nil, nil, err
 	}
-	return nil, nil, nil, err
+
+	candidateNode := pickOneNodeForPreemption(nodeToVictims)
+	if candidateNode == nil {
+		return nil, nil, nil, err
+	}
+
+	// Lower priority pods nominated to run on this node, may no longer fit on
+	// this node. So, we should remove their nomination. Removing their
+	// nomination updates these pods and moves them to the active queue. It
+	// lets scheduler find another place for them.
+	nominatedPods := g.getLowerPriorityNominatedPods(pod, candidateNode.Name)
+	if nodeInfo, ok := g.cachedNodeInfoMap[candidateNode.Name]; ok {
+		return nodeInfo.Node(), nodeToVictims[candidateNode].Pods, nominatedPods, err
+	}
+
+	return nil, nil, nil, fmt.Errorf(
+		"preemption failed: the target node %s has been deleted from scheduler cache",
+		candidateNode.Name)
 }
 
-// GetLowerPriorityNominatedPods returns pods whose priority is smaller than the
+// processPreemptionWithExtenders processes preemption with extenders
+func (g *genericScheduler) processPreemptionWithExtenders(
+	pod *v1.Pod,
+	nodeToVictims map[*v1.Node]*schedulerapi.Victims,
+) (map[*v1.Node]*schedulerapi.Victims, error) {
+	if len(nodeToVictims) > 0 {
+		for _, extender := range g.extenders {
+			if extender.SupportsPreemption() {
+				var err error
+				// Replace nodeToVictims with result after preemption from extender.
+				if nodeToVictims, err = extender.ProcessPreemption(pod, nodeToVictims, g.cachedNodeInfoMap); err != nil {
+					return nil, err
+				}
+				// If node list is empty, no preemption will happen, skip other extenders.
+				if len(nodeToVictims) == 0 {
+					break
+				}
+			}
+		}
+	}
+
+	return nodeToVictims, nil
+}
+
+// getLowerPriorityNominatedPods returns pods whose priority is smaller than the
 // priority of the given "pod" and are nominated to run on the given node.
 // Note: We could possibly check if the nominated lower priority pods still fit
 // and return those that no longer fit, but that would require lots of
@@ -270,6 +294,7 @@ func (g *genericScheduler) Preempt(pod *v1.Pod, nodeLister algorithm.NodeLister,
 // small number of nominated pods per node.
 func (g *genericScheduler) getLowerPriorityNominatedPods(pod *v1.Pod, nodeName string) []*v1.Pod {
 	pods := g.schedulingQueue.WaitingPodsForNode(nodeName)
+
 	if len(pods) == 0 {
 		return nil
 	}
@@ -683,7 +708,7 @@ func EqualPriorityMap(_ *v1.Pod, _ interface{}, nodeInfo *schedulercache.NodeInf
 // 5. If there are still ties, the first such node is picked (sort of randomly).
 // The 'minNodes1' and 'minNodes2' are being reused here to save the memory
 // allocation and garbage collection time.
-func pickOneNodeForPreemption(nodesToVictims map[*v1.Node]*Victims) *v1.Node {
+func pickOneNodeForPreemption(nodesToVictims map[*v1.Node]*schedulerapi.Victims) *v1.Node {
 	if len(nodesToVictims) == 0 {
 		return nil
 	}
@@ -691,14 +716,14 @@ func pickOneNodeForPreemption(nodesToVictims map[*v1.Node]*Victims) *v1.Node {
 	var minNodes1 []*v1.Node
 	lenNodes1 := 0
 	for node, victims := range nodesToVictims {
-		if len(victims.pods) == 0 {
+		if len(victims.Pods) == 0 {
 			// We found a node that doesn't need any preemption. Return it!
 			// This should happen rarely when one or more pods are terminated between
 			// the time that scheduler tries to schedule the pod and the time that
 			// preemption logic tries to find nodes for preemption.
 			return node
 		}
-		numPDBViolatingPods := victims.numPDBViolations
+		numPDBViolatingPods := victims.NumPDBViolations
 		if numPDBViolatingPods < minNumPDBViolatingPods {
 			minNumPDBViolatingPods = numPDBViolatingPods
 			minNodes1 = nil
@@ -722,7 +747,7 @@ func pickOneNodeForPreemption(nodesToVictims map[*v1.Node]*Victims) *v1.Node {
 		node := minNodes1[i]
 		victims := nodesToVictims[node]
 		// highestPodPriority is the highest priority among the victims on this node.
-		highestPodPriority := util.GetPodPriority(victims.pods[0])
+		highestPodPriority := util.GetPodPriority(victims.Pods[0])
 		if highestPodPriority < minHighestPriority {
 			minHighestPriority = highestPodPriority
 			lenNodes2 = 0
@@ -743,7 +768,7 @@ func pickOneNodeForPreemption(nodesToVictims map[*v1.Node]*Victims) *v1.Node {
 	for i := 0; i < lenNodes2; i++ {
 		var sumPriorities int64
 		node := minNodes2[i]
-		for _, pod := range nodesToVictims[node].pods {
+		for _, pod := range nodesToVictims[node].Pods {
 			// We add MaxInt32+1 to all priorities to make all of them >= 0. This is
 			// needed so that a node with a few pods with negative priority is not
 			// picked over a node with a smaller number of pods with the same negative
@@ -769,7 +794,7 @@ func pickOneNodeForPreemption(nodesToVictims map[*v1.Node]*Victims) *v1.Node {
 	lenNodes2 = 0
 	for i := 0; i < lenNodes1; i++ {
 		node := minNodes1[i]
-		numPods := len(nodesToVictims[node].pods)
+		numPods := len(nodesToVictims[node].Pods)
 		if numPods < minNumPods {
 			minNumPods = numPods
 			lenNodes2 = 0
@@ -797,9 +822,9 @@ func selectNodesForPreemption(pod *v1.Pod,
 	metadataProducer algorithm.PredicateMetadataProducer,
 	queue SchedulingQueue,
 	pdbs []*policy.PodDisruptionBudget,
-) (map[*v1.Node]*Victims, error) {
+) (map[*v1.Node]*schedulerapi.Victims, error) {
 
-	nodeNameToVictims := map[*v1.Node]*Victims{}
+	nodeToVictims := map[*v1.Node]*schedulerapi.Victims{}
 	var resultLock sync.Mutex
 
 	// We can use the same metadata producer for all nodes.
@@ -813,50 +838,16 @@ func selectNodesForPreemption(pod *v1.Pod,
 		pods, numPDBViolations, fits := selectVictimsOnNode(pod, metaCopy, nodeNameToInfo[nodeName], predicates, queue, pdbs)
 		if fits {
 			resultLock.Lock()
-			victims := Victims{
-				pods:             pods,
-				numPDBViolations: numPDBViolations,
+			victims := schedulerapi.Victims{
+				Pods:             pods,
+				NumPDBViolations: numPDBViolations,
 			}
-			nodeNameToVictims[potentialNodes[i]] = &victims
+			nodeToVictims[potentialNodes[i]] = &victims
 			resultLock.Unlock()
 		}
 	}
 	workqueue.Parallelize(16, len(potentialNodes), checkNode)
-	return nodeNameToVictims, nil
-}
-
-func nodePassesExtendersForPreemption(
-	pod *v1.Pod,
-	nodeName string,
-	victims []*v1.Pod,
-	nodeNameToInfo map[string]*schedulercache.NodeInfo,
-	extenders []algorithm.SchedulerExtender) (bool, error) {
-	// If there are any extenders, run them and filter the list of candidate nodes.
-	if len(extenders) == 0 {
-		return true, nil
-	}
-	// Remove the victims from the corresponding nodeInfo and send nodes to the
-	// extenders for filtering.
-	originalNodeInfo := nodeNameToInfo[nodeName]
-	nodeInfoCopy := nodeNameToInfo[nodeName].Clone()
-	for _, victim := range victims {
-		nodeInfoCopy.RemovePod(victim)
-	}
-	nodeNameToInfo[nodeName] = nodeInfoCopy
-	defer func() { nodeNameToInfo[nodeName] = originalNodeInfo }()
-	filteredNodes := []*v1.Node{nodeInfoCopy.Node()}
-	for _, extender := range extenders {
-		var err error
-		var failedNodesMap map[string]string
-		filteredNodes, failedNodesMap, err = extender.Filter(pod, filteredNodes, nodeNameToInfo)
-		if err != nil {
-			return false, err
-		}
-		if _, found := failedNodesMap[nodeName]; found || len(filteredNodes) == 0 {
-			return false, nil
-		}
-	}
-	return true, nil
+	return nodeToVictims, nil
 }
 
 // filterPodsWithPDBViolation groups the given "pods" into two groups of "violatingPods"
@@ -996,6 +987,7 @@ func nodesWherePreemptionMightHelp(pod *v1.Pod, nodes []*v1.Node, failedPredicat
 		// (which is the case today), the !found case should never happen, but we'd prefer
 		// to rely less on such assumptions in the code when checking does not impose
 		// significant overhead.
+		// Also, we currently assume all failures returned by extender as resolvable.
 		for _, failedPredicate := range failedPredicates {
 			switch failedPredicate {
 			case

pkg/scheduler/core/generic_scheduler_test.go

@@ -660,11 +660,11 @@ func TestZeroRequest(t *testing.T) {
 	}
 }
 
-func printNodeToVictims(nodeToVictims map[*v1.Node]*Victims) string {
+func printNodeToVictims(nodeToVictims map[*v1.Node]*schedulerapi.Victims) string {
 	var output string
 	for node, victims := range nodeToVictims {
 		output += node.Name + ": ["
-		for _, pod := range victims.pods {
+		for _, pod := range victims.Pods {
 			output += pod.Name + ", "
 		}
 		output += "]"
@@ -672,15 +672,15 @@ func printNodeToVictims(nodeToVictims map[*v1.Node]*Victims) string {
 	return output
 }
 
-func checkPreemptionVictims(testName string, expected map[string]map[string]bool, nodeToPods map[*v1.Node]*Victims) error {
+func checkPreemptionVictims(testName string, expected map[string]map[string]bool, nodeToPods map[*v1.Node]*schedulerapi.Victims) error {
 	if len(expected) == len(nodeToPods) {
 		for k, victims := range nodeToPods {
 			if expPods, ok := expected[k.Name]; ok {
-				if len(victims.pods) != len(expPods) {
+				if len(victims.Pods) != len(expPods) {
 					return fmt.Errorf("test [%v]: unexpected number of pods. expected: %v, got: %v", testName, expected, printNodeToVictims(nodeToPods))
 				}
 				prevPriority := int32(math.MaxInt32)
-				for _, p := range victims.pods {
+				for _, p := range victims.Pods {
 					// Check that pods are sorted by their priority.
 					if *p.Spec.Priority > prevPriority {
 						return fmt.Errorf("test [%v]: pod %v of node %v was not sorted by priority", testName, p.Name, k)
@@ -883,7 +883,7 @@ func TestSelectNodesForPreemption(t *testing.T) {
 	for _, test := range tests {
 		nodes := []*v1.Node{}
 		for _, n := range test.nodes {
-			node := makeNode(n, priorityutil.DefaultMilliCPURequest*5, priorityutil.DefaultMemoryRequest*5)
+			node := makeNode(n, 1000*5, priorityutil.DefaultMemoryRequest*5)
 			node.ObjectMeta.Labels = map[string]string{"hostname": node.Name}
 			nodes = append(nodes, node)
 		}
@@ -1282,11 +1282,20 @@ func TestPreempt(t *testing.T) {
 		for _, pod := range test.pods {
 			cache.AddPod(pod)
 		}
+		cachedNodeInfoMap := map[string]*schedulercache.NodeInfo{}
 		for _, name := range nodeNames {
-			cache.AddNode(makeNode(name, priorityutil.DefaultMilliCPURequest*5, priorityutil.DefaultMemoryRequest*5))
+			node := makeNode(name, 1000*5, priorityutil.DefaultMemoryRequest*5)
+			cache.AddNode(node)
+
+			// Set nodeInfo to extenders to mock extenders' cache for preemption.
+			cachedNodeInfo := schedulercache.NewNodeInfo()
+			cachedNodeInfo.SetNode(node)
+			cachedNodeInfoMap[name] = cachedNodeInfo
 		}
 		extenders := []algorithm.SchedulerExtender{}
 		for _, extender := range test.extenders {
+			// Set nodeInfoMap as extenders cached node information.
+			extender.cachedNodeNameToInfo = cachedNodeInfoMap
 			extenders = append(extenders, extender)
 		}
 		scheduler := NewGenericScheduler(

pkg/scheduler/factory/factory_test.go

@@ -541,11 +541,30 @@ type fakeExtender struct {
 	interestedPodName string
 }
 
-func (f *fakeExtender) Filter(pod *v1.Pod, nodes []*v1.Node, nodeNameToInfo map[string]*schedulercache.NodeInfo) (filteredNodes []*v1.Node, failedNodesMap schedulerapi.FailedNodesMap, err error) {
+func (f *fakeExtender) ProcessPreemption(
+	pod *v1.Pod,
+	nodeToVictims map[*v1.Node]*schedulerapi.Victims,
+	nodeNameToInfo map[string]*schedulercache.NodeInfo,
+) (map[*v1.Node]*schedulerapi.Victims, error) {
+	return nil, nil
+}
+
+func (f *fakeExtender) SupportsPreemption() bool {
+	return false
+}
+
+func (f *fakeExtender) Filter(
+	pod *v1.Pod,
+	nodes []*v1.Node,
+	nodeNameToInfo map[string]*schedulercache.NodeInfo,
+) (filteredNodes []*v1.Node, failedNodesMap schedulerapi.FailedNodesMap, err error) {
 	return nil, nil, nil
 }
 
-func (f *fakeExtender) Prioritize(pod *v1.Pod, nodes []*v1.Node) (hostPriorities *schedulerapi.HostPriorityList, weight int, err error) {
+func (f *fakeExtender) Prioritize(
+	pod *v1.Pod,
+	nodes []*v1.Node,
+) (hostPriorities *schedulerapi.HostPriorityList, weight int, err error) {
 	return nil, 0, nil
 }
 

test/integration/scheduler/extender_test.go

@@ -130,7 +130,7 @@ func (e *Extender) filterUsingNodeCache(args *schedulerapi.ExtenderArgs) (*sched
 	for _, nodeName := range *args.NodeNames {
 		fits := true
 		for _, predicate := range e.predicates {
-			fit, err := predicate(&args.Pod,
+			fit, err := predicate(args.Pod,
 				&v1.Node{ObjectMeta: metav1.ObjectMeta{Name: nodeName}})
 			if err != nil {
 				return &schedulerapi.ExtenderFilterResult{
@@ -169,7 +169,7 @@ func (e *Extender) Filter(args *schedulerapi.ExtenderArgs) (*schedulerapi.Extend
 		for _, node := range args.Nodes.Items {
 			fits := true
 			for _, predicate := range e.predicates {
-				fit, err := predicate(&args.Pod, &node)
+				fit, err := predicate(args.Pod, &node)
 				if err != nil {
 					return &schedulerapi.ExtenderFilterResult{
 						Nodes:       &v1.NodeList{},
@@ -217,7 +217,7 @@ func (e *Extender) Prioritize(args *schedulerapi.ExtenderArgs) (*schedulerapi.Ho
 			continue
 		}
 		priorityFunc := prioritizer.function
-		prioritizedList, err := priorityFunc(&args.Pod, nodes)
+		prioritizedList, err := priorityFunc(args.Pod, nodes)
 		if err != nil {
 			return &schedulerapi.HostPriorityList{}, err
 		}