PodLifeTime: allow custom podStatusPhases

README.md

@@ -261,18 +261,36 @@ strategies:
 
 ### PodLifeTime
 
-This strategy evicts pods that are older than `.strategies.PodLifeTime.params.maxPodLifeTimeSeconds` The policy
+This strategy evicts pods that are older than `maxPodLifeTimeSeconds`. The policy file should look like:
-file should look like:
 
-````
+```
 apiVersion: "descheduler/v1alpha1"
 kind: "DeschedulerPolicy"
 strategies:
   "PodLifeTime":
      enabled: true
      params:
-        maxPodLifeTimeSeconds: 86400
+       podLifeTime:
-````
+         maxPodLifeTimeSeconds: 86400
+```
+
+You can specify `podStatusPhases` to `only` evict pods with specific `StatusPhases`, currently this parameter is limited
+to `Running` and `Pending`. E.g.
+
+```
+apiVersion: "descheduler/v1alpha1"
+kind: "DeschedulerPolicy"
+strategies:
+  "PodLifeTime":
+     enabled: true
+     params:
+       podLifeTime:
+         maxPodLifeTimeSeconds: 86400
+         podStatusPhases:
+         - "Pending"
+```
+
+Only `Pending` pods will get evicted in this example.
 
 ## Filter Pods
 
@@ -290,7 +308,8 @@ strategies:
   "PodLifeTime":
      enabled: true
      params:
-        maxPodLifeTimeSeconds: 86400
+        podLifeTime:
+          maxPodLifeTimeSeconds: 86400
         namespaces:
           include:
           - "namespace1"
@@ -307,7 +326,8 @@ strategies:
   "PodLifeTime":
      enabled: true
      params:
-        maxPodLifeTimeSeconds: 86400
+        podLifeTime:
+          maxPodLifeTimeSeconds: 86400
         namespaces:
           exclude:
           - "namespace1"
@@ -334,7 +354,8 @@ strategies:
   "PodLifeTime":
      enabled: true
      params:
-        maxPodLifeTimeSeconds: 86400
+        podLifeTime:
+          maxPodLifeTimeSeconds: 86400
         thresholdPriority: 10000
 ```
 
@@ -346,7 +367,8 @@ strategies:
   "PodLifeTime":
      enabled: true
      params:
-        maxPodLifeTimeSeconds: 86400
+        podLifeTime:
+          maxPodLifeTimeSeconds: 86400
         thresholdPriorityClassName: "priorityclass1"
 ```
 

examples/pod-life-time.yml

@@ -17,4 +17,5 @@ strategies:
   "PodLifeTime":
     enabled: true
     params:
-      maxPodLifeTimeSeconds: 604800 # 7 days
+      podLifeTime:
+        maxPodLifeTimeSeconds: 604800 # 7 days

pkg/api/types.go

@@ -67,7 +67,7 @@ type StrategyParameters struct {
 	NodeResourceUtilizationThresholds *NodeResourceUtilizationThresholds
 	NodeAffinityType                  []string
 	PodsHavingTooManyRestarts         *PodsHavingTooManyRestarts
-	MaxPodLifeTimeSeconds             *uint
+	PodLifeTime                       *PodLifeTime
 	RemoveDuplicates                  *RemoveDuplicates
 	Namespaces                        *Namespaces
 	ThresholdPriority                 *int32
@@ -91,3 +91,8 @@ type PodsHavingTooManyRestarts struct {
 type RemoveDuplicates struct {
 	ExcludeOwnerKinds []string
 }
+
+type PodLifeTime struct {
+	MaxPodLifeTimeSeconds *uint
+	PodStatusPhases       []string
+}

pkg/api/v1alpha1/types.go

@@ -65,7 +65,7 @@ type StrategyParameters struct {
 	NodeResourceUtilizationThresholds *NodeResourceUtilizationThresholds `json:"nodeResourceUtilizationThresholds,omitempty"`
 	NodeAffinityType                  []string                           `json:"nodeAffinityType,omitempty"`
 	PodsHavingTooManyRestarts         *PodsHavingTooManyRestarts         `json:"podsHavingTooManyRestarts,omitempty"`
-	MaxPodLifeTimeSeconds             *uint                              `json:"maxPodLifeTimeSeconds,omitempty"`
+	PodLifeTime                       *PodLifeTime                       `json:"podLifeTime,omitempty"`
 	RemoveDuplicates                  *RemoveDuplicates                  `json:"removeDuplicates,omitempty"`
 	Namespaces                        *Namespaces                        `json:"namespaces"`
 	ThresholdPriority                 *int32                             `json:"thresholdPriority"`
@@ -89,3 +89,8 @@ type PodsHavingTooManyRestarts struct {
 type RemoveDuplicates struct {
 	ExcludeOwnerKinds []string `json:"excludeOwnerKinds,omitempty"`
 }
+
+type PodLifeTime struct {
+	MaxPodLifeTimeSeconds *uint    `json:"maxPodLifeTimeSeconds,omitempty"`
+	PodStatusPhases       []string `json:"podStatusPhases,omitempty"`
+}

pkg/api/v1alpha1/zz_generated.conversion.go

@@ -75,6 +75,16 @@ func RegisterConversions(s *runtime.Scheme) error {
 	}); err != nil {
 		return err
 	}
+	if err := s.AddGeneratedConversionFunc((*PodLifeTime)(nil), (*api.PodLifeTime)(nil), func(a, b interface{}, scope conversion.Scope) error {
+		return Convert_v1alpha1_PodLifeTime_To_api_PodLifeTime(a.(*PodLifeTime), b.(*api.PodLifeTime), scope)
+	}); err != nil {
+		return err
+	}
+	if err := s.AddGeneratedConversionFunc((*api.PodLifeTime)(nil), (*PodLifeTime)(nil), func(a, b interface{}, scope conversion.Scope) error {
+		return Convert_api_PodLifeTime_To_v1alpha1_PodLifeTime(a.(*api.PodLifeTime), b.(*PodLifeTime), scope)
+	}); err != nil {
+		return err
+	}
 	if err := s.AddGeneratedConversionFunc((*PodsHavingTooManyRestarts)(nil), (*api.PodsHavingTooManyRestarts)(nil), func(a, b interface{}, scope conversion.Scope) error {
 		return Convert_v1alpha1_PodsHavingTooManyRestarts_To_api_PodsHavingTooManyRestarts(a.(*PodsHavingTooManyRestarts), b.(*api.PodsHavingTooManyRestarts), scope)
 	}); err != nil {
@@ -204,6 +214,28 @@ func Convert_api_NodeResourceUtilizationThresholds_To_v1alpha1_NodeResourceUtili
 	return autoConvert_api_NodeResourceUtilizationThresholds_To_v1alpha1_NodeResourceUtilizationThresholds(in, out, s)
 }
 
+func autoConvert_v1alpha1_PodLifeTime_To_api_PodLifeTime(in *PodLifeTime, out *api.PodLifeTime, s conversion.Scope) error {
+	out.MaxPodLifeTimeSeconds = (*uint)(unsafe.Pointer(in.MaxPodLifeTimeSeconds))
+	out.PodStatusPhases = *(*[]string)(unsafe.Pointer(&in.PodStatusPhases))
+	return nil
+}
+
+// Convert_v1alpha1_PodLifeTime_To_api_PodLifeTime is an autogenerated conversion function.
+func Convert_v1alpha1_PodLifeTime_To_api_PodLifeTime(in *PodLifeTime, out *api.PodLifeTime, s conversion.Scope) error {
+	return autoConvert_v1alpha1_PodLifeTime_To_api_PodLifeTime(in, out, s)
+}
+
+func autoConvert_api_PodLifeTime_To_v1alpha1_PodLifeTime(in *api.PodLifeTime, out *PodLifeTime, s conversion.Scope) error {
+	out.MaxPodLifeTimeSeconds = (*uint)(unsafe.Pointer(in.MaxPodLifeTimeSeconds))
+	out.PodStatusPhases = *(*[]string)(unsafe.Pointer(&in.PodStatusPhases))
+	return nil
+}
+
+// Convert_api_PodLifeTime_To_v1alpha1_PodLifeTime is an autogenerated conversion function.
+func Convert_api_PodLifeTime_To_v1alpha1_PodLifeTime(in *api.PodLifeTime, out *PodLifeTime, s conversion.Scope) error {
+	return autoConvert_api_PodLifeTime_To_v1alpha1_PodLifeTime(in, out, s)
+}
+
 func autoConvert_v1alpha1_PodsHavingTooManyRestarts_To_api_PodsHavingTooManyRestarts(in *PodsHavingTooManyRestarts, out *api.PodsHavingTooManyRestarts, s conversion.Scope) error {
 	out.PodRestartThreshold = in.PodRestartThreshold
 	out.IncludingInitContainers = in.IncludingInitContainers
@@ -250,7 +282,7 @@ func autoConvert_v1alpha1_StrategyParameters_To_api_StrategyParameters(in *Strat
 	out.NodeResourceUtilizationThresholds = (*api.NodeResourceUtilizationThresholds)(unsafe.Pointer(in.NodeResourceUtilizationThresholds))
 	out.NodeAffinityType = *(*[]string)(unsafe.Pointer(&in.NodeAffinityType))
 	out.PodsHavingTooManyRestarts = (*api.PodsHavingTooManyRestarts)(unsafe.Pointer(in.PodsHavingTooManyRestarts))
-	out.MaxPodLifeTimeSeconds = (*uint)(unsafe.Pointer(in.MaxPodLifeTimeSeconds))
+	out.PodLifeTime = (*api.PodLifeTime)(unsafe.Pointer(in.PodLifeTime))
 	out.RemoveDuplicates = (*api.RemoveDuplicates)(unsafe.Pointer(in.RemoveDuplicates))
 	out.Namespaces = (*api.Namespaces)(unsafe.Pointer(in.Namespaces))
 	out.ThresholdPriority = (*int32)(unsafe.Pointer(in.ThresholdPriority))
@@ -267,7 +299,7 @@ func autoConvert_api_StrategyParameters_To_v1alpha1_StrategyParameters(in *api.S
 	out.NodeResourceUtilizationThresholds = (*NodeResourceUtilizationThresholds)(unsafe.Pointer(in.NodeResourceUtilizationThresholds))
 	out.NodeAffinityType = *(*[]string)(unsafe.Pointer(&in.NodeAffinityType))
 	out.PodsHavingTooManyRestarts = (*PodsHavingTooManyRestarts)(unsafe.Pointer(in.PodsHavingTooManyRestarts))
-	out.MaxPodLifeTimeSeconds = (*uint)(unsafe.Pointer(in.MaxPodLifeTimeSeconds))
+	out.PodLifeTime = (*PodLifeTime)(unsafe.Pointer(in.PodLifeTime))
 	out.RemoveDuplicates = (*RemoveDuplicates)(unsafe.Pointer(in.RemoveDuplicates))
 	out.Namespaces = (*Namespaces)(unsafe.Pointer(in.Namespaces))
 	out.ThresholdPriority = (*int32)(unsafe.Pointer(in.ThresholdPriority))

pkg/api/v1alpha1/zz_generated.deepcopy.go

@@ -148,6 +148,32 @@ func (in *NodeResourceUtilizationThresholds) DeepCopy() *NodeResourceUtilization
 	return out
 }
 
+// DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
+func (in *PodLifeTime) DeepCopyInto(out *PodLifeTime) {
+	*out = *in
+	if in.MaxPodLifeTimeSeconds != nil {
+		in, out := &in.MaxPodLifeTimeSeconds, &out.MaxPodLifeTimeSeconds
+		*out = new(uint)
+		**out = **in
+	}
+	if in.PodStatusPhases != nil {
+		in, out := &in.PodStatusPhases, &out.PodStatusPhases
+		*out = make([]string, len(*in))
+		copy(*out, *in)
+	}
+	return
+}
+
+// DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new PodLifeTime.
+func (in *PodLifeTime) DeepCopy() *PodLifeTime {
+	if in == nil {
+		return nil
+	}
+	out := new(PodLifeTime)
+	in.DeepCopyInto(out)
+	return out
+}
+
 // DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
 func (in *PodsHavingTooManyRestarts) DeepCopyInto(out *PodsHavingTooManyRestarts) {
 	*out = *in
@@ -247,10 +273,10 @@ func (in *StrategyParameters) DeepCopyInto(out *StrategyParameters) {
 		*out = new(PodsHavingTooManyRestarts)
 		**out = **in
 	}
-	if in.MaxPodLifeTimeSeconds != nil {
+	if in.PodLifeTime != nil {
-		in, out := &in.MaxPodLifeTimeSeconds, &out.MaxPodLifeTimeSeconds
+		in, out := &in.PodLifeTime, &out.PodLifeTime
-		*out = new(uint)
+		*out = new(PodLifeTime)
-		**out = **in
+		(*in).DeepCopyInto(*out)
 	}
 	if in.RemoveDuplicates != nil {
 		in, out := &in.RemoveDuplicates, &out.RemoveDuplicates

pkg/api/zz_generated.deepcopy.go

@@ -148,6 +148,32 @@ func (in *NodeResourceUtilizationThresholds) DeepCopy() *NodeResourceUtilization
 	return out
 }
 
+// DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
+func (in *PodLifeTime) DeepCopyInto(out *PodLifeTime) {
+	*out = *in
+	if in.MaxPodLifeTimeSeconds != nil {
+		in, out := &in.MaxPodLifeTimeSeconds, &out.MaxPodLifeTimeSeconds
+		*out = new(uint)
+		**out = **in
+	}
+	if in.PodStatusPhases != nil {
+		in, out := &in.PodStatusPhases, &out.PodStatusPhases
+		*out = make([]string, len(*in))
+		copy(*out, *in)
+	}
+	return
+}
+
+// DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new PodLifeTime.
+func (in *PodLifeTime) DeepCopy() *PodLifeTime {
+	if in == nil {
+		return nil
+	}
+	out := new(PodLifeTime)
+	in.DeepCopyInto(out)
+	return out
+}
+
 // DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
 func (in *PodsHavingTooManyRestarts) DeepCopyInto(out *PodsHavingTooManyRestarts) {
 	*out = *in
@@ -247,10 +273,10 @@ func (in *StrategyParameters) DeepCopyInto(out *StrategyParameters) {
 		*out = new(PodsHavingTooManyRestarts)
 		**out = **in
 	}
-	if in.MaxPodLifeTimeSeconds != nil {
+	if in.PodLifeTime != nil {
-		in, out := &in.MaxPodLifeTimeSeconds, &out.MaxPodLifeTimeSeconds
+		in, out := &in.PodLifeTime, &out.PodLifeTime
-		*out = new(uint)
+		*out = new(PodLifeTime)
-		**out = **in
+		(*in).DeepCopyInto(*out)
 	}
 	if in.RemoveDuplicates != nil {
 		in, out := &in.RemoveDuplicates, &out.RemoveDuplicates

pkg/descheduler/strategies/pod_lifetime.go

@@ -32,10 +32,18 @@ import (
 )
 
 func validatePodLifeTimeParams(params *api.StrategyParameters) error {
-	if params == nil || params.MaxPodLifeTimeSeconds == nil {
+	if params == nil || params.PodLifeTime == nil || params.PodLifeTime.MaxPodLifeTimeSeconds == nil {
 		return fmt.Errorf("MaxPodLifeTimeSeconds not set")
 	}
 
+	if params.PodLifeTime.PodStatusPhases != nil {
+		for _, phase := range params.PodLifeTime.PodStatusPhases {
+			if phase != string(v1.PodPending) && phase != string(v1.PodRunning) {
+				return fmt.Errorf("only Pending and Running phases are supported in PodLifeTime")
+			}
+		}
+	}
+
 	// At most one of include/exclude can be set
 	if params.Namespaces != nil && len(params.Namespaces.Include) > 0 && len(params.Namespaces.Exclude) > 0 {
 		return fmt.Errorf("only one of Include/Exclude namespaces can be set")
@@ -68,14 +76,26 @@ func PodLifeTime(ctx context.Context, client clientset.Interface, strategy api.D
 
 	evictable := podEvictor.Evictable(evictions.WithPriorityThreshold(thresholdPriority))
 
+	filter := evictable.IsEvictable
+	if strategy.Params.PodLifeTime.PodStatusPhases != nil {
+		filter = func(pod *v1.Pod) bool {
+			for _, phase := range strategy.Params.PodLifeTime.PodStatusPhases {
+				if string(pod.Status.Phase) == phase {
+					return evictable.IsEvictable(pod)
+				}
+			}
+			return false
+		}
+	}
+
 	for _, node := range nodes {
 		klog.V(1).InfoS("Processing node", "node", klog.KObj(node))
 
-		pods := listOldPodsOnNode(ctx, client, node, includedNamespaces, excludedNamespaces, *strategy.Params.MaxPodLifeTimeSeconds, evictable.IsEvictable)
+		pods := listOldPodsOnNode(ctx, client, node, includedNamespaces, excludedNamespaces, *strategy.Params.PodLifeTime.MaxPodLifeTimeSeconds, filter)
 		for _, pod := range pods {
 			success, err := podEvictor.EvictPod(ctx, pod, node, "PodLifeTime")
 			if success {
-				klog.V(1).InfoS("Evicted pod because it exceeded its lifetime", "pod", klog.KObj(pod), "maxPodLifeTime", *strategy.Params.MaxPodLifeTimeSeconds)
+				klog.V(1).InfoS("Evicted pod because it exceeded its lifetime", "pod", klog.KObj(pod), "maxPodLifeTime", *strategy.Params.PodLifeTime.MaxPodLifeTimeSeconds)
 			}
 
 			if err != nil {

pkg/descheduler/strategies/pod_lifetime_test.go

@@ -85,6 +85,19 @@ func TestPodLifeTime(t *testing.T) {
 	p5.ObjectMeta.OwnerReferences = ownerRef4
 	p6.ObjectMeta.OwnerReferences = ownerRef4
 
+	// Setup two old pods with different status phases
+	p9 := test.BuildTestPod("p9", 100, 0, node.Name, nil)
+	p9.Namespace = "dev"
+	p9.ObjectMeta.CreationTimestamp = olderPodCreationTime
+	p10 := test.BuildTestPod("p10", 100, 0, node.Name, nil)
+	p10.Namespace = "dev"
+	p10.ObjectMeta.CreationTimestamp = olderPodCreationTime
+
+	p9.Status.Phase = "Pending"
+	p10.Status.Phase = "Running"
+	p9.ObjectMeta.OwnerReferences = ownerRef1
+	p10.ObjectMeta.OwnerReferences = ownerRef1
+
 	var maxLifeTime uint = 600
 	testCases := []struct {
 		description             string
@@ -98,7 +111,7 @@ func TestPodLifeTime(t *testing.T) {
 			strategy: api.DeschedulerStrategy{
 				Enabled: true,
 				Params: &api.StrategyParameters{
-					MaxPodLifeTimeSeconds: &maxLifeTime,
+					PodLifeTime: &api.PodLifeTime{MaxPodLifeTimeSeconds: &maxLifeTime},
 				},
 			},
 			maxPodsToEvictPerNode:   5,
@@ -110,7 +123,7 @@ func TestPodLifeTime(t *testing.T) {
 			strategy: api.DeschedulerStrategy{
 				Enabled: true,
 				Params: &api.StrategyParameters{
-					MaxPodLifeTimeSeconds: &maxLifeTime,
+					PodLifeTime: &api.PodLifeTime{MaxPodLifeTimeSeconds: &maxLifeTime},
 				},
 			},
 			maxPodsToEvictPerNode:   5,
@@ -122,7 +135,7 @@ func TestPodLifeTime(t *testing.T) {
 			strategy: api.DeschedulerStrategy{
 				Enabled: true,
 				Params: &api.StrategyParameters{
-					MaxPodLifeTimeSeconds: &maxLifeTime,
+					PodLifeTime: &api.PodLifeTime{MaxPodLifeTimeSeconds: &maxLifeTime},
 				},
 			},
 			maxPodsToEvictPerNode:   5,
@@ -134,13 +147,28 @@ func TestPodLifeTime(t *testing.T) {
 			strategy: api.DeschedulerStrategy{
 				Enabled: true,
 				Params: &api.StrategyParameters{
-					MaxPodLifeTimeSeconds: &maxLifeTime,
+					PodLifeTime: &api.PodLifeTime{MaxPodLifeTimeSeconds: &maxLifeTime},
 				},
 			},
 			maxPodsToEvictPerNode:   5,
 			pods:                    []v1.Pod{*p7, *p8},
 			expectedEvictedPodCount: 0,
 		},
+		{
+			description: "Two old pods with different status phases. 1 should be evicted.",
+			strategy: api.DeschedulerStrategy{
+				Enabled: true,
+				Params: &api.StrategyParameters{
+					PodLifeTime: &api.PodLifeTime{
+						MaxPodLifeTimeSeconds: &maxLifeTime,
+						PodStatusPhases:       []string{"Pending"},
+					},
+				},
+			},
+			maxPodsToEvictPerNode:   5,
+			pods:                    []v1.Pod{*p9, *p10},
+			expectedEvictedPodCount: 1,
+		},
 	}
 
 	for _, tc := range testCases {

test/e2e/e2e_test.go

@@ -206,7 +206,7 @@ func runPodLifetimeStrategy(ctx context.Context, clientset clientset.Interface,
 		deschedulerapi.DeschedulerStrategy{
 			Enabled: true,
 			Params: &deschedulerapi.StrategyParameters{
-				MaxPodLifeTimeSeconds:      &maxPodLifeTimeSeconds,
+				PodLifeTime:                &deschedulerapi.PodLifeTime{MaxPodLifeTimeSeconds: &maxPodLifeTimeSeconds},
 				Namespaces:                 namespaces,
 				ThresholdPriority:          priority,
 				ThresholdPriorityClassName: priorityClass,