scheduler: add pod Overhead support for requests

Signed-off-by: Eric Ernst <eric.ernst@intel.com>

pkg/scheduler/algorithm/predicates/predicates.go

@@ -726,6 +726,9 @@ func (c *VolumeZoneChecker) predicate(pod *v1.Pod, meta PredicateMetadata, nodeI
 // the max in each dimension iteratively. In contrast, we sum the resource vectors for
 // regular containers since they run simultaneously.
 //
+// If Pod Overhead is specified and the feature gate is set, the resources defined for Overhead
+// are added to the calculated Resource request sum
+//
 // Example:
 //
 // Pod:
@@ -756,6 +759,11 @@ func GetResourceRequest(pod *v1.Pod) *schedulernodeinfo.Resource {
 		result.SetMaxResource(container.Resources.Requests)
 	}
 
+	// If Overhead is being utilized, add to the total requests for the pod
+	if pod.Spec.Overhead != nil && utilfeature.DefaultFeatureGate.Enabled(features.PodOverhead) {
+		result.Add(pod.Spec.Overhead)
+	}
+
 	return result
 }
 

pkg/scheduler/algorithm/predicates/predicates_test.go

@@ -29,7 +29,10 @@ import (
 	"k8s.io/apimachinery/pkg/api/resource"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/util/sets"
+	utilfeature "k8s.io/apiserver/pkg/util/feature"
+	featuregatetesting "k8s.io/component-base/featuregate/testing"
 	v1helper "k8s.io/kubernetes/pkg/apis/core/v1/helper"
+	"k8s.io/kubernetes/pkg/features"
 	schedulerapi "k8s.io/kubernetes/pkg/scheduler/api"
 	schedulernodeinfo "k8s.io/kubernetes/pkg/scheduler/nodeinfo"
 	schedulertesting "k8s.io/kubernetes/pkg/scheduler/testing"
@@ -86,12 +89,20 @@ func newResourceInitPod(pod *v1.Pod, usage ...schedulernodeinfo.Resource) *v1.Po
 	return pod
 }
 
+func newResourceOverheadPod(pod *v1.Pod, overhead v1.ResourceList) *v1.Pod {
+	pod.Spec.Overhead = overhead
+	return pod
+}
+
 func GetPredicateMetadata(p *v1.Pod, nodeInfo map[string]*schedulernodeinfo.NodeInfo) PredicateMetadata {
 	pm := PredicateMetadataFactory{schedulertesting.FakePodLister{p}}
 	return pm.GetMetadata(p, nodeInfo)
 }
 
 func TestPodFitsResources(t *testing.T) {
+
+	defer featuregatetesting.SetFeatureGateDuringTest(t, utilfeature.DefaultFeatureGate, features.PodOverhead, true)()
+
 	enoughPodsTests := []struct {
 		pod                      *v1.Pod
 		nodeInfo                 *schedulernodeinfo.NodeInfo
@@ -351,6 +362,31 @@ func TestPodFitsResources(t *testing.T) {
 			ignoredExtendedResources: sets.NewString(string(extendedResourceB)),
 			name:                     "skip checking ignored extended resource",
 		},
+		{
+			pod: newResourceOverheadPod(
+				newResourcePod(schedulernodeinfo.Resource{MilliCPU: 1, Memory: 1}),
+				v1.ResourceList{v1.ResourceCPU: resource.MustParse("3m"), v1.ResourceMemory: resource.MustParse("13")},
+			),
+			nodeInfo: schedulernodeinfo.NewNodeInfo(
+				newResourcePod(schedulernodeinfo.Resource{MilliCPU: 5, Memory: 5})),
+			fits:                     true,
+			ignoredExtendedResources: sets.NewString(string(extendedResourceB)),
+			name:                     "resources + pod overhead fits",
+		},
+		{
+			pod: newResourceOverheadPod(
+				newResourcePod(schedulernodeinfo.Resource{MilliCPU: 1, Memory: 1}),
+				v1.ResourceList{v1.ResourceCPU: resource.MustParse("1m"), v1.ResourceMemory: resource.MustParse("15")},
+			),
+			nodeInfo: schedulernodeinfo.NewNodeInfo(
+				newResourcePod(schedulernodeinfo.Resource{MilliCPU: 5, Memory: 5})),
+			fits:                     false,
+			ignoredExtendedResources: sets.NewString(string(extendedResourceB)),
+			name:                     "requests + overhead does not fit for memory",
+			reasons: []PredicateFailureReason{
+				NewInsufficientResourceError(v1.ResourceMemory, 16, 5, 20),
+			},
+		},
 	}
 
 	for _, test := range enoughPodsTests {

pkg/scheduler/algorithm/priorities/resource_allocation.go

@@ -91,6 +91,8 @@ func (r *ResourceAllocationPriority) PriorityMap(
 	}, nil
 }
 
+// getNonZeroRequests returns the total non-zero requests. If Overhead is defined for the pod and the
+// PodOverhead feature is enabled, the Overhead is added to the result.
 func getNonZeroRequests(pod *v1.Pod) *schedulernodeinfo.Resource {
 	result := &schedulernodeinfo.Resource{}
 	for i := range pod.Spec.Containers {
@@ -99,5 +101,17 @@ func getNonZeroRequests(pod *v1.Pod) *schedulernodeinfo.Resource {
 		result.MilliCPU += cpu
 		result.Memory += memory
 	}
+
+	// If Overhead is being utilized, add to the total requests for the pod
+	if pod.Spec.Overhead != nil && utilfeature.DefaultFeatureGate.Enabled(features.PodOverhead) {
+		if _, found := pod.Spec.Overhead[v1.ResourceCPU]; found {
+			result.MilliCPU += pod.Spec.Overhead.Cpu().MilliValue()
+		}
+
+		if _, found := pod.Spec.Overhead[v1.ResourceMemory]; found {
+			result.Memory += pod.Spec.Overhead.Memory().Value()
+		}
+	}
+
 	return result
 }

pkg/scheduler/nodeinfo/node_info.go

@@ -25,8 +25,10 @@ import (
 	v1 "k8s.io/api/core/v1"
 	storagev1beta1 "k8s.io/api/storage/v1beta1"
 	"k8s.io/apimachinery/pkg/api/resource"
+	utilfeature "k8s.io/apiserver/pkg/util/feature"
 	"k8s.io/klog"
 	v1helper "k8s.io/kubernetes/pkg/apis/core/v1/helper"
+	"k8s.io/kubernetes/pkg/features"
 	priorityutil "k8s.io/kubernetes/pkg/scheduler/algorithm/priorities/util"
 	volumeutil "k8s.io/kubernetes/pkg/volume/util"
 )
@@ -613,6 +615,19 @@ func calculateResource(pod *v1.Pod) (res Resource, non0CPU int64, non0Mem int64)
 		// No non-zero resources for GPUs or opaque resources.
 	}
 
+	// If Overhead is being utilized, add to the total requests for the pod
+	if pod.Spec.Overhead != nil && utilfeature.DefaultFeatureGate.Enabled(features.PodOverhead) {
+		resPtr.Add(pod.Spec.Overhead)
+
+		if _, found := pod.Spec.Overhead[v1.ResourceCPU]; found {
+			non0CPU += pod.Spec.Overhead.Cpu().MilliValue()
+		}
+
+		if _, found := pod.Spec.Overhead[v1.ResourceMemory]; found {
+			non0Mem += pod.Spec.Overhead.Memory().Value()
+		}
+	}
+
 	return
 }
 

pkg/scheduler/nodeinfo/node_info_test.go

@@ -26,6 +26,9 @@ import (
 	"k8s.io/apimachinery/pkg/api/resource"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/types"
+	utilfeature "k8s.io/apiserver/pkg/util/feature"
+	featuregatetesting "k8s.io/component-base/featuregate/testing"
+	"k8s.io/kubernetes/pkg/features"
 )
 
 func TestNewResource(t *testing.T) {
@@ -540,6 +543,9 @@ func TestNodeInfoClone(t *testing.T) {
 }
 
 func TestNodeInfoAddPod(t *testing.T) {
+
+	defer featuregatetesting.SetFeatureGateDuringTest(t, utilfeature.DefaultFeatureGate, features.PodOverhead, true)()
+
 	nodeName := "test-node"
 	pods := []*v1.Pod{
 		{
@@ -567,6 +573,9 @@ func TestNodeInfoAddPod(t *testing.T) {
 					},
 				},
 				NodeName: nodeName,
+				Overhead: v1.ResourceList{
+					v1.ResourceCPU: resource.MustParse("500m"),
+				},
 			},
 		},
 		{
@@ -580,8 +589,7 @@ func TestNodeInfoAddPod(t *testing.T) {
 					{
 						Resources: v1.ResourceRequirements{
 							Requests: v1.ResourceList{
-								v1.ResourceCPU:    resource.MustParse("200m"),
+								v1.ResourceCPU: resource.MustParse("200m"),
-								v1.ResourceMemory: resource.MustParse("1Ki"),
 							},
 						},
 						Ports: []v1.ContainerPort{
@@ -594,6 +602,10 @@ func TestNodeInfoAddPod(t *testing.T) {
 					},
 				},
 				NodeName: nodeName,
+				Overhead: v1.ResourceList{
+					v1.ResourceCPU:    resource.MustParse("500m"),
+					v1.ResourceMemory: resource.MustParse("500"),
+				},
 			},
 		},
 	}
@@ -604,15 +616,15 @@ func TestNodeInfoAddPod(t *testing.T) {
 			},
 		},
 		requestedResource: &Resource{
-			MilliCPU:         300,
+			MilliCPU:         1300,
-			Memory:           1524,
+			Memory:           1000,
 			EphemeralStorage: 0,
 			AllowedPodNumber: 0,
 			ScalarResources:  map[v1.ResourceName]int64(nil),
 		},
 		nonzeroRequest: &Resource{
-			MilliCPU:         300,
+			MilliCPU:         1300,
-			Memory:           1524,
+			Memory:           209716200, //200MB + 1000 specified in requests/overhead
 			EphemeralStorage: 0,
 			AllowedPodNumber: 0,
 			ScalarResources:  map[v1.ResourceName]int64(nil),
@@ -653,6 +665,9 @@ func TestNodeInfoAddPod(t *testing.T) {
 						},
 					},
 					NodeName: nodeName,
+					Overhead: v1.ResourceList{
+						v1.ResourceCPU: resource.MustParse("500m"),
+					},
 				},
 			},
 			{
@@ -666,8 +681,7 @@ func TestNodeInfoAddPod(t *testing.T) {
 						{
 							Resources: v1.ResourceRequirements{
 								Requests: v1.ResourceList{
-									v1.ResourceCPU:    resource.MustParse("200m"),
+									v1.ResourceCPU: resource.MustParse("200m"),
-									v1.ResourceMemory: resource.MustParse("1Ki"),
 								},
 							},
 							Ports: []v1.ContainerPort{
@@ -680,6 +694,10 @@ func TestNodeInfoAddPod(t *testing.T) {
 						},
 					},
 					NodeName: nodeName,
+					Overhead: v1.ResourceList{
+						v1.ResourceCPU:    resource.MustParse("500m"),
+						v1.ResourceMemory: resource.MustParse("500"),
+					},
 				},
 			},
 		},
@@ -702,12 +720,23 @@ func TestNodeInfoAddPod(t *testing.T) {
 }
 
 func TestNodeInfoRemovePod(t *testing.T) {
+
+	defer featuregatetesting.SetFeatureGateDuringTest(t, utilfeature.DefaultFeatureGate, features.PodOverhead, true)()
+
 	nodeName := "test-node"
 	pods := []*v1.Pod{
 		makeBasePod(t, nodeName, "test-1", "100m", "500", "", []v1.ContainerPort{{HostIP: "127.0.0.1", HostPort: 80, Protocol: "TCP"}}),
 		makeBasePod(t, nodeName, "test-2", "200m", "1Ki", "", []v1.ContainerPort{{HostIP: "127.0.0.1", HostPort: 8080, Protocol: "TCP"}}),
 	}
 
+	// add pod Overhead
+	for _, pod := range pods {
+		pod.Spec.Overhead = v1.ResourceList{
+			v1.ResourceCPU:    resource.MustParse("500m"),
+			v1.ResourceMemory: resource.MustParse("500"),
+		}
+	}
+
 	tests := []struct {
 		pod              *v1.Pod
 		errExpected      bool
@@ -723,15 +752,15 @@ func TestNodeInfoRemovePod(t *testing.T) {
 					},
 				},
 				requestedResource: &Resource{
-					MilliCPU:         300,
+					MilliCPU:         1300,
-					Memory:           1524,
+					Memory:           2524,
 					EphemeralStorage: 0,
 					AllowedPodNumber: 0,
 					ScalarResources:  map[v1.ResourceName]int64(nil),
 				},
 				nonzeroRequest: &Resource{
-					MilliCPU:         300,
+					MilliCPU:         1300,
-					Memory:           1524,
+					Memory:           2524,
 					EphemeralStorage: 0,
 					AllowedPodNumber: 0,
 					ScalarResources:  map[v1.ResourceName]int64(nil),
@@ -772,6 +801,10 @@ func TestNodeInfoRemovePod(t *testing.T) {
 								},
 							},
 							NodeName: nodeName,
+							Overhead: v1.ResourceList{
+								v1.ResourceCPU:    resource.MustParse("500m"),
+								v1.ResourceMemory: resource.MustParse("500"),
+							},
 						},
 					},
 					{
@@ -799,6 +832,10 @@ func TestNodeInfoRemovePod(t *testing.T) {
 								},
 							},
 							NodeName: nodeName,
+							Overhead: v1.ResourceList{
+								v1.ResourceCPU:    resource.MustParse("500m"),
+								v1.ResourceMemory: resource.MustParse("500"),
+							},
 						},
 					},
 				},
@@ -830,6 +867,10 @@ func TestNodeInfoRemovePod(t *testing.T) {
 						},
 					},
 					NodeName: nodeName,
+					Overhead: v1.ResourceList{
+						v1.ResourceCPU:    resource.MustParse("500m"),
+						v1.ResourceMemory: resource.MustParse("500"),
+					},
 				},
 			},
 			errExpected: false,
@@ -840,15 +881,15 @@ func TestNodeInfoRemovePod(t *testing.T) {
 					},
 				},
 				requestedResource: &Resource{
-					MilliCPU:         200,
+					MilliCPU:         700,
-					Memory:           1024,
+					Memory:           1524,
 					EphemeralStorage: 0,
 					AllowedPodNumber: 0,
 					ScalarResources:  map[v1.ResourceName]int64(nil),
 				},
 				nonzeroRequest: &Resource{
-					MilliCPU:         200,
+					MilliCPU:         700,
-					Memory:           1024,
+					Memory:           1524,
 					EphemeralStorage: 0,
 					AllowedPodNumber: 0,
 					ScalarResources:  map[v1.ResourceName]int64(nil),
@@ -888,6 +929,10 @@ func TestNodeInfoRemovePod(t *testing.T) {
 								},
 							},
 							NodeName: nodeName,
+							Overhead: v1.ResourceList{
+								v1.ResourceCPU:    resource.MustParse("500m"),
+								v1.ResourceMemory: resource.MustParse("500"),
+							},
 						},
 					},
 				},