Merge pull request #89222 from fengzixu/master

bugfix: initcontainer is also taken into account when calculating resource requests

pkg/scheduler/framework/plugins/noderesources/resource_allocation.go

@@ -114,6 +114,7 @@ func calculateResourceAllocatableRequest(nodeInfo *framework.NodeInfo, pod *v1.P
 
 // calculatePodResourceRequest 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.
+// podResourceRequest = max(sum(podSpec.Containers), podSpec.InitContainers) + overHead
 func calculatePodResourceRequest(pod *v1.Pod, resource v1.ResourceName) int64 {
 	var podRequest int64
 	for i := range pod.Spec.Containers {
@@ -122,11 +123,20 @@ func calculatePodResourceRequest(pod *v1.Pod, resource v1.ResourceName) int64 {
 		podRequest += value
 	}
 
+	for i := range pod.Spec.InitContainers {
+		initContainer := &pod.Spec.InitContainers[i]
+		value := schedutil.GetNonzeroRequestForResource(resource, &initContainer.Resources.Requests)
+		if podRequest < value {
+			podRequest = value
+		}
+	}
+
 	// If Overhead is being utilized, add to the total requests for the pod
 	if pod.Spec.Overhead != nil && utilfeature.DefaultFeatureGate.Enabled(features.PodOverhead) {
 		if quantity, found := pod.Spec.Overhead[resource]; found {
 			podRequest += quantity.Value()
 		}
 	}
+
 	return podRequest
 }

pkg/scheduler/framework/v1alpha1/types.go

@@ -208,7 +208,10 @@ func (r *Resource) Add(rl v1.ResourceList) {
 		case v1.ResourcePods:
 			r.AllowedPodNumber += int(rQuant.Value())
 		case v1.ResourceEphemeralStorage:
+			if utilfeature.DefaultFeatureGate.Enabled(features.LocalStorageCapacityIsolation) {
+				// if the local storage capacity isolation feature gate is disabled, pods request 0 disk.
 				r.EphemeralStorage += rQuant.Value()
+			}
 		default:
 			if v1helper.IsScalarResourceName(rName) {
 				r.AddScalar(rName, rQuant.Value())
@@ -458,21 +461,32 @@ func (n *NodeInfo) resetSlicesIfEmpty() {
 	}
 }
 
+// resourceRequest = max(sum(podSpec.Containers), podSpec.InitContainers) + overHead
 func calculateResource(pod *v1.Pod) (res Resource, non0CPU int64, non0Mem int64) {
 	resPtr := &res
 	for _, c := range pod.Spec.Containers {
 		resPtr.Add(c.Resources.Requests)
-
 		non0CPUReq, non0MemReq := schedutil.GetNonzeroRequests(&c.Resources.Requests)
 		non0CPU += non0CPUReq
 		non0Mem += non0MemReq
 		// No non-zero resources for GPUs or opaque resources.
 	}
 
+	for _, ic := range pod.Spec.InitContainers {
+		resPtr.SetMaxResource(ic.Resources.Requests)
+		non0CPUReq, non0MemReq := schedutil.GetNonzeroRequests(&ic.Resources.Requests)
+		if non0CPU < non0CPUReq {
+			non0CPU = non0CPUReq
+		}
+
+		if non0Mem < non0MemReq {
+			non0Mem = non0MemReq
+		}
+	}
+
 	// 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()
 		}

pkg/scheduler/framework/v1alpha1/types_test.go

@@ -614,6 +614,46 @@ func TestNodeInfoAddPod(t *testing.T) {
 				},
 			},
 		},
+		{
+			ObjectMeta: metav1.ObjectMeta{
+				Namespace: "node_info_cache_test",
+				Name:      "test-3",
+				UID:       types.UID("test-3"),
+			},
+			Spec: v1.PodSpec{
+				Containers: []v1.Container{
+					{
+						Resources: v1.ResourceRequirements{
+							Requests: v1.ResourceList{
+								v1.ResourceCPU: resource.MustParse("200m"),
+							},
+						},
+						Ports: []v1.ContainerPort{
+							{
+								HostIP:   "127.0.0.1",
+								HostPort: 8080,
+								Protocol: "TCP",
+							},
+						},
+					},
+				},
+				InitContainers: []v1.Container{
+					{
+						Resources: v1.ResourceRequirements{
+							Requests: v1.ResourceList{
+								v1.ResourceCPU:    resource.MustParse("500m"),
+								v1.ResourceMemory: resource.MustParse("200Mi"),
+							},
+						},
+					},
+				},
+				NodeName: nodeName,
+				Overhead: v1.ResourceList{
+					v1.ResourceCPU:    resource.MustParse("500m"),
+					v1.ResourceMemory: resource.MustParse("500"),
+				},
+			},
+		},
 	}
 	expected := &NodeInfo{
 		node: &v1.Node{
@@ -622,15 +662,15 @@ func TestNodeInfoAddPod(t *testing.T) {
 			},
 		},
 		Requested: &Resource{
-			MilliCPU:         1300,
+			MilliCPU:         2300,
-			Memory:           1000,
+			Memory:           209716700, //1500 + 200MB in initContainers
 			EphemeralStorage: 0,
 			AllowedPodNumber: 0,
 			ScalarResources:  map[v1.ResourceName]int64(nil),
 		},
 		NonZeroRequested: &Resource{
-			MilliCPU:         1300,
+			MilliCPU:         2300,
-			Memory:           209716200, //200MB + 1000 specified in requests/overhead
+			Memory:           419431900, //200MB(initContainers) + 200MB(default memory value) + 1500 specified in requests/overhead
 			EphemeralStorage: 0,
 			AllowedPodNumber: 0,
 			ScalarResources:  map[v1.ResourceName]int64(nil),
@@ -710,6 +750,48 @@ func TestNodeInfoAddPod(t *testing.T) {
 					},
 				},
 			},
+			{
+				Pod: &v1.Pod{
+					ObjectMeta: metav1.ObjectMeta{
+						Namespace: "node_info_cache_test",
+						Name:      "test-3",
+						UID:       types.UID("test-3"),
+					},
+					Spec: v1.PodSpec{
+						Containers: []v1.Container{
+							{
+								Resources: v1.ResourceRequirements{
+									Requests: v1.ResourceList{
+										v1.ResourceCPU: resource.MustParse("200m"),
+									},
+								},
+								Ports: []v1.ContainerPort{
+									{
+										HostIP:   "127.0.0.1",
+										HostPort: 8080,
+										Protocol: "TCP",
+									},
+								},
+							},
+						},
+						InitContainers: []v1.Container{
+							{
+								Resources: v1.ResourceRequirements{
+									Requests: v1.ResourceList{
+										v1.ResourceCPU:    resource.MustParse("500m"),
+										v1.ResourceMemory: resource.MustParse("200Mi"),
+									},
+								},
+							},
+						},
+						NodeName: nodeName,
+						Overhead: v1.ResourceList{
+							v1.ResourceCPU:    resource.MustParse("500m"),
+							v1.ResourceMemory: resource.MustParse("500"),
+						},
+					},
+				},
+			},
 		},
 	}
 

pkg/scheduler/util/BUILD

@@ -38,10 +38,12 @@ go_library(
     deps = [
         "//pkg/api/v1/pod:go_default_library",
         "//pkg/apis/core/v1/helper:go_default_library",
+        "//pkg/features:go_default_library",
         "//staging/src/k8s.io/api/core/v1:go_default_library",
         "//staging/src/k8s.io/apimachinery/pkg/apis/meta/v1:go_default_library",
         "//staging/src/k8s.io/apimachinery/pkg/labels:go_default_library",
         "//staging/src/k8s.io/apimachinery/pkg/util/sets:go_default_library",
+        "//staging/src/k8s.io/apiserver/pkg/util/feature:go_default_library",
         "//staging/src/k8s.io/kube-scheduler/extender/v1:go_default_library",
         "//vendor/k8s.io/klog:go_default_library",
     ],

pkg/scheduler/util/non_zero.go

@@ -18,7 +18,9 @@ package util
 
 import (
 	v1 "k8s.io/api/core/v1"
+	utilfeature "k8s.io/apiserver/pkg/util/feature"
 	v1helper "k8s.io/kubernetes/pkg/apis/core/v1/helper"
+	"k8s.io/kubernetes/pkg/features"
 )
 
 // For each of these resources, a pod that doesn't request the resource explicitly
@@ -60,6 +62,11 @@ func GetNonzeroRequestForResource(resource v1.ResourceName, requests *v1.Resourc
 		}
 		return requests.Memory().Value()
 	case v1.ResourceEphemeralStorage:
+		// if the local storage capacity isolation feature gate is disabled, pods request 0 disk.
+		if !utilfeature.DefaultFeatureGate.Enabled(features.LocalStorageCapacityIsolation) {
+			return 0
+		}
+
 		quantity, found := (*requests)[v1.ResourceEphemeralStorage]
 		if !found {
 			return 0