VPA: Implement startup CPU boost support

vertical-pod-autoscaler/deploy/vpa-v1-crd-gen.yaml

@@ -372,9 +372,82 @@ spec:
                           - Auto
                           - "Off"
                           type: string
+                        startupBoost:
+                          description: |-
+                            StartupBoost specifies the startup boost policy for the container.
+                            This overrides any pod-level startup boost policy.
+                          properties:
+                            cpu:
+                              description: CPU specifies the CPU startup boost policy.
+                              properties:
+                                duration:
+                                  description: |-
+                                    Duration indicates for how long to keep the pod boosted after it goes to Ready.
+                                    Defaults to 0s.
+                                  type: string
+                                factor:
+                                  description: Factor specifies the factor to apply
+                                    to the CPU request.
+                                  format: int32
+                                  type: integer
+                                quantity:
+                                  anyOf:
+                                  - type: integer
+                                  - type: string
+                                  description: Quantity specifies the absolute CPU
+                                    resource quantity.
+                                  pattern: ^(\+|-)?(([0-9]+(\.[0-9]*)?)|(\.[0-9]+))(([KMGTPE]i)|[numkMGTPE]|([eE](\+|-)?(([0-9]+(\.[0-9]*)?)|(\.[0-9]+))))?$
+                                  x-kubernetes-int-or-string: true
+                                type:
+                                  description: |-
+                                    Type specifies the kind of boost to apply.
+                                    Supported values are: "Factor", "Quantity".
+                                    Defaults to "Factor".
+                                  enum:
+                                  - Factor
+                                  - Quantity
+                                  type: string
+                              type: object
+                          type: object
                       type: object
                     type: array
                 type: object
+              startupBoost:
+                description: StartupBoost specifies the startup boost policy for the
+                  pod.
+                properties:
+                  cpu:
+                    description: CPU specifies the CPU startup boost policy.
+                    properties:
+                      duration:
+                        description: |-
+                          Duration indicates for how long to keep the pod boosted after it goes to Ready.
+                          Defaults to 0s.
+                        type: string
+                      factor:
+                        description: Factor specifies the factor to apply to the CPU
+                          request.
+                        format: int32
+                        type: integer
+                      quantity:
+                        anyOf:
+                        - type: integer
+                        - type: string
+                        description: Quantity specifies the absolute CPU resource
+                          quantity.
+                        pattern: ^(\+|-)?(([0-9]+(\.[0-9]*)?)|(\.[0-9]+))(([KMGTPE]i)|[numkMGTPE]|([eE](\+|-)?(([0-9]+(\.[0-9]*)?)|(\.[0-9]+))))?$
+                        x-kubernetes-int-or-string: true
+                      type:
+                        description: |-
+                          Type specifies the kind of boost to apply.
+                          Supported values are: "Factor", "Quantity".
+                          Defaults to "Factor".
+                        enum:
+                        - Factor
+                        - Quantity
+                        type: string
+                    type: object
+                type: object
               targetRef:
                 description: |-
                   TargetRef points to the controller managing the set of pods for the

vertical-pod-autoscaler/e2e/v1/actuation.go

@@ -35,6 +35,7 @@ import (
 	"k8s.io/apimachinery/pkg/util/wait"
 	"k8s.io/autoscaler/vertical-pod-autoscaler/e2e/utils"
 	vpa_types "k8s.io/autoscaler/vertical-pod-autoscaler/pkg/apis/autoscaling.k8s.io/v1"
+	"k8s.io/autoscaler/vertical-pod-autoscaler/pkg/features"
 	restriction "k8s.io/autoscaler/vertical-pod-autoscaler/pkg/updater/restriction"
 	updaterutils "k8s.io/autoscaler/vertical-pod-autoscaler/pkg/updater/utils"
 	"k8s.io/autoscaler/vertical-pod-autoscaler/pkg/utils/annotations"
@@ -57,7 +58,7 @@ var _ = ActuationSuiteE2eDescribe("Actuation", ginkgo.Label("FG:InPlaceOrRecreat
 	f.NamespacePodSecurityEnforceLevel = podsecurity.LevelBaseline
 
 	ginkgo.BeforeEach(func() {
-		checkInPlaceOrRecreateTestsEnabled(f, true, true)
+		checkFeatureGateTestsEnabled(f, features.InPlaceOrRecreate, true, true)
 	})
 
 	ginkgo.It("still applies recommendations on restart when update mode is InPlaceOrRecreate", func() {

vertical-pod-autoscaler/e2e/v1/admission_controller.go

@@ -28,6 +28,7 @@ import (
 	apiv1 "k8s.io/api/core/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	vpa_types "k8s.io/autoscaler/vertical-pod-autoscaler/pkg/apis/autoscaling.k8s.io/v1"
+	"k8s.io/autoscaler/vertical-pod-autoscaler/pkg/features"
 	"k8s.io/autoscaler/vertical-pod-autoscaler/pkg/utils/test"
 	"k8s.io/kubernetes/test/e2e/framework"
 	framework_deployment "k8s.io/kubernetes/test/e2e/framework/deployment"
@@ -47,7 +48,7 @@ var _ = AdmissionControllerE2eDescribe("Admission-controller", ginkgo.Label("FG:
 	f.NamespacePodSecurityEnforceLevel = podsecurity.LevelBaseline
 
 	ginkgo.BeforeEach(func() {
-		checkInPlaceOrRecreateTestsEnabled(f, true, false)
+		checkFeatureGateTestsEnabled(f, features.InPlaceOrRecreate, true, false)
 		waitForVpaWebhookRegistration(f)
 	})
 
@@ -961,6 +962,123 @@ var _ = AdmissionControllerE2eDescribe("Admission-controller", func() {
 	})
 })
 
+var _ = AdmissionControllerE2eDescribe("Admission-controller", ginkgo.Label("FG:CPUStartupBoost"), func() {
+	f := framework.NewDefaultFramework("vertical-pod-autoscaling")
+	f.NamespacePodSecurityEnforceLevel = podsecurity.LevelBaseline
+
+	ginkgo.BeforeEach(func() {
+		checkFeatureGateTestsEnabled(f, features.CPUStartupBoost, true, false)
+		waitForVpaWebhookRegistration(f)
+	})
+
+	ginkgo.It("boosts CPU by factor on pod creation", func() {
+		initialCPU := ParseQuantityOrDie("100m")
+		boostedCPU := ParseQuantityOrDie("200m")
+		d := NewHamsterDeploymentWithResources(f, initialCPU, ParseQuantityOrDie("100Mi"))
+
+		ginkgo.By("Setting up a VPA with a startup boost policy (factor)")
+		containerName := GetHamsterContainerNameByIndex(0)
+		factor := int32(2)
+		vpaCRD := test.VerticalPodAutoscaler().
+			WithName("hamster-vpa").
+			WithNamespace(f.Namespace.Name).
+			WithTargetRef(hamsterTargetRef).
+			WithContainer(containerName).
+			WithCPUStartupBoost(&factor, nil, "15s").
+			AppendRecommendation(
+				test.Recommendation().
+					WithContainer(containerName).
+					WithTarget("100m", "100Mi").
+					GetContainerResources(),
+			).
+			Get()
+		InstallVPA(f, vpaCRD)
+
+		ginkgo.By("Starting the deployment and verifying the pod is boosted")
+		podList := startDeploymentPods(f, d)
+		pod := podList.Items[0]
+		gomega.Expect(pod.Spec.Containers[0].Resources.Requests.Cpu().Cmp(boostedCPU)).To(gomega.Equal(0))
+		gomega.Expect(pod.Spec.Containers[0].Resources.Limits.Cpu().Cmp(boostedCPU)).To(gomega.Equal(0))
+	})
+
+	ginkgo.It("boosts CPU by quantity on pod creation", func() {
+		initialCPU := ParseQuantityOrDie("100m")
+		boostedCPU := ParseQuantityOrDie("500m")
+		d := NewHamsterDeploymentWithResources(f, initialCPU, ParseQuantityOrDie("100Mi"))
+
+		ginkgo.By("Setting up a VPA with a startup boost policy (quantity)")
+		containerName := GetHamsterContainerNameByIndex(0)
+		quantity := ParseQuantityOrDie("500m")
+		vpaCRD := test.VerticalPodAutoscaler().
+			WithName("hamster-vpa").
+			WithNamespace(f.Namespace.Name).
+			WithTargetRef(hamsterTargetRef).
+			WithContainer(containerName).
+			WithCPUStartupBoost(nil, &quantity, "15s").
+			AppendRecommendation(
+				test.Recommendation().
+					WithContainer(containerName).
+					WithTarget("100m", "100Mi").
+					GetContainerResources(),
+			).
+			Get()
+		InstallVPA(f, vpaCRD)
+
+		ginkgo.By("Starting the deployment and verifying the pod is boosted")
+		podList := startDeploymentPods(f, d)
+		pod := podList.Items[0]
+		gomega.Expect(pod.Spec.Containers[0].Resources.Requests.Cpu().Cmp(boostedCPU)).To(gomega.Equal(0))
+		gomega.Expect(pod.Spec.Containers[0].Resources.Limits.Cpu().Cmp(boostedCPU)).To(gomega.Equal(0))
+	})
+
+	ginkgo.It("boosts CPU on pod creation when VPA update mode is Off", func() {
+		initialCPU := ParseQuantityOrDie("100m")
+		boostedCPU := ParseQuantityOrDie("200m")
+		d := NewHamsterDeploymentWithResources(f, initialCPU, ParseQuantityOrDie("100Mi"))
+
+		ginkgo.By("Setting up a VPA with updateMode Off and a startup boost policy")
+		containerName := GetHamsterContainerNameByIndex(0)
+		factor := int32(2)
+		vpaCRD := test.VerticalPodAutoscaler().
+			WithName("hamster-vpa").
+			WithNamespace(f.Namespace.Name).
+			WithTargetRef(hamsterTargetRef).
+			WithContainer(containerName).
+			WithUpdateMode(vpa_types.UpdateModeOff). // VPA is off, but boost should still work
+			WithCPUStartupBoost(&factor, nil, "15s").
+			Get()
+		InstallVPA(f, vpaCRD)
+
+		ginkgo.By("Starting the deployment and verifying the pod is boosted")
+		podList := startDeploymentPods(f, d)
+		pod := podList.Items[0]
+		gomega.Expect(pod.Spec.Containers[0].Resources.Requests.Cpu().Cmp(boostedCPU)).To(gomega.Equal(0))
+	})
+
+	ginkgo.It("doesn't boost CPU on pod creation when scaling mode is Off", func() {
+		initialCPU := ParseQuantityOrDie("100m")
+		d := NewHamsterDeploymentWithResources(f, initialCPU, ParseQuantityOrDie("100Mi"))
+
+		ginkgo.By("Setting up a VPA with a startup boost policy and scaling mode Off")
+		containerName := GetHamsterContainerNameByIndex(0)
+		factor := int32(2)
+		vpaCRD := test.VerticalPodAutoscaler().
+			WithName("hamster-vpa").
+			WithNamespace(f.Namespace.Name).
+			WithTargetRef(hamsterTargetRef).
+			WithContainer(containerName).
+			WithCPUStartupBoost(&factor, nil, "15s").
+			WithScalingMode(containerName, vpa_types.ContainerScalingModeOff).
+			Get()
+		InstallVPA(f, vpaCRD)
+
+		ginkgo.By("Starting the deployment and verifying the pod is NOT boosted")
+		podList := startDeploymentPods(f, d)
+		pod := podList.Items[0]
+		gomega.Expect(pod.Spec.Containers[0].Resources.Requests.Cpu().Cmp(initialCPU)).To(gomega.Equal(0))
+	})
+})
+
 func startDeploymentPods(f *framework.Framework, deployment *appsv1.Deployment) *apiv1.PodList {
 	// Apiserver watch can lag depending on cached object count and apiserver resource usage.
 	// We assume that watch can lag up to 5 seconds.

vertical-pod-autoscaler/e2e/v1/common.go

@@ -37,8 +37,8 @@ import (
 	"k8s.io/apimachinery/pkg/util/wait"
 	vpa_types "k8s.io/autoscaler/vertical-pod-autoscaler/pkg/apis/autoscaling.k8s.io/v1"
 	vpa_clientset "k8s.io/autoscaler/vertical-pod-autoscaler/pkg/client/clientset/versioned"
-	"k8s.io/autoscaler/vertical-pod-autoscaler/pkg/features"
 	clientset "k8s.io/client-go/kubernetes"
+	"k8s.io/component-base/featuregate"
 	"k8s.io/kubernetes/test/e2e/framework"
 	framework_deployment "k8s.io/kubernetes/test/e2e/framework/deployment"
 )
@@ -359,14 +359,30 @@ func PatchVpaRecommendation(f *framework.Framework, vpa *vpa_types.VerticalPodAu
 
 // AnnotatePod adds annotation for an existing pod.
 func AnnotatePod(f *framework.Framework, podName, annotationName, annotationValue string) {
-	bytes, err := json.Marshal([]patchRecord{{
+	pod, err := f.ClientSet.CoreV1().Pods(f.Namespace.Name).Get(context.TODO(), podName, metav1.GetOptions{})
+	gomega.Expect(err).NotTo(gomega.HaveOccurred(), "Failed to get pod.")
+
+	patches := []patchRecord{}
+	if pod.Annotations == nil {
+		patches = append(patches, patchRecord{
+			Op:    "add",
+			Path:  "/metadata/annotations",
+			Value: make(map[string]string),
+		})
+	}
+
+	patches = append(patches, patchRecord{
 		Op:    "add",
-		Path:  fmt.Sprintf("/metadata/annotations/%v", annotationName),
+		Path:  fmt.Sprintf("/metadata/annotations/%s", annotationName),
 		Value: annotationValue,
-	}})
-	pod, err := f.ClientSet.CoreV1().Pods(f.Namespace.Name).Patch(context.TODO(), podName, types.JSONPatchType, bytes, metav1.PatchOptions{})
+	})
+
+	bytes, err := json.Marshal(patches)
+	gomega.Expect(err).NotTo(gomega.HaveOccurred())
+
+	patchedPod, err := f.ClientSet.CoreV1().Pods(f.Namespace.Name).Patch(context.TODO(), podName, types.JSONPatchType, bytes, metav1.PatchOptions{})
 	gomega.Expect(err).NotTo(gomega.HaveOccurred(), "Failed to patch pod.")
-	gomega.Expect(pod.Annotations[annotationName]).To(gomega.Equal(annotationValue))
+	gomega.Expect(patchedPod.Annotations[annotationName]).To(gomega.Equal(annotationValue))
 }
 
 // ParseQuantityOrDie parses quantity from string and dies with an error if
@@ -613,35 +629,35 @@ func WaitForPodsUpdatedWithoutEviction(f *framework.Framework, initialPods *apiv
 	return err
 }
 
-// checkInPlaceOrRecreateTestsEnabled check for enabled feature gates in the cluster used for the
-// InPlaceOrRecreate VPA feature.
-// Use this in a "beforeEach" call before any suites that use InPlaceOrRecreate featuregate.
-func checkInPlaceOrRecreateTestsEnabled(f *framework.Framework, checkAdmission, checkUpdater bool) {
-	ginkgo.By("Checking InPlacePodVerticalScaling cluster feature gate is on")
+// checkFeatureGateTestsEnabled check for enabled feature gates in the cluster used for the
+// given VPA feature.
+// Use this in a "beforeEach" call before any suites that use a featuregate.
+func checkFeatureGateTestsEnabled(f *framework.Framework, feature featuregate.Feature, checkAdmission, checkUpdater bool) {
+	ginkgo.By(fmt.Sprintf("Checking %s cluster feature gate is on", feature))
 
 	if checkUpdater {
-		ginkgo.By("Checking InPlaceOrRecreate VPA feature gate is enabled for updater")
+		ginkgo.By(fmt.Sprintf("Checking %s VPA feature gate is enabled for updater", feature))
 
 		deploy, err := f.ClientSet.AppsV1().Deployments(VpaNamespace).Get(context.TODO(), "vpa-updater", metav1.GetOptions{})
 		gomega.Expect(err).NotTo(gomega.HaveOccurred())
 		gomega.Expect(deploy.Spec.Template.Spec.Containers).To(gomega.HaveLen(1))
 		vpaUpdaterPod := deploy.Spec.Template.Spec.Containers[0]
 		gomega.Expect(vpaUpdaterPod.Name).To(gomega.Equal("updater"))
-		if !anyContainsSubstring(vpaUpdaterPod.Args, fmt.Sprintf("%s=true", string(features.InPlaceOrRecreate))) {
-			ginkgo.Skip("Skipping suite: InPlaceOrRecreate feature gate is not enabled for the VPA updater")
+		if !anyContainsSubstring(vpaUpdaterPod.Args, fmt.Sprintf("%s=true", string(feature))) {
+			ginkgo.Skip(fmt.Sprintf("Skipping suite: %s feature gate is not enabled for the VPA updater", feature))
 		}
 	}
 
 	if checkAdmission {
-		ginkgo.By("Checking InPlaceOrRecreate VPA feature gate is enabled for admission controller")
+		ginkgo.By(fmt.Sprintf("Checking %s VPA feature gate is enabled for admission controller", feature))
 
 		deploy, err := f.ClientSet.AppsV1().Deployments(VpaNamespace).Get(context.TODO(), "vpa-admission-controller", metav1.GetOptions{})
 		gomega.Expect(err).NotTo(gomega.HaveOccurred())
 		gomega.Expect(deploy.Spec.Template.Spec.Containers).To(gomega.HaveLen(1))
 		vpaAdmissionPod := deploy.Spec.Template.Spec.Containers[0]
 		gomega.Expect(vpaAdmissionPod.Name).To(gomega.Equal("admission-controller"))
-		if !anyContainsSubstring(vpaAdmissionPod.Args, fmt.Sprintf("%s=true", string(features.InPlaceOrRecreate))) {
-			ginkgo.Skip("Skipping suite: InPlaceOrRecreate feature gate is not enabled for VPA admission controller")
+		if !anyContainsSubstring(vpaAdmissionPod.Args, fmt.Sprintf("%s=true", string(feature))) {
+			ginkgo.Skip(fmt.Sprintf("Skipping suite: %s feature gate is not enabled for VPA admission controller", feature))
 		}
 	}
 }