Inject top level QoS cgroup creation in the Kubelet

Makefile

@@ -164,6 +164,7 @@ test-e2e: ginkgo generated_files
 # Example:
 #   make test-e2e-node FOCUS=kubelet SKIP=container
 #   make test-e2e-node REMOTE=true DELETE_INSTANCES=true
+#   make test-e2e-node TEST_ARGS="--cgroups-per-qos=true"
 # Build and run tests.
 .PHONY: test-e2e-node
 test-e2e-node: ginkgo generated_files

cmd/kubelet/app/options/options.go

@@ -133,6 +133,7 @@ func (s *KubeletServer) AddFlags(fs *pflag.FlagSet) {
 	fs.MarkDeprecated("system-container", "Use --system-cgroups instead. Will be removed in a future version.")
 	fs.StringVar(&s.SystemCgroups, "system-cgroups", s.SystemCgroups, "Optional absolute name of cgroups in which to place all non-kernel processes that are not already inside a cgroup under `/`. Empty for no container. Rolling back the flag requires a reboot. (Default: \"\").")
 
+	fs.BoolVar(&s.CgroupsPerQOS, "cgroups-per-qos", s.CgroupsPerQOS, "Enable creation of QoS cgroup hierarchy, if true top level QoS and pod cgroups are created.")
 	fs.StringVar(&s.CgroupRoot, "cgroup-root", s.CgroupRoot, "Optional root cgroup to use for pods. This is handled by the container runtime on a best effort basis. Default: '', which means use the container runtime default.")
 	fs.StringVar(&s.ContainerRuntime, "container-runtime", s.ContainerRuntime, "The container runtime to use. Possible values: 'docker', 'rkt'. Default: 'docker'.")
 	fs.DurationVar(&s.RuntimeRequestTimeout.Duration, "runtime-request-timeout", s.RuntimeRequestTimeout.Duration, "Timeout of all runtime requests except long running request - pull, logs, exec and attach. When timeout exceeded, kubelet will cancel the request, throw out an error and retry later. Default: 2m0s")

cmd/kubelet/app/server.go

@@ -220,6 +220,7 @@ func UnsecuredKubeletConfig(s *options.KubeletServer) (*KubeletConfig, error) {
 		EnableControllerAttachDetach: s.EnableControllerAttachDetach,
 		EnableCustomMetrics:          s.EnableCustomMetrics,
 		EnableDebuggingHandlers:      s.EnableDebuggingHandlers,
+		CgroupsPerQOS:                s.CgroupsPerQOS,
 		EnableServer:                 s.EnableServer,
 		EventBurst:                   int(s.EventBurst),
 		EventRecordQPS:               float32(s.EventRecordQPS),
@@ -363,12 +364,13 @@ func run(s *options.KubeletServer, kcfg *KubeletConfig) (err error) {
 		if kcfg.SystemCgroups != "" && kcfg.CgroupRoot == "" {
 			return fmt.Errorf("invalid configuration: system container was specified and cgroup root was not specified")
 		}
-
 		kcfg.ContainerManager, err = cm.NewContainerManager(kcfg.Mounter, kcfg.CAdvisorInterface, cm.NodeConfig{
 			RuntimeCgroupsName: kcfg.RuntimeCgroups,
 			SystemCgroupsName:  kcfg.SystemCgroups,
 			KubeletCgroupsName: kcfg.KubeletCgroups,
 			ContainerRuntime:   kcfg.ContainerRuntime,
+			CgroupsPerQOS:      kcfg.CgroupsPerQOS,
+			CgroupRoot:         kcfg.CgroupRoot,
 		})
 		if err != nil {
 			return err
@@ -575,6 +577,7 @@ func SimpleKubelet(client *clientset.Clientset,
 		EnableCustomMetrics:          false,
 		EnableDebuggingHandlers:      true,
 		EnableServer:                 true,
+		CgroupsPerQOS:                false,
 		FileCheckFrequency:           fileCheckFrequency,
 		// Since this kubelet runs with --configure-cbr0=false, it needs to use
 		// hairpin-veth to allow hairpin packets. Note that this deviates from
@@ -798,6 +801,7 @@ type KubeletConfig struct {
 	EnableControllerAttachDetach   bool
 	EnableCustomMetrics            bool
 	EnableDebuggingHandlers        bool
+	CgroupsPerQOS                  bool
 	EnableServer                   bool
 	EventClient                    *clientset.Clientset
 	EventBurst                     int
@@ -929,6 +933,7 @@ func CreateAndInitKubelet(kc *KubeletConfig) (k KubeletBootstrap, pc *config.Pod
 		kc.NodeLabels,
 		kc.NodeStatusUpdateFrequency,
 		kc.OSInterface,
+		kc.CgroupsPerQOS,
 		kc.CgroupRoot,
 		kc.ContainerRuntime,
 		kc.RuntimeRequestTimeout,

docs/devel/e2e-node-tests.md

@@ -205,6 +205,14 @@ less useful for catching flakes related creating the instance from an image.**
 make test-e2e-node REMOTE=true RUN_UNTIL_FAILURE=true
 ```
 
+## Additional QoS Cgroups Hierarchy level testing
+
+For testing with the QoS Cgroup Hierarchy enabled, you can pass --cgroups-per-qos flag as an argument into Ginkgo using TEST_ARGS
+
+```sh
+make test_e2e_node TEST_ARGS="--cgroups-per-qos=true"
+```
+
 # Notes on tests run by the Kubernetes project during pre-, post- submit.
 
 The node e2e tests are run by the PR builder for each Pull Request and the results published at

hack/make-rules/test-e2e-node.sh

@@ -33,6 +33,7 @@ cleanup=${CLEANUP:-"true"}
 delete_instances=${DELETE_INSTANCES:-"false"}
 run_until_failure=${RUN_UNTIL_FAILURE:-"false"}
 list_images=${LIST_IMAGES:-"false"}
+test_args=${TEST_ARGS:-""}
 
 if  [[ $list_images == "true" ]]; then
   gcloud compute images list --project="${image_project}" | grep "e2e-node"
@@ -117,7 +118,7 @@ if [ $remote = true ] ; then
     --hosts="$hosts" --images="$images" --cleanup="$cleanup" \
     --results-dir="$artifacts" --ginkgo-flags="$ginkgoflags" \
     --image-project="$image_project" --instance-name-prefix="$instance_prefix" --setup-node="true" \
-    --delete-instances="$delete_instances"
+    --delete-instances="$delete_instances" --test_args="$test_args"
   exit $?
 
 else
@@ -129,6 +130,7 @@ else
   # Test using the host the script was run on
   # Provided for backwards compatibility
   "${ginkgo}" --focus=$focus --skip=$skip "${KUBE_ROOT}/test/e2e_node/" --report-dir=${report} \
-    -- --alsologtostderr --v 2 --node-name $(hostname) --disable-kubenet=true --build-services=true --start-services=true --stop-services=true
+    -- --alsologtostderr --v 2 --node-name $(hostname) --disable-kubenet=true --build-services=true \
+    --start-services=true --stop-services=true "$test_args"
   exit $?
 fi

hack/verify-flags/known-flags.txt

@@ -42,6 +42,7 @@ build-tag
 cadvisor-port
 cert-dir
 certificate-authority
+cgroups-per-qos
 cgroup-root
 chaos-chance
 clean-start

pkg/apis/componentconfig/types.go

@@ -258,14 +258,18 @@ type KubeletConfiguration struct {
 	CloudConfigFile string `json:"cloudConfigFile,omitempty"`
 	// KubeletCgroups is the absolute name of cgroups to isolate the kubelet in.
 	KubeletCgroups string `json:"kubeletCgroups,omitempty"`
+	// Enable QoS based Cgroup hierarchy: top level cgroups for QoS Classes
+	// And all Burstable and BestEffort pods are brought up under their
+	// specific top level QoS cgroup.
+	CgroupsPerQOS bool `json:"CgroupsPerQOS,omitempty"`
 	// Cgroups that container runtime is expected to be isolated in.
 	RuntimeCgroups string `json:"runtimeCgroups,omitempty"`
 	// SystemCgroups is absolute name of cgroups in which to place
 	// all non-kernel processes that are not already in a container. Empty
 	// for no container. Rolling back the flag requires a reboot.
 	SystemCgroups string `json:"systemCgroups,omitempty"`
-	// cgroupRoot is the root cgroup to use for pods. This is handled by the
+	// CgroupRoot is the root cgroup to use for pods.
-	// container runtime on a best effort basis.
+	// If CgroupsPerQOS is enabled, this is the root of the QoS cgroup hierarchy.
 	CgroupRoot string `json:"cgroupRoot,omitempty"`
 	// containerRuntime is the container runtime to use.
 	ContainerRuntime string `json:"containerRuntime"`

pkg/apis/componentconfig/v1alpha1/defaults.go

@@ -154,6 +154,9 @@ func SetDefaults_KubeletConfiguration(obj *KubeletConfiguration) {
 	if obj.ConfigureCBR0 == nil {
 		obj.ConfigureCBR0 = boolVar(false)
 	}
+	if obj.CgroupsPerQOS == nil {
+		obj.CgroupsPerQOS = boolVar(false)
+	}
 	if obj.ContainerRuntime == "" {
 		obj.ContainerRuntime = "docker"
 	}

pkg/apis/componentconfig/v1alpha1/types.go

@@ -322,6 +322,10 @@ type KubeletConfiguration struct {
 	// cgroupRoot is the root cgroup to use for pods. This is handled by the
 	// container runtime on a best effort basis.
 	CgroupRoot string `json:"cgroupRoot"`
+	// Enable QoS based Cgroup hierarchy: top level cgroups for QoS Classes
+	// And all Burstable and BestEffort pods are brought up under their
+	// specific top level QoS cgroup.
+	CgroupsPerQOS *bool `json:"CgroupsPerQOS,omitempty"`
 	// containerRuntime is the container runtime to use.
 	ContainerRuntime string `json:"containerRuntime"`
 	// runtimeRequestTimeout is the timeout for all runtime requests except long running

pkg/kubelet/cm/container_manager.go

@@ -43,6 +43,8 @@ type NodeConfig struct {
 	SystemCgroupsName  string
 	KubeletCgroupsName string
 	ContainerRuntime   string
+	CgroupsPerQOS      bool
+	CgroupRoot         string
 }
 
 type Status struct {

pkg/kubelet/cm/container_manager_linux.go

@@ -37,6 +37,7 @@ import (
 	"k8s.io/kubernetes/pkg/api"
 	"k8s.io/kubernetes/pkg/api/resource"
 	"k8s.io/kubernetes/pkg/kubelet/cadvisor"
+	"k8s.io/kubernetes/pkg/kubelet/qos"
 	"k8s.io/kubernetes/pkg/util"
 	utilerrors "k8s.io/kubernetes/pkg/util/errors"
 	"k8s.io/kubernetes/pkg/util/mount"
@@ -97,7 +98,10 @@ type containerManagerImpl struct {
 	status Status
 	// External containers being managed.
 	systemContainers []*systemContainer
+	qosContainers    QOSContainersInfo
 	periodicTasks    []func()
+	// holds all the mounted cgroup subsystems
+	subsystems *cgroupSubsystems
 }
 
 type features struct {
@@ -161,10 +165,24 @@ func validateSystemRequirements(mountUtil mount.Interface) (features, error) {
 // Takes the absolute name of the specified containers.
 // Empty container name disables use of the specified container.
 func NewContainerManager(mountUtil mount.Interface, cadvisorInterface cadvisor.Interface, nodeConfig NodeConfig) (ContainerManager, error) {
+	// Check if Cgroup-root actually exists on the node
+	if nodeConfig.CgroupsPerQOS {
+		if nodeConfig.CgroupRoot == "" {
+			return nil, fmt.Errorf("invalid configuration: cgroups-per-qos was specified and cgroup-root was not specified. To enable the QoS cgroup hierarchy you need to specify a valid cgroup-root")
+		}
+		if _, err := os.Stat(nodeConfig.CgroupRoot); err != nil {
+			return nil, fmt.Errorf("invalid configuration: cgroup-root doesn't exist : %v", err)
+		}
+	}
+	subsystems, err := getCgroupSubsystems()
+	if err != nil {
+		return nil, fmt.Errorf("failed to get mounted subsystems: %v", err)
+	}
 	return &containerManagerImpl{
 		cadvisorInterface: cadvisorInterface,
 		mountUtil:         mountUtil,
 		NodeConfig:        nodeConfig,
+		subsystems:        subsystems,
 	}, nil
 }
 
@@ -190,6 +208,41 @@ const (
 	KernelTunableModify KernelTunableBehavior = "modify"
 )
 
+// InitQOS creates the top level qos cgroup containers
+// We create top level QoS containers for only Burstable and Best Effort
+// and not Guaranteed QoS class. All guaranteed pods are nested under the
+// RootContainer by default. InitQOS is called only once during kubelet bootstrapping.
+// TODO(@dubstack) Add support for cgroup-root to work on both systemd and cgroupfs
+// drivers. Currently we only support systems running cgroupfs driver
+func InitQOS(rootContainer string, subsystems *cgroupSubsystems) (QOSContainersInfo, error) {
+	cm := NewCgroupManager(subsystems)
+	// Top level for Qos containers are created only for Burstable
+	// and Best Effort classes
+	qosClasses := [2]qos.QOSClass{qos.Burstable, qos.BestEffort}
+
+	// Create containers for both qos classes
+	for _, qosClass := range qosClasses {
+		// get the container's absolute name
+		absoluteContainerName := path.Join(rootContainer, string(qosClass))
+		// containerConfig object stores the cgroup specifications
+		containerConfig := &CgroupConfig{
+			Name:               absoluteContainerName,
+			ResourceParameters: &ResourceConfig{},
+		}
+		// TODO(@dubstack) Add support on systemd cgroups driver
+		if err := cm.Create(containerConfig); err != nil {
+			return QOSContainersInfo{}, fmt.Errorf("failed to create top level %v QOS cgroup : %v", qosClass, err)
+		}
+	}
+	// Store the top level qos container names
+	qosContainersInfo := QOSContainersInfo{
+		Guaranteed: rootContainer,
+		Burstable:  path.Join(rootContainer, string(qos.Burstable)),
+		BestEffort: path.Join(rootContainer, string(qos.BestEffort)),
+	}
+	return qosContainersInfo, nil
+}
+
 // setupKernelTunables validates kernel tunable flags are set as expected
 // depending upon the specified option, it will either warn, error, or modify the kernel tunable flags
 func setupKernelTunables(option KernelTunableBehavior) error {
@@ -240,6 +293,15 @@ func (cm *containerManagerImpl) setupNode() error {
 		return err
 	}
 
+	// Setup top level qos containers only if CgroupsPerQOS flag is specified as true
+	if cm.NodeConfig.CgroupsPerQOS {
+		qosContainersInfo, err := InitQOS(cm.NodeConfig.CgroupRoot, cm.subsystems)
+		if err != nil {
+			return fmt.Errorf("failed to initialise top level QOS containers: %v", err)
+		}
+		cm.qosContainers = qosContainersInfo
+	}
+
 	systemContainers := []*systemContainer{}
 	if cm.ContainerRuntime == "docker" {
 		if cm.RuntimeCgroupsName != "" {

pkg/kubelet/cm/types.go

@@ -56,3 +56,10 @@ type CgroupManager interface {
 	// Exists checks if the cgroup already exists
 	Exists(string) bool
 }
+
+// QOSContainersInfo hold the names of containers per qos
+type QOSContainersInfo struct {
+	Guaranteed string
+	BestEffort string
+	Burstable  string
+}

pkg/kubelet/kubelet.go

@@ -206,6 +206,7 @@ func NewMainKubelet(
 	nodeLabels map[string]string,
 	nodeStatusUpdateFrequency time.Duration,
 	osInterface kubecontainer.OSInterface,
+	CgroupsPerQOS bool,
 	cgroupRoot string,
 	containerRuntime string,
 	runtimeRequestTimeout time.Duration,
@@ -339,6 +340,7 @@ func NewMainKubelet(
 		nodeStatusUpdateFrequency: nodeStatusUpdateFrequency,
 		os:                         osInterface,
 		oomWatcher:                 oomWatcher,
+		CgroupsPerQOS:              CgroupsPerQOS,
 		cgroupRoot:                 cgroupRoot,
 		mounter:                    mounter,
 		writer:                     writer,
@@ -706,6 +708,9 @@ type Kubelet struct {
 	// Monitor resource usage
 	resourceAnalyzer stats.ResourceAnalyzer
 
+	// Whether or not we should have the QOS cgroup hierarchy for resource management
+	CgroupsPerQOS bool
+
 	// If non-empty, pass this to the container runtime as the root cgroup.
 	cgroupRoot string
 

test/e2e/framework/test_context.go

@@ -67,6 +67,8 @@ type TestContextType struct {
 	DumpLogsOnFailure bool
 	// Name of the node to run tests on (node e2e suite only).
 	NodeName string
+	// Whether to enable the QoS Cgroup Hierarchy or not
+	CgroupsPerQOS bool
 }
 
 type CloudConfig struct {
@@ -148,4 +150,5 @@ func RegisterClusterFlags() {
 // Register flags specific to the node e2e test suite.
 func RegisterNodeFlags() {
 	flag.StringVar(&TestContext.NodeName, "node-name", "", "Name of the node to run tests on (node e2e suite only).")
+	flag.BoolVar(&TestContext.CgroupsPerQOS, "cgroups-per-qos", false, "Enable creation of QoS cgroup hierarchy, if true top level QoS and pod cgroups are created.")
 }

test/e2e_node/cgroup_manager_test.go

@@ -0,0 +1,77 @@
+/*
+Copyright 2016 The Kubernetes Authors.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+
+package e2e_node
+
+import (
+	"k8s.io/kubernetes/pkg/api"
+	"k8s.io/kubernetes/pkg/util"
+	"k8s.io/kubernetes/test/e2e/framework"
+
+	. "github.com/onsi/ginkgo"
+	. "github.com/onsi/gomega"
+)
+
+var _ = framework.KubeDescribe("Kubelet Cgroup Manager", func() {
+	f := NewDefaultFramework("kubelet-cgroup-manager")
+	Describe("QOS containers", func() {
+		Context("On enabling QOS cgroup hierarchy", func() {
+			It("Top level QoS containers should have been created", func() {
+				if framework.TestContext.CgroupsPerQOS {
+					podName := "qos-pod" + string(util.NewUUID())
+					contName := "qos-container" + string(util.NewUUID())
+					pod := &api.Pod{
+						ObjectMeta: api.ObjectMeta{
+							Name:      podName,
+							Namespace: f.Namespace.Name,
+						},
+						Spec: api.PodSpec{
+							// Don't restart the Pod since it is expected to exit
+							RestartPolicy: api.RestartPolicyNever,
+							Containers: []api.Container{
+								{
+									Image:   "gcr.io/google_containers/busybox:1.24",
+									Name:    contName,
+									Command: []string{"sh", "-c", "if [ -d /tmp/memory/Burstable ] && [ -d /tmp/memory/BestEffort ]; then exit 0; else exit 1; fi"},
+									VolumeMounts: []api.VolumeMount{
+										{
+											Name:      "sysfscgroup",
+											MountPath: "/tmp",
+										},
+									},
+								},
+							},
+							Volumes: []api.Volume{
+								{
+									Name: "sysfscgroup",
+									VolumeSource: api.VolumeSource{
+										HostPath: &api.HostPathVolumeSource{Path: "/sys/fs/cgroup"},
+									},
+								},
+							},
+						},
+					}
+					f.MungePodSpec(pod)
+					podClient := f.Client.Pods(f.Namespace.Name)
+					_, err := podClient.Create(pod)
+					Expect(err).NotTo(HaveOccurred())
+					err = framework.WaitForPodSuccessInNamespace(f.Client, podName, contName, f.Namespace.Name)
+					Expect(err).NotTo(HaveOccurred())
+				}
+			})
+		})
+	})
+})

test/e2e_node/e2e_node_suite_test.go

@@ -96,7 +96,7 @@ var _ = BeforeSuite(func() {
 	maskLocksmithdOnCoreos()
 
 	if *startServices {
-		e2es = newE2eService(framework.TestContext.NodeName)
+		e2es = newE2eService(framework.TestContext.NodeName, framework.TestContext.CgroupsPerQOS)
 		if err := e2es.start(); err != nil {
 			Fail(fmt.Sprintf("Unable to start node services.\n%v", err))
 		}

test/e2e_node/e2e_remote.go

@@ -146,7 +146,7 @@ func CreateTestArchive() (string, error) {
 }
 
 // Returns the command output, whether the exit was ok, and any errors
-func RunRemote(archive string, host string, cleanup bool, junitFileNumber int, setupNode bool) (string, bool, error) {
+func RunRemote(archive string, host string, cleanup bool, junitFileNumber int, setupNode bool, testArgs string) (string, bool, error) {
 	if setupNode {
 		uname, err := user.Current()
 		if err != nil {
@@ -211,11 +211,10 @@ func RunRemote(archive string, host string, cleanup bool, junitFileNumber int, s
 		// Exit failure with the error
 		return "", false, err
 	}
-
 	// Run the tests
 	cmd = getSshCommand(" && ",
 		fmt.Sprintf("cd %s", tmp),
-		fmt.Sprintf("timeout -k 30s %ds ./ginkgo %s ./e2e_node.test -- --logtostderr --v 2 --build-services=false --stop-services=%t --node-name=%s --report-dir=%s/results --junit-file-number=%d", *testTimeoutSeconds, *ginkgoFlags, cleanup, host, tmp, junitFileNumber),
+		fmt.Sprintf("timeout -k 30s %ds ./ginkgo %s ./e2e_node.test -- --logtostderr --v 2 --build-services=false --stop-services=%t --node-name=%s --report-dir=%s/results --junit-file-number=%d %s", *testTimeoutSeconds, *ginkgoFlags, cleanup, host, tmp, junitFileNumber, testArgs),
 	)
 	aggErrs := []error{}
 

test/e2e_node/e2e_service.go

@@ -46,6 +46,7 @@ type e2eService struct {
 	kubeletStaticPodDir string
 	nodeName            string
 	logFiles            map[string]logFileData
+	cgroupsPerQOS       bool
 }
 
 type logFileData struct {
@@ -58,14 +59,18 @@ const (
 	LOG_VERBOSITY_LEVEL = "4"
 )
 
-func newE2eService(nodeName string) *e2eService {
+func newE2eService(nodeName string, cgroupsPerQOS bool) *e2eService {
 	// Special log files that need to be collected for additional debugging.
 	var logFiles = map[string]logFileData{
 		"kern.log":   {[]string{"/var/log/kern.log"}, []string{"-k"}},
 		"docker.log": {[]string{"/var/log/docker.log", "/var/log/upstart/docker.log"}, []string{"-u", "docker"}},
 	}
 
-	return &e2eService{nodeName: nodeName, logFiles: logFiles}
+	return &e2eService{
+		nodeName:      nodeName,
+		logFiles:      logFiles,
+		cgroupsPerQOS: cgroupsPerQOS,
+	}
 }
 
 func (es *e2eService) start() error {
@@ -236,6 +241,12 @@ func (es *e2eService) startKubeletServer() (*killCmd, error) {
 		"--v", LOG_VERBOSITY_LEVEL, "--logtostderr",
 		"--pod-cidr=10.180.0.0/24", // Assign a fixed CIDR to the node because there is no node controller.
 	)
+	if es.cgroupsPerQOS {
+		cmdArgs = append(cmdArgs,
+			"--cgroups-per-qos", "true",
+			"--cgroup-root", "/",
+		)
+	}
 	if !*disableKubenet {
 		cwd, err := os.Getwd()
 		if err != nil {
@@ -245,6 +256,7 @@ func (es *e2eService) startKubeletServer() (*killCmd, error) {
 			"--network-plugin=kubenet",
 			"--network-plugin-dir", filepath.Join(cwd, CNIDirectory, "bin")) // Enable kubenet
 	}
+
 	cmd := exec.Command("sudo", cmdArgs...)
 	hcc := newHealthCheckCommand(
 		"http://127.0.0.1:10255/healthz",

test/e2e_node/jenkins/e2e-node-jenkins.sh

@@ -39,4 +39,4 @@ go run test/e2e_node/runner/run_e2e.go  --logtostderr --vmodule=*=2 --ssh-env="g
   --zone="$GCE_ZONE" --project="$GCE_PROJECT" --hosts="$GCE_HOSTS" \
   --image-config-file="$GCE_IMAGE_CONFIG_PATH" --cleanup="$CLEANUP" \
   --results-dir="$ARTIFACTS" --ginkgo-flags="$GINKGO_FLAGS" \
-  --setup-node="$SETUP_NODE" --instance-metadata="$GCE_INSTANCE_METADATA"
+  --setup-node="$SETUP_NODE" --test_args="$TEST_ARGS" --instance-metadata="$GCE_INSTANCE_METADATA"

test/e2e_node/jenkins/jenkins-ci.properties

@@ -5,4 +5,4 @@ GCE_PROJECT=kubernetes-jenkins
 CLEANUP=true
 GINKGO_FLAGS=--skip=FLAKY
 SETUP_NODE=false
-
+TEST_ARGS=--cgroups-per-qos=true

test/e2e_node/jenkins/jenkins-pull.properties

@@ -5,3 +5,4 @@ GCE_PROJECT=kubernetes-jenkins-pull
 CLEANUP=true
 GINKGO_FLAGS=--skip=FLAKY
 SETUP_NODE=false
+TEST_ARGS=--cgroups-per-qos=true

test/e2e_node/jenkins/template.properties

@@ -17,3 +17,5 @@ GCE_IMAGE_PROJECT=
 CLEANUP=true
 # If true, current user will be added to the docker group on test node
 SETUP_NODE=false
+# If true QoS Cgroup Hierarchy is created and tests specifc to the cgroup hierarchy run
+TEST_ARGS=--cgroups-per-qos=true

test/e2e_node/runner/run_e2e.go

@@ -41,6 +41,7 @@ import (
 	"google.golang.org/api/compute/v1"
 )
 
+var testArgs = flag.String("test_args", "", "Space-separated list of arguments to pass to Ginkgo test runner.")
 var instanceNamePrefix = flag.String("instance-name-prefix", "", "prefix for instance names")
 var zone = flag.String("zone", "", "gce zone the hosts live in")
 var project = flag.String("project", "", "gce project the hosts live in")
@@ -254,7 +255,7 @@ func testHost(host string, deleteFiles bool, junitFileNum int, setupNode bool) *
 		}
 	}
 
-	output, exitOk, err := e2e_node.RunRemote(path, host, deleteFiles, junitFileNum, setupNode)
+	output, exitOk, err := e2e_node.RunRemote(path, host, deleteFiles, junitFileNum, setupNode, *testArgs)
 	return &TestResult{
 		output: output,
 		err:    err,