Merge pull request kubernetes#125328 from harche/v1_warning

[KEP-4569] Add a warning log, an event for cgroup v1 usage and a metric for cgroup version

Author: k8s-ci-robot

pkg/kubelet/cm/cgroup_manager_unsupported.go

@@ -41,6 +41,10 @@ func NewCgroupManager(_ interface{}) CgroupManager {
 	return &unsupportedCgroupManager{}
 }
 
+func (m *unsupportedCgroupManager) Version() int {
+	return 0
+}
+
 func (m *unsupportedCgroupManager) Name(_ CgroupName) string {
 	return ""
 }

pkg/kubelet/cm/cgroup_v1_manager_linux.go

@@ -48,6 +48,11 @@ func NewCgroupV1Manager(cs *CgroupSubsystems, cgroupDriver string) CgroupManager
 	}
 }
 
+// Version of the cgroup implementation on the host
+func (c *cgroupV1impl) Version() int {
+	return 1
+}
+
 // Validate checks if all subsystem cgroups are valid
 func (c *cgroupV1impl) Validate(name CgroupName) error {
 	// Get map of all cgroup paths on the system for the particular cgroup

pkg/kubelet/cm/cgroup_v2_manager_linux.go

@@ -47,6 +47,11 @@ func NewCgroupV2Manager(cs *CgroupSubsystems, cgroupDriver string) CgroupManager
 	}
 }
 
+// Version of the cgroup implementation on the host
+func (c *cgroupV2impl) Version() int {
+	return 2
+}
+
 // Validate checks if all subsystem cgroups are valid
 func (c *cgroupV2impl) Validate(name CgroupName) error {
 	cgroupPath := c.buildCgroupUnifiedPath(name)

pkg/kubelet/cm/container_manager.go

@@ -42,6 +42,11 @@ import (
 	"k8s.io/utils/cpuset"
 )
 
+const (
+	// Warning message for the users still using cgroup v1
+	CgroupV1MaintenanceModeWarning = "Cgroup v1 support is in maintenance mode, please migrate to Cgroup v2."
+)
+
 type ActivePodsFunc func() []*v1.Pod
 
 // Manages the containers running on a machine.
@@ -159,6 +164,7 @@ type NodeConfig struct {
 	CPUCFSQuotaPeriod                       time.Duration
 	TopologyManagerPolicy                   string
 	TopologyManagerPolicyOptions            map[string]string
+	CgroupVersion                           int
 }
 
 type NodeAllocatableConfig struct {

pkg/kubelet/cm/container_manager_linux.go

@@ -245,6 +245,7 @@ func NewContainerManager(mountUtil mount.Interface, cadvisorInterface cadvisor.I
 	// Turn CgroupRoot from a string (in cgroupfs path format) to internal CgroupName
 	cgroupRoot := ParseCgroupfsToCgroupName(nodeConfig.CgroupRoot)
 	cgroupManager := NewCgroupManager(subsystems, nodeConfig.CgroupDriver)
+	nodeConfig.CgroupVersion = cgroupManager.Version()
 	// Check if Cgroup-root actually exists on the node
 	if nodeConfig.CgroupsPerQOS {
 		// this does default to / when enabled, but this tests against regressions.

pkg/kubelet/cm/types.go

@@ -88,6 +88,8 @@ type CgroupManager interface {
 	GetCgroupConfig(name CgroupName, resource v1.ResourceName) (*ResourceConfig, error)
 	// Set resource config for the specified resource type on the cgroup
 	SetCgroupConfig(name CgroupName, resource v1.ResourceName, resourceConfig *ResourceConfig) error
+	// Version of the cgroup implementation on the host
+	Version() int
 }
 
 // QOSContainersInfo stores the names of containers per qos

pkg/kubelet/events/event.go

@@ -76,6 +76,7 @@ const (
 	FailedMountOnFilesystemMismatch      = "FailedMountOnFilesystemMismatch"
 	FailedPrepareDynamicResources        = "FailedPrepareDynamicResources"
 	PossibleMemoryBackedVolumesOnDisk    = "PossibleMemoryBackedVolumesOnDisk"
+	CgroupV1                             = "CgroupV1"
 )
 
 // Image manager event reason list

pkg/kubelet/kubelet.go

@@ -1636,6 +1636,8 @@ func (kl *Kubelet) Run(updates <-chan kubetypes.PodUpdate) {
 		os.Exit(1)
 	}
 
+	kl.warnCgroupV1Usage()
+
 	// Start volume manager
 	go kl.volumeManager.Run(kl.sourcesReady, wait.NeverStop)
 
@@ -3021,3 +3023,12 @@ func (kl *Kubelet) PrepareDynamicResources(pod *v1.Pod) error {
 func (kl *Kubelet) UnprepareDynamicResources(pod *v1.Pod) error {
 	return kl.containerManager.UnprepareDynamicResources(pod)
 }
+
+func (kl *Kubelet) warnCgroupV1Usage() {
+	cgroupVersion := kl.containerManager.GetNodeConfig().CgroupVersion
+	if cgroupVersion == 1 {
+		kl.recorder.Eventf(kl.nodeRef, v1.EventTypeWarning, events.CgroupV1, cm.CgroupV1MaintenanceModeWarning)
+		klog.V(2).InfoS("Warning: cgroup v1", "message", cm.CgroupV1MaintenanceModeWarning)
+	}
+	metrics.CgroupVersion.Set(float64(cgroupVersion))
+}

pkg/kubelet/metrics/metrics.go

@@ -72,6 +72,7 @@ const (
 	OrphanedRuntimePodTotalKey         = "orphaned_runtime_pods_total"
 	RestartedPodTotalKey               = "restarted_pods_total"
 	ImagePullDurationKey               = "image_pull_duration_seconds"
+	CgroupVersionKey                   = "cgroup_version"
 
 	// Metrics keys of remote runtime operations
 	RuntimeOperationsKey         = "runtime_operations_total"
@@ -907,6 +908,15 @@ var (
 			StabilityLevel: metrics.ALPHA,
 		},
 	)
+
+	CgroupVersion = metrics.NewGauge(
+		&metrics.GaugeOpts{
+			Subsystem:      KubeletSubsystem,
+			Name:           CgroupVersionKey,
+			Help:           "cgroup version on the hosts.",
+			StabilityLevel: metrics.ALPHA,
+		},
+	)
 )
 
 var registerMetrics sync.Once
@@ -996,6 +1006,7 @@ func Register(collectors ...metrics.StableCollector) {
 
 		legacyregistry.MustRegister(LifecycleHandlerHTTPFallbacks)
 		legacyregistry.MustRegister(LifecycleHandlerSleepTerminated)
+		legacyregistry.MustRegister(CgroupVersion)
 	})
 }