Merge pull request #87650 from nolancon/beta-feature-gate

Update TopologyManager Feature Gate

pkg/features/kube_features.go

@@ -140,6 +140,7 @@ const (
 
 	// owner: @lmdaly
 	// alpha: v1.16
+	// beta: v1.18
 	//
 	// Enable resource managers to make NUMA aligned decisions
 	TopologyManager featuregate.Feature = "TopologyManager"
@@ -598,7 +599,7 @@ var defaultKubernetesFeatureGates = map[featuregate.Feature]featuregate.FeatureS
 	AttachVolumeLimit:              {Default: true, PreRelease: featuregate.GA, LockToDefault: true}, // remove in 1.19
 	CPUManager:                     {Default: true, PreRelease: featuregate.Beta},
 	CPUCFSQuotaPeriod:              {Default: false, PreRelease: featuregate.Alpha},
-	TopologyManager:                {Default: false, PreRelease: featuregate.Alpha},
+	TopologyManager:                {Default: true, PreRelease: featuregate.Beta},
 	ServiceNodeExclusion:           {Default: false, PreRelease: featuregate.Alpha},
 	NodeDisruptionExclusion:        {Default: false, PreRelease: featuregate.Alpha},
 	CSIDriverRegistry:              {Default: true, PreRelease: featuregate.Beta},

pkg/kubelet/apis/config/validation/validation_test.go

@@ -53,7 +53,6 @@ func TestValidateKubeletConfiguration(t *testing.T) {
 		HairpinMode:                 kubeletconfig.PromiscuousBridge,
 		NodeLeaseDurationSeconds:    1,
 		CPUCFSQuotaPeriod:           metav1.Duration{Duration: 100 * time.Millisecond},
-		TopologyManagerPolicy:       "none",
 	}
 	if allErrors := ValidateKubeletConfiguration(successCase1); allErrors != nil {
 		t.Errorf("expect no errors, got %v", allErrors)
@@ -86,7 +85,6 @@ func TestValidateKubeletConfiguration(t *testing.T) {
 		HairpinMode:                 kubeletconfig.PromiscuousBridge,
 		NodeLeaseDurationSeconds:    1,
 		CPUCFSQuotaPeriod:           metav1.Duration{Duration: 100 * time.Millisecond},
-		TopologyManagerPolicy:       "none",
 		ReservedSystemCPUs:          "0-3",
 	}
 	if allErrors := ValidateKubeletConfiguration(successCase2); allErrors != nil {
@@ -118,9 +116,8 @@ func TestValidateKubeletConfiguration(t *testing.T) {
 		HairpinMode:                 "foo",
 		NodeLeaseDurationSeconds:    -1,
 		CPUCFSQuotaPeriod:           metav1.Duration{Duration: 0},
-		TopologyManagerPolicy:       "",
 	}
-	const numErrsErrorCase1 = 26
+	const numErrsErrorCase1 = 25
 	if allErrors := ValidateKubeletConfiguration(errorCase1); len(allErrors.(utilerrors.Aggregate).Errors()) != numErrsErrorCase1 {
 		t.Errorf("expect %d errors, got %v", numErrsErrorCase1, len(allErrors.(utilerrors.Aggregate).Errors()))
 	}
@@ -152,7 +149,6 @@ func TestValidateKubeletConfiguration(t *testing.T) {
 		HairpinMode:                 kubeletconfig.PromiscuousBridge,
 		NodeLeaseDurationSeconds:    1,
 		CPUCFSQuotaPeriod:           metav1.Duration{Duration: 100 * time.Millisecond},
-		TopologyManagerPolicy:       "none",
 		ReservedSystemCPUs:          "0-3",
 	}
 	const numErrsErrorCase2 = 1

pkg/kubelet/cm/topologymanager/topology_manager.go

@@ -18,6 +18,7 @@ package topologymanager
 
 import (
 	"fmt"
+	"sync"
 
 	"k8s.io/api/core/v1"
 	"k8s.io/klog"
@@ -54,6 +55,7 @@ type Manager interface {
 }
 
 type manager struct {
+	mutex sync.Mutex
 	//The list of components registered with the Manager
 	hintProviders []HintProvider
 	//Mapping of a Pods mapping of Containers and their TopologyHints
@@ -203,13 +205,18 @@ func (m *manager) AddHintProvider(h HintProvider) {
 }
 
 func (m *manager) AddContainer(pod *v1.Pod, containerID string) error {
+	m.mutex.Lock()
+	defer m.mutex.Unlock()
+
 	m.podMap[containerID] = string(pod.UID)
 	return nil
 }
 
 func (m *manager) RemoveContainer(containerID string) error {
-	klog.Infof("[topologymanager] RemoveContainer - Container ID: %v", containerID)
+	m.mutex.Lock()
+	defer m.mutex.Unlock()
 
+	klog.Infof("[topologymanager] RemoveContainer - Container ID: %v", containerID)
 	podUIDString := m.podMap[containerID]
 	delete(m.podMap, containerID)
 	if _, exists := m.podTopologyHints[podUIDString]; exists {