Merge pull request kubernetes#128556 from AnishShah/kubelet-reject-metric

Introduce a metric to track kubelet admission failure.

Author: k8s-ci-robot

pkg/kubelet/kubelet.go

@@ -29,6 +29,7 @@ import (
 	sysruntime "runtime"
 	"slices"
 	"sort"
+	"strings"
 	"sync"
 	"sync/atomic"
 	"time"
@@ -81,6 +82,7 @@ import (
 	"k8s.io/kubernetes/pkg/kubelet/cloudresource"
 	"k8s.io/kubernetes/pkg/kubelet/clustertrustbundle"
 	"k8s.io/kubernetes/pkg/kubelet/cm"
+	"k8s.io/kubernetes/pkg/kubelet/cm/topologymanager"
 	"k8s.io/kubernetes/pkg/kubelet/config"
 	"k8s.io/kubernetes/pkg/kubelet/configmap"
 	kubecontainer "k8s.io/kubernetes/pkg/kubelet/container"
@@ -120,6 +122,7 @@ import (
 	"k8s.io/kubernetes/pkg/kubelet/volumemanager"
 	"k8s.io/kubernetes/pkg/kubelet/watchdog"
 	httpprobe "k8s.io/kubernetes/pkg/probe/http"
+	"k8s.io/kubernetes/pkg/scheduler/framework/plugins/tainttoleration"
 	"k8s.io/kubernetes/pkg/security/apparmor"
 	"k8s.io/kubernetes/pkg/util/oom"
 	"k8s.io/kubernetes/pkg/volume"
@@ -220,6 +223,26 @@ var (
 	// ContainerLogsDir can be overwritten for testing usage
 	ContainerLogsDir = DefaultContainerLogsDir
 	etcHostsPath     = getContainerEtcHostsPath()
+
+	admissionRejectionReasons = sets.New[string](
+		lifecycle.AppArmorNotAdmittedReason,
+		lifecycle.PodOSSelectorNodeLabelDoesNotMatch,
+		lifecycle.PodOSNotSupported,
+		lifecycle.InvalidNodeInfo,
+		lifecycle.InitContainerRestartPolicyForbidden,
+		lifecycle.UnexpectedAdmissionError,
+		lifecycle.UnknownReason,
+		lifecycle.UnexpectedPredicateFailureType,
+		lifecycle.OutOfCPU,
+		lifecycle.OutOfMemory,
+		lifecycle.OutOfEphemeralStorage,
+		lifecycle.OutOfPods,
+		tainttoleration.ErrReasonNotMatch,
+		eviction.Reason,
+		sysctl.ForbiddenReason,
+		topologymanager.ErrorTopologyAffinity,
+		nodeshutdown.NodeShutdownNotAdmittedReason,
+	)
 )
 
 func getContainerEtcHostsPath() string {
@@ -2310,7 +2333,6 @@ func (kl *Kubelet) canAdmitPod(allocatedPods []*v1.Pod, pod *v1.Pod) (bool, stri
 	attrs := &lifecycle.PodAdmitAttributes{Pod: pod, OtherPods: allocatedPods}
 	for _, podAdmitHandler := range kl.admitHandlers {
 		if result := podAdmitHandler.Admit(attrs); !result.Admit {
-
 			klog.InfoS("Pod admission denied", "podUID", attrs.Pod.UID, "pod", klog.KObj(attrs.Pod), "reason", result.Reason, "message", result.Message)
 
 			return false, result.Reason, result.Message
@@ -2320,6 +2342,22 @@ func (kl *Kubelet) canAdmitPod(allocatedPods []*v1.Pod, pod *v1.Pod) (bool, stri
 	return true, "", ""
 }
 
+func recordAdmissionRejection(reason string) {
+	// It is possible that the "reason" label can have high cardinality.
+	// To avoid this metric from exploding, we create an allowlist of known
+	// reasons, and only record reasons from this list. Use "Other" reason
+	// for the rest.
+	if admissionRejectionReasons.Has(reason) {
+		metrics.AdmissionRejectionsTotal.WithLabelValues(reason).Inc()
+	} else if strings.HasPrefix(reason, lifecycle.InsufficientResourcePrefix) {
+		// non-extended resources (like cpu, memory, ephemeral-storage, pods)
+		// are already included in admissionRejectionReasons.
+		metrics.AdmissionRejectionsTotal.WithLabelValues("OutOfExtendedResources").Inc()
+	} else {
+		metrics.AdmissionRejectionsTotal.WithLabelValues("Other").Inc()
+	}
+}
+
 // syncLoop is the main loop for processing changes. It watches for changes from
 // three channels (file, apiserver, and http) and creates a union of them. For
 // any new change seen, will run a sync against desired state and running state. If
@@ -2590,6 +2628,11 @@ func (kl *Kubelet) HandlePodAdditions(pods []*v1.Pod) {
 				// Check if we can admit the pod; if not, reject it.
 				if ok, reason, message := kl.canAdmitPod(allocatedPods, allocatedPod); !ok {
 					kl.rejectPod(pod, reason, message)
+					// We avoid recording the metric in canAdmitPod because it's called
+					// repeatedly during a resize, which would inflate the metric.
+					// Instead, we record the metric here in HandlePodAdditions for new pods
+					// and capture resize events separately.
+					recordAdmissionRejection(reason)
 					continue
 				}
 				// For new pod, checkpoint the resource values at which the Pod has been admitted
@@ -2601,6 +2644,11 @@ func (kl *Kubelet) HandlePodAdditions(pods []*v1.Pod) {
 				// Check if we can admit the pod; if not, reject it.
 				if ok, reason, message := kl.canAdmitPod(allocatedPods, pod); !ok {
 					kl.rejectPod(pod, reason, message)
+					// We avoid recording the metric in canAdmitPod because it's called
+					// repeatedly during a resize, which would inflate the metric.
+					// Instead, we record the metric here in HandlePodAdditions for new pods
+					// and capture resize events separately.
+					recordAdmissionRejection(reason)
 					continue
 				}
 			}

pkg/kubelet/kubelet_test.go

@@ -55,6 +55,7 @@ import (
 	"k8s.io/client-go/tools/record"
 	"k8s.io/client-go/util/flowcontrol"
 	featuregatetesting "k8s.io/component-base/featuregate/testing"
+	"k8s.io/component-base/metrics/testutil"
 	internalapi "k8s.io/cri-api/pkg/apis"
 	runtimeapi "k8s.io/cri-api/pkg/apis/runtime/v1"
 	remote "k8s.io/cri-client/pkg"
@@ -65,6 +66,7 @@ import (
 	cadvisortest "k8s.io/kubernetes/pkg/kubelet/cadvisor/testing"
 	"k8s.io/kubernetes/pkg/kubelet/clustertrustbundle"
 	"k8s.io/kubernetes/pkg/kubelet/cm"
+	"k8s.io/kubernetes/pkg/kubelet/cm/topologymanager"
 	"k8s.io/kubernetes/pkg/kubelet/config"
 	"k8s.io/kubernetes/pkg/kubelet/configmap"
 	kubecontainer "k8s.io/kubernetes/pkg/kubelet/container"
@@ -74,6 +76,7 @@ import (
 	"k8s.io/kubernetes/pkg/kubelet/kuberuntime"
 	"k8s.io/kubernetes/pkg/kubelet/lifecycle"
 	"k8s.io/kubernetes/pkg/kubelet/logs"
+	"k8s.io/kubernetes/pkg/kubelet/metrics"
 	"k8s.io/kubernetes/pkg/kubelet/network/dns"
 	"k8s.io/kubernetes/pkg/kubelet/nodeshutdown"
 	"k8s.io/kubernetes/pkg/kubelet/pleg"
@@ -89,12 +92,14 @@ import (
 	"k8s.io/kubernetes/pkg/kubelet/status"
 	"k8s.io/kubernetes/pkg/kubelet/status/state"
 	statustest "k8s.io/kubernetes/pkg/kubelet/status/testing"
+	"k8s.io/kubernetes/pkg/kubelet/sysctl"
 	"k8s.io/kubernetes/pkg/kubelet/token"
 	kubetypes "k8s.io/kubernetes/pkg/kubelet/types"
 	kubeletutil "k8s.io/kubernetes/pkg/kubelet/util"
 	"k8s.io/kubernetes/pkg/kubelet/util/queue"
 	kubeletvolume "k8s.io/kubernetes/pkg/kubelet/volumemanager"
 	schedulerframework "k8s.io/kubernetes/pkg/scheduler/framework"
+	"k8s.io/kubernetes/pkg/scheduler/framework/plugins/tainttoleration"
 	"k8s.io/kubernetes/pkg/util/oom"
 	"k8s.io/kubernetes/pkg/volume"
 	_ "k8s.io/kubernetes/pkg/volume/hostpath"
@@ -3461,3 +3466,200 @@ func TestIsPodResizeInProgress(t *testing.T) {
 		})
 	}
 }
+
+func TestRecordAdmissionRejection(t *testing.T) {
+	metrics.Register()
+
+	testCases := []struct {
+		name   string
+		reason string
+		wants  string
+	}{
+		{
+			name:   "AppArmor",
+			reason: lifecycle.AppArmorNotAdmittedReason,
+			wants: `
+				# HELP kubelet_admission_rejections_total [ALPHA] Cumulative number pod admission rejections by the Kubelet.
+				# TYPE kubelet_admission_rejections_total counter
+				kubelet_admission_rejections_total{reason="AppArmor"} 1
+			`,
+		},
+		{
+			name:   "PodOSSelectorNodeLabelDoesNotMatch",
+			reason: lifecycle.PodOSSelectorNodeLabelDoesNotMatch,
+			wants: `
+                # HELP kubelet_admission_rejections_total [ALPHA] Cumulative number pod admission rejections by the Kubelet.
+                # TYPE kubelet_admission_rejections_total counter
+                kubelet_admission_rejections_total{reason="PodOSSelectorNodeLabelDoesNotMatch"} 1
+            `,
+		},
+		{
+			name:   "PodOSNotSupported",
+			reason: lifecycle.PodOSNotSupported,
+			wants: `
+                # HELP kubelet_admission_rejections_total [ALPHA] Cumulative number pod admission rejections by the Kubelet.
+                # TYPE kubelet_admission_rejections_total counter
+                kubelet_admission_rejections_total{reason="PodOSNotSupported"} 1
+            `,
+		},
+		{
+			name:   "InvalidNodeInfo",
+			reason: lifecycle.InvalidNodeInfo,
+			wants: `
+                # HELP kubelet_admission_rejections_total [ALPHA] Cumulative number pod admission rejections by the Kubelet.
+                # TYPE kubelet_admission_rejections_total counter
+                kubelet_admission_rejections_total{reason="InvalidNodeInfo"} 1
+            `,
+		},
+		{
+			name:   "InitContainerRestartPolicyForbidden",
+			reason: lifecycle.InitContainerRestartPolicyForbidden,
+			wants: `
+                # HELP kubelet_admission_rejections_total [ALPHA] Cumulative number pod admission rejections by the Kubelet.
+                # TYPE kubelet_admission_rejections_total counter
+                kubelet_admission_rejections_total{reason="InitContainerRestartPolicyForbidden"} 1
+            `,
+		},
+		{
+			name:   "UnexpectedAdmissionError",
+			reason: lifecycle.UnexpectedAdmissionError,
+			wants: `
+                # HELP kubelet_admission_rejections_total [ALPHA] Cumulative number pod admission rejections by the Kubelet.
+                # TYPE kubelet_admission_rejections_total counter
+                kubelet_admission_rejections_total{reason="UnexpectedAdmissionError"} 1
+            `,
+		},
+		{
+			name:   "UnknownReason",
+			reason: lifecycle.UnknownReason,
+			wants: `
+                # HELP kubelet_admission_rejections_total [ALPHA] Cumulative number pod admission rejections by the Kubelet.
+                # TYPE kubelet_admission_rejections_total counter
+                kubelet_admission_rejections_total{reason="UnknownReason"} 1
+            `,
+		},
+		{
+			name:   "UnexpectedPredicateFailureType",
+			reason: lifecycle.UnexpectedPredicateFailureType,
+			wants: `
+                # HELP kubelet_admission_rejections_total [ALPHA] Cumulative number pod admission rejections by the Kubelet.
+                # TYPE kubelet_admission_rejections_total counter
+                kubelet_admission_rejections_total{reason="UnexpectedPredicateFailureType"} 1
+            `,
+		},
+		{
+			name:   "node(s) had taints that the pod didn't tolerate",
+			reason: tainttoleration.ErrReasonNotMatch,
+			wants: `
+                # HELP kubelet_admission_rejections_total [ALPHA] Cumulative number pod admission rejections by the Kubelet.
+                # TYPE kubelet_admission_rejections_total counter
+                kubelet_admission_rejections_total{reason="node(s) had taints that the pod didn't tolerate"} 1
+            `,
+		},
+		{
+			name:   "Evicted",
+			reason: eviction.Reason,
+			wants: `
+                # HELP kubelet_admission_rejections_total [ALPHA] Cumulative number pod admission rejections by the Kubelet.
+                # TYPE kubelet_admission_rejections_total counter
+                kubelet_admission_rejections_total{reason="Evicted"} 1
+            `,
+		},
+		{
+			name:   "SysctlForbidden",
+			reason: sysctl.ForbiddenReason,
+			wants: `
+                # HELP kubelet_admission_rejections_total [ALPHA] Cumulative number pod admission rejections by the Kubelet.
+                # TYPE kubelet_admission_rejections_total counter
+                kubelet_admission_rejections_total{reason="SysctlForbidden"} 1
+            `,
+		},
+		{
+			name:   "TopologyAffinityError",
+			reason: topologymanager.ErrorTopologyAffinity,
+			wants: `
+                # HELP kubelet_admission_rejections_total [ALPHA] Cumulative number pod admission rejections by the Kubelet.
+                # TYPE kubelet_admission_rejections_total counter
+                kubelet_admission_rejections_total{reason="TopologyAffinityError"} 1
+            `,
+		},
+		{
+			name:   "NodeShutdown",
+			reason: nodeshutdown.NodeShutdownNotAdmittedReason,
+			wants: `
+                # HELP kubelet_admission_rejections_total [ALPHA] Cumulative number pod admission rejections by the Kubelet.
+                # TYPE kubelet_admission_rejections_total counter
+                kubelet_admission_rejections_total{reason="NodeShutdown"} 1
+            `,
+		},
+		{
+			name:   "OutOfcpu",
+			reason: "OutOfcpu",
+			wants: `
+                # HELP kubelet_admission_rejections_total [ALPHA] Cumulative number pod admission rejections by the Kubelet.
+                # TYPE kubelet_admission_rejections_total counter
+                kubelet_admission_rejections_total{reason="OutOfcpu"} 1
+            `,
+		},
+		{
+			name:   "OutOfmemory",
+			reason: "OutOfmemory",
+			wants: `
+                # HELP kubelet_admission_rejections_total [ALPHA] Cumulative number pod admission rejections by the Kubelet.
+                # TYPE kubelet_admission_rejections_total counter
+                kubelet_admission_rejections_total{reason="OutOfmemory"} 1
+            `,
+		},
+		{
+			name:   "OutOfephemeral-storage",
+			reason: "OutOfephemeral-storage",
+			wants: `
+                # HELP kubelet_admission_rejections_total [ALPHA] Cumulative number pod admission rejections by the Kubelet.
+                # TYPE kubelet_admission_rejections_total counter
+                kubelet_admission_rejections_total{reason="OutOfephemeral-storage"} 1
+            `,
+		},
+		{
+			name:   "OutOfpods",
+			reason: "OutOfpods",
+			wants: `
+                # HELP kubelet_admission_rejections_total [ALPHA] Cumulative number pod admission rejections by the Kubelet.
+                # TYPE kubelet_admission_rejections_total counter
+                kubelet_admission_rejections_total{reason="OutOfpods"} 1
+            `,
+		},
+		{
+			name:   "OutOfgpu",
+			reason: "OutOfgpu",
+			wants: `
+                # HELP kubelet_admission_rejections_total [ALPHA] Cumulative number pod admission rejections by the Kubelet.
+                # TYPE kubelet_admission_rejections_total counter
+                kubelet_admission_rejections_total{reason="OutOfExtendedResources"} 1
+            `,
+		},
+		{
+			name:   "OtherReason",
+			reason: "OtherReason",
+			wants: `
+                # HELP kubelet_admission_rejections_total [ALPHA] Cumulative number pod admission rejections by the Kubelet.
+                # TYPE kubelet_admission_rejections_total counter
+                kubelet_admission_rejections_total{reason="Other"} 1
+            `,
+		},
+	}
+
+	// Run tests.
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			// Clear the metrics after the test.
+			metrics.AdmissionRejectionsTotal.Reset()
+
+			// Call the function.
+			recordAdmissionRejection(tc.reason)
+
+			if err := testutil.GatherAndCompare(metrics.GetGather(), strings.NewReader(tc.wants), "kubelet_admission_rejections_total"); err != nil {
+				t.Error(err)
+			}
+		})
+	}
+}

pkg/kubelet/lifecycle/handlers.go

@@ -44,6 +44,8 @@ import (
 
 const (
 	maxRespBodyLength = 10 * 1 << 10 // 10KB
+
+	AppArmorNotAdmittedReason = "AppArmor"
 )
 
 type handlerRunner struct {
@@ -224,7 +226,7 @@ func (a *appArmorAdmitHandler) Admit(attrs *PodAdmitAttributes) PodAdmitResult {
 	}
 	return PodAdmitResult{
 		Admit:   false,
-		Reason:  "AppArmor",
+		Reason:  AppArmorNotAdmittedReason,
 		Message: fmt.Sprintf("Cannot enforce AppArmor: %v", err),
 	}
 }