Surface cadvisor PSI metrics to summary API and prometheus metrics

Author: roycaihw

Diff for: pkg/kubelet/cadvisor/cadvisor_linux.go

@@ -39,7 +39,9 @@ import (
 	cadvisorapiv2 "github.com/google/cadvisor/info/v2"
 	"github.com/google/cadvisor/manager"
 	"github.com/google/cadvisor/utils/sysfs"
+	utilfeature "k8s.io/apiserver/pkg/util/feature"
 	"k8s.io/klog/v2"
+	"k8s.io/kubernetes/pkg/features"
 	"k8s.io/utils/ptr"
 )
 
@@ -93,6 +95,10 @@ func New(imageFsInfoProvider ImageFsInfoProvider, rootPath string, cgroupRoots [
 		cadvisormetrics.OOMMetrics:          struct{}{},
 	}
 
+	if utilfeature.DefaultFeatureGate.Enabled(features.KubeletPSI) {
+		includedMetrics[cadvisormetrics.PressureMetrics] = struct{}{}
+	}
+
 	if usingLegacyStats || localStorageCapacityIsolation {
 		includedMetrics[cadvisormetrics.DiskUsageMetrics] = struct{}{}
 	}

Diff for: pkg/kubelet/server/server.go

@@ -455,6 +455,11 @@ func (s *Server) InstallAuthNotRequiredHandlers() {
 		cadvisormetrics.ProcessMetrics:      struct{}{},
 		cadvisormetrics.OOMMetrics:          struct{}{},
 	}
+
+	if utilfeature.DefaultFeatureGate.Enabled(features.KubeletPSI) {
+		includedMetrics[cadvisormetrics.PressureMetrics] = struct{}{}
+	}
+
 	// cAdvisor metrics are exposed under the secured handler as well
 	r := compbasemetrics.NewKubeRegistry()
 	r.RawMustRegister(metrics.NewPrometheusMachineCollector(prometheusHostAdapter{s.host}, includedMetrics))

Diff for: pkg/kubelet/server/stats/summary.go

@@ -24,7 +24,9 @@ import (
 	"k8s.io/klog/v2"
 
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	utilfeature "k8s.io/apiserver/pkg/util/feature"
 	statsapi "k8s.io/kubelet/pkg/apis/stats/v1alpha1"
+	"k8s.io/kubernetes/pkg/features"
 	"k8s.io/kubernetes/pkg/kubelet/util"
 )
 
@@ -113,6 +115,9 @@ func (sp *summaryProviderImpl) Get(ctx context.Context, updateStats bool) (*stat
 		Rlimit:           rlimit,
 		SystemContainers: sp.GetSystemContainersStats(nodeConfig, podStats, updateStats),
 	}
+	if utilfeature.DefaultFeatureGate.Enabled(features.KubeletPSI) {
+		nodeStats.IO = rootStats.IO
+	}
 	summary := statsapi.Summary{
 		Node: nodeStats,
 		Pods: podStats,

Diff for: pkg/kubelet/stats/cadvisor_stats_provider.go

@@ -156,6 +156,9 @@ func (p *cadvisorStatsProvider) ListPodStats(ctx context.Context) ([]statsapi.Po
 			podStats.CPU = cpu
 			podStats.Memory = memory
 			podStats.Swap = cadvisorInfoToSwapStats(podInfo)
+			if utilfeature.DefaultFeatureGate.Enabled(features.KubeletPSI) {
+				podStats.IO = cadvisorInfoToIOStats(podInfo)
+			}
 			// ProcessStats were accumulated as the containers were iterated.
 		}
 

Diff for: pkg/kubelet/stats/helper.go

@@ -24,8 +24,10 @@ import (
 	cadvisorapiv2 "github.com/google/cadvisor/info/v2"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/types"
+	utilfeature "k8s.io/apiserver/pkg/util/feature"
 	"k8s.io/klog/v2"
 	statsapi "k8s.io/kubelet/pkg/apis/stats/v1alpha1"
+	"k8s.io/kubernetes/pkg/features"
 	"k8s.io/kubernetes/pkg/kubelet/cadvisor"
 	"k8s.io/kubernetes/pkg/kubelet/server/stats"
 )
@@ -53,6 +55,9 @@ func cadvisorInfoToCPUandMemoryStats(info *cadvisorapiv2.ContainerInfo) (*statsa
 		}
 		if cstat.Cpu != nil {
 			cpuStats.UsageCoreNanoSeconds = &cstat.Cpu.Usage.Total
+			if utilfeature.DefaultFeatureGate.Enabled(features.KubeletPSI) {
+				cpuStats.PSI = cadvisorPSIToStatsPSI(&cstat.Cpu.PSI)
+			}
 		}
 	}
 	if info.Spec.HasMemory && cstat.Memory != nil {
@@ -71,6 +76,9 @@ func cadvisorInfoToCPUandMemoryStats(info *cadvisorapiv2.ContainerInfo) (*statsa
 			availableBytes := info.Spec.Memory.Limit - cstat.Memory.WorkingSet
 			memoryStats.AvailableBytes = &availableBytes
 		}
+		if utilfeature.DefaultFeatureGate.Enabled(features.KubeletPSI) {
+			memoryStats.PSI = cadvisorPSIToStatsPSI(&cstat.Memory.PSI)
+		}
 	} else {
 		memoryStats = &statsapi.MemoryStats{
 			Time:            metav1.NewTime(cstat.Timestamp),
@@ -96,6 +104,9 @@ func cadvisorInfoToContainerStats(name string, info *cadvisorapiv2.ContainerInfo
 	result.CPU = cpu
 	result.Memory = memory
 	result.Swap = cadvisorInfoToSwapStats(info)
+	if utilfeature.DefaultFeatureGate.Enabled(features.KubeletPSI) {
+		result.IO = cadvisorInfoToIOStats(info)
+	}
 
 	// NOTE: if they can be found, log stats will be overwritten
 	// by the caller, as it knows more information about the pod,
@@ -307,6 +318,24 @@ func cadvisorInfoToSwapStats(info *cadvisorapiv2.ContainerInfo) *statsapi.SwapSt
 	return swapStats
 }
 
+func cadvisorInfoToIOStats(info *cadvisorapiv2.ContainerInfo) *statsapi.IOStats {
+	cstat, found := latestContainerStats(info)
+	if !found {
+		return nil
+	}
+
+	var ioStats *statsapi.IOStats
+
+	if info.Spec.HasDiskIo && cstat.DiskIo != nil {
+		ioStats = &statsapi.IOStats{
+			Time: metav1.NewTime(cstat.Timestamp),
+			PSI:  cadvisorPSIToStatsPSI(&cstat.DiskIo.PSI),
+		}
+	}
+
+	return ioStats
+}
+
 // latestContainerStats returns the latest container stats from cadvisor, or nil if none exist
 func latestContainerStats(info *cadvisorapiv2.ContainerInfo) (*cadvisorapiv2.ContainerStats, bool) {
 	stats := info.Stats
@@ -493,3 +522,23 @@ func makePodStorageStats(s *statsapi.PodStats, rootFsInfo *cadvisorapiv2.FsInfo,
 	}
 	s.EphemeralStorage = calcEphemeralStorage(s.Containers, ephemeralStats, rootFsInfo, logStats, etcHostsStats, isCRIStatsProvider)
 }
+
+func cadvisorPSIToStatsPSI(psi *cadvisorapiv1.PSIStats) *statsapi.PSIStats {
+	if psi == nil {
+		return nil
+	}
+	return &statsapi.PSIStats{
+		Full: statsapi.PSIData{
+			Total:  psi.Full.Total,
+			Avg10:  psi.Full.Avg10,
+			Avg60:  psi.Full.Avg60,
+			Avg300: psi.Full.Avg300,
+		},
+		Some: statsapi.PSIData{
+			Total:  psi.Some.Total,
+			Avg10:  psi.Some.Avg10,
+			Avg60:  psi.Some.Avg60,
+			Avg300: psi.Some.Avg300,
+		},
+	}
+}