scottk/kubernetes
1
0
Fork 0
mirror of https://github.com/optim-enterprises-bv/kubernetes.git synced 2025-10-31 02:08:13 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #129950 from ffromani/alignment-error-detail-metrics

Browse Source 
node: metrics for alignment failures
This commit is contained in:
Kubernetes Prow Robot
2025-03-12 18:03:46 -07:00
committed by GitHub
parent 2261137135 04129d1dc8
commit 05bfdbc6dd
7 changed files with 75 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
pkg/kubelet/cm/cpumanager/policy_static.go
View File

@@ -325,13 +325,15 @@ func (p *staticPolicy) Allocate(s state.State, pod *v1.Pod, container *v1.Contai
defer func() {
if rerr != nil {
metrics.CPUManagerPinningErrorsTotal.Inc()
if p.options.FullPhysicalCPUsOnly {
metrics.ContainerAlignedComputeResourcesFailure.WithLabelValues(metrics.AlignScopeContainer, metrics.AlignedPhysicalCPU).Inc()
}
return
}
if !p.options.FullPhysicalCPUsOnly {
if p.options.FullPhysicalCPUsOnly {
// increment only if we know we allocate aligned resources
return
}
metrics.ContainerAlignedComputeResources.WithLabelValues(metrics.AlignScopeContainer, metrics.AlignedPhysicalCPU).Inc()
}
}()
if p.options.FullPhysicalCPUsOnly {
@@ -752,6 +754,7 @@ func (p *staticPolicy) getAlignedCPUs(numaAffinity bitmask.BitMask, allocatableC
func (p *staticPolicy) initializeMetrics(s state.State) {
metrics.CPUManagerSharedPoolSizeMilliCores.Set(float64(p.GetAvailableCPUs(s).Size() * 1000))
metrics.CPUManagerExclusiveCPUsAllocationCount.Set(float64(countExclusiveCPUs(s)))
metrics.ContainerAlignedComputeResourcesFailure.WithLabelValues(metrics.AlignScopeContainer, metrics.AlignedPhysicalCPU).Add(0) // ensure the value exists
}
func (p *staticPolicy) updateMetricsOnAllocate(cset cpuset.CPUSet) {

3
pkg/kubelet/cm/topologymanager/scope_container.go
View File

@@ -50,6 +50,9 @@ func (s *containerScope) Admit(pod *v1.Pod) lifecycle.PodAdmitResult {
klog.InfoS("Best TopologyHint", "bestHint", bestHint, "pod", klog.KObj(pod), "containerName", container.Name)
if !admit {
if IsAlignmentGuaranteed(s.policy) {
metrics.ContainerAlignedComputeResourcesFailure.WithLabelValues(metrics.AlignScopeContainer, metrics.AlignedNUMANode).Inc()
}
metrics.TopologyManagerAdmissionErrorsTotal.Inc()
return admission.GetPodAdmitResult(&TopologyAffinityError{})
}

4
pkg/kubelet/cm/topologymanager/scope_pod.go
View File

@@ -48,6 +48,10 @@ func (s *podScope) Admit(pod *v1.Pod) lifecycle.PodAdmitResult {
bestHint, admit := s.calculateAffinity(pod)
klog.InfoS("Best TopologyHint", "bestHint", bestHint, "pod", klog.KObj(pod))
if !admit {
if IsAlignmentGuaranteed(s.policy) {
// increment only if we know we allocate aligned resources.
metrics.ContainerAlignedComputeResourcesFailure.WithLabelValues(metrics.AlignScopePod, metrics.AlignedNUMANode).Inc()
}
metrics.TopologyManagerAdmissionErrorsTotal.Inc()
return admission.GetPodAdmitResult(&TopologyAffinityError{})
}

10
pkg/kubelet/cm/topologymanager/topology_manager.go
View File

@@ -188,9 +188,19 @@ func NewManager(topology []cadvisorapi.Node, topologyPolicyName string, topology
scope: scope,
}
manager.initializeMetrics()
return manager, nil
}
func (m *manager) initializeMetrics() {
// ensure the values exist
metrics.ContainerAlignedComputeResources.WithLabelValues(metrics.AlignScopeContainer, metrics.AlignedNUMANode).Add(0)
metrics.ContainerAlignedComputeResources.WithLabelValues(metrics.AlignScopePod, metrics.AlignedNUMANode).Add(0)
metrics.ContainerAlignedComputeResourcesFailure.WithLabelValues(metrics.AlignScopeContainer, metrics.AlignedNUMANode).Add(0)
metrics.ContainerAlignedComputeResourcesFailure.WithLabelValues(metrics.AlignScopePod, metrics.AlignedNUMANode).Add(0)
}
func (m *manager) GetAffinity(podUID string, containerName string) TopologyHint {
return m.scope.GetAffinity(podUID, containerName)
}

15
pkg/kubelet/metrics/metrics.go
View File

@@ -132,6 +132,7 @@ const (
// Metric for tracking aligment of compute resources
ContainerAlignedComputeResourcesNameKey = "container_aligned_compute_resources_count"
ContainerAlignedComputeResourcesFailureNameKey = "container_aligned_compute_resources_failure_count"
ContainerAlignedComputeResourcesScopeLabelKey = "scope"
ContainerAlignedComputeResourcesBoundaryLabelKey = "boundary"
@@ -818,7 +819,18 @@ var (
},
[]string{ContainerAlignedComputeResourcesScopeLabelKey, ContainerAlignedComputeResourcesBoundaryLabelKey},
)
// MemoryManagerPinningRequestTotal tracks the number of times the pod spec required the memory manager to pin memory pages
// ContainerAlignedComputeResourcesFailure reports the count of resources allocation attempts which failed to align resources, per alignment boundary
ContainerAlignedComputeResourcesFailure = metrics.NewCounterVec(
&metrics.CounterOpts{
Subsystem: KubeletSubsystem,
Name: ContainerAlignedComputeResourcesFailureNameKey,
Help: "Cumulative number of failures to allocate aligned compute resources to containers by alignment type.",
StabilityLevel: metrics.ALPHA,
},
[]string{ContainerAlignedComputeResourcesScopeLabelKey, ContainerAlignedComputeResourcesBoundaryLabelKey},
)
MemoryManagerPinningRequestTotal = metrics.NewCounter(
&metrics.CounterOpts{
Subsystem: KubeletSubsystem,
@@ -1079,6 +1091,7 @@ func Register(collectors ...metrics.StableCollector) {
legacyregistry.MustRegister(CPUManagerSharedPoolSizeMilliCores)
legacyregistry.MustRegister(CPUManagerExclusiveCPUsAllocationCount)
legacyregistry.MustRegister(ContainerAlignedComputeResources)
legacyregistry.MustRegister(ContainerAlignedComputeResourcesFailure)
legacyregistry.MustRegister(MemoryManagerPinningRequestTotal)
legacyregistry.MustRegister(MemoryManagerPinningErrorsTotal)
legacyregistry.MustRegister(TopologyManagerAdmissionRequestsTotal)

15
test/e2e_node/cpu_manager_metrics_test.go
View File

@@ -104,6 +104,7 @@ var _ = SIGDescribe("CPU Manager Metrics", framework.WithSerial(), feature.CPUMa
// being [Serial], we can also assume noone else but us is running pods.
ginkgo.By("Checking the cpumanager metrics right after the kubelet restart, with no pods running")
idFn := makeCustomPairID("scope", "boundary")
matchResourceMetrics := gstruct.MatchKeys(gstruct.IgnoreExtras, gstruct.Keys{
"kubelet_cpu_manager_pinning_requests_total": gstruct.MatchAllElements(nodeID, gstruct.Elements{
"": timelessSample(0),
@@ -111,6 +112,9 @@ var _ = SIGDescribe("CPU Manager Metrics", framework.WithSerial(), feature.CPUMa
"kubelet_cpu_manager_pinning_errors_total": gstruct.MatchAllElements(nodeID, gstruct.Elements{
"": timelessSample(0),
}),
"kubelet_container_aligned_compute_resources_failure_count": gstruct.MatchElements(idFn, gstruct.IgnoreExtras, gstruct.Elements{
"container::physical_cpu": timelessSample(0),
}),
})
ginkgo.By("Giving the Kubelet time to start up and produce metrics")
@@ -127,6 +131,7 @@ var _ = SIGDescribe("CPU Manager Metrics", framework.WithSerial(), feature.CPUMa
// being [Serial], we can also assume noone else but us is running pods.
ginkgo.By("Checking the cpumanager metrics right after the kubelet restart, with pod failed to admit")
idFn := makeCustomPairID("scope", "boundary")
matchResourceMetrics := gstruct.MatchKeys(gstruct.IgnoreExtras, gstruct.Keys{
"kubelet_cpu_manager_pinning_requests_total": gstruct.MatchAllElements(nodeID, gstruct.Elements{
"": timelessSample(1),
@@ -134,6 +139,9 @@ var _ = SIGDescribe("CPU Manager Metrics", framework.WithSerial(), feature.CPUMa
"kubelet_cpu_manager_pinning_errors_total": gstruct.MatchAllElements(nodeID, gstruct.Elements{
"": timelessSample(1),
}),
"kubelet_container_aligned_compute_resources_failure_count": gstruct.MatchElements(idFn, gstruct.IgnoreExtras, gstruct.Elements{
"container::physical_cpu": timelessSample(1),
}),
})
ginkgo.By("Giving the Kubelet time to start up and produce metrics")
@@ -150,6 +158,7 @@ var _ = SIGDescribe("CPU Manager Metrics", framework.WithSerial(), feature.CPUMa
// being [Serial], we can also assume noone else but us is running pods.
ginkgo.By("Checking the cpumanager metrics right after the kubelet restart, with pod should be admitted")
idFn := makeCustomPairID("scope", "boundary")
matchResourceMetrics := gstruct.MatchKeys(gstruct.IgnoreExtras, gstruct.Keys{
"kubelet_cpu_manager_pinning_requests_total": gstruct.MatchAllElements(nodeID, gstruct.Elements{
"": timelessSample(1),
@@ -157,6 +166,9 @@ var _ = SIGDescribe("CPU Manager Metrics", framework.WithSerial(), feature.CPUMa
"kubelet_cpu_manager_pinning_errors_total": gstruct.MatchAllElements(nodeID, gstruct.Elements{
"": timelessSample(0),
}),
"kubelet_container_aligned_compute_resources_failure_count": gstruct.MatchElements(idFn, gstruct.IgnoreExtras, gstruct.Elements{
"container::physical_cpu": timelessSample(0),
}),
})
ginkgo.By("Giving the Kubelet time to start up and produce metrics")
@@ -178,6 +190,9 @@ var _ = SIGDescribe("CPU Manager Metrics", framework.WithSerial(), feature.CPUMa
"kubelet_container_aligned_compute_resources_count": gstruct.MatchElements(idFn, gstruct.IgnoreExtras, gstruct.Elements{
"container::physical_cpu": timelessSample(1),
}),
"kubelet_container_aligned_compute_resources_failure_count": gstruct.MatchElements(idFn, gstruct.IgnoreExtras, gstruct.Elements{
"container::physical_cpu": timelessSample(0),
}),
})
ginkgo.By("Giving the Kubelet time to update the alignment metrics")

26
test/e2e_node/topology_manager_metrics_test.go
View File

@@ -28,7 +28,6 @@ import (
v1 "k8s.io/api/core/v1"
kubeletconfig "k8s.io/kubernetes/pkg/kubelet/apis/config"
"k8s.io/kubernetes/pkg/kubelet/cm/topologymanager"
"k8s.io/kubernetes/pkg/kubelet/metrics"
"k8s.io/kubernetes/test/e2e/feature"
"k8s.io/kubernetes/test/e2e/framework"
e2epod "k8s.io/kubernetes/test/e2e/framework/pod"
@@ -84,6 +83,7 @@ var _ = SIGDescribe("Topology Manager Metrics", framework.WithSerial(), feature.
// being [Serial], we can also assume noone else but us is running pods.
ginkgo.By("Checking the topologymanager metrics right after the kubelet restart, with no pods running")
idFn := makeCustomPairID("scope", "boundary")
matchResourceMetrics := gstruct.MatchKeys(gstruct.IgnoreExtras, gstruct.Keys{
"kubelet_topology_manager_admission_requests_total": gstruct.MatchAllElements(nodeID, gstruct.Elements{
"": timelessSample(0),
@@ -91,6 +91,10 @@ var _ = SIGDescribe("Topology Manager Metrics", framework.WithSerial(), feature.
"kubelet_topology_manager_admission_errors_total": gstruct.MatchAllElements(nodeID, gstruct.Elements{
"": timelessSample(0),
}),
"kubelet_container_aligned_compute_resources_failure_count": gstruct.MatchElements(idFn, gstruct.IgnoreExtras, gstruct.Elements{
"container::numa_node": timelessSample(0),
"pod::numa_node": timelessSample(0),
}),
"kubelet_topology_manager_admission_duration_ms_count": gstruct.MatchElements(nodeID, gstruct.IgnoreExtras, gstruct.Elements{
"": timelessSample(0),
}),
@@ -110,6 +114,7 @@ var _ = SIGDescribe("Topology Manager Metrics", framework.WithSerial(), feature.
// being [Serial], we can also assume noone else but us is running pods.
ginkgo.By("Checking the topologymanager metrics right after the kubelet restart, with pod failed to admit")
idFn := makeCustomPairID("scope", "boundary")
matchResourceMetrics := gstruct.MatchKeys(gstruct.IgnoreExtras, gstruct.Keys{
"kubelet_topology_manager_admission_requests_total": gstruct.MatchAllElements(nodeID, gstruct.Elements{
"": timelessSample(1),
@@ -117,6 +122,10 @@ var _ = SIGDescribe("Topology Manager Metrics", framework.WithSerial(), feature.
"kubelet_topology_manager_admission_errors_total": gstruct.MatchAllElements(nodeID, gstruct.Elements{
"": timelessSample(1),
}),
"kubelet_container_aligned_compute_resources_failure_count": gstruct.MatchElements(idFn, gstruct.IgnoreExtras, gstruct.Elements{
"container::numa_node": timelessSample(0),
"pod::numa_node": timelessSample(1),
}),
"kubelet_topology_manager_admission_duration_ms_count": gstruct.MatchElements(nodeID, gstruct.IgnoreExtras, gstruct.Elements{
"": checkMetricValueGreaterThan(0),
}),
@@ -136,6 +145,7 @@ var _ = SIGDescribe("Topology Manager Metrics", framework.WithSerial(), feature.
// being [Serial], we can also assume noone else but us is running pods.
ginkgo.By("Checking the topologymanager metrics right after the kubelet restart, with pod should be admitted")
idFn := makeCustomPairID("scope", "boundary")
matchResourceMetrics := gstruct.MatchKeys(gstruct.IgnoreExtras, gstruct.Keys{
"kubelet_topology_manager_admission_requests_total": gstruct.MatchAllElements(nodeID, gstruct.Elements{
"": timelessSample(1),
@@ -143,6 +153,10 @@ var _ = SIGDescribe("Topology Manager Metrics", framework.WithSerial(), feature.
"kubelet_topology_manager_admission_errors_total": gstruct.MatchAllElements(nodeID, gstruct.Elements{
"": timelessSample(0),
}),
"kubelet_container_aligned_compute_resources_failure_count": gstruct.MatchElements(idFn, gstruct.IgnoreExtras, gstruct.Elements{
"container::numa_node": timelessSample(0),
"pod::numa_node": timelessSample(0),
}),
"kubelet_topology_manager_admission_duration_ms_count": gstruct.MatchElements(nodeID, gstruct.IgnoreExtras, gstruct.Elements{
"": checkMetricValueGreaterThan(0),
}),
@@ -162,9 +176,15 @@ var _ = SIGDescribe("Topology Manager Metrics", framework.WithSerial(), feature.
// being [Serial], we can also assume noone else but us is running pods.
ginkgo.By("Checking the cpumanager metrics right after the kubelet restart, with pod should be admitted")
idFn := makeCustomPairID("scope", "boundary")
matchAlignmentMetrics := gstruct.MatchKeys(gstruct.IgnoreExtras, gstruct.Keys{
"kubelet_container_aligned_compute_resources_count": gstruct.MatchAllElements(nodeID, gstruct.Elements{
metrics.AlignedNUMANode: timelessSample(1),
"kubelet_container_aligned_compute_resources_count": gstruct.MatchAllElements(idFn, gstruct.Elements{
"container::numa_node": timelessSample(0),
"pod::numa_node": timelessSample(1),
}),
"kubelet_container_aligned_compute_resources_failure_count": gstruct.MatchElements(idFn, gstruct.IgnoreExtras, gstruct.Elements{
"container::numa_node": timelessSample(0),
"pod::numa_node": timelessSample(0),
}),
})