Merge pull request #96873 from ruiwen-zhao/accelerator_stats

Add AcceleratorStats to cri_stats_provider

pkg/kubelet/stats/cri_stats_provider.go

@@ -292,7 +292,7 @@ func (p *criStatsProvider) ListPodCPUAndMemoryStats() ([]statsapi.PodStats, erro
 		if !caFound {
 			klog.V(4).Infof("Unable to find cadvisor stats for %q", containerID)
 		} else {
-			p.addCadvisorContainerStats(cs, &caStats)
+			p.addCadvisorContainerCPUAndMemoryStats(cs, &caStats)
 		}
 		ps.Containers = append(ps.Containers, *cs)
 	}
@@ -785,6 +785,25 @@ func (p *criStatsProvider) addCadvisorContainerStats(
 		cs.UserDefinedMetrics = cadvisorInfoToUserDefinedMetrics(caPodStats)
 	}
 
+	cpu, memory := cadvisorInfoToCPUandMemoryStats(caPodStats)
+	if cpu != nil {
+		cs.CPU = cpu
+	}
+	if memory != nil {
+		cs.Memory = memory
+	}
+	accelerators := cadvisorInfoToAcceleratorStats(caPodStats)
+	cs.Accelerators = accelerators
+}
+
+func (p *criStatsProvider) addCadvisorContainerCPUAndMemoryStats(
+	cs *statsapi.ContainerStats,
+	caPodStats *cadvisorapiv2.ContainerInfo,
+) {
+	if caPodStats.Spec.HasCustomMetrics {
+		cs.UserDefinedMetrics = cadvisorInfoToUserDefinedMetrics(caPodStats)
+	}
+
 	cpu, memory := cadvisorInfoToCPUandMemoryStats(caPodStats)
 	if cpu != nil {
 		cs.CPU = cpu

pkg/kubelet/stats/cri_stats_provider_test.go

@@ -260,11 +260,14 @@ func TestCRIListPodStats(t *testing.T) {
 	c0 := containerStatsMap[cName0]
 	assert.Equal(container0.CreatedAt, c0.StartTime.UnixNano())
 	checkCRICPUAndMemoryStats(assert, c0, infos[container0.ContainerStatus.Id].Stats[0])
+	checkCRIAcceleratorStats(assert, c0, infos[container0.ContainerStatus.Id].Stats[0])
 	checkCRIRootfsStats(assert, c0, containerStats0, &imageFsInfo)
 	checkCRILogsStats(assert, c0, &rootFsInfo, containerLogStats0)
+
 	c1 := containerStatsMap[cName1]
 	assert.Equal(container1.CreatedAt, c1.StartTime.UnixNano())
 	checkCRICPUAndMemoryStats(assert, c1, infos[container1.ContainerStatus.Id].Stats[0])
+	checkCRIAcceleratorStats(assert, c1, infos[container1.ContainerStatus.Id].Stats[0])
 	checkCRIRootfsStats(assert, c1, containerStats1, nil)
 	checkCRILogsStats(assert, c1, &rootFsInfo, containerLogStats1)
 	checkCRINetworkStats(assert, p0.Network, infos[sandbox0.PodSandboxStatus.Id].Stats[0].Network)
@@ -280,6 +283,7 @@ func TestCRIListPodStats(t *testing.T) {
 	assert.Equal(cName2, c2.Name)
 	assert.Equal(container2.CreatedAt, c2.StartTime.UnixNano())
 	checkCRICPUAndMemoryStats(assert, c2, infos[container2.ContainerStatus.Id].Stats[0])
+	checkCRIAcceleratorStats(assert, c2, infos[container2.ContainerStatus.Id].Stats[0])
 	checkCRIRootfsStats(assert, c2, containerStats2, &imageFsInfo)
 	checkCRILogsStats(assert, c2, &rootFsInfo, containerLogStats2)
 	checkCRINetworkStats(assert, p1.Network, infos[sandbox1.PodSandboxStatus.Id].Stats[0].Network)
@@ -296,6 +300,7 @@ func TestCRIListPodStats(t *testing.T) {
 	assert.Equal(cName3, c3.Name)
 	assert.Equal(container4.CreatedAt, c3.StartTime.UnixNano())
 	checkCRICPUAndMemoryStats(assert, c3, infos[container4.ContainerStatus.Id].Stats[0])
+	checkCRIAcceleratorStats(assert, c3, infos[container4.ContainerStatus.Id].Stats[0])
 	checkCRIRootfsStats(assert, c3, containerStats4, &imageFsInfo)
 
 	checkCRILogsStats(assert, c3, &rootFsInfo, containerLogStats4)
@@ -692,6 +697,18 @@ func checkCRICPUAndMemoryStats(assert *assert.Assertions, actual statsapi.Contai
 	assert.Equal(cs.Memory.ContainerData.Pgmajfault, *actual.Memory.MajorPageFaults)
 }
 
+func checkCRIAcceleratorStats(assert *assert.Assertions, actual statsapi.ContainerStats, cs *cadvisorapiv2.ContainerStats) {
+	assert.Equal(len(cs.Accelerators), len(actual.Accelerators))
+	for i := range cs.Accelerators {
+		assert.Equal(cs.Accelerators[i].Make, actual.Accelerators[i].Make)
+		assert.Equal(cs.Accelerators[i].Model, actual.Accelerators[i].Model)
+		assert.Equal(cs.Accelerators[i].ID, actual.Accelerators[i].ID)
+		assert.Equal(cs.Accelerators[i].MemoryTotal, actual.Accelerators[i].MemoryTotal)
+		assert.Equal(cs.Accelerators[i].MemoryUsed, actual.Accelerators[i].MemoryUsed)
+		assert.Equal(cs.Accelerators[i].DutyCycle, actual.Accelerators[i].DutyCycle)
+	}
+}
+
 func checkCRIRootfsStats(assert *assert.Assertions, actual statsapi.ContainerStats, cs *runtimeapi.ContainerStats, imageFsInfo *cadvisorapiv2.FsInfo) {
 	assert.Equal(cs.WritableLayer.Timestamp, actual.Rootfs.Time.UnixNano())
 	if imageFsInfo != nil {

pkg/kubelet/stats/helper.go

@@ -153,6 +153,27 @@ func cadvisorInfoToContainerCPUAndMemoryStats(name string, info *cadvisorapiv2.C
 	return result
 }
 
+// cadvisorInfoToAcceleratorStats returns the statsapi.AcceleratorStats converted from
+// the container info from cadvisor.
+func cadvisorInfoToAcceleratorStats(info *cadvisorapiv2.ContainerInfo) []statsapi.AcceleratorStats {
+	cstat, found := latestContainerStats(info)
+	if !found || cstat.Accelerators == nil {
+		return nil
+	}
+	var result []statsapi.AcceleratorStats
+	for _, acc := range cstat.Accelerators {
+		result = append(result, statsapi.AcceleratorStats{
+			Make:        acc.Make,
+			Model:       acc.Model,
+			ID:          acc.ID,
+			MemoryTotal: acc.MemoryTotal,
+			MemoryUsed:  acc.MemoryUsed,
+			DutyCycle:   acc.DutyCycle,
+		})
+	}
+	return result
+}
+
 func cadvisorInfoToProcessStats(info *cadvisorapiv2.ContainerInfo) *statsapi.ProcessStats {
 	cstat, found := latestContainerStats(info)
 	if !found || cstat.Processes == nil {

pkg/kubelet/stats/provider_test.go

@@ -59,6 +59,7 @@ const (
 	offsetFsTotalUsageBytes
 	offsetFsBaseUsageBytes
 	offsetFsInodeUsage
+	offsetAcceleratorDutyCycle
 )
 
 var (
@@ -499,6 +500,16 @@ func getTestContainerInfo(seed int, podName string, podNamespace string, contain
 			BaseUsageBytes:  &baseUsageBytes,
 			InodeUsage:      &inodeUsage,
 		},
+		Accelerators: []cadvisorapiv1.AcceleratorStats{
+			{
+				Make:        "nvidia",
+				Model:       "Tesla K80",
+				ID:          "foobar",
+				MemoryTotal: uint64(seed + offsetMemUsageBytes),
+				MemoryUsed:  uint64(seed + offsetMemUsageBytes),
+				DutyCycle:   uint64(seed + offsetAcceleratorDutyCycle),
+			},
+		},
 	}
 	stats.Cpu.Usage.Total = uint64(seed + offsetCPUUsageCoreSeconds)
 	stats.CpuInst.Usage.Total = uint64(seed + offsetCPUUsageCores)