Merge pull request #251 from sustainable-computing-io/perf-metrics

add cpu instructions perf metrics

Author: wangchen615

pkg/attacher/bcc_attacher.go

@@ -49,14 +49,14 @@ type BpfModuleTables struct {
 
 const (
 	CPUCycleLable       = "cpu_cycles"
-	CPUInstructionLable = "cpu_instr"
+	CPUInstructionLabel = "cpu_instr"
 	CacheMissLabel      = "cache_miss"
 )
 
 var (
 	Counters = map[string]perfCounter{
 		CPUCycleLable:       {unix.PERF_TYPE_HARDWARE, unix.PERF_COUNT_HW_CPU_CYCLES, true},
-		CPUInstructionLable: {unix.PERF_TYPE_HARDWARE, unix.PERF_COUNT_HW_INSTRUCTIONS, true},
+		CPUInstructionLabel: {unix.PERF_TYPE_HARDWARE, unix.PERF_COUNT_HW_INSTRUCTIONS, true},
 		CacheMissLabel:      {unix.PERF_TYPE_HARDWARE, unix.PERF_COUNT_HW_CACHE_MISSES, true},
 	}
 	EnableCPUFreq = true

pkg/attacher/bcc_attacher_stub.go

@@ -25,7 +25,7 @@ import (
 
 const (
 	CPUCycleLable       = "cpu_cycles"
-	CPUInstructionLable = "cpu_instr"
+	CPUInstructionLabel = "cpu_instr"
 	CacheMissLabel      = "cache_miss"
 )
 

pkg/collector/collector.go

@@ -26,12 +26,13 @@ import (
 )
 
 const (
-	nodeEnergyStatMetric = "node_energy_stat"
-	podEnergyStatMetric  = "pod_energy_stat"
-	nodeEnergyMetric     = "node_hwmon_energy_joule_total"
-	freqMetric           = "node_cpu_scaling_frequency_hertz"
-	pkgEnergyMetric      = "node_package_energy_millijoule"
-	perCPUStatMetric     = "pod_cpu_cpu_time_us"
+	nodeEnergyStatMetric    = "node_energy_stat"
+	podEnergyStatMetric     = "pod_energy_stat"
+	nodeEnergyMetric        = "node_hwmon_energy_joule_total"
+	freqMetric              = "node_cpu_scaling_frequency_hertz"
+	pkgEnergyMetric         = "node_package_energy_millijoule"
+	perCPUStatMetric        = "pod_cpu_cpu_time_us"
+	podCPUInstructionMetric = "pod_cpu_instructions"
 
 	podLabelPrefix  = "pod_"
 	nodeLabelPrefix = "node_"
@@ -265,6 +266,18 @@ func (c *Collector) getPodDetailedCPUTimeDescription() *prometheus.Desc {
 	)
 }
 
+func (c *Collector) getPodInstructionDescription() *prometheus.Desc {
+	return prometheus.NewDesc(
+		podCPUInstructionMetric,
+		"Recorded CPU instructions during the period.",
+		[]string{
+			"pod_name",
+			"pod_namespace",
+		},
+		nil,
+	)
+}
+
 func (c *Collector) Describe(ch chan<- *prometheus.Desc) {
 	lock.Lock()
 	defer lock.Unlock()
@@ -284,6 +297,9 @@ func (c *Collector) Describe(ch chan<- *prometheus.Desc) {
 	ch <- c.getFreqDescription()
 	ch <- c.getPackageEnergyDescription()
 	ch <- c.getPodDetailedCPUTimeDescription()
+	if cpuInstrCounterEnabled {
+		ch <- c.getPodInstructionDescription()
+	}
 }
 
 func (c *Collector) Collect(ch chan<- prometheus.Metric) {
@@ -390,6 +406,16 @@ func (c *Collector) Collect(ch chan<- prometheus.Metric) {
 			)
 			ch <- metric
 		}
+		if cpuInstrCounterEnabled &&
+			v.CounterStats[attacher.CPUInstructionLabel] != nil {
+			// all curr pod cpu instructions
+			edesc = prometheus.MustNewConstMetric(
+				c.getPodInstructionDescription(),
+				prometheus.GaugeValue,
+				float64(v.CounterStats[attacher.CPUInstructionLabel].Curr),
+				v.PodName, v.Namespace)
+			ch <- edesc
+		}
 	}
 
 	// de_node_energy and desc_node_energy give indexable values for total energy consumptions of a node

pkg/collector/metrics.go

@@ -47,6 +47,8 @@ var (
 	uintFeatures  []string = getUIntFeatures()
 	features      []string = append(FloatFeatures, uintFeatures...)
 	metricNames   []string = getEstimatorMetrics()
+
+	cpuInstrCounterEnabled = isCounterStatEnabled(attacher.CPUInstructionLabel)
 )
 
 func getUIntFeatures() []string {
@@ -88,3 +90,12 @@ func getEstimatorMetrics() []string {
 	model.InitMetricIndexes(names)
 	return names
 }
+
+func isCounterStatEnabled(label string) bool {
+	for _, counter := range availableCounters {
+		if counter == label {
+			return true
+		}
+	}
+	return false
+}

pkg/collector/reader.go

@@ -160,7 +160,7 @@ func (c *Collector) readBPFEvent() (pidPodName map[uint32]string, containerIDPod
 			switch counterKey {
 			case attacher.CPUCycleLable:
 				val = ct.CPUCycles
-			case attacher.CPUInstructionLable:
+			case attacher.CPUInstructionLabel:
 				val = ct.CPUInstr
 			case attacher.CacheMissLabel:
 				val = ct.CacheMisses