Merge pull request #1718 from maryamtahhan/config-globals-cleanup

chore: cleanup pkg/config globals

cmd/exporter/exporter.go

@@ -98,9 +98,9 @@ func healthProbe(w http.ResponseWriter, req *http.Request) {
 func main() {
 	start := time.Now()
 	klog.InitFlags(nil)
-	appConfig := newAppConfig()
-	flag.Parse()
-
+	appConfig := newAppConfig() // Initialize appConfig and define flags
+	flag.Parse()                // Parse command-line flags
+	config.GetConfig()          // Initialize the configuration
 	klog.Infof("Kepler running on version: %s", build.Version)
 
 	registry := metrics.GetRegistry()
@@ -149,7 +149,7 @@ func main() {
 
 	stats.InitAvailableParamAndMetrics()
 
-	if config.EnabledGPU {
+	if config.EnabledGPU() {
 		r := accelerator.GetRegistry()
 		if a, err := accelerator.New(accelerator.GPU, true); err == nil {
 			r.MustRegister(a) // Register the accelerator with the registry

pkg/bpf/exporter.go

@@ -91,7 +91,7 @@ func (e *exporter) attach() error {
 
 	// Set program global variables
 	err = specs.RewriteConstants(map[string]interface{}{
-		"SAMPLE_RATE": int32(config.BPFSampleRate),
+		"SAMPLE_RATE": int32(config.GetBPFSampleRate()),
 	})
 	if err != nil {
 		return fmt.Errorf("error rewriting program constants: %v", err)
@@ -112,7 +112,7 @@ func (e *exporter) attach() error {
 	}
 	e.enabledSoftwareCounters[config.CPUTime] = struct{}{}
 
-	if config.ExposeIRQCounterMetrics {
+	if config.ExposeIRQCounterMetrics() {
 		e.irqLink, err = link.AttachTracing(link.TracingOptions{
 			Program:    e.bpfObjects.KeplerIrqTrace,
 			AttachType: ebpf.AttachTraceRawTp,
@@ -148,7 +148,7 @@ func (e *exporter) attach() error {
 	}
 
 	// Return early if hardware counters are not enabled
-	if !config.ExposeHardwareCounterMetrics {
+	if !config.ExposeHardwareCounterMetrics() {
 		klog.Infof("Hardware counter metrics are disabled")
 		return nil
 	}

pkg/bpf/test_utils.go

@@ -23,7 +23,7 @@ func defaultHardwareCounters() sets.Set[string] {
 
 func defaultSoftwareCounters() sets.Set[string] {
 	swCounters := sets.New(config.CPUTime, config.PageCacheHit)
-	if config.ExposeIRQCounterMetrics {
+	if config.ExposeIRQCounterMetrics() {
 		swCounters.Insert(config.IRQNetTXLabel, config.IRQNetRXLabel, config.IRQBlockLabel)
 	}
 	return swCounters

pkg/collector/energy/node_energy_collector.go

@@ -66,7 +66,7 @@ func UpdateNodeComponentsEnergy(nodeStats *stats.NodeStats, wg *sync.WaitGroup)
 // UpdateNodeGPUEnergy updates each GPU power consumption. Right now we don't support other types of accelerators
 func UpdateNodeGPUEnergy(nodeStats *stats.NodeStats, wg *sync.WaitGroup) {
 	defer wg.Done()
-	if config.EnabledGPU {
+	if config.EnabledGPU() {
 		if gpu := acc.GetRegistry().ActiveAcceleratorByType(acc.GPU); gpu != nil {
 			gpuEnergy := gpu.Device().AbsEnergyFromDevice()
 			for gpu, energy := range gpuEnergy {

pkg/collector/metric_collector.go

@@ -79,8 +79,8 @@ func (c *Collector) Initialize() error {
 	// model component decide whether/how to init
 	model.CreatePowerEstimatorModels(
 		stats.GetProcessFeatureNames(c.bpfSupportedMetrics),
-		stats.NodeMetadataFeatureNames,
-		stats.NodeMetadataFeatureValues,
+		stats.NodeMetadataFeatureNames(),
+		stats.NodeMetadataFeatureValues(),
 		c.bpfSupportedMetrics,
 	)
 
@@ -161,7 +161,7 @@ func (c *Collector) updateProcessResourceUtilizationMetrics(wg *sync.WaitGroup)
 	// update process metrics regarding the resource utilization to be used to calculate the energy consumption
 	// we first updates the bpf which is responsible to include new processes in the ProcessStats collection
 	resourceBpf.UpdateProcessBPFMetrics(c.bpfExporter, c.ProcessStats)
-	if config.EnabledGPU {
+	if config.EnabledGPU() {
 		if acc.GetRegistry().ActiveAcceleratorByType(acc.GPU) != nil {
 			accelerator.UpdateProcessGPUUtilizationMetrics(c.ProcessStats, c.bpfSupportedMetrics)
 		}
@@ -184,7 +184,7 @@ func (c *Collector) AggregateProcessResourceUtilizationMetrics() {
 				// aggregate metrics per container
 				if config.IsExposeContainerStatsEnabled() {
 					if process.ContainerID != "" {
-						c.createContainerStatsIfNotExist(process.ContainerID, process.CGroupID, process.PID, config.EnabledEBPFCgroupID)
+						c.createContainerStatsIfNotExist(process.ContainerID, process.CGroupID, process.PID, config.EnabledEBPFCgroupID())
 						c.ContainerStats[process.ContainerID].ResourceUsage[metricName].AddDeltaStat(id, delta)
 						foundContainer[process.ContainerID] = true
 					}
@@ -270,7 +270,7 @@ func (c *Collector) AggregateProcessEnergyUtilizationMetrics() {
 				// aggregate metrics per container
 				if config.IsExposeContainerStatsEnabled() {
 					if process.ContainerID != "" {
-						c.createContainerStatsIfNotExist(process.ContainerID, process.CGroupID, process.PID, config.EnabledEBPFCgroupID)
+						c.createContainerStatsIfNotExist(process.ContainerID, process.CGroupID, process.PID, config.EnabledEBPFCgroupID())
 						c.ContainerStats[process.ContainerID].EnergyUsage[metricName].AddDeltaStat(id, delta)
 					}
 				}

pkg/collector/metric_collector_test.go

@@ -45,7 +45,7 @@ var _ = Describe("Test Collector Unit", func() {
 		metricCollector := newMockCollector(bpfExporter)
 		// The default estimator model is the ratio
 		bpfSupportedMetrics := bpfExporter.SupportedMetrics()
-		model.CreatePowerEstimatorModels(stats.GetProcessFeatureNames(bpfSupportedMetrics), stats.NodeMetadataFeatureNames, stats.NodeMetadataFeatureValues, bpfSupportedMetrics)
+		model.CreatePowerEstimatorModels(stats.GetProcessFeatureNames(bpfSupportedMetrics), stats.NodeMetadataFeatureNames(), stats.NodeMetadataFeatureValues(), bpfSupportedMetrics)
 		// update container and node metrics
 		metricCollector.UpdateProcessEnergyUtilizationMetrics()
 		metricCollector.AggregateProcessEnergyUtilizationMetrics()

pkg/collector/resourceutilization/bpf/process_bpf_collector.go

@@ -99,7 +99,7 @@ func UpdateProcessBPFMetrics(bpfExporter bpf.Exporter, processStats map[uint64]*
 		}
 
 		// if the pid is within a container, it will have a container ID
-		containerID, err := cgroup.GetContainerID(ct.CgroupId, ct.Pid, config.EnabledEBPFCgroupID)
+		containerID, err := cgroup.GetContainerID(ct.CgroupId, ct.Pid, config.EnabledEBPFCgroupID())
 		if err != nil {
 			klog.V(6).Infof("failed to resolve container for PID %v (command=%s): %v, set containerID=%s", ct.Pid, comm, err, utils.SystemProcessName)
 		}
@@ -114,7 +114,7 @@ func UpdateProcessBPFMetrics(bpfExporter bpf.Exporter, processStats map[uint64]*
 		}
 
 		mapKey := ct.Pid
-		if ct.CgroupId == 1 && config.EnabledEBPFCgroupID {
+		if ct.CgroupId == 1 && config.EnabledEBPFCgroupID() {
 			// we aggregate all kernel process to minimize overhead
 			// all kernel process has cgroup id as 1 and pid 1 is also a kernel process
 			mapKey = 1

pkg/collector/stats/benchmark_test.go

@@ -39,7 +39,7 @@ func benchmarkNtesting(b *testing.B, processNumber int) {
 
 	// The default estimator model is the ratio
 	bpfSupportedMetrics := bpf.DefaultSupportedMetrics()
-	model.CreatePowerEstimatorModels(stats.GetProcessFeatureNames(bpfSupportedMetrics), stats.NodeMetadataFeatureNames, stats.NodeMetadataFeatureValues, bpfSupportedMetrics)
+	model.CreatePowerEstimatorModels(stats.GetProcessFeatureNames(bpfSupportedMetrics), stats.NodeMetadataFeatureNames(), stats.NodeMetadataFeatureValues(), bpfSupportedMetrics)
 
 	// update container and node metrics
 	b.ReportAllocs()

pkg/collector/stats/node_stats.go

@@ -25,24 +25,33 @@ import (
 	"github.com/sustainable-computing-io/kepler/pkg/utils"
 )
 
-var (
-	NodeName            = GetNodeName()
-	NodeCPUArchitecture = getCPUArch()
-	NodeCPUPackageMap   = getCPUPackageMap()
-
-	// NodeMetricNames holds the name of the system metadata information.
-	NodeMetadataFeatureNames []string = []string{"cpu_architecture"}
-	// SystemMetadata holds the metadata regarding the system information
-	NodeMetadataFeatureValues []string = []string{NodeCPUArchitecture}
-)
-
 type NodeStats struct {
 	Stats
 
 	// IdleResUtilization is used to determine idle pmap[string]eriods
 	IdleResUtilization map[string]uint64
 }
 
+// NodeCPUArchitecture returns the CPU architecture
+func NodeCPUArchitecture() string {
+	return getCPUArch()
+}
+
+// NodeCPUPackageMap returns the CPU package map
+func NodeCPUPackageMap() map[int32]string {
+	return getCPUPackageMap()
+}
+
+// NodeMetadataFeatureNames returns the feature names for metadata
+func NodeMetadataFeatureNames() []string {
+	return []string{"cpu_architecture"}
+}
+
+// NodeMetadataFeatureValues returns the feature values for metadata
+func NodeMetadataFeatureValues() []string {
+	return []string{NodeCPUArchitecture()}
+}
+
 func NewNodeStats(bpfSupportedMetrics bpf.SupportedMetrics) *NodeStats {
 	return &NodeStats{
 		Stats:              *NewStats(bpfSupportedMetrics),
@@ -57,7 +66,7 @@ func (ne *NodeStats) ResetDeltaValues() {
 
 func (ne *NodeStats) UpdateIdleEnergyWithMinValue(isComponentsSystemCollectionSupported bool) {
 	// gpu metric
-	if config.EnabledGPU {
+	if config.EnabledGPU() {
 		if acc.GetRegistry().ActiveAcceleratorByType(acc.GPU) != nil {
 			ne.CalcIdleEnergy(config.AbsEnergyInGPU, config.IdleEnergyInGPU, config.GPUComputeUtilization)
 		}

pkg/collector/stats/stats.go

@@ -69,7 +69,7 @@ func NewStats(bpfSupportedMetrics bpf.SupportedMetrics) *Stats {
 		m.ResourceUsage[metricName] = types.NewUInt64StatCollection()
 	}
 
-	if config.EnabledGPU {
+	if config.EnabledGPU() {
 		if acc.GetRegistry().ActiveAcceleratorByType(acc.GPU) != nil {
 			m.ResourceUsage[config.GPUComputeUtilization] = types.NewUInt64StatCollection()
 			m.ResourceUsage[config.GPUMemUtilization] = types.NewUInt64StatCollection()
@@ -127,7 +127,7 @@ func (m *Stats) UpdateDynEnergy() {
 		m.CalcDynEnergy(config.AbsEnergyInPlatform, config.IdleEnergyInPlatform, config.DynEnergyInPlatform, sensorID)
 	}
 	// gpu metric
-	if config.EnabledGPU {
+	if config.EnabledGPU() {
 		if acc.GetRegistry().ActiveAcceleratorByType(acc.GPU) != nil {
 			for gpuID := range m.EnergyUsage[config.AbsEnergyInGPU] {
 				m.CalcDynEnergy(config.AbsEnergyInGPU, config.IdleEnergyInGPU, config.DynEnergyInGPU, gpuID)
@@ -162,7 +162,7 @@ func calcDynEnergy(totalE, idleE uint64) uint64 {
 
 func normalize(val float64, shouldNormalize bool) float64 {
 	if shouldNormalize {
-		return val / float64(config.SamplePeriodSec)
+		return val / float64(config.SamplePeriodSec())
 	}
 	return val
 }
@@ -181,7 +181,7 @@ func (m *Stats) ToEstimatorValues(featuresName []string, shouldNormalize bool) [
 		}
 		// some features are not related to resource utilization, such as power metrics
 		switch feature {
-		case config.GeneralUsageMetric: // is an empty string for UNCORE and OTHER resource usage
+		case config.GeneralUsageMetric(): // is an empty string for UNCORE and OTHER resource usage
 			featureValues = append(featureValues, 0)
 
 		case config.DynEnergyInPkg: // for dynamic PKG power consumption

pkg/collector/stats/stats_test.go

@@ -9,7 +9,8 @@ import (
 
 var _ = Describe("Stats", func() {
 	It("Test InitAvailableParamAndMetrics", func() {
-		config.ExposeHardwareCounterMetrics = false
+		config.GetConfig()
+		config.SetEnabledHardwareCounterMetrics(false)
 		supportedMetrics := bpf.DefaultSupportedMetrics()
 		InitAvailableParamAndMetrics()
 		exp := []string{}

pkg/collector/stats/test_utils.go

@@ -32,6 +32,7 @@ const (
 // SetMockedCollectorMetrics adds all metric to a process, otherwise it will not create the right usageMetric with all elements. The usageMetric is used in the Prediction Power Models
 // TODO: do not use a fixed usageMetric array in the power models, a structured data is more disarable.
 func SetMockedCollectorMetrics() {
+	config.GetConfig()
 	if gpu := acc.GetRegistry().ActiveAcceleratorByType(acc.GPU); gpu != nil {
 		err := gpu.Device().Init() // create structure instances that will be accessed to create a processMetric
 		klog.Fatalln(err)
@@ -50,9 +51,6 @@ func SetMockedCollectorMetrics() {
 		config.IdleEnergyInCore, config.IdleEnergyInDRAM, config.IdleEnergyInUnCore, config.IdleEnergyInPkg,
 		config.IdleEnergyInGPU, config.IdleEnergyInOther, config.IdleEnergyInPlatform,
 	}
-
-	NodeMetadataFeatureNames = []string{"cpu_architecture"}
-	NodeMetadataFeatureValues = []string{"Sandy Bridge"}
 }
 
 // CreateMockedProcessStats adds two containers with all metrics initialized

pkg/collector/stats/utils.go

@@ -81,7 +81,7 @@ func GetProcessFeatureNames(bpfSupportedMetrics bpf.SupportedMetrics) []string {
 	klog.V(3).Infof("Available ebpf counters: %v", metrics)
 
 	// gpu metric
-	if config.EnabledGPU {
+	if config.EnabledGPU() {
 		if acc.GetRegistry().ActiveAcceleratorByType(acc.GPU) != nil {
 			gpuMetrics := []string{config.GPUComputeUtilization, config.GPUMemUtilization}
 			metrics = append(metrics, gpuMetrics...)
@@ -92,7 +92,7 @@ func GetProcessFeatureNames(bpfSupportedMetrics bpf.SupportedMetrics) []string {
 	return metrics
 }
 
-func GetNodeName() string {
+func NodeName() string {
 	if nodeName := os.Getenv("NODE_NAME"); nodeName != "" {
 		return nodeName
 	}
@@ -298,7 +298,7 @@ func getCPUPmuName() (pmuName string, err error) {
 
 func getCPUArchitecture() (string, error) {
 	// check if there is a CPU architecture override
-	cpuArchOverride := config.CPUArchOverride
+	cpuArchOverride := config.CPUArchOverride()
 	if cpuArchOverride != "" {
 		klog.V(2).Infof("cpu arch override: %v\n", cpuArchOverride)
 		return cpuArchOverride, nil