Merge pull request #1704 from maryamtahhan/cleanup-gpu

chore: gpu cleanup

Makefile

@@ -43,14 +43,8 @@ CTR_CMD_PUSH_OPTIONS ?=
 
 GENERAL_TAGS := include_gcs include_oss containers_image_openpgp gssapi providerless netgo osusergo
 GPU_TAGS :=
-ifeq ($(shell ldconfig -p | grep -q libnvml_injection.so && echo exists),exists)
-	GPU_TAGS := nvml
-endif
-ifeq ($(shell ldconfig -p | grep -q libdcgm.so && echo exists),exists)
-	GPU_TAGS := dcgm
-endif
 ifeq ($(shell ldconfig -p | grep -q libhlml.so && echo exists),exists)
-	GPU_TAGS := habana
+    GPU_TAGS := habana
 endif
 
 # set GOENV

cmd/exporter/exporter.go

@@ -148,7 +148,7 @@ func main() {
 
 	if config.EnabledGPU() {
 		r := accelerator.GetRegistry()
-		if a, err := accelerator.New(accelerator.GPU, true); err == nil {
+		if a, err := accelerator.New(config.GPU, true); err == nil {
 			r.MustRegister(a) // Register the accelerator with the registry
 		} else {
 			klog.Errorf("failed to init GPU accelerators: %v", err)

pkg/collector/energy/node_energy_collector.go

@@ -67,7 +67,7 @@ func UpdateNodeComponentsEnergy(nodeStats *stats.NodeStats, wg *sync.WaitGroup)
 func UpdateNodeGPUEnergy(nodeStats *stats.NodeStats, wg *sync.WaitGroup) {
 	defer wg.Done()
 	if config.EnabledGPU() {
-		if gpu := acc.GetRegistry().ActiveAcceleratorByType(acc.GPU); gpu != nil {
+		if gpu := acc.GetActiveAcceleratorByType(config.GPU); gpu != nil {
 			gpuEnergy := gpu.Device().AbsEnergyFromDevice()
 			for gpu, energy := range gpuEnergy {
 				nodeStats.EnergyUsage[config.AbsEnergyInGPU].SetDeltaStat(fmt.Sprintf("%d", gpu), uint64(energy))

pkg/collector/metric_collector.go

@@ -159,7 +159,7 @@ func (c *Collector) updateProcessResourceUtilizationMetrics(wg *sync.WaitGroup)
 	// 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 acc.GetRegistry().ActiveAcceleratorByType(acc.GPU) != nil {
+		if acc.GetActiveAcceleratorByType(config.GPU) != nil {
 			accelerator.UpdateProcessGPUUtilizationMetrics(c.ProcessStats)
 		}
 	}

pkg/collector/metric_collector_test.go

@@ -16,7 +16,7 @@ import (
 )
 
 func newMockCollector(mockAttacher bpf.Exporter) *Collector {
-	if gpu := acc.GetRegistry().ActiveAcceleratorByType(acc.GPU); gpu != nil {
+	if gpu := acc.GetActiveAcceleratorByType(config.GPU); gpu != nil {
 		d := gpu.Device()
 		err := d.Init() // create structure instances that will be accessed to create a containerMetric
 		Expect(err).NotTo(HaveOccurred())

pkg/collector/resourceutilization/accelerator/process_gpu_collector.go

@@ -26,7 +26,7 @@ import (
 	"github.com/sustainable-computing-io/kepler/pkg/config"
 	"github.com/sustainable-computing-io/kepler/pkg/libvirt"
 	acc "github.com/sustainable-computing-io/kepler/pkg/sensors/accelerator"
-	dev "github.com/sustainable-computing-io/kepler/pkg/sensors/accelerator/device"
+	dev "github.com/sustainable-computing-io/kepler/pkg/sensors/accelerator/devices"
 	"k8s.io/klog/v2"
 
 	"github.com/sustainable-computing-io/kepler/pkg/utils"
@@ -43,7 +43,7 @@ var (
 
 // UpdateProcessGPUUtilizationMetrics reads the GPU metrics of each process using the GPU
 func UpdateProcessGPUUtilizationMetrics(processStats map[uint64]*stats.ProcessStats) {
-	if gpu := acc.GetRegistry().ActiveAcceleratorByType(acc.GPU); gpu != nil {
+	if gpu := acc.GetActiveAcceleratorByType(config.GPU); gpu != nil {
 		d := gpu.Device()
 		migDevices := d.DeviceInstances()
 		for _, _device := range d.DevicesByID() {

pkg/collector/stats/node_stats.go

@@ -51,7 +51,7 @@ func (ne *NodeStats) ResetDeltaValues() {
 func (ne *NodeStats) UpdateIdleEnergyWithMinValue(isComponentsSystemCollectionSupported bool) {
 	// gpu metric
 	if config.EnabledGPU() {
-		if acc.GetRegistry().ActiveAcceleratorByType(acc.GPU) != nil {
+		if acc.GetActiveAcceleratorByType(config.GPU) != nil {
 			ne.CalcIdleEnergy(config.AbsEnergyInGPU, config.IdleEnergyInGPU, config.GPUComputeUtilization)
 		}
 	}

pkg/collector/stats/stats.go

@@ -82,7 +82,7 @@ func NewStats() *Stats {
 	}
 
 	if config.EnabledGPU() {
-		if acc.GetRegistry().ActiveAcceleratorByType(acc.GPU) != nil {
+		if acc.GetActiveAcceleratorByType(config.GPU) != nil {
 			stats.ResourceUsage[config.GPUComputeUtilization] = types.NewUInt64StatCollection()
 			stats.ResourceUsage[config.GPUMemUtilization] = types.NewUInt64StatCollection()
 			stats.ResourceUsage[config.IdleEnergyInGPU] = types.NewUInt64StatCollection()
@@ -141,7 +141,7 @@ func (s *Stats) UpdateDynEnergy() {
 	}
 	// GPU metric
 	if config.EnabledGPU() {
-		if acc.GetRegistry().ActiveAcceleratorByType(acc.GPU) != nil {
+		if acc.GetActiveAcceleratorByType(config.GPU) != nil {
 			for gpuID := range s.EnergyUsage[config.AbsEnergyInGPU] {
 				s.CalcDynEnergy(config.AbsEnergyInGPU, config.IdleEnergyInGPU, config.DynEnergyInGPU, gpuID)
 			}

pkg/collector/stats/test_utils.go

@@ -32,7 +32,7 @@ const (
 // 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 {
+	if gpu := acc.GetActiveAcceleratorByType(config.GPU); gpu != nil {
 		err := gpu.Device().Init() // create structure instances that will be accessed to create a processMetric
 		klog.Fatalln(err)
 	}

pkg/collector/stats/utils.go

@@ -31,7 +31,7 @@ func GetProcessFeatureNames() []string {
 
 	// gpu metric
 	if config.EnabledGPU() {
-		if acc.GetRegistry().ActiveAcceleratorByType(acc.GPU) != nil {
+		if acc.GetActiveAcceleratorByType(config.GPU) != nil {
 			gpuMetrics := []string{config.GPUComputeUtilization, config.GPUMemUtilization}
 			metrics = append(metrics, gpuMetrics...)
 			klog.V(3).Infof("Available GPU metrics: %v", gpuMetrics)

pkg/metrics/metricfactory/metric_factory.go

@@ -36,7 +36,7 @@ func EnergyMetricsPromDesc(context string) (descriptions map[string]*prometheus.
 		// set the default source to trained power model
 		source := modeltypes.TrainedPowerModelSource
 		if strings.Contains(name, config.GPU) {
-			if gpu := acc.GetRegistry().ActiveAcceleratorByType(acc.GPU); gpu != nil {
+			if gpu := acc.GetActiveAcceleratorByType(config.GPU); gpu != nil {
 				source = gpu.Device().Name()
 			}
 		} else if strings.Contains(name, config.PLATFORM) && platform.IsSystemCollectionSupported() {
@@ -87,7 +87,7 @@ func SCMetricsPromDesc(context string, bpfSupportedMetrics bpf.SupportedMetrics)
 func GPUUsageMetricsPromDesc(context string) (descriptions map[string]*prometheus.Desc) {
 	descriptions = make(map[string]*prometheus.Desc)
 	if config.EnabledGPU() {
-		if gpu := acc.GetRegistry().ActiveAcceleratorByType(acc.GPU); gpu != nil {
+		if gpu := acc.GetActiveAcceleratorByType(config.GPU); gpu != nil {
 			for _, name := range consts.GPUMetricNames {
 				descriptions[name] = resMetricsPromDesc(context, name, gpu.Device().Name())
 			}

pkg/metrics/prometheus_collector_test.go

@@ -33,6 +33,7 @@ import (
 	"github.com/sustainable-computing-io/kepler/pkg/bpf"
 	"github.com/sustainable-computing-io/kepler/pkg/collector"
 	"github.com/sustainable-computing-io/kepler/pkg/collector/stats"
+	"github.com/sustainable-computing-io/kepler/pkg/config"
 	"github.com/sustainable-computing-io/kepler/pkg/model"
 
 	acc "github.com/sustainable-computing-io/kepler/pkg/sensors/accelerator"
@@ -63,7 +64,7 @@ var _ = Describe("Test Prometheus Collector Unit", func() {
 		// we need to disable the system real time power metrics for testing since we add mock values or use power model estimator
 		components.SetIsSystemCollectionSupported(false)
 		platform.SetIsSystemCollectionSupported(false)
-		if gpu := acc.GetRegistry().ActiveAcceleratorByType(acc.GPU); gpu != nil {
+		if gpu := acc.GetActiveAcceleratorByType(config.GPU); gpu != nil {
 			err := gpu.Device().Init() // create structure instances that will be accessed to create a containerMetric
 			Expect(err).NotTo(HaveOccurred())
 		}

pkg/metrics/utils/utils.go

@@ -58,7 +58,7 @@ func CollectResUtilizationMetrics(ch chan<- prometheus.Metric, instance interfac
 			CollectResUtil(ch, instance, collectorName, collectors[collectorName])
 		}
 		if config.EnabledGPU() {
-			if gpu := acc.GetRegistry().ActiveAcceleratorByType(acc.GPU); gpu != nil {
+			if gpu := acc.GetActiveAcceleratorByType(config.GPU); gpu != nil {
 				for _, collectorName := range consts.GPUMetricNames {
 					CollectResUtil(ch, instance, collectorName, collectors[collectorName])
 				}

pkg/model/process_energy.go

@@ -201,7 +201,7 @@ func addEstimatedEnergy(processIDList []uint64, processesMetrics map[uint64]*sta
 		}
 		// estimate the associated power consumption of GPU for each process
 		if config.EnabledGPU() {
-			if gpu := acc.GetRegistry().ActiveAcceleratorByType(acc.GPU); gpu != nil {
+			if gpu := acc.GetActiveAcceleratorByType(config.GPU); gpu != nil {
 				processGPUPower, errGPU = processComponentPowerModel.GetGPUPower(isIdlePower)
 				if errGPU != nil {
 					klog.V(5).Infoln("Could not estimate the Process GPU Power")