feat(metrics): expose platform and components power source (#1133)

This commit adds `platform_power_source` and `components_power_source`
to `kepler_node_info`. This allows users to filter metrics by a particular
power source by joining.

Signed-off-by: Sunil Thaha <[email protected]>

cmd/validator/validator.go

@@ -251,7 +251,7 @@ func main() {
 			}
 		}
 		if platform.IsSystemCollectionSupported() {
-			powerSource := platform.GetPowerSource()
+			powerSource := platform.GetSourceName()
 			switch powerSource {
 			case "hmc":
 				hmcEnable = true

pkg/metrics/node/metrics.go

@@ -24,6 +24,8 @@ import (
 	"github.com/sustainable-computing-io/kepler/pkg/config"
 	"github.com/sustainable-computing-io/kepler/pkg/metrics/metricfactory"
 	"github.com/sustainable-computing-io/kepler/pkg/metrics/utils"
+	"github.com/sustainable-computing-io/kepler/pkg/sensors/components"
+	"github.com/sustainable-computing-io/kepler/pkg/sensors/platform"
 )
 
 const (
@@ -70,7 +72,9 @@ func (c *collector) initMetrics() {
 	}
 
 	// TODO: prometheus metric should be "node_info"
-	desc := metricfactory.MetricsPromDesc(context, "", "info", "os", []string{"cpu_architecture"})
+	desc := metricfactory.MetricsPromDesc(context, "", "info", "os", []string{
+		"cpu_architecture", "components_power_source", "platform_power_source",
+	})
 	c.descriptions["info"] = desc
 	c.collectors["info"] = metricfactory.NewPromCounter(desc)
 }
@@ -91,5 +95,9 @@ func (c *collector) Collect(ch chan<- prometheus.Metric) {
 	c.Mx.Unlock()
 
 	// update node info
-	ch <- c.collectors["info"].MustMetric(1, stats.NodeCPUArchitecture)
+	ch <- c.collectors["info"].MustMetric(1,
+		stats.NodeCPUArchitecture,
+		components.GetSourceName(),
+		platform.GetSourceName(),
+	)
 }

pkg/sensors/accelerator/gpu/power.go

@@ -31,6 +31,9 @@ var (
 )
 
 type acceleratorInterface interface {
+	// GetName returns the name of the collector
+	GetName() string
+
 	// Init initizalize and start the GPU metric collector
 	Init() error
 	// Shutdown stops the GPU metric collector

pkg/sensors/accelerator/gpu/source/gpu_dummy.go

@@ -29,6 +29,10 @@ type GPUDummy struct {
 	collectionSupported bool
 }
 
+func (d *GPUDummy) GetName() string {
+	return "dummy"
+}
+
 // todo: refactor logic at invoking side, if gpu is not set?
 func (d *GPUDummy) Init() error {
 	d.collectionSupported = false

pkg/sensors/accelerator/gpu/source/gpu_nvml.go

@@ -37,6 +37,10 @@ type GPUNvml struct {
 	collectionSupported bool
 }
 
+func (GPUNvml) GetName() string {
+	return "nvidia-nvml"
+}
+
 // Init initizalize and start the GPU metric collector
 // the nvml only works if the container has support to GPU, e.g., it is using nvidia-docker2
 // otherwise it will fail to load the libnvidia-ml.so.1

pkg/sensors/accelerator/qat/source/qat_telemetry.go

@@ -44,6 +44,10 @@ type QATTelemetry struct {
 	collectionSupported bool
 }
 
+func (QATTelemetry) GetName() string {
+	return "qat"
+}
+
 // Init initizalize and start the QAT metric collector
 func (q *QATTelemetry) Init() (err error) {
 	defer func() {

pkg/sensors/components/power.go

@@ -24,6 +24,8 @@ import (
 )
 
 type powerInterface interface {
+	// GetName() returns the name of the source / impl used for estimation
+	GetName() string
 	// GetAbsEnergyFromDram returns mJ in DRAM. Absolute energy is the sum of Idle + Dynamic energy.
 	GetAbsEnergyFromDram() (uint64, error)
 	// GetAbsEnergyFromCore returns mJ in CPU cores
@@ -41,37 +43,41 @@ type powerInterface interface {
 }
 
 var (
-	estimateImpl                     = &source.PowerEstimate{}
-	sysfsImpl                        = &source.PowerSysfs{}
-	msrImpl                          = &source.PowerMSR{}
-	apmXgeneSysfsImpl                = &source.ApmXgeneSysfs{}
-	powerImpl         powerInterface = sysfsImpl
-	enabled                          = true
+	powerImpl powerInterface = &source.PowerSysfs{}
+	enabled                  = true
 )
 
 func InitPowerImpl() {
+	sysfsImpl := &source.PowerSysfs{}
 	if sysfsImpl.IsSystemCollectionSupported() /*&& false*/ {
 		klog.V(1).Infoln("use sysfs to obtain power")
 		powerImpl = sysfsImpl
 		return
 	}
 
+	msrImpl := &source.PowerMSR{}
 	if msrImpl.IsSystemCollectionSupported() && config.EnabledMSR {
 		klog.V(1).Infoln("use MSR to obtain power")
 		powerImpl = msrImpl
 		return
 	}
 
+	apmXgeneSysfsImpl := &source.ApmXgeneSysfs{}
 	if apmXgeneSysfsImpl.IsSystemCollectionSupported() {
 		klog.V(1).Infoln("use Ampere Xgene sysfs to obtain power")
 		powerImpl = apmXgeneSysfsImpl
 		return
 	}
 
 	klog.V(1).Infoln("Unable to obtain power, use estimate method")
+	estimateImpl := &source.PowerEstimate{}
 	powerImpl = estimateImpl
 }
 
+func GetSourceName() string {
+	return powerImpl.GetName()
+}
+
 func GetAbsEnergyFromDram() (uint64, error) {
 	return powerImpl.GetAbsEnergyFromDram()
 }

pkg/sensors/components/source/apm_xgene_sysfs.go

@@ -41,6 +41,10 @@ type ApmXgeneSysfs struct {
 	currTime time.Time
 }
 
+func (ApmXgeneSysfs) GetName() string {
+	return "ampere-xgene-hwmon"
+}
+
 func (r *ApmXgeneSysfs) IsSystemCollectionSupported() bool {
 	labelFiles, err := filepath.Glob(powerLabelPathTemplate)
 	if err != nil {

pkg/sensors/components/source/dummy.go

@@ -23,6 +23,10 @@ var (
 
 type PowerDummy struct{}
 
+func (PowerDummy) GetName() string {
+	return "dummy"
+}
+
 func (r *PowerDummy) IsSystemCollectionSupported() bool {
 	return SystemCollectionSupported
 }

pkg/sensors/components/source/estimate.go

@@ -37,6 +37,10 @@ type PowerEstimateData struct {
 	PerGBWatts   float64 `csv:"GB/Chip"`
 }
 
+func (PowerEstimate) GetName() string {
+	return "estimator"
+}
+
 // If the Estimated Power is being used, it means that the system does not support Components Power Measurement
 func (r *PowerEstimate) IsSystemCollectionSupported() bool {
 	return false

pkg/sensors/components/source/rapl_msr.go

@@ -18,6 +18,10 @@ package source
 
 type PowerMSR struct{}
 
+func (PowerMSR) GetName() string {
+	return "rapl-msr"
+}
+
 func (r *PowerMSR) IsSystemCollectionSupported() bool {
 	return InitUnits() == nil
 }

pkg/sensors/components/source/rapl_sysfs.go

@@ -120,6 +120,10 @@ func getMaxEnergyRange(eventName string) (uint64, error) {
 
 type PowerSysfs struct{}
 
+func (PowerSysfs) GetName() string {
+	return "rapl-sysfs"
+}
+
 func (r *PowerSysfs) IsSystemCollectionSupported() bool {
 	path := fmt.Sprintf(packageNamePathTemplate, 0)
 	_, err := os.ReadFile(path + energyFile)

pkg/sensors/platform/power.go

@@ -25,6 +25,8 @@ import (
 )
 
 type powerInterface interface {
+	// GetName() returns the name of the platform power source
+	GetName() string
 	// GetAbsEnergyFromPlatform returns mJ in DRAM. Absolute energy is the sum of Idle + Dynamic energy.
 	GetAbsEnergyFromPlatform() (map[string]float64, error)
 	// StopPower stops the collection
@@ -33,48 +35,57 @@ type powerInterface interface {
 	IsSystemCollectionSupported() bool
 }
 
+// dummy satisfies the powerInterface and can be used as the default NOP source
+type dummy struct {
+}
+
+func (dummy) GetName() string {
+	return "none"
+}
+
+func (dummy) IsSystemCollectionSupported() bool {
+	return false
+}
+func (dummy) StopPower() {
+}
+
+func (dummy) GetAbsEnergyFromPlatform() (map[string]float64, error) {
+	return nil, fmt.Errorf("dummy power source")
+}
+
 var (
-	powerImpl   powerInterface
-	redfishImpl *source.RedFishClient
-	hmcImpl     = &source.PowerHMC{}
-	powerSource = "none"
-	enabled     = true
+	powerImpl powerInterface = &dummy{}
+	enabled                  = true
 )
 
 func InitPowerImpl() {
 	// switch the platform power collector source to hmc if the system architecture is s390x
 	// TODO: add redfish or ipmi as well.
 	if runtime.GOARCH == "s390x" {
-		klog.V(1).Infoln("use hmc to obtain power")
-		powerImpl = hmcImpl
-		powerSource = "hmc"
-	} else if redfishImpl = source.NewRedfishClient(); redfishImpl != nil && redfishImpl.IsSystemCollectionSupported() {
-		klog.V(1).Infoln("use redfish to obtain power")
-		powerImpl = redfishImpl
-		powerSource = "redfish"
-	} else if powerImpl = source.NewACPIPowerMeter(); powerImpl != nil && powerImpl.IsSystemCollectionSupported() {
-		klog.V(1).Infoln("use acpi to obtain power")
-		powerSource = "acpi"
+		powerImpl = &source.PowerHMC{}
+	} else if redfish := source.NewRedfishClient(); redfish != nil && redfish.IsSystemCollectionSupported() {
+		powerImpl = redfish
+	} else if acpi := source.NewACPIPowerMeter(); acpi != nil && acpi.CollectEnergy {
+		powerImpl = acpi
 	}
+
+	klog.V(1).Infof("using %s to obtain power", powerImpl.GetName())
 }
 
-func GetPowerSource() string {
-	return powerSource
+func GetSourceName() string {
+	return powerImpl.GetName()
 }
 
 // GetAbsEnergyFromPlatform returns the absolute energy, which is the sum of Idle + Dynamic energy.
 func GetAbsEnergyFromPlatform() (map[string]float64, error) {
-	if powerImpl != nil {
-		return powerImpl.GetAbsEnergyFromPlatform()
-	}
-	return nil, fmt.Errorf("powerImpl is nil")
+	return powerImpl.GetAbsEnergyFromPlatform()
 }
 
 func IsSystemCollectionSupported() bool {
-	if powerImpl != nil && enabled {
-		return powerImpl.IsSystemCollectionSupported()
+	if !enabled {
+		return false
 	}
-	return false
+	return powerImpl.IsSystemCollectionSupported()
 }
 
 // SetIsSystemCollectionSupported is used to enable or disable the system power collection.
@@ -84,7 +95,5 @@ func SetIsSystemCollectionSupported(enable bool) {
 }
 
 func StopPower() {
-	if powerImpl != nil {
-		powerImpl.StopPower()
-	}
+	powerImpl.StopPower()
 }

pkg/sensors/platform/source/acpi.go

@@ -97,6 +97,10 @@ func findACPIPowerPath() string {
 	return powerPath
 }
 
+func (ACPI) GetName() string {
+	return "acpi"
+}
+
 func (a *ACPI) StopPower() {
 }
 

pkg/sensors/platform/source/hmc.go

@@ -18,6 +18,10 @@ package source
 
 type PowerHMC struct{}
 
+func (a *PowerHMC) GetName() string {
+	return "hmc"
+}
+
 func (a *PowerHMC) StopPower() {
 }
 

pkg/sensors/platform/source/redfish.go

@@ -203,6 +203,10 @@ func NewRedfishClient() *RedFishClient {
 	return nil
 }
 
+func (*RedFishClient) GetName() string {
+	return "redfish"
+}
+
 func (rf *RedFishClient) IsSystemCollectionSupported() bool {
 	// goroutine for collecting power info from Redfish already exists
 	if rf.ticker != nil {