Add a Sample Rate config (#942)

* Add a sample rate config Signed-off-by: Michael Mercier <[email protected]> * Fix: use static PATH in the container build Signed-off-by: Michael Mercier <[email protected]> * Fix comments Signed-off-by: Michael Mercier <[email protected]> * Fix lint error and cast sampleRate to uint64 Signed-off-by: Michael Mercier <[email protected]> --------- Signed-off-by: Michael Mercier <[email protected]>
sustainable-computing-io · Sep 21, 2023 · 8db29a4 · 8db29a4
1 parent 5272ee2
commit 8db29a4
Show file tree

Hide file tree

Showing 6 changed files with 12 additions and 10 deletions.
diff --git a/Makefile b/Makefile
@@ -201,7 +201,7 @@ container_build:
 	$(CTR_CMD) run --rm \
 		--network host \
 		-v $(base_dir):/kepler:Z -w /kepler \
-		-e GOROOT=/usr/local/go -e PATH=$(PATH):/usr/local/go/bin \
+		-e GOROOT=/usr/local/go -e PATH=/usr/bin:/bin:/sbin:/usr/local/bin:/usr/local/go/bin \
 		$(BUILDER_IMAGE) \
 		git config --global --add safe.directory /kepler && make build
 

diff --git a/pkg/config/config.go b/pkg/config/config.go
@@ -55,8 +55,8 @@ const (
 	// MaxIRQ is the maximum number of IRQs to be monitored
 	MaxIRQ = 10
 
-	// SamplePeriodSec is the time in seconds that the reader will wait before reading the metrics again
-	SamplePeriodSec = 3
+	// defaultSamplePeriodSec is the time in seconds that the reader will wait before reading the metrics again
+	defaultSamplePeriodSec = 3
 )
 
 var (
@@ -93,6 +93,8 @@ var (
 	GpuUsageMetric        = getConfig("GPU_USAGE_METRIC", GPUSMUtilization)       // no metric (evenly divided)
 	GeneralUsageMetric    = getConfig("GENERAL_USAGE_METRIC", defaultMetricValue) // for uncategorized energy
 
+	SamplePeriodSec = uint64(getIntConfig("SAMPLE_PERIOD_SEC", defaultSamplePeriodSec))
+
 	versionRegex = regexp.MustCompile(`^(\d+)\.(\d+).`)
 
 	configPath = "/etc/kepler/kepler.config"

diff --git a/pkg/manager/manager.go b/pkg/manager/manager.go
@@ -24,8 +24,8 @@ import (
 	"github.com/sustainable-computing-io/kepler/pkg/kubernetes"
 )
 
-const (
-	samplePeriod = config.SamplePeriodSec * 1000 * time.Millisecond
+var (
+	samplePeriod = time.Duration(config.SamplePeriodSec * 1000 * uint64(time.Millisecond))
 )
 
 type CollectorManager struct {
@@ -48,7 +48,7 @@ func New() *CollectorManager {
 	manager.PrometheusCollector.ContainersMetrics = &manager.MetricCollector.ContainersMetrics
 	manager.PrometheusCollector.ProcessMetrics = &manager.MetricCollector.ProcessMetrics
 	manager.PrometheusCollector.VMMetrics = &manager.MetricCollector.VMMetrics
-	manager.PrometheusCollector.SamplePeriodSec = config.SamplePeriodSec
+	manager.PrometheusCollector.SamplePeriodSec = float64(config.SamplePeriodSec)
 	// configure the wather
 	manager.Watcher = kubernetes.NewObjListWatcher()
 	manager.Watcher.Mx = &manager.PrometheusCollector.Mx

diff --git a/pkg/model/node_platform_energy.go b/pkg/model/node_platform_energy.go
@@ -94,13 +94,13 @@ func UpdateNodePlatformEnergy(nodeMetrics *collector_metric.NodeMetrics) {
 	platformPower := GetNodePlatformPower(nodeMetrics, absPower)
 	for id, power := range platformPower {
 		// convert power to energy
-		platformPower[id] = power * config.SamplePeriodSec
+		platformPower[id] = power * float64(config.SamplePeriodSec)
 	}
 	nodeMetrics.SetNodePlatformEnergy(platformPower, gauge, absPower)
 	platformPower = GetNodePlatformPower(nodeMetrics, idlePower)
 	for id, power := range platformPower {
 		// convert power to energy
-		platformPower[id] = power * config.SamplePeriodSec
+		platformPower[id] = power * float64(config.SamplePeriodSec)
 	}
 	nodeMetrics.SetNodePlatformEnergy(platformPower, gauge, idlePower)
 }
diff --git a/pkg/power/accelerator/gpu/source/gpu_nvml.go b/pkg/power/accelerator/gpu/source/gpu_nvml.go
@@ -98,7 +98,7 @@ func (n *GPUNvml) GetAbsEnergyFromGPU() []uint32 {
 		}
 		// since Kepler collects metrics at intervals of SamplePeriodSec, which is greater than 1 second, it is
 		// necessary to calculate the energy consumption for the entire waiting period
-		energy := uint32(float64(power) * config.SamplePeriodSec)
+		energy := uint32(uint64(power) * config.SamplePeriodSec)
 		gpuEnergy = append(gpuEnergy, energy)
 	}
 	return gpuEnergy

diff --git a/pkg/power/platform/source/acpi.go b/pkg/power/platform/source/acpi.go
@@ -163,7 +163,7 @@ func (a *ACPI) GetAbsEnergyFromPlatform() (map[string]float64, error) {
 		if err == nil {
 			// since Kepler collects metrics at intervals of SamplePeriodSec, which is greater than 1 second, it is
 			// necessary to calculate the energy consumption for the entire waiting period
-			power[sensorIDPrefix+strconv.Itoa(int(i))] = float64(currPower) / 1000 * config.SamplePeriodSec /*miliJoules*/
+			power[sensorIDPrefix+strconv.Itoa(int(i))] = float64(currPower / 1000 * config.SamplePeriodSec) /*miliJoules*/
 		} else {
 			return power, err
 		}