From 2b56b40e6d2b69d49f666f0753e131f34a13aa83 Mon Sep 17 00:00:00 2001 From: Christoph Kluge Date: Thu, 6 Mar 2025 12:46:25 +0100 Subject: [PATCH] Review energyFootprint calculation, fix missing numNodes factor, add log --- internal/importer/handleImport.go | 24 ++++++++++++++++-------- internal/importer/initDB.go | 24 ++++++++++++++++-------- internal/repository/job.go | 24 +++++++++++++++--------- 3 files changed, 47 insertions(+), 25 deletions(-) diff --git a/internal/importer/handleImport.go b/internal/importer/handleImport.go index 01773a5..623291c 100644 --- a/internal/importer/handleImport.go +++ b/internal/importer/handleImport.go @@ -96,27 +96,35 @@ func HandleImportFlag(flag string) error { } job.EnergyFootprint = make(map[string]float64) - var totalEnergy float64 - var energy float64 + // Total Job Energy Outside Loop + totalEnergy := 0.0 for _, fp := range sc.EnergyFootprint { + // Always Init Metric Energy Inside Loop + metricEnergy := 0.0 if i, err := archive.MetricIndex(sc.MetricConfig, fp); err == nil { // Note: For DB data, calculate and save as kWh - // Energy: Power (in Watts) * Time (in Seconds) if sc.MetricConfig[i].Energy == "energy" { // this metric has energy as unit (Joules) + log.Warnf("Update EnergyFootprint for Job %d and Metric %s on cluster %s: Set to 'energy' in cluster.json: Not implemented, will return 0.0", job.JobID, job.Cluster, fp) + // FIXME: Needs sum as stats type } else if sc.MetricConfig[i].Energy == "power" { // this metric has power as unit (Watt) - // Unit: ( W * s ) / 3600 / 1000 = kWh ; Rounded to 2 nearest digits - energy = math.Round(((repository.LoadJobStat(&job, fp, "avg")*float64(job.Duration))/3600/1000)*100) / 100 + // Energy: Power (in Watts) * Time (in Seconds) + // Unit: (W * (s / 3600)) / 1000 = kWh + // Round 2 Digits: round(Energy * 100) / 100 + // Here: (All-Node Metric Average * Number of Nodes) * (Job Duration in Seconds / 3600) / 1000 + // Note: Shared Jobs handled correctly since "Node Average" is based on partial resources, while "numNodes" factor is 1 + rawEnergy := ((repository.LoadJobStat(&job, fp, "avg") * float64(job.NumNodes)) * (float64(job.Duration) / 3600.0)) / 1000.0 + metricEnergy = math.Round(rawEnergy*100.0) / 100.0 } } else { log.Warnf("Error while collecting energy metric %s for job, DB ID '%v', return '0.0'", fp, job.ID) } - job.EnergyFootprint[fp] = energy - totalEnergy += energy + job.EnergyFootprint[fp] = metricEnergy + totalEnergy += metricEnergy } - job.Energy = (math.Round(totalEnergy*100) / 100) + job.Energy = (math.Round(totalEnergy*100.0) / 100.0) if job.RawEnergyFootprint, err = json.Marshal(job.EnergyFootprint); err != nil { log.Warnf("Error while marshaling energy footprint for job INTO BYTES, DB ID '%v'", job.ID) return err diff --git a/internal/importer/initDB.go b/internal/importer/initDB.go index fa2ee6e..9a2ccdf 100644 --- a/internal/importer/initDB.go +++ b/internal/importer/initDB.go @@ -93,27 +93,35 @@ func InitDB() error { } job.EnergyFootprint = make(map[string]float64) - var totalEnergy float64 - var energy float64 + // Total Job Energy Outside Loop + totalEnergy := 0.0 for _, fp := range sc.EnergyFootprint { + // Always Init Metric Energy Inside Loop + metricEnergy := 0.0 if i, err := archive.MetricIndex(sc.MetricConfig, fp); err == nil { // Note: For DB data, calculate and save as kWh - // Energy: Power (in Watts) * Time (in Seconds) if sc.MetricConfig[i].Energy == "energy" { // this metric has energy as unit (Joules) + log.Warnf("Update EnergyFootprint for Job %d and Metric %s on cluster %s: Set to 'energy' in cluster.json: Not implemented, will return 0.0", jobMeta.JobID, jobMeta.Cluster, fp) + // FIXME: Needs sum as stats type } else if sc.MetricConfig[i].Energy == "power" { // this metric has power as unit (Watt) - // Unit: ( W * s ) / 3600 / 1000 = kWh ; Rounded to 2 nearest digits - energy = math.Round(((repository.LoadJobStat(jobMeta, fp, "avg")*float64(jobMeta.Duration))/3600/1000)*100) / 100 + // Energy: Power (in Watts) * Time (in Seconds) + // Unit: (W * (s / 3600)) / 1000 = kWh + // Round 2 Digits: round(Energy * 100) / 100 + // Here: (All-Node Metric Average * Number of Nodes) * (Job Duration in Seconds / 3600) / 1000 + // Note: Shared Jobs handled correctly since "Node Average" is based on partial resources, while "numNodes" factor is 1 + rawEnergy := ((repository.LoadJobStat(jobMeta, fp, "avg") * float64(jobMeta.NumNodes)) * (float64(jobMeta.Duration) / 3600.0)) / 1000.0 + metricEnergy = math.Round(rawEnergy*100.0) / 100.0 } } else { log.Warnf("Error while collecting energy metric %s for job, DB ID '%v', return '0.0'", fp, jobMeta.ID) } - job.EnergyFootprint[fp] = energy - totalEnergy += energy + job.EnergyFootprint[fp] = metricEnergy + totalEnergy += metricEnergy } - job.Energy = (math.Round(totalEnergy*100) / 100) + job.Energy = (math.Round(totalEnergy*100.0) / 100.0) if job.RawEnergyFootprint, err = json.Marshal(job.EnergyFootprint); err != nil { log.Warnf("Error while marshaling energy footprint for job INTO BYTES, DB ID '%v'", jobMeta.ID) return err diff --git a/internal/repository/job.go b/internal/repository/job.go index 020c3c2..84de6f7 100644 --- a/internal/repository/job.go +++ b/internal/repository/job.go @@ -590,28 +590,34 @@ func (r *JobRepository) UpdateEnergy( return stmt, err } energyFootprint := make(map[string]float64) - var totalEnergy float64 - var energy float64 + // Total Job Energy Outside Loop + totalEnergy := 0.0 for _, fp := range sc.EnergyFootprint { + // Always Init Metric Energy Inside Loop + metricEnergy := 0.0 if i, err := archive.MetricIndex(sc.MetricConfig, fp); err == nil { // Note: For DB data, calculate and save as kWh if sc.MetricConfig[i].Energy == "energy" { // this metric has energy as unit (Joules or Wh) + log.Warnf("Update EnergyFootprint for Job %d and Metric %s on cluster %s: Set to 'energy' in cluster.json: Not implemented, will return 0.0", jobMeta.JobID, jobMeta.Cluster, fp) // FIXME: Needs sum as stats type } else if sc.MetricConfig[i].Energy == "power" { // this metric has power as unit (Watt) // Energy: Power (in Watts) * Time (in Seconds) - // Unit: (( W * s ) / 3600) / 1000 = kWh ; Rounded to 2 nearest digits: (Energy * 100) / 100 - // Here: All-Node Metric Average * Number of Nodes * Job Runtime + // Unit: (W * (s / 3600)) / 1000 = kWh + // Round 2 Digits: round(Energy * 100) / 100 + // Here: (All-Node Metric Average * Number of Nodes) * (Job Duration in Seconds / 3600) / 1000 // Note: Shared Jobs handled correctly since "Node Average" is based on partial resources, while "numNodes" factor is 1 - metricNodeSum := LoadJobStat(jobMeta, fp, "avg") * float64(jobMeta.NumNodes) * float64(jobMeta.Duration) - energy = math.Round(((metricNodeSum/3600)/1000)*100) / 100 + rawEnergy := ((LoadJobStat(jobMeta, fp, "avg") * float64(jobMeta.NumNodes)) * (float64(jobMeta.Duration) / 3600.0)) / 1000.0 + metricEnergy = math.Round(rawEnergy*100.0) / 100.0 } } else { log.Warnf("Error while collecting energy metric %s for job, DB ID '%v', return '0.0'", fp, jobMeta.ID) } - energyFootprint[fp] = energy - totalEnergy += energy + energyFootprint[fp] = metricEnergy + totalEnergy += metricEnergy + + // log.Infof("Metric %s Average %f -> %f kWh | Job %d Total -> %f kWh", fp, LoadJobStat(jobMeta, fp, "avg"), energy, jobMeta.JobID, totalEnergy) } var rawFootprint []byte @@ -620,7 +626,7 @@ func (r *JobRepository) UpdateEnergy( return stmt, err } - return stmt.Set("energy_footprint", string(rawFootprint)).Set("energy", (math.Round(totalEnergy*100) / 100)), nil + return stmt.Set("energy_footprint", string(rawFootprint)).Set("energy", (math.Round(totalEnergy*100.0) / 100.0)), nil } func (r *JobRepository) UpdateFootprint(