ClusterCockpit/cc-backend
1
0
Fork 0
mirror of https://github.com/ClusterCockpit/cc-backend synced 2025-03-14 18:05:56 +01:00
Code Issues Packages Releases Activity

Review energyFootprint calculation, fix missing numNodes factor, add log

Browse Source
This commit is contained in:
Christoph Kluge 2025-03-06 12:46:25 +01:00
parent 4b2d7068b3
commit 2b56b40e6d
3 changed files with 47 additions and 25 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

24
internal/importer/handleImport.go
View File

@ -96,27 +96,35 @@ func HandleImportFlag(flag string) error {
} }
job.EnergyFootprint = make(map[string]float64) 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 { for _, fp := range sc.EnergyFootprint {
// Always Init Metric Energy Inside Loop
metricEnergy := 0.0
if i, err := archive.MetricIndex(sc.MetricConfig, fp); err == nil { if i, err := archive.MetricIndex(sc.MetricConfig, fp); err == nil {
// Note: For DB data, calculate and save as kWh // 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) 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) } 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: Power (in Watts) * Time (in Seconds)
energy = math.Round(((repository.LoadJobStat(&job, fp, "avg")*float64(job.Duration))/3600/1000)*100) / 100 // 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 { } else {
log.Warnf("Error while collecting energy metric %s for job, DB ID '%v', return '0.0'", fp, job.ID) log.Warnf("Error while collecting energy metric %s for job, DB ID '%v', return '0.0'", fp, job.ID)
} }
job.EnergyFootprint[fp] = energy job.EnergyFootprint[fp] = metricEnergy
totalEnergy += energy 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 { 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) log.Warnf("Error while marshaling energy footprint for job INTO BYTES, DB ID '%v'", job.ID)
return err return err

24
internal/importer/initDB.go
View File

@ -93,27 +93,35 @@ func InitDB() error {
} }
job.EnergyFootprint = make(map[string]float64) 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 { for _, fp := range sc.EnergyFootprint {
// Always Init Metric Energy Inside Loop
metricEnergy := 0.0
if i, err := archive.MetricIndex(sc.MetricConfig, fp); err == nil { if i, err := archive.MetricIndex(sc.MetricConfig, fp); err == nil {
// Note: For DB data, calculate and save as kWh // 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) 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) } 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: Power (in Watts) * Time (in Seconds)
energy = math.Round(((repository.LoadJobStat(jobMeta, fp, "avg")*float64(jobMeta.Duration))/3600/1000)*100) / 100 // 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 { } else {
log.Warnf("Error while collecting energy metric %s for job, DB ID '%v', return '0.0'", fp, jobMeta.ID) log.Warnf("Error while collecting energy metric %s for job, DB ID '%v', return '0.0'", fp, jobMeta.ID)
} }
job.EnergyFootprint[fp] = energy job.EnergyFootprint[fp] = metricEnergy
totalEnergy += energy 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 { 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) log.Warnf("Error while marshaling energy footprint for job INTO BYTES, DB ID '%v'", jobMeta.ID)
return err return err

24
internal/repository/job.go
View File

@ -590,28 +590,34 @@ func (r *JobRepository) UpdateEnergy(
return stmt, err return stmt, err
} }
energyFootprint := make(map[string]float64) energyFootprint := make(map[string]float64)
var totalEnergy float64
var energy float64
// Total Job Energy Outside Loop
totalEnergy := 0.0
for _, fp := range sc.EnergyFootprint { for _, fp := range sc.EnergyFootprint {
// Always Init Metric Energy Inside Loop
metricEnergy := 0.0
if i, err := archive.MetricIndex(sc.MetricConfig, fp); err == nil { if i, err := archive.MetricIndex(sc.MetricConfig, fp); err == nil {
// Note: For DB data, calculate and save as kWh // Note: For DB data, calculate and save as kWh
if sc.MetricConfig[i].Energy == "energy" { // this metric has energy as unit (Joules or Wh) 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 // FIXME: Needs sum as stats type
} else if sc.MetricConfig[i].Energy == "power" { // this metric has power as unit (Watt) } else if sc.MetricConfig[i].Energy == "power" { // this metric has power as unit (Watt)
// Energy: Power (in Watts) * Time (in Seconds) // Energy: Power (in Watts) * Time (in Seconds)
// Unit: (( W * s ) / 3600) / 1000 = kWh ; Rounded to 2 nearest digits: (Energy * 100) / 100 // Unit: (W * (s / 3600)) / 1000 = kWh
// Here: All-Node Metric Average * Number of Nodes * Job Runtime // 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 // 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) rawEnergy := ((LoadJobStat(jobMeta, fp, "avg") * float64(jobMeta.NumNodes)) * (float64(jobMeta.Duration) / 3600.0)) / 1000.0
energy = math.Round(((metricNodeSum/3600)/1000)*100) / 100 metricEnergy = math.Round(rawEnergy*100.0) / 100.0
} }
} else { } else {
log.Warnf("Error while collecting energy metric %s for job, DB ID '%v', return '0.0'", fp, jobMeta.ID) log.Warnf("Error while collecting energy metric %s for job, DB ID '%v', return '0.0'", fp, jobMeta.ID)
} }
energyFootprint[fp] = energy energyFootprint[fp] = metricEnergy
totalEnergy += energy 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 var rawFootprint []byte
@ -620,7 +626,7 @@ func (r *JobRepository) UpdateEnergy(
return stmt, err 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( func (r *JobRepository) UpdateFootprint(