Merge branch 'hotfix' of github.com:ClusterCockpit/cc-backend into hotfix

This commit is contained in:
Jan Eitzinger 2023-08-31 16:17:17 +02:00
commit 47a82bf843
3 changed files with 22 additions and 30 deletions

View File

@ -6,7 +6,6 @@ package graph
import ( import (
"context" "context"
"errors"
"fmt" "fmt"
"math" "math"
@ -33,7 +32,7 @@ func (r *queryResolver) rooflineHeatmap(
return nil, err return nil, err
} }
if len(jobs) > MAX_JOBS_FOR_ANALYSIS { if len(jobs) > MAX_JOBS_FOR_ANALYSIS {
return nil, fmt.Errorf("GRAPH/STATS > too many jobs matched (max: %d)", MAX_JOBS_FOR_ANALYSIS) return nil, fmt.Errorf("GRAPH/UTIL > too many jobs matched (max: %d)", MAX_JOBS_FOR_ANALYSIS)
} }
fcols, frows := float64(cols), float64(rows) fcols, frows := float64(cols), float64(rows)
@ -50,20 +49,24 @@ func (r *queryResolver) rooflineHeatmap(
jobdata, err := metricdata.LoadData(job, []string{"flops_any", "mem_bw"}, []schema.MetricScope{schema.MetricScopeNode}, ctx) jobdata, err := metricdata.LoadData(job, []string{"flops_any", "mem_bw"}, []schema.MetricScope{schema.MetricScopeNode}, ctx)
if err != nil { if err != nil {
log.Error("Error while loading metrics for roofline") log.Errorf("Error while loading roofline metrics for job %d", job.ID)
return nil, err return nil, err
} }
flops_, membw_ := jobdata["flops_any"], jobdata["mem_bw"] flops_, membw_ := jobdata["flops_any"], jobdata["mem_bw"]
if flops_ == nil && membw_ == nil { if flops_ == nil && membw_ == nil {
return nil, fmt.Errorf("GRAPH/STATS > 'flops_any' or 'mem_bw' missing for job %d", job.ID) log.Infof("rooflineHeatmap(): 'flops_any' or 'mem_bw' missing for job %d", job.ID)
continue
// return nil, fmt.Errorf("GRAPH/UTIL > 'flops_any' or 'mem_bw' missing for job %d", job.ID)
} }
flops, ok1 := flops_["node"] flops, ok1 := flops_["node"]
membw, ok2 := membw_["node"] membw, ok2 := membw_["node"]
if !ok1 || !ok2 { if !ok1 || !ok2 {
log.Info("rooflineHeatmap() query not implemented for where flops_any or mem_bw not available at 'node' level")
continue
// TODO/FIXME: // TODO/FIXME:
return nil, errors.New("GRAPH/STATS > todo: rooflineHeatmap() query not implemented for where flops_any or mem_bw not available at 'node' level") // return nil, errors.New("GRAPH/UTIL > todo: rooflineHeatmap() query not implemented for where flops_any or mem_bw not available at 'node' level")
} }
for n := 0; n < len(flops.Series); n++ { for n := 0; n < len(flops.Series); n++ {
@ -99,7 +102,7 @@ func (r *queryResolver) jobsFootprints(ctx context.Context, filter []*model.JobF
return nil, err return nil, err
} }
if len(jobs) > MAX_JOBS_FOR_ANALYSIS { if len(jobs) > MAX_JOBS_FOR_ANALYSIS {
return nil, fmt.Errorf("GRAPH/STATS > too many jobs matched (max: %d)", MAX_JOBS_FOR_ANALYSIS) return nil, fmt.Errorf("GRAPH/UTIL > too many jobs matched (max: %d)", MAX_JOBS_FOR_ANALYSIS)
} }
avgs := make([][]schema.Float, len(metrics)) avgs := make([][]schema.Float, len(metrics))

View File

@ -533,7 +533,9 @@ func (ccms *CCMetricStore) LoadStats(
metric := ccms.toLocalName(query.Metric) metric := ccms.toLocalName(query.Metric)
data := res[0] data := res[0]
if data.Error != nil { if data.Error != nil {
return nil, fmt.Errorf("METRICDATA/CCMS > fetching %s for node %s failed: %s", metric, query.Hostname, *data.Error) log.Infof("fetching %s for node %s failed: %s", metric, query.Hostname, *data.Error)
continue
// return nil, fmt.Errorf("METRICDATA/CCMS > fetching %s for node %s failed: %s", metric, query.Hostname, *data.Error)
} }
metricdata, ok := stats[metric] metricdata, ok := stats[metric]
@ -543,7 +545,9 @@ func (ccms *CCMetricStore) LoadStats(
} }
if data.Avg.IsNaN() || data.Min.IsNaN() || data.Max.IsNaN() { if data.Avg.IsNaN() || data.Min.IsNaN() || data.Max.IsNaN() {
return nil, fmt.Errorf("METRICDATA/CCMS > fetching %s for node %s failed: %s", metric, query.Hostname, "avg/min/max is NaN") log.Infof("fetching %s for node %s failed: one of avg/min/max is NaN", metric, query.Hostname)
continue
// return nil, fmt.Errorf("METRICDATA/CCMS > fetching %s for node %s failed: %s", metric, query.Hostname, "avg/min/max is NaN")
} }
metricdata[query.Hostname] = schema.MetricStatistics{ metricdata[query.Hostname] = schema.MetricStatistics{

View File

@ -350,31 +350,16 @@ export function binsFromFootprint(weights, scope, values, numBins) {
scopeWeights = weights.nodeHours scopeWeights = weights.nodeHours
} }
const bins = new Array(numBins).fill(0) const rawBins = new Array(numBins).fill(0)
for (let i = 0; i < values.length; i++) for (let i = 0; i < values.length; i++)
bins[Math.floor(((values[i] - min) / (max - min)) * numBins)] += scopeWeights ? scopeWeights[i] : 1 rawBins[Math.floor(((values[i] - min) / (max - min)) * numBins)] += scopeWeights ? scopeWeights[i] : 1
// Manual Canvas Original const bins = rawBins.map((count, idx) => ({
// return { value: Math.floor(min + ((idx + 1) / numBins) * (max - min)),
// label: idx => { count: count
// let start = min + (idx / numBins) * (max - min) }))
// let stop = min + ((idx + 1) / numBins) * (max - min)
// return `${formatNumber(start)} - ${formatNumber(stop)}`
// },
// bins: bins.map((count, idx) => ({ value: idx, count: count })),
// min: min,
// max: max
// }
return { return {
bins: bins.map((count, idx) => ({ bins: bins
value: idx => { // Use bins' max value instead of mean
// let start = min + (idx / numBins) * (max - min)
let stop = min + ((idx + 1) / numBins) * (max - min)
// return `${formatNumber(Math.floor((start+stop)/2))}`
return Math.floor(stop)
},
count: count
}))
} }
} }