From b52330ebf038ca7b1a910160b05b8d77e9c5af34 Mon Sep 17 00:00:00 2001
From: Christoph Kluge <christoph.kluge@fau.de>
Date: Thu, 31 Aug 2023 14:48:19 +0200
Subject: [PATCH 1/4] simplify bins return
---
web/frontend/src/utils.js | 27 ++++++---------------------
1 file changed, 6 insertions(+), 21 deletions(-)
diff --git a/web/frontend/src/utils.js b/web/frontend/src/utils.js
index 5e9cdaef..42942fd0 100644
--- a/web/frontend/src/utils.js
+++ b/web/frontend/src/utils.js
@@ -350,31 +350,16 @@ export function binsFromFootprint(weights, scope, values, numBins) {
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++)
bins[Math.floor(((values[i] - min) / (max - min)) * numBins)] += scopeWeights ? scopeWeights[i] : 1
- // Manual Canvas Original
- // return {
- // label: idx => {
- // let start = min + (idx / numBins) * (max - min)
- // let stop = min + ((idx + 1) / numBins) * (max - min)
- // return `${formatNumber(start)} - ${formatNumber(stop)}`
- // },
- // bins: bins.map((count, idx) => ({ value: idx, count: count })),
- // min: min,
- // max: max
- // }
+ const bins = rawBins.map((count, idx) => ({
+ value: Math.floor(min + ((idx + 1) / numBins) * (max - min)),
+ count: count
+ }))
return {
- bins: bins.map((count, idx) => ({
- value: idx => { // Use bins' max value instead of mean
- // let start = min + (idx / numBins) * (max - min)
- let stop = min + ((idx + 1) / numBins) * (max - min)
- // return `${formatNumber(Math.floor((start+stop)/2))}`
- return Math.floor(stop)
- },
- count: count
- }))
+ bins: bins
}
}
From 64796519c6eeae8c5aa982857ad2b954ba3b2fa8 Mon Sep 17 00:00:00 2001
From: Christoph Kluge <christoph.kluge@fau.de>
Date: Thu, 31 Aug 2023 15:10:57 +0200
Subject: [PATCH 2/4] change: use continue for rooflineHeatmap errors
- hard errors blocked rendering in frontend
---
internal/graph/util.go | 15 +++++++++------
1 file changed, 9 insertions(+), 6 deletions(-)
diff --git a/internal/graph/util.go b/internal/graph/util.go
index b61bcc73..3a2c3b16 100644
--- a/internal/graph/util.go
+++ b/internal/graph/util.go
@@ -6,7 +6,6 @@ package graph
import (
"context"
- "errors"
"fmt"
"math"
@@ -33,7 +32,7 @@ func (r *queryResolver) rooflineHeatmap(
return nil, err
}
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)
@@ -50,20 +49,24 @@ func (r *queryResolver) rooflineHeatmap(
jobdata, err := metricdata.LoadData(job, []string{"flops_any", "mem_bw"}, []schema.MetricScope{schema.MetricScopeNode}, ctx)
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
}
flops_, membw_ := jobdata["flops_any"], jobdata["mem_bw"]
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"]
membw, ok2 := membw_["node"]
if !ok1 || !ok2 {
+ log.Info("rooflineHeatmap() query not implemented for where flops_any or mem_bw not available at 'node' level")
+ continue
// 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++ {
@@ -99,7 +102,7 @@ func (r *queryResolver) jobsFootprints(ctx context.Context, filter []*model.JobF
return nil, err
}
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))
From cffdd055c9c2b2c28871ccc6078c38c6af82d977 Mon Sep 17 00:00:00 2001
From: Christoph Kluge <christoph.kluge@fau.de>
Date: Thu, 31 Aug 2023 15:17:40 +0200
Subject: [PATCH 3/4] change: use continue for ccms.loadStats errors
---
internal/metricdata/cc-metric-store.go | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)
diff --git a/internal/metricdata/cc-metric-store.go b/internal/metricdata/cc-metric-store.go
index cfaa6fd1..4874975c 100644
--- a/internal/metricdata/cc-metric-store.go
+++ b/internal/metricdata/cc-metric-store.go
@@ -533,7 +533,9 @@ func (ccms *CCMetricStore) LoadStats(
metric := ccms.toLocalName(query.Metric)
data := res[0]
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]
@@ -543,7 +545,9 @@ func (ccms *CCMetricStore) LoadStats(
}
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{
From a2a4b2e6c2a0631d95adb0270e63dbde1a1002f7 Mon Sep 17 00:00:00 2001
From: Christoph Kluge <christoph.kluge@fau.de>
Date: Thu, 31 Aug 2023 15:19:54 +0200
Subject: [PATCH 4/4] fix variable
---
web/frontend/src/utils.js | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/web/frontend/src/utils.js b/web/frontend/src/utils.js
index 42942fd0..06509167 100644
--- a/web/frontend/src/utils.js
+++ b/web/frontend/src/utils.js
@@ -352,7 +352,7 @@ export function binsFromFootprint(weights, scope, values, numBins) {
const rawBins = new Array(numBins).fill(0)
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
const bins = rawBins.map((count, idx) => ({
value: Math.floor(min + ((idx + 1) / numBins) * (max - min)),