ClusterCockpit/cc-backend
1
0
Fork 0
mirror of https://github.com/ClusterCockpit/cc-backend synced 2026-01-31 16:21:46 +01:00
Code Issues Packages Releases Activity

Streamline missing data warnings, review logging

Browse Source
This commit is contained in:
Christoph Kluge
2026-01-29 15:17:33 +01:00
parent 7101d2bb3b
commit f26cabbdf1
11 changed files with 147 additions and 88 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
internal/graph/util.go
View File

@@ -57,13 +57,13 @@ func (r *queryResolver) rooflineHeatmap(
jobdata, err := metricdispatch.LoadData(job, []string{"flops_any", "mem_bw"}, []schema.MetricScope{schema.MetricScopeNode}, ctx, 0)
if err != nil {
cclog.Errorf("Error while loading roofline metrics for job %d", job.ID)
cclog.Warnf("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 {
cclog.Infof("rooflineHeatmap(): 'flops_any' or 'mem_bw' missing for job %d", job.ID)
cclog.Warnf("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)
}

52
internal/metricdispatch/dataLoader.go
View File

@@ -97,8 +97,8 @@ func LoadData(job *schema.Job,
ms, err := GetMetricDataRepo(job.Cluster, job.SubCluster)
if err != nil {
cclog.Errorf("failed to load job data from metric store for job %d (user: %s, project: %s): %s",
job.JobID, job.User, job.Project, err.Error())
cclog.Errorf("failed to access metricDataRepo for cluster %s-%s: %s",
job.Cluster, job.SubCluster, err.Error())
return err, 0, 0
}
@@ -116,11 +116,11 @@ func LoadData(job *schema.Job,
jd, err = ms.LoadData(job, metrics, scopes, ctx, resolution)
if err != nil {
if len(jd) != 0 {
cclog.Warnf("partial error loading metrics from store for job %d (user: %s, project: %s): %s",
job.JobID, job.User, job.Project, err.Error())
cclog.Warnf("partial error loading metrics from store for job %d (user: %s, project: %s, cluster: %s-%s): %s",
job.JobID, job.User, job.Project, job.Cluster, job.SubCluster, err.Error())
} else {
cclog.Errorf("failed to load job data from metric store for job %d (user: %s, project: %s): %s",
job.JobID, job.User, job.Project, err.Error())
cclog.Warnf("failed to load job data from metric store for job %d (user: %s, project: %s, cluster: %s-%s): %s",
job.JobID, job.User, job.Project, job.Cluster, job.SubCluster, err.Error())
return err, 0, 0
}
}
@@ -129,8 +129,8 @@ func LoadData(job *schema.Job,
var jdTemp schema.JobData
jdTemp, err = archive.GetHandle().LoadJobData(job)
if err != nil {
cclog.Errorf("failed to load job data from archive for job %d (user: %s, project: %s): %s",
job.JobID, job.User, job.Project, err.Error())
cclog.Warnf("failed to load job data from archive for job %d (user: %s, project: %s, cluster: %s-%s): %s",
job.JobID, job.User, job.Project, job.Cluster, job.SubCluster, err.Error())
return err, 0, 0
}
@@ -244,15 +244,15 @@ func LoadAverages(
ms, err := GetMetricDataRepo(job.Cluster, job.SubCluster)
if err != nil {
cclog.Errorf("failed to load job data from metric store for job %d (user: %s, project: %s): %s",
job.JobID, job.User, job.Project, err.Error())
cclog.Errorf("failed to access metricDataRepo for cluster %s-%s: %s",
job.Cluster, job.SubCluster, err.Error())
return err
}
stats, err := ms.LoadStats(job, metrics, ctx)
if err != nil {
cclog.Errorf("failed to load statistics from metric store for job %d (user: %s, project: %s): %s",
job.JobID, job.User, job.Project, err.Error())
cclog.Warnf("failed to load statistics from metric store for job %d (user: %s, project: %s, cluster: %s-%s): %s",
job.JobID, job.User, job.Project, job.Cluster, job.SubCluster, err.Error())
return err
}
@@ -288,15 +288,15 @@ func LoadScopedJobStats(
ms, err := GetMetricDataRepo(job.Cluster, job.SubCluster)
if err != nil {
cclog.Errorf("failed to load job data from metric store for job %d (user: %s, project: %s): %s",
job.JobID, job.User, job.Project, err.Error())
cclog.Errorf("failed to access metricDataRepo for cluster %s-%s: %s",
job.Cluster, job.SubCluster, err.Error())
return nil, err
}
scopedStats, err := ms.LoadScopedStats(job, metrics, scopes, ctx)
if err != nil {
cclog.Errorf("failed to load scoped statistics from metric store for job %d (user: %s, project: %s): %s",
job.JobID, job.User, job.Project, err.Error())
cclog.Warnf("failed to load scoped statistics from metric store for job %d (user: %s, project: %s, cluster: %s-%s): %s",
job.JobID, job.User, job.Project, job.Cluster, job.SubCluster, err.Error())
return nil, err
}
@@ -320,8 +320,8 @@ func LoadJobStats(
ms, err := GetMetricDataRepo(job.Cluster, job.SubCluster)
if err != nil {
cclog.Errorf("failed to load job data from metric store for job %d (user: %s, project: %s): %s",
job.JobID, job.User, job.Project, err.Error())
cclog.Errorf("failed to access metricDataRepo for cluster %s-%s: %s",
job.Cluster, job.SubCluster, err.Error())
return nil, err
}
@@ -329,8 +329,8 @@ func LoadJobStats(
stats, err := ms.LoadStats(job, metrics, ctx)
if err != nil {
cclog.Errorf("failed to load statistics from metric store for job %d (user: %s, project: %s): %s",
job.JobID, job.User, job.Project, err.Error())
cclog.Warnf("failed to load statistics from metric store for job %d (user: %s, project: %s, cluster: %s-%s): %s",
job.JobID, job.User, job.Project, job.Cluster, job.SubCluster, err.Error())
return data, err
}
@@ -379,8 +379,8 @@ func LoadNodeData(
ms, err := GetMetricDataRepo(cluster, "")
if err != nil {
cclog.Errorf("failed to load node data from metric store: %s",
err.Error())
cclog.Errorf("failed to access metricDataRepo for cluster %s: %s",
cluster, err.Error())
return nil, err
}
@@ -389,7 +389,7 @@ func LoadNodeData(
if len(data) != 0 {
cclog.Warnf("partial error loading node data from metric store for cluster %s: %s", cluster, err.Error())
} else {
cclog.Errorf("failed to load node data from metric store for cluster %s: %s", cluster, err.Error())
cclog.Warnf("failed to load node data from metric store for cluster %s: %s", cluster, err.Error())
return nil, err
}
}
@@ -423,8 +423,8 @@ func LoadNodeListData(
ms, err := GetMetricDataRepo(cluster, subCluster)
if err != nil {
cclog.Errorf("failed to load node data from metric store: %s",
err.Error())
cclog.Errorf("failed to access metricDataRepo for cluster %s-%s: %s",
cluster, subCluster, err.Error())
return nil, err
}
@@ -434,7 +434,7 @@ func LoadNodeListData(
cclog.Warnf("partial error loading node list data from metric store for cluster %s, subcluster %s: %s",
cluster, subCluster, err.Error())
} else {
cclog.Errorf("failed to load node list data from metric store for cluster %s, subcluster %s: %s",
cclog.Warnf("failed to load node list data from metric store for cluster %s, subcluster %s: %s",
cluster, subCluster, err.Error())
return nil, err
}

4
internal/metricstoreclient/cc-metric-store.go
View File

@@ -329,7 +329,7 @@ func (ccms *CCMetricStore) LoadStats(
metric := query.Metric
data := res[0]
if data.Error != nil {
cclog.Errorf("fetching %s for node %s failed: %s", metric, query.Hostname, *data.Error)
cclog.Warnf("fetching %s for node %s failed: %s", metric, query.Hostname, *data.Error)
continue
}
@@ -556,7 +556,7 @@ func (ccms *CCMetricStore) LoadNodeListData(
) (map[string]schema.JobData, error) {
queries, assignedScope, err := ccms.buildNodeQueries(cluster, subCluster, nodes, metrics, scopes, resolution)
if err != nil {
cclog.Errorf("Error while building node queries for Cluster %s, SubCLuster %s, Metrics %v, Scopes %v: %s", cluster, subCluster, metrics, scopes, err.Error())
cclog.Errorf("Error while building node queries for Cluster %s, SubCluster %s, Metrics %v, Scopes %v: %s", cluster, subCluster, metrics, scopes, err.Error())
return nil, err
}

4
internal/taskmanager/updateFootprintService.go
View File

@@ -68,8 +68,8 @@ func RegisterFootprintWorker() {
ms, err := metricdispatch.GetMetricDataRepo(job.Cluster, job.SubCluster)
if err != nil {
cclog.Errorf("failed to load job data from metric store for job %d (user: %s, project: %s): %s",
job.JobID, job.User, job.Project, err.Error())
cclog.Errorf("failed to access metricDataRepo for cluster %s-%s: %s",
job.Cluster, job.SubCluster, err.Error())
continue
}