Merge branch 'feat/565-add-metric-tooltip' into metric-store-tickets

Resolve conflicts in the generated GraphQL code by regenerating it against
the merged schema. cc-lib v2.13.0 adds Tooltip to schema.MetricConfig and
schema.GlobalMetricListItem, so gqlgen now binds the tooltip field directly
and the hand-written globalMetricListItem/metricConfig resolvers introduced
on the tooltip branch are no longer needed.

Also migrate to the cc-lib v2.13.0 metric container types, which changed
from bare maps to structs carrying array-valued metric groups:

  schema.JobData          map -> {Metrics, Groups}
  schema.ScopedJobStats   map -> {Metrics, Groups}
  job.Statistics          map -> schema.JobStatisticsSet{Metrics, Groups}

Callers index .Metrics, return the zero struct instead of nil, and
deepCopy/DecodeJobStats now also carry the Groups payload through.
archive.GetStatistics returns the full JobStatisticsSet so group
statistics survive the round trip.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
2026-08-27 10:22:22 +02:00
co-authored by Claude Opus 5
35 changed files with 411 additions and 212 deletions
+6 -6
View File
@@ -196,8 +196,8 @@ func cleanup() {
func TestRestApi(t *testing.T) {
restapi := setup(t)
t.Cleanup(cleanup)
testData := schema.JobData{
"load_one": map[schema.MetricScope]*schema.JobMetric{
testData := schema.JobData{Metrics: map[string]schema.ScopedMetrics{
"load_one": {
schema.MetricScopeNode: {
Unit: schema.Unit{Base: "load"},
Timestep: 60,
@@ -210,7 +210,7 @@ func TestRestApi(t *testing.T) {
},
},
},
}
}}
metricstore.TestLoadDataCallback = func(job *schema.Job, metrics []string, scopes []schema.MetricScope, ctx context.Context, resolution int) (schema.JobData, error) {
return testData, nil
@@ -497,8 +497,8 @@ func TestStopJobWithReusedJobId(t *testing.T) {
restapi := setup(t)
t.Cleanup(cleanup)
testData := schema.JobData{
"load_one": map[schema.MetricScope]*schema.JobMetric{
testData := schema.JobData{Metrics: map[string]schema.ScopedMetrics{
"load_one": {
schema.MetricScopeNode: {
Unit: schema.Unit{Base: "load"},
Timestep: 60,
@@ -511,7 +511,7 @@ func TestStopJobWithReusedJobId(t *testing.T) {
},
},
},
}
}}
metricstore.TestLoadDataCallback = func(job *schema.Job, metrics []string, scopes []schema.MetricScope, ctx context.Context, resolution int) (schema.JobData, error) {
return testData, nil
+1 -1
View File
@@ -412,7 +412,7 @@ func (api *RestAPI) getJobByID(rw http.ResponseWriter, r *http.Request) {
}
res := []*JobMetricWithName{}
for name, md := range data {
for name, md := range data.Metrics {
for scope, metric := range md {
res = append(res, &JobMetricWithName{
Name: name,
+3 -3
View File
@@ -531,8 +531,8 @@ func TestNatsHandleStopJob(t *testing.T) {
},
}
testData := schema.JobData{
"load_one": map[schema.MetricScope]*schema.JobMetric{
testData := schema.JobData{Metrics: map[string]schema.ScopedMetrics{
"load_one": {
schema.MetricScopeNode: {
Unit: schema.Unit{Base: "load"},
Timestep: 60,
@@ -545,7 +545,7 @@ func TestNatsHandleStopJob(t *testing.T) {
},
},
},
}
}}
metricstore.TestLoadDataCallback = func(job *schema.Job, metrics []string, scopes []schema.MetricScope, ctx context.Context, resolution int) (schema.JobData, error) {
return testData, nil
+3 -3
View File
@@ -65,9 +65,9 @@ func ArchiveJob(job *schema.Job, ctx context.Context) (*schema.Job, error) {
return nil, err
}
job.Statistics = make(map[string]schema.JobStatistics)
job.Statistics = schema.JobStatisticsSet{Metrics: make(map[string]schema.JobStatistics, len(jobData.Metrics))}
for metric, data := range jobData {
for metric, data := range jobData.Metrics {
avg, min, max := 0.0, math.MaxFloat32, -math.MaxFloat32
nodeData, ok := data["node"]
if !ok {
@@ -82,7 +82,7 @@ func ArchiveJob(job *schema.Job, ctx context.Context) (*schema.Job, error) {
}
// Round AVG Result to 2 Digits
job.Statistics[metric] = schema.JobStatistics{
job.Statistics.Metrics[metric] = schema.JobStatistics{
Unit: schema.Unit{
Prefix: archive.GetMetricConfig(job.Cluster, metric).Unit.Prefix,
Base: archive.GetMetricConfig(job.Cluster, metric).Unit.Base,
+76
View File
@@ -100,6 +100,7 @@ type ComplexityRoot struct {
Footprint func(childComplexity int) int
Name func(childComplexity int) int
Scope func(childComplexity int) int
Tooltip func(childComplexity int) int
Unit func(childComplexity int) int
}
@@ -219,6 +220,7 @@ type ComplexityRoot struct {
Scope func(childComplexity int) int
SubClusters func(childComplexity int) int
Timestep func(childComplexity int) int
Tooltip func(childComplexity int) int
Unit func(childComplexity int) int
}
@@ -698,6 +700,12 @@ func (e *executableSchema) Complexity(ctx context.Context, typeName, field strin
}
return e.ComplexityRoot.GlobalMetricListItem.Scope(childComplexity), true
case "GlobalMetricListItem.tooltip":
if e.ComplexityRoot.GlobalMetricListItem.Tooltip == nil {
break
}
return e.ComplexityRoot.GlobalMetricListItem.Tooltip(childComplexity), true
case "GlobalMetricListItem.unit":
if e.ComplexityRoot.GlobalMetricListItem.Unit == nil {
break
@@ -1225,6 +1233,12 @@ func (e *executableSchema) Complexity(ctx context.Context, typeName, field strin
}
return e.ComplexityRoot.MetricConfig.Timestep(childComplexity), true
case "MetricConfig.tooltip":
if e.ComplexityRoot.MetricConfig.Tooltip == nil {
break
}
return e.ComplexityRoot.MetricConfig.Tooltip(childComplexity), true
case "MetricConfig.unit":
if e.ComplexityRoot.MetricConfig.Unit == nil {
break
@@ -2426,6 +2440,7 @@ type MetricConfig {
alert: Float!
lowerIsBetter: Boolean
subClusters: [SubClusterConfig!]!
tooltip: String
}
type Tag {
@@ -2589,6 +2604,7 @@ type GlobalMetricListItem {
unit: Unit!
scope: MetricScope!
footprint: String
tooltip: String
availability: [ClusterSupport!]!
}
@@ -2973,6 +2989,8 @@ func (ec *executionContext) childFields_GlobalMetricListItem(ctx context.Context
return ec.fieldContext_GlobalMetricListItem_scope(ctx, field)
case "footprint":
return ec.fieldContext_GlobalMetricListItem_footprint(ctx, field)
case "tooltip":
return ec.fieldContext_GlobalMetricListItem_tooltip(ctx, field)
case "availability":
return ec.fieldContext_GlobalMetricListItem_availability(ctx, field)
}
@@ -3203,6 +3221,8 @@ func (ec *executionContext) childFields_MetricConfig(ctx context.Context, field
return ec.fieldContext_MetricConfig_lowerIsBetter(ctx, field)
case "subClusters":
return ec.fieldContext_MetricConfig_subClusters(ctx, field)
case "tooltip":
return ec.fieldContext_MetricConfig_tooltip(ctx, field)
}
return nil, fmt.Errorf("no field named %q was found under type MetricConfig", field.Name)
}
@@ -5187,6 +5207,29 @@ func (ec *executionContext) fieldContext_GlobalMetricListItem_footprint(_ contex
return graphql.NewScalarFieldContext("GlobalMetricListItem", field, false, false, errors.New("field of type String does not have child fields"))
}
func (ec *executionContext) _GlobalMetricListItem_tooltip(ctx context.Context, field graphql.CollectedField, obj *schema.GlobalMetricListItem) (ret graphql.Marshaler) {
return graphql.ResolveField(
ctx,
ec.OperationContext,
field,
func(ctx context.Context, field graphql.CollectedField) (*graphql.FieldContext, error) {
return ec.fieldContext_GlobalMetricListItem_tooltip(ctx, field)
},
func(ctx context.Context) (any, error) {
return obj.Tooltip, nil
},
nil,
func(ctx context.Context, selections ast.SelectionSet, v string) graphql.Marshaler {
return ec.marshalOString2string(ctx, selections, v)
},
true,
false,
)
}
func (ec *executionContext) fieldContext_GlobalMetricListItem_tooltip(_ context.Context, field graphql.CollectedField) (fc *graphql.FieldContext, err error) {
return graphql.NewScalarFieldContext("GlobalMetricListItem", field, false, false, errors.New("field of type String does not have child fields"))
}
func (ec *executionContext) _GlobalMetricListItem_availability(ctx context.Context, field graphql.CollectedField, obj *schema.GlobalMetricListItem) (ret graphql.Marshaler) {
return graphql.ResolveField(
ctx,
@@ -7377,6 +7420,29 @@ func (ec *executionContext) fieldContext_MetricConfig_subClusters(_ context.Cont
return fc, nil
}
func (ec *executionContext) _MetricConfig_tooltip(ctx context.Context, field graphql.CollectedField, obj *schema.MetricConfig) (ret graphql.Marshaler) {
return graphql.ResolveField(
ctx,
ec.OperationContext,
field,
func(ctx context.Context, field graphql.CollectedField) (*graphql.FieldContext, error) {
return ec.fieldContext_MetricConfig_tooltip(ctx, field)
},
func(ctx context.Context) (any, error) {
return obj.Tooltip, nil
},
nil,
func(ctx context.Context, selections ast.SelectionSet, v string) graphql.Marshaler {
return ec.marshalOString2string(ctx, selections, v)
},
true,
false,
)
}
func (ec *executionContext) fieldContext_MetricConfig_tooltip(_ context.Context, field graphql.CollectedField) (fc *graphql.FieldContext, err error) {
return graphql.NewScalarFieldContext("MetricConfig", field, false, false, errors.New("field of type String does not have child fields"))
}
func (ec *executionContext) _MetricFootprints_metric(ctx context.Context, field graphql.CollectedField, obj *model.MetricFootprints) (ret graphql.Marshaler) {
return graphql.ResolveField(
ctx,
@@ -13311,6 +13377,11 @@ func (ec *executionContext) _GlobalMetricListItem(ctx context.Context, sel ast.S
if out.Values[i] == graphql.RequiredNull {
out.Invalids++
}
case "tooltip":
out.Values[i] = ec._GlobalMetricListItem_tooltip(ctx, field, obj)
if out.Values[i] == graphql.RequiredNull {
out.Invalids++
}
case "availability":
out.Values[i] = ec._GlobalMetricListItem_availability(ctx, field, obj)
if out.Values[i] == graphql.Null {
@@ -14340,6 +14411,11 @@ func (ec *executionContext) _MetricConfig(ctx context.Context, sel ast.Selection
if out.Values[i] == graphql.Null {
out.Invalids++
}
case "tooltip":
out.Values[i] = ec._MetricConfig_tooltip(ctx, field, obj)
if out.Values[i] == graphql.RequiredNull {
out.Invalids++
}
default:
panic("unknown field " + strconv.Quote(field.Name))
}
+3 -3
View File
@@ -528,7 +528,7 @@ func (r *queryResolver) JobMetrics(ctx context.Context, id string, metrics []str
}
res := []*model.JobMetricWithName{}
for name, md := range data {
for name, md := range data.Metrics {
for scope, metric := range md {
res = append(res, &model.JobMetricWithName{
Name: name,
@@ -581,7 +581,7 @@ func (r *queryResolver) ScopedJobStats(ctx context.Context, id string, metrics [
}
res := make([]*model.NamedStatsWithScope, 0)
for name, scoped := range data {
for name, scoped := range data.Metrics {
for scope, stats := range scoped {
mdlStats := make([]*model.ScopedStats, 0)
@@ -939,7 +939,7 @@ func (r *queryResolver) NodeMetricsList(ctx context.Context, cluster string, sub
cclog.Warnf("error in nodeMetrics resolver: %s", err)
}
for metric, scopedMetrics := range data[hostname] {
for metric, scopedMetrics := range data[hostname].Metrics {
for scope, scopedMetric := range scopedMetrics {
host.Metrics = append(host.Metrics, &model.JobMetricWithName{
Name: metric,
+1 -1
View File
@@ -61,7 +61,7 @@ func (r *queryResolver) rooflineHeatmap(
return nil, err
}
flops_, membw_ := jobdata["flops_any"], jobdata["mem_bw"]
flops_, membw_ := jobdata.Metrics["flops_any"], jobdata.Metrics["mem_bw"]
if flops_ == nil && membw_ == nil {
cclog.Warnf("rooflineHeatmap(): 'flops_any' or 'mem_bw' missing for job %d", *job.ID)
continue
+1 -1
View File
@@ -290,7 +290,7 @@ func SanityChecks(job *schema.Job) error {
//
// TODO: Either implement the metric normalization or remove this dead code.
func checkJobData(d *schema.JobData) error {
for _, scopes := range *d {
for _, scopes := range d.Metrics {
// var newUnit schema.Unit
// TODO Add node scope if missing
for _, metric := range scopes {
+38 -17
View File
@@ -117,7 +117,7 @@ func LoadData(job *schema.Job,
jd, err = ms.LoadData(job, metrics, scopes, ctx, resolution)
if err != nil {
if len(jd) != 0 {
if len(jd.Metrics) != 0 {
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 {
@@ -144,7 +144,7 @@ func LoadData(job *schema.Job,
if rfErr != nil {
return rfErr, 0, 0
}
for _, v := range jd {
for _, v := range jd.Metrics {
for _, v_ := range v {
timestep := int64(0)
for i := 0; i < len(v_.Series); i += 1 {
@@ -160,17 +160,20 @@ func LoadData(job *schema.Job,
// Filter job data to only include requested metrics and scopes, avoiding unnecessary data transfer.
if metrics != nil || scopes != nil {
if metrics == nil {
metrics = make([]string, 0, len(jd))
for k := range jd {
metrics = make([]string, 0, len(jd.Metrics))
for k := range jd.Metrics {
metrics = append(metrics, k)
}
}
res := schema.JobData{}
res := schema.JobData{
Metrics: make(map[string]schema.ScopedMetrics, len(metrics)),
Groups: jd.Groups,
}
for _, metric := range metrics {
if perscope, ok := jd[metric]; ok {
if perscope, ok := jd.Metrics[metric]; ok {
if len(perscope) > 1 {
subset := make(map[schema.MetricScope]*schema.JobMetric)
subset := make(schema.ScopedMetrics)
for _, scope := range scopes {
if jm, ok := perscope[scope]; ok {
subset[scope] = jm
@@ -182,7 +185,7 @@ func LoadData(job *schema.Job,
}
}
res[metric] = perscope
res.Metrics[metric] = perscope
}
}
jd = res
@@ -199,7 +202,7 @@ func LoadData(job *schema.Job,
// instead of overwhelming the UI with individual node lines. Note that newly calculated
// statistics use min/median/max, while archived statistics may use min/mean/max.
const maxSeriesSize int = 8
for _, scopes := range jd {
for _, scopes := range jd.Metrics {
for _, jm := range scopes {
if jm.StatisticsSeries != nil || len(jm.Series) < maxSeriesSize {
continue
@@ -229,7 +232,7 @@ func LoadData(job *schema.Job,
if err, ok := data.(error); ok {
cclog.Errorf("error in cached dataset for job %d: %s", job.JobID, err.Error())
return nil, err
return schema.JobData{}, err
}
return data.(schema.JobData), nil
@@ -296,14 +299,14 @@ func LoadScopedJobStats(
if err != nil {
cclog.Errorf("failed to access metricDataRepo for cluster %s-%s: %s",
job.Cluster, job.SubCluster, err.Error())
return nil, err
return schema.ScopedJobStats{}, err
}
scopedStats, err := ms.LoadScopedStats(job, metrics, scopes, ctx)
if err != nil {
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
return schema.ScopedJobStats{}, err
}
// Round Resulting Stat Values
@@ -451,7 +454,7 @@ func LoadNodeListData(
// Statistics are calculated as min/median/max.
const maxSeriesSize int = 8
for _, jd := range data {
for _, scopes := range jd {
for _, scopes := range jd.Metrics {
for _, jm := range scopes {
if jm.StatisticsSeries != nil || len(jm.Series) < maxSeriesSize {
continue
@@ -472,14 +475,32 @@ func LoadNodeListData(
// archived data (e.g., during resampling). This ensures the cached archive data remains
// immutable while allowing per-request transformations.
func deepCopy(source schema.JobData) schema.JobData {
result := make(schema.JobData, len(source))
result := schema.JobData{Metrics: copyScopedMetrics(source.Metrics)}
for _, group := range source.Groups {
copied := schema.MetricGroup{Key: group.Key}
for _, inst := range group.Instances {
copied.Instances = append(copied.Instances, schema.MetricGroupInstance{
Name: inst.Name,
Type: inst.Type,
Metrics: copyScopedMetrics(inst.Metrics),
})
}
result.Groups = append(result.Groups, copied)
}
return result
}
func copyScopedMetrics(source map[string]schema.ScopedMetrics) map[string]schema.ScopedMetrics {
result := make(map[string]schema.ScopedMetrics, len(source))
for metricName, scopeMap := range source {
result[metricName] = make(map[schema.MetricScope]*schema.JobMetric, len(scopeMap))
scopes := make(schema.ScopedMetrics, len(scopeMap))
for scope, jobMetric := range scopeMap {
result[metricName][scope] = copyJobMetric(jobMetric)
scopes[scope] = copyJobMetric(jobMetric)
}
result[metricName] = scopes
}
return result
+22 -22
View File
@@ -13,7 +13,7 @@ import (
func TestDeepCopy(t *testing.T) {
nodeId := "0"
original := schema.JobData{
original := schema.JobData{Metrics: map[string]schema.ScopedMetrics{
"cpu_load": {
schema.MetricScopeNode: &schema.JobMetric{
Timestep: 60,
@@ -42,42 +42,42 @@ func TestDeepCopy(t *testing.T) {
},
},
},
}
}}
copied := deepCopy(original)
original["cpu_load"][schema.MetricScopeNode].Series[0].Data[0] = 999.0
original["cpu_load"][schema.MetricScopeNode].StatisticsSeries.Min[0] = 888.0
original["cpu_load"][schema.MetricScopeNode].StatisticsSeries.Percentiles[25][0] = 777.0
original.Metrics["cpu_load"][schema.MetricScopeNode].Series[0].Data[0] = 999.0
original.Metrics["cpu_load"][schema.MetricScopeNode].StatisticsSeries.Min[0] = 888.0
original.Metrics["cpu_load"][schema.MetricScopeNode].StatisticsSeries.Percentiles[25][0] = 777.0
if copied["cpu_load"][schema.MetricScopeNode].Series[0].Data[0] != 1.0 {
if copied.Metrics["cpu_load"][schema.MetricScopeNode].Series[0].Data[0] != 1.0 {
t.Errorf("Series data was not deeply copied: got %v, want 1.0",
copied["cpu_load"][schema.MetricScopeNode].Series[0].Data[0])
copied.Metrics["cpu_load"][schema.MetricScopeNode].Series[0].Data[0])
}
if copied["cpu_load"][schema.MetricScopeNode].StatisticsSeries.Min[0] != 1.0 {
if copied.Metrics["cpu_load"][schema.MetricScopeNode].StatisticsSeries.Min[0] != 1.0 {
t.Errorf("StatisticsSeries was not deeply copied: got %v, want 1.0",
copied["cpu_load"][schema.MetricScopeNode].StatisticsSeries.Min[0])
copied.Metrics["cpu_load"][schema.MetricScopeNode].StatisticsSeries.Min[0])
}
if copied["cpu_load"][schema.MetricScopeNode].StatisticsSeries.Percentiles[25][0] != 1.5 {
if copied.Metrics["cpu_load"][schema.MetricScopeNode].StatisticsSeries.Percentiles[25][0] != 1.5 {
t.Errorf("Percentiles was not deeply copied: got %v, want 1.5",
copied["cpu_load"][schema.MetricScopeNode].StatisticsSeries.Percentiles[25][0])
copied.Metrics["cpu_load"][schema.MetricScopeNode].StatisticsSeries.Percentiles[25][0])
}
if copied["cpu_load"][schema.MetricScopeNode].Timestep != 60 {
if copied.Metrics["cpu_load"][schema.MetricScopeNode].Timestep != 60 {
t.Errorf("Timestep not copied correctly: got %v, want 60",
copied["cpu_load"][schema.MetricScopeNode].Timestep)
copied.Metrics["cpu_load"][schema.MetricScopeNode].Timestep)
}
if copied["cpu_load"][schema.MetricScopeNode].Series[0].Hostname != "node001" {
if copied.Metrics["cpu_load"][schema.MetricScopeNode].Series[0].Hostname != "node001" {
t.Errorf("Hostname not copied correctly: got %v, want node001",
copied["cpu_load"][schema.MetricScopeNode].Series[0].Hostname)
copied.Metrics["cpu_load"][schema.MetricScopeNode].Series[0].Hostname)
}
}
func TestDeepCopyNilStatisticsSeries(t *testing.T) {
original := schema.JobData{
original := schema.JobData{Metrics: map[string]schema.ScopedMetrics{
"mem_used": {
schema.MetricScopeNode: &schema.JobMetric{
Timestep: 60,
@@ -90,18 +90,18 @@ func TestDeepCopyNilStatisticsSeries(t *testing.T) {
StatisticsSeries: nil,
},
},
}
}}
copied := deepCopy(original)
if copied["mem_used"][schema.MetricScopeNode].StatisticsSeries != nil {
if copied.Metrics["mem_used"][schema.MetricScopeNode].StatisticsSeries != nil {
t.Errorf("StatisticsSeries should be nil, got %v",
copied["mem_used"][schema.MetricScopeNode].StatisticsSeries)
copied.Metrics["mem_used"][schema.MetricScopeNode].StatisticsSeries)
}
}
func TestDeepCopyEmptyPercentiles(t *testing.T) {
original := schema.JobData{
original := schema.JobData{Metrics: map[string]schema.ScopedMetrics{
"cpu_load": {
schema.MetricScopeNode: &schema.JobMetric{
Timestep: 60,
@@ -115,11 +115,11 @@ func TestDeepCopyEmptyPercentiles(t *testing.T) {
},
},
},
}
}}
copied := deepCopy(original)
if copied["cpu_load"][schema.MetricScopeNode].StatisticsSeries.Percentiles != nil {
if copied.Metrics["cpu_load"][schema.MetricScopeNode].StatisticsSeries.Percentiles != nil {
t.Errorf("Percentiles should be nil when source is nil/empty")
}
}
+27 -27
View File
@@ -235,7 +235,7 @@ func (ccms *CCMetricStore) LoadData(
queries, assignedScope, err := ccms.buildQueries(job, metrics, scopes, resolution)
if err != nil {
cclog.Errorf("Error while building queries for jobId %d, Metrics %v, Scopes %v: %s", job.JobID, metrics, scopes, err.Error())
return nil, err
return schema.JobData{}, err
}
// Verify assignment is correct - log any inconsistencies for debugging
@@ -256,11 +256,11 @@ func (ccms *CCMetricStore) LoadData(
resBody, err := ccms.doRequest(ctx, &req)
if err != nil {
cclog.Errorf("Error while performing request for job %d: %s", job.JobID, err.Error())
return nil, err
return schema.JobData{}, err
}
var errors []string
jobData := make(schema.JobData)
jobData := schema.JobData{Metrics: make(map[string]schema.ScopedMetrics)}
// Add safety check for potential index out of range errors
if len(resBody.Results) != len(req.Queries) || len(assignedScope) != len(req.Queries) {
@@ -285,8 +285,8 @@ func (ccms *CCMetricStore) LoadData(
continue
}
if _, ok := jobData[metric]; !ok {
jobData[metric] = make(map[schema.MetricScope]*schema.JobMetric)
if _, ok := jobData.Metrics[metric]; !ok {
jobData.Metrics[metric] = make(schema.ScopedMetrics)
}
res := mc.Timestep
@@ -294,14 +294,14 @@ func (ccms *CCMetricStore) LoadData(
res = row[0].Resolution
}
jobMetric, ok := jobData[metric][scope]
jobMetric, ok := jobData.Metrics[metric][scope]
if !ok {
jobMetric = &schema.JobMetric{
Unit: mc.Unit,
Timestep: res,
Series: make([]schema.Series, 0),
}
jobData[metric][scope] = jobMetric
jobData.Metrics[metric][scope] = jobMetric
}
for ndx, res := range row {
@@ -329,9 +329,9 @@ func (ccms *CCMetricStore) LoadData(
// So that one can later check len(jobData):
if len(jobMetric.Series) == 0 {
delete(jobData[metric], scope)
if len(jobData[metric]) == 0 {
delete(jobData, metric)
delete(jobData.Metrics[metric], scope)
if len(jobData.Metrics[metric]) == 0 {
delete(jobData.Metrics, metric)
}
}
}
@@ -426,7 +426,7 @@ func (ccms *CCMetricStore) LoadScopedStats(
queries, assignedScope, err := ccms.buildQueries(job, metrics, scopes, 0)
if err != nil {
cclog.Errorf("Error while building queries for jobId %d, Metrics %v, Scopes %v: %s", job.JobID, metrics, scopes, err.Error())
return nil, err
return schema.ScopedJobStats{}, err
}
req := APIQueryRequest{
@@ -441,23 +441,23 @@ func (ccms *CCMetricStore) LoadScopedStats(
resBody, err := ccms.doRequest(ctx, &req)
if err != nil {
cclog.Errorf("Error while performing request for job %d: %s", job.JobID, err.Error())
return nil, err
return schema.ScopedJobStats{}, err
}
var errors []string
scopedJobStats := make(schema.ScopedJobStats)
scopedJobStats := schema.ScopedJobStats{Metrics: make(map[string]schema.ScopedMetricStats)}
for i, row := range resBody.Results {
query := req.Queries[i]
metric := query.Metric
scope := assignedScope[i]
if _, ok := scopedJobStats[metric]; !ok {
scopedJobStats[metric] = make(map[schema.MetricScope][]*schema.ScopedStats)
if _, ok := scopedJobStats.Metrics[metric]; !ok {
scopedJobStats.Metrics[metric] = make(schema.ScopedMetricStats)
}
if _, ok := scopedJobStats[metric][scope]; !ok {
scopedJobStats[metric][scope] = make([]*schema.ScopedStats, 0)
if _, ok := scopedJobStats.Metrics[metric][scope]; !ok {
scopedJobStats.Metrics[metric][scope] = make([]*schema.ScopedStats, 0)
}
for ndx, res := range row {
@@ -471,7 +471,7 @@ func (ccms *CCMetricStore) LoadScopedStats(
ms.SanitizeStats(&res.Avg, &res.Min, &res.Max)
scopedJobStats[metric][scope] = append(scopedJobStats[metric][scope], &schema.ScopedStats{
scopedJobStats.Metrics[metric][scope] = append(scopedJobStats.Metrics[metric][scope], &schema.ScopedStats{
Hostname: query.Hostname,
ID: id,
Data: &schema.MetricStatistics{
@@ -483,10 +483,10 @@ func (ccms *CCMetricStore) LoadScopedStats(
}
// So that one can later check len(scopedJobStats[metric][scope]): Remove from map if empty
if len(scopedJobStats[metric][scope]) == 0 {
delete(scopedJobStats[metric], scope)
if len(scopedJobStats[metric]) == 0 {
delete(scopedJobStats, metric)
if len(scopedJobStats.Metrics[metric][scope]) == 0 {
delete(scopedJobStats.Metrics[metric], scope)
if len(scopedJobStats.Metrics[metric]) == 0 {
delete(scopedJobStats.Metrics, metric)
}
}
}
@@ -691,14 +691,14 @@ func (ccms *CCMetricStore) LoadNodeListData(
// Init Nested Map Data Structures If Not Found
hostData, ok := data[query.Hostname]
if !ok {
hostData = make(schema.JobData)
hostData = schema.JobData{Metrics: make(map[string]schema.ScopedMetrics)}
data[query.Hostname] = hostData
}
metricData, ok := hostData[metric]
metricData, ok := hostData.Metrics[metric]
if !ok {
metricData = make(map[schema.MetricScope]*schema.JobMetric)
data[query.Hostname][metric] = metricData
metricData = make(schema.ScopedMetrics)
hostData.Metrics[metric] = metricData
}
scopeData, ok := metricData[scope]
@@ -708,7 +708,7 @@ func (ccms *CCMetricStore) LoadNodeListData(
Timestep: res,
Series: make([]schema.Series, 0),
}
data[query.Hostname][metric][scope] = scopeData
metricData[scope] = scopeData
}
for ndx, res := range row {
+1 -1
View File
@@ -403,7 +403,7 @@ func (r *JobRepository) JobsStats(
//
// Returns the requested statistic value or 0.0 if not found.
func LoadJobStat(job *schema.Job, metric string, statType string) float64 {
if stats, ok := job.Statistics[metric]; ok {
if stats, ok := job.Statistics.Metrics[metric]; ok {
switch statType {
case "avg":
return stats.Avg
+2 -2
View File
@@ -107,7 +107,7 @@ type JobClassTagger struct {
// repo provides access to job database operations
repo JobRepository
// getStatistics retrieves job statistics for analysis
getStatistics func(job *schema.Job) (map[string]schema.JobStatistics, error)
getStatistics func(job *schema.Job) (schema.JobStatisticsSet, error)
// getMetricConfig retrieves metric configuration (limits) for a cluster
getMetricConfig func(cluster, subCluster string) map[string]*schema.Metric
}
@@ -361,7 +361,7 @@ func (t *JobClassTagger) Match(job *schema.Job) {
// add metrics to env
skipRule := false
for _, m := range ri.metrics {
stats, ok := jobStats[m]
stats, ok := jobStats.Metrics[m]
if !ok {
cclog.Debugf("job classification: missing metric '%s' for rule %s on job %d", m, tag, job.JobID)
skipRule = true
+6 -6
View File
@@ -64,10 +64,10 @@ func TestClassifyJobMatch(t *testing.T) {
parameters: make(map[string]any),
tagType: "jobClass",
repo: mockRepo,
getStatistics: func(job *schema.Job) (map[string]schema.JobStatistics, error) {
return map[string]schema.JobStatistics{
getStatistics: func(job *schema.Job) (schema.JobStatisticsSet, error) {
return schema.JobStatisticsSet{Metrics: map[string]schema.JobStatistics{
"flops_any": {Min: 0, Max: 200, Avg: 150},
}, nil
}}, nil
},
getMetricConfig: func(cluster, subCluster string) map[string]*schema.Metric {
return map[string]*schema.Metric{
@@ -120,10 +120,10 @@ func TestMatch_NoMatch(t *testing.T) {
parameters: make(map[string]any),
tagType: "jobClass",
repo: mockRepo,
getStatistics: func(job *schema.Job) (map[string]schema.JobStatistics, error) {
return map[string]schema.JobStatistics{
getStatistics: func(job *schema.Job) (schema.JobStatisticsSet, error) {
return schema.JobStatisticsSet{Metrics: map[string]schema.JobStatistics{
"flops_any": {Min: 0, Max: 50, Avg: 20}, // Avg 20 < 100
}, nil
}}, nil
},
getMetricConfig: func(cluster, subCluster string) map[string]*schema.Metric {
return map[string]*schema.Metric{
@@ -80,7 +80,7 @@ func RegisterFootprintWorker() {
continue
}
job.Statistics = make(map[string]schema.JobStatistics)
job.Statistics = schema.JobStatisticsSet{Metrics: make(map[string]schema.JobStatistics, len(allMetrics))}
for _, metric := range allMetrics {
avg, min, max := 0.0, 0.0, 0.0
@@ -98,7 +98,7 @@ func RegisterFootprintWorker() {
}
// Add values rounded to 2 digits: repo.LoadStats may return unrounded
job.Statistics[metric] = schema.JobStatistics{
job.Statistics.Metrics[metric] = schema.JobStatistics{
Unit: schema.Unit{
Prefix: archive.GetMetricConfig(job.Cluster, metric).Unit.Prefix,
Base: archive.GetMetricConfig(job.Cluster, metric).Unit.Base,