Merge branch 'master' into hotfix

Author: moebiusband73

.gitignore

@@ -1,21 +1,23 @@
 /cc-backend
+/.env
+/config.json
 
 /var/job-archive
-/var/*.db
 /var/machine-state
-
-/.env
-/config.json
+/var/job.db-shm
+/var/job.db-wal
+/var/*.db
+/var/*.txt
 
 /web/frontend/public/build
 /web/frontend/node_modules
-/.vscode/*
+
 /archive-migration
 /archive-manager
-var/job.db-shm
-var/job.db-wal
 
+/internal/repository/testdata/job.db-shm
+/internal/repository/testdata/job.db-wal
+
+/.vscode/*
 dist/
 *.db
-internal/repository/testdata/job.db-shm
-internal/repository/testdata/job.db-wal

Makefile

@@ -82,7 +82,7 @@ tags:
 	@ctags -R
 
 $(VAR):
-	@mkdir $(VAR)
+	@mkdir -p $(VAR)
 
 config.json:
 	$(info ===>  Initialize config.json file)

internal/archiver/archiver.go

@@ -60,12 +60,13 @@ func ArchiveJob(job *schema.Job, ctx context.Context) (*schema.JobMeta, error) {
 			max = math.Max(max, series.Statistics.Max)
 		}
 
+		// Round AVG Result to 2 Digits
 		jobMeta.Statistics[metric] = schema.JobStatistics{
 			Unit: schema.Unit{
 				Prefix: archive.GetMetricConfig(job.Cluster, metric).Unit.Prefix,
 				Base:   archive.GetMetricConfig(job.Cluster, metric).Unit.Base,
 			},
-			Avg: avg / float64(job.NumNodes),
+			Avg: (math.Round((avg/float64(job.NumNodes))*100) / 100),
 			Min: min,
 			Max: max,
 		}

internal/graph/schema.resolvers.go

@@ -170,6 +170,9 @@ func LoadData(job *schema.Job,
 			jd.AddNodeScope("mem_bw")
 		}
 
+		// Round Resulting Stat Values
+		jd.RoundMetricStats()
+
 		return jd, ttl, size
 	})
 

internal/metricDataDispatcher/dataLoader.go

@@ -440,6 +440,23 @@ func (ccms *CCMetricStore) buildQueries(
 					continue
 				}
 
+				// Core -> Socket
+				if nativeScope == schema.MetricScopeCore && scope == schema.MetricScopeSocket {
+					sockets, _ := topology.GetSocketsFromCores(hwthreads)
+					for _, socket := range sockets {
+						queries = append(queries, ApiQuery{
+							Metric:     remoteName,
+							Hostname:   host.Hostname,
+							Aggregate:  true,
+							Type:       &coreString,
+							TypeIds:    intToStringSlice(topology.Socket[socket]),
+							Resolution: resolution,
+						})
+						assignedScope = append(assignedScope, scope)
+					}
+					continue
+				}
+
 				// Core -> Node
 				if nativeScope == schema.MetricScopeCore && scope == schema.MetricScopeNode {
 					cores, _ := topology.GetCoresFromHWThreads(hwthreads)
@@ -627,7 +644,7 @@ func (ccms *CCMetricStore) LoadNodeData(
 				req.Queries = append(req.Queries, ApiQuery{
 					Hostname:   node,
 					Metric:     ccms.toRemoteName(metric),
-					Resolution: 60, // Default for Node Queries
+					Resolution: 0, // Default for Node Queries: Will return metric $Timestep Resolution
 				})
 			}
 		}
@@ -1038,6 +1055,23 @@ func (ccms *CCMetricStore) buildNodeQueries(
 					continue
 				}
 
+				// Core -> Socket
+				if nativeScope == schema.MetricScopeCore && scope == schema.MetricScopeSocket {
+					sockets, _ := topology.GetSocketsFromCores(topology.Node)
+					for _, socket := range sockets {
+						queries = append(queries, ApiQuery{
+							Metric:     remoteName,
+							Hostname:   hostname,
+							Aggregate:  true,
+							Type:       &coreString,
+							TypeIds:    intToStringSlice(topology.Socket[socket]),
+							Resolution: resolution,
+						})
+						assignedScope = append(assignedScope, scope)
+					}
+					continue
+				}
+
 				// Core -> Node
 				if nativeScope == schema.MetricScopeCore && scope == schema.MetricScopeNode {
 					cores, _ := topology.GetCoresFromHWThreads(topology.Node)

internal/metricdata/cc-metric-store.go

@@ -217,11 +217,6 @@ func (r *JobRepository) UpdateMetadata(job *schema.Job, key, val string) (err er
 
 func (r *JobRepository) FetchFootprint(job *schema.Job) (map[string]float64, error) {
 	start := time.Now()
-	cachekey := fmt.Sprintf("footprint:%d", job.ID)
-	if cached := r.cache.Get(cachekey, nil); cached != nil {
-		job.Footprint = cached.(map[string]float64)
-		return job.Footprint, nil
-	}
 
 	if err := sq.Select("job.footprint").From("job").Where("job.id = ?", job.ID).
 		RunWith(r.stmtCache).QueryRow().Scan(&job.RawFootprint); err != nil {
@@ -238,7 +233,6 @@ func (r *JobRepository) FetchFootprint(job *schema.Job) (map[string]float64, err
 		return nil, err
 	}
 
-	r.cache.Put(cachekey, job.Footprint, len(job.Footprint), 24*time.Hour)
 	log.Debugf("Timer FetchFootprint %s", time.Since(start))
 	return job.Footprint, nil
 }
@@ -606,8 +600,11 @@ func (r *JobRepository) UpdateEnergy(
 				// FIXME: Needs sum as stats type
 			} else if sc.MetricConfig[i].Energy == "power" { // this metric has power as unit (Watt)
 				// Energy: Power (in Watts) * Time (in Seconds)
-				// Unit: ( W * s ) / 3600 / 1000 = kWh ; Rounded to 2 nearest digits
-				energy = math.Round(((LoadJobStat(jobMeta, fp, "avg")*float64(jobMeta.Duration))/3600/1000)*100) / 100
+				// Unit: (( W * s ) / 3600) / 1000 = kWh ; Rounded to 2 nearest digits: (Energy * 100) / 100
+				// Here: All-Node Metric Average * Number of Nodes * Job Runtime
+				// 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)
+				energy = math.Round(((metricNodeSum/3600)/1000)*100) / 100
 			}
 		} else {
 			log.Warnf("Error while collecting energy metric %s for job, DB ID '%v', return '0.0'", fp, jobMeta.ID)

internal/repository/job.go

@@ -94,7 +94,7 @@ func RegisterFootprintWorker() {
 								}
 							}
 
-							// Add values rounded to 2 digits
+							// Add values rounded to 2 digits: repo.LoadStats may return unrounded
 							jobMeta.Statistics[metric] = schema.JobStatistics{
 								Unit: schema.Unit{
 									Prefix: archive.GetMetricConfig(job.Cluster, metric).Unit.Prefix,

internal/taskManager/updateFootprintService.go

@@ -122,6 +122,38 @@ func (topo *Topology) GetSocketsFromHWThreads(
 	return sockets, exclusive
 }
 
+// Return a list of socket IDs given a list of core IDs.  Even if just one
+// core is in that socket, add it to the list.  If no cores other than
+// those in the argument list are assigned to one of the sockets in the first
+// return value, return true as the second value.  TODO: Optimize this, there
+// must be a more efficient way/algorithm.
+func (topo *Topology) GetSocketsFromCores (
+	cores []int,
+) (sockets []int, exclusive bool) {
+	socketsMap := map[int]int{}
+	for _, core := range cores {
+		for _, hwthreadInCore := range topo.Core[core] {
+			for socket, hwthreadsInSocket := range topo.Socket {
+				for _, hwthreadInSocket := range hwthreadsInSocket {
+					if hwthreadInCore == hwthreadInSocket {
+						socketsMap[socket] += 1
+					}
+				}
+			}
+		}
+	}
+
+	exclusive = true
+	hwthreadsPerSocket := len(topo.Node) / len(topo.Socket)
+	sockets = make([]int, 0, len(socketsMap))
+	for socket, count := range socketsMap {
+		sockets = append(sockets, socket)
+		exclusive = exclusive && count == hwthreadsPerSocket
+	}
+
+	return sockets, exclusive
+}
+
 // Return a list of core IDs given a list of hwthread IDs.  Even if just one
 // hwthread is in that core, add it to the list.  If no hwthreads other than
 // those in the argument list are assigned to one of the cores in the first

pkg/schema/cluster.go

@@ -291,6 +291,21 @@ func (jd *JobData) AddNodeScope(metric string) bool {
 	return true
 }
 
+func (jd *JobData) RoundMetricStats() {
+	// TODO: Make Digit-Precision Configurable? (Currently: Fixed to 2 Digits)
+	for _, scopes := range *jd {
+		for _, jm := range scopes {
+			for index := range jm.Series {
+				jm.Series[index].Statistics = MetricStatistics{
+					Avg: (math.Round(jm.Series[index].Statistics.Avg*100) / 100),
+					Min: (math.Round(jm.Series[index].Statistics.Min*100) / 100),
+					Max: (math.Round(jm.Series[index].Statistics.Max*100) / 100),
+				}
+			}
+		}
+	}
+}
+
 func (jm *JobMetric) AddPercentiles(ps []int) bool {
 	if jm.StatisticsSeries == nil {
 		jm.AddStatisticsSeries()

pkg/schema/metrics.go

@@ -446,7 +446,7 @@
           }
         },
         "job_view_selectedMetrics": {
-          "description": "",
+          "description": "Initial metrics shown as plots in single job view",
           "type": "array",
           "items": {
             "type": "string",

pkg/schema/schemas/config.schema.json

@@ -117,27 +117,41 @@
     }
   `;
 
+const roofQuery = gql`
+    query ($dbid: ID!, $selectedMetrics: [String!]!, $selectedScopes: [MetricScope!]!, $selectedResolution: Int) {
+      jobMetrics(id: $dbid, metrics: $selectedMetrics, scopes: $selectedScopes, resolution: $selectedResolution) {
+        name
+        scope
+        metric {
+          series {
+            data
+          }
+        }
+      }
+    }
+  `;
+
   $: jobMetrics = queryStore({
     client: client,
     query: query,
     variables: { dbid, selectedMetrics, selectedScopes },
   });
 
+  // Roofline: Always load roofMetrics with configured timestep (Resolution: 0)
+  $: roofMetrics = queryStore({
+    client: client,
+    query: roofQuery,
+    variables: { dbid, selectedMetrics: ["flops_any", "mem_bw"], selectedScopes: ["node"], selectedResolution: 0 },
+  });
+
   // Handle Job Query on Init -> is not executed anymore
   getContext("on-init")(() => {
     let job = $initq.data.job;
     if (!job) return;
 
     const pendingMetrics = [
-      "flops_any",
-      "mem_bw",
       ...(ccconfig[`job_view_selectedMetrics:${job.cluster}`] ||
-        $initq.data.globalMetrics.reduce((names, gm) => {
-          if (gm.availability.find((av) => av.cluster === job.cluster)) {
-            names.push(gm.name);
-          }
-          return names;
-        }, [])
+        ccconfig[`job_view_selectedMetrics`]
       ),
       ...(ccconfig[`job_view_nodestats_selectedMetrics:${job.cluster}`] ||
         ccconfig[`job_view_nodestats_selectedMetrics`]
@@ -276,12 +290,12 @@
 
   <!-- Column 3: Job Roofline; If footprint Enabled: full width, else half width -->
   <Col xs={12} md={12} xl={5} xxl={6}>
-    {#if $initq.error || $jobMetrics.error}
+    {#if $initq.error || $roofMetrics.error}
       <Card body color="danger">
         <p>Initq Error: {$initq.error?.message}</p>
-        <p>jobMetrics Error: {$jobMetrics.error?.message}</p>
+        <p>roofMetrics (jobMetrics) Error: {$roofMetrics.error?.message}</p>
       </Card>
-    {:else if $initq?.data && $jobMetrics?.data}
+    {:else if $initq?.data && $roofMetrics?.data}
       <Card style="height: 400px;">
         <div bind:clientWidth={roofWidth}>
           <Roofline
@@ -292,10 +306,10 @@
               .find((c) => c.name == $initq.data.job.cluster)
               .subClusters.find((sc) => sc.name == $initq.data.job.subCluster)}
             data={transformDataForRoofline(
-              $jobMetrics.data?.jobMetrics?.find(
+              $roofMetrics.data?.jobMetrics?.find(
                 (m) => m.name == "flops_any" && m.scope == "node",
               )?.metric,
-              $jobMetrics.data?.jobMetrics?.find(
+              $roofMetrics.data?.jobMetrics?.find(
                 (m) => m.name == "mem_bw" && m.scope == "node",
               )?.metric,
             )}

web/frontend/src/Job.root.svelte

@@ -80,6 +80,7 @@
     : ccconfig.user_view_histogramMetrics || [];
 
   const client = getContextClient();
+  // Note: nodeMetrics are requested on configured $timestep resolution
   $: mainQuery = queryStore({
     client: client,
     query: gql`

web/frontend/src/Status.root.svelte

@@ -77,6 +77,7 @@
     for (let sm of systemMetrics) {
       systemUnits[sm.name] = (sm?.unit?.prefix ? sm.unit.prefix : "") + (sm?.unit?.base ? sm.unit.base : "")
     }
+    if (!selectedMetric) selectedMetric = systemMetrics[0].name
   }
 
   $: loadMetrics($initialized)