Merge branch 'Refactor-job-footprint' into sample_resolution_select

cmd/cc-backend/server.go

@@ -38,6 +38,15 @@ var (
 	apiHandle *api.RestApi
 )
 
+func onFailureResponse(rw http.ResponseWriter, r *http.Request, err error) {
+	rw.Header().Add("Content-Type", "application/json")
+	rw.WriteHeader(http.StatusUnauthorized)
+	json.NewEncoder(rw).Encode(map[string]string{
+		"status": http.StatusText(http.StatusUnauthorized),
+		"error":  err.Error(),
+	})
+}
+
 func serverInit() {
 	// Setup the http.Handler/Router used by the server
 	graph.Init()
@@ -166,64 +175,32 @@ func serverInit() {
 			return authHandle.AuthApi(
 				// On success;
 				next,
-
 				// On failure: JSON Response
-				func(rw http.ResponseWriter, r *http.Request, err error) {
+				onFailureResponse)
-					rw.Header().Add("Content-Type", "application/json")
-					rw.WriteHeader(http.StatusUnauthorized)
-					json.NewEncoder(rw).Encode(map[string]string{
-						"status": http.StatusText(http.StatusUnauthorized),
-						"error":  err.Error(),
-					})
-				})
 		})
 
 		userapi.Use(func(next http.Handler) http.Handler {
 			return authHandle.AuthUserApi(
 				// On success;
 				next,
-
 				// On failure: JSON Response
-				func(rw http.ResponseWriter, r *http.Request, err error) {
+				onFailureResponse)
-					rw.Header().Add("Content-Type", "application/json")
-					rw.WriteHeader(http.StatusUnauthorized)
-					json.NewEncoder(rw).Encode(map[string]string{
-						"status": http.StatusText(http.StatusUnauthorized),
-						"error":  err.Error(),
-					})
-				})
 		})
 
 		configapi.Use(func(next http.Handler) http.Handler {
 			return authHandle.AuthConfigApi(
 				// On success;
 				next,
-
 				// On failure: JSON Response
-				func(rw http.ResponseWriter, r *http.Request, err error) {
+				onFailureResponse)
-					rw.Header().Add("Content-Type", "application/json")
-					rw.WriteHeader(http.StatusUnauthorized)
-					json.NewEncoder(rw).Encode(map[string]string{
-						"status": http.StatusText(http.StatusUnauthorized),
-						"error":  err.Error(),
-					})
-				})
 		})
 
 		frontendapi.Use(func(next http.Handler) http.Handler {
 			return authHandle.AuthFrontendApi(
 				// On success;
 				next,
-
 				// On failure: JSON Response
-				func(rw http.ResponseWriter, r *http.Request, err error) {
+				onFailureResponse)
-					rw.Header().Add("Content-Type", "application/json")
-					rw.WriteHeader(http.StatusUnauthorized)
-					json.NewEncoder(rw).Encode(map[string]string{
-						"status": http.StatusText(http.StatusUnauthorized),
-						"error":  err.Error(),
-					})
-				})
 		})
 	}
 

internal/metricdata/metricdata.go

@@ -307,6 +307,10 @@ func ArchiveJob(job *schema.Job, ctx context.Context) (*schema.JobMeta, error) {
 		scopes = append(scopes, schema.MetricScopeCore)
 	}
 
+	if job.NumAcc > 0 {
+		scopes = append(scopes, schema.MetricScopeAccelerator)
+	}
+
 	jobData, err := LoadData(job, allMetrics, scopes, ctx)
 	if err != nil {
 		log.Error("Error wile loading job data for archiving")

internal/repository/migrations/sqlite3/08_add-footprint.down.sql

@@ -0,0 +1,21 @@
+ALTER TABLE job DROP energy;
+ALTER TABLE job DROP energy_footprint;
+ALTER TABLE job ADD COLUMN flops_any_avg;
+ALTER TABLE job ADD COLUMN mem_bw_avg;
+ALTER TABLE job ADD COLUMN mem_used_max;
+ALTER TABLE job ADD COLUMN load_avg;
+ALTER TABLE job ADD COLUMN net_bw_avg;
+ALTER TABLE job ADD COLUMN net_data_vol_total;
+ALTER TABLE job ADD COLUMN file_bw_avg;
+ALTER TABLE job ADD COLUMN file_data_vol_total;
+
+UPDATE job SET flops_any_avg = json_extract(footprint, '$.flops_any_avg');
+UPDATE job SET mem_bw_avg = json_extract(footprint, '$.mem_bw_avg');
+UPDATE job SET mem_used_max = json_extract(footprint, '$.mem_used_max');
+UPDATE job SET load_avg = json_extract(footprint, '$.cpu_load_avg');
+UPDATE job SET net_bw_avg = json_extract(footprint, '$.net_bw_avg');
+UPDATE job SET net_data_vol_total = json_extract(footprint, '$.net_data_vol_total');
+UPDATE job SET file_bw_avg = json_extract(footprint, '$.file_bw_avg');
+UPDATE job SET file_data_vol_total = json_extract(footprint, '$.file_data_vol_total');
+
+ALTER TABLE job DROP footprint;

internal/repository/migrations/sqlite3/08_add-footprint.up.sql

@@ -1,4 +1,5 @@
 ALTER TABLE job ADD COLUMN energy REAL NOT NULL DEFAULT 0.0;
+ALTER TABLE job ADD COLUMN energy_footprint TEXT DEFAULT NULL;
 
 ALTER TABLE job ADD COLUMN footprint TEXT DEFAULT NULL;
 UPDATE job SET footprint = '{"flops_any_avg": 0.0}';
@@ -6,7 +7,16 @@ UPDATE job SET footprint = json_replace(footprint, '$.flops_any_avg', job.flops_
 UPDATE job SET footprint = json_insert(footprint, '$.mem_bw_avg', job.mem_bw_avg);
 UPDATE job SET footprint = json_insert(footprint, '$.mem_used_max', job.mem_used_max);
 UPDATE job SET footprint = json_insert(footprint, '$.cpu_load_avg', job.load_avg);
+UPDATE job SET footprint = json_insert(footprint, '$.net_bw_avg', job.net_bw_avg) WHERE job.net_bw_avg != 0;
+UPDATE job SET footprint = json_insert(footprint, '$.net_data_vol_total', job.net_data_vol_total) WHERE job.net_data_vol_total != 0;
+UPDATE job SET footprint = json_insert(footprint, '$.file_bw_avg', job.file_bw_avg) WHERE job.file_bw_avg != 0;
+UPDATE job SET footprint = json_insert(footprint, '$.file_data_vol_total', job.file_data_vol_total) WHERE job.file_data_vol_total != 0;
+
 ALTER TABLE job DROP flops_any_avg;
 ALTER TABLE job DROP mem_bw_avg;
 ALTER TABLE job DROP mem_used_max;
 ALTER TABLE job DROP load_avg;
+ALTER TABLE job DROP net_bw_avg;
+ALTER TABLE job DROP net_data_vol_total;
+ALTER TABLE job DROP file_bw_avg;
+ALTER TABLE job DROP file_data_vol_total;

pkg/schema/cluster.go

@@ -47,11 +47,11 @@ type SubCluster struct {
 
 type SubClusterConfig struct {
 	Name          string  `json:"name"`
+	Footprint     string  `json:"footprint,omitempty"`
 	Peak          float64 `json:"peak"`
 	Normal        float64 `json:"normal"`
 	Caution       float64 `json:"caution"`
 	Alert         float64 `json:"alert"`
-	Footprint     string  `json:"footprint,omitempty"`
 	Remove        bool    `json:"remove"`
 	LowerIsBetter bool    `json:"lowerIsBetter"`
 	Energy        bool    `json:"energy"`
@@ -62,14 +62,14 @@ type MetricConfig struct {
 	Name          string              `json:"name"`
 	Scope         MetricScope         `json:"scope"`
 	Aggregation   string              `json:"aggregation"`
+	Footprint     string              `json:"footprint,omitempty"`
 	SubClusters   []*SubClusterConfig `json:"subClusters,omitempty"`
-	Timestep      int                 `json:"timestep"`
 	Peak          float64             `json:"peak"`
 	Normal        float64             `json:"normal"`
 	Caution       float64             `json:"caution"`
 	Alert         float64             `json:"alert"`
+	Timestep      int                 `json:"timestep"`
 	LowerIsBetter bool                `json:"lowerIsBetter"`
-	Footprint     string              `json:"footprint,omitempty"`
 	Energy        bool                `json:"energy"`
 }
 

pkg/schema/job.go

@@ -32,7 +32,7 @@ type BaseJob struct {
 	Footprint          map[string]float64 `json:"footprint"`
 	MetaData           map[string]string  `json:"metaData"`
 	ConcurrentJobs     JobLinkResultList  `json:"concurrentJobs"`
-	Energy             float64            `json:"energy"`
+	Energy             float64            `json:"energy" db:"energy"`
 	ArrayJobId         int64              `json:"arrayJobId,omitempty" db:"array_job_id" example:"123000"`
 	Walltime           int64              `json:"walltime,omitempty" db:"walltime" example:"86400" minimum:"1"`
 	JobID              int64              `json:"jobId" db:"job_id" example:"123000"`

pkg/schema/schemas/cluster.schema.json

@@ -39,6 +39,23 @@
               "avg"
             ]
           },
+          "footprint": {
+            "description": "Is it a footprint metric and what type",
+            "type": "string",
+            "enum": [
+              "avg",
+              "max",
+              "min"
+            ]
+          },
+          "energy": {
+            "description": "Is it used to calculate job energy",
+            "type": "boolean"
+          },
+          "lowerIsBetter": {
+            "description": "Is lower better.",
+            "type": "boolean"
+          },
           "peak": {
             "description": "Metric peak threshold (Upper metric limit)",
             "type": "number"
@@ -65,6 +82,23 @@
                   "description": "Hardware partition name",
                   "type": "string"
                 },
+                "footprint": {
+                  "description": "Is it a footprint metric and what type. Overwrite global setting",
+                  "type": "string",
+                  "enum": [
+                    "avg",
+                    "max",
+                    "min"
+                  ]
+                },
+                "energy": {
+                  "description": "Is it used to calculate job energy. Overwrite global",
+                  "type": "boolean"
+                },
+                "lowerIsBetter": {
+                  "description": "Is lower better. Overwrite global",
+                  "type": "boolean"
+                },
                 "peak": {
                   "type": "number"
                 },
@@ -78,6 +112,7 @@
                   "type": "number"
                 },
                 "remove": {
+                  "description": "Remove this metric for this subcluster",
                   "type": "boolean"
                 }
               },

web/frontend/src/generic/helper/JobFootprint.svelte

@@ -67,8 +67,9 @@
   export let height = "310px";
 
   const footprintData = job?.footprint?.map((jf) => {
-    // Unit
     const fmc = getContext("getMetricConfig")(job.cluster, job.subCluster, jf.name);
+    if (fmc) {
+      // Unit
       const unit = (fmc?.unit?.prefix ? fmc.unit.prefix : "") + (fmc?.unit?.base ? fmc.unit.base : "")
 
       // Threshold / -Differences
@@ -122,7 +123,18 @@
           impact: -1,
         };
       }
-  });
+    } else { // No matching metric config: display as single value
+      return {
+        name: jf.name + ' (' + jf.stat + ')',
+        avg: jf.value,
+        message:
+          `No config for metric ${jf.name} found.`,
+        impact: 4,
+      };
+    }
+  }).sort(function (a, b) { // Sort by impact value primarily, within impact sort name alphabetically
+    return a.impact - b.impact || ((a.name > b.name) ? 1 : ((b.name > a.name) ? -1 : 0));
+  });;
 
   function evalFootprint(mean, thresholds, lowerIsBetter, level) {
     // Handle Metrics in which less value is better
@@ -159,6 +171,7 @@
   {/if}
   <CardBody>
     {#each footprintData as fpd, index}
+      {#if fpd.impact !== 4}
         <div class="mb-1 d-flex justify-content-between">
           <div>&nbsp;<b>{fpd.name}</b></div>
           <!-- For symmetry, see below ...-->
@@ -213,6 +226,29 @@
           </Col>
           {/if}
         </Row>
+      {:else}
+        <div class="mb-1 d-flex justify-content-between">
+          <div>
+            &nbsp;<b>{fpd.name}</b>
+          </div>
+          <div
+            class="cursor-help d-inline-flex"
+            id={`footprint-${job.jobId}-${index}`}
+          >
+            <div class="mx-1">
+              <Icon name="info-circle"/>
+            </div>
+            <div>
+              {fpd.avg}&nbsp;
+            </div>
+          </div>
+        </div>
+        <Tooltip
+          target={`footprint-${job.jobId}-${index}`}
+          placement="right"
+          offset={[0, 20]}>{fpd.message}</Tooltip
+        >
+      {/if}
     {/each}
     {#if job?.metaData?.message}
       <hr class="mt-1 mb-2" />