Introduce hierarchical metriclists and thresholds

pkg/schema/cluster.go

@@ -38,6 +38,7 @@ type SubCluster struct {
 	FlopRateSimd    *MetricValue    `json:"flopRateSimd"`
 	MemoryBandwidth *MetricValue    `json:"memoryBandwidth"`
 	Topology        *Topology       `json:"topology"`
+	MetricConfig    []*MetricConfig `json:"metricConfig"`
 }
 
 type SubClusterConfig struct {
@@ -54,6 +55,10 @@ type MetricConfig struct {
 	Scope       MetricScope         `json:"scope"`
 	Aggregation *string             `json:"aggregation"`
 	Timestep    int                 `json:"timestep"`
+	Peak        *float64            `json:"peak"`
+	Normal      *float64            `json:"normal"`
+	Caution     *float64            `json:"caution"`
+	Alert       *float64            `json:"alert"`
 	SubClusters []*SubClusterConfig `json:"subClusters"`
 }
 

pkg/schema/schemas/cluster.schema.json

@@ -39,6 +39,22 @@
                             "avg"
                         ]
                     },
+                    "peak": {
+                        "description": "Metric peak threshold (Upper metric limit)",
+                        "type": "number"
+                    },
+                    "normal": {
+                        "description": "Metric normal threshold",
+                        "type": "number"
+                    },
+                    "caution": {
+                        "description": "Metric caution threshold (Suspicious but does not require immediate action)",
+                        "type": "number"
+                    },
+                    "alert": {
+                        "description": "Metric alert threshold (Requires immediate action)",
+                        "type": "number"
+                    },
                     "subClusters": {
                         "description": "Array of cluster hardware partition metric thresholds",
                         "type": "array",
@@ -76,7 +92,10 @@
                     "unit",
                     "scope",
                     "timestep",
-                    "subClusters"
+                    "aggregation",
+                    "peak",
+                    "caution",
+                    "alert"
                 ]
             },
             "minItems": 1