Imporve monitoring 4

infrastructure/signoz/dashboards/system-health.json

@@ -0,0 +1,92 @@
+{
+  "dashboard": {
+    "title": "Bakery IA - System Health",
+    "description": "Comprehensive system health monitoring dashboard",
+    "tags": ["system", "health", "monitoring"],
+    "panels": [
+      {
+        "title": "System Availability",
+        "type": "stat",
+        "query": {
+          "metric": "system_availability",
+          "aggregate": "avg",
+          "filters": [
+            {
+              "key": "namespace",
+              "operator": "=",
+              "value": "${namespace}"
+            }
+          ]
+        },
+        "unit": "percent"
+      },
+      {
+        "title": "Service Health Score",
+        "type": "stat",
+        "query": {
+          "metric": "service_health_score",
+          "aggregate": "avg",
+          "filters": [
+            {
+              "key": "namespace",
+              "operator": "=",
+              "value": "${namespace}"
+            }
+          ]
+        },
+        "unit": "number"
+      },
+      {
+        "title": "CPU Usage",
+        "type": "timeseries",
+        "query": {
+          "metric": "system_cpu_usage",
+          "aggregate": "avg",
+          "filters": [
+            {
+              "key": "namespace",
+              "operator": "=",
+              "value": "${namespace}"
+            }
+          ]
+        },
+        "unit": "percent"
+      },
+      {
+        "title": "Memory Usage",
+        "type": "timeseries",
+        "query": {
+          "metric": "system_memory_usage",
+          "aggregate": "avg",
+          "filters": [
+            {
+              "key": "namespace",
+              "operator": "=",
+              "value": "${namespace}"
+            }
+          ]
+        },
+        "unit": "percent"
+      }
+    ],
+    "variables": [
+      {
+        "name": "namespace",
+        "label": "Namespace",
+        "type": "dropdown",
+        "default": "bakery-ia",
+        "values": ["bakery-ia", "default"]
+      }
+    ],
+    "layout": {
+      "type": "grid",
+      "columns": 12,
+      "gap": [16, 16]
+    },
+    "refresh": "30s",
+    "time": {
+      "from": "now-1h",
+      "to": "now"
+    }
+  }
+}