Imporve monitoring 4

infrastructure/signoz/dashboards/alert-management.json

@@ -0,0 +1,104 @@
+{
+  "dashboard": {
+    "title": "Bakery IA - Alert Management",
+    "description": "Alert monitoring and management dashboard",
+    "tags": ["alerts", "monitoring", "management"],
+    "panels": [
+      {
+        "title": "Active Alerts",
+        "type": "stat",
+        "query": {
+          "metric": "alerts_active",
+          "aggregate": "sum",
+          "filters": [
+            {
+              "key": "severity",
+              "operator": "=",
+              "value": "${severity}"
+            },
+            {
+              "key": "status",
+              "operator": "=",
+              "value": "firing"
+            }
+          ]
+        },
+        "unit": "number"
+      },
+      {
+        "title": "Alert Rate",
+        "type": "timeseries",
+        "query": {
+          "metric": "alerts_total",
+          "aggregate": "rate",
+          "filters": [
+            {
+              "key": "severity",
+              "operator": "=",
+              "value": "${severity}"
+            }
+          ]
+        },
+        "unit": "alerts/s"
+      },
+      {
+        "title": "Alerts by Severity",
+        "type": "pie",
+        "query": {
+          "metric": "alerts_total",
+          "aggregate": "sum",
+          "groupBy": ["severity"],
+          "filters": [
+            {
+              "key": "severity",
+              "operator": "=",
+              "value": "${severity}"
+            }
+          ]
+        }
+      },
+      {
+        "title": "Alerts by Status",
+        "type": "pie",
+        "query": {
+          "metric": "alerts_total",
+          "aggregate": "sum",
+          "groupBy": ["status"],
+          "filters": [
+            {
+              "key": "status",
+              "operator": "=",
+              "value": "${status}"
+            }
+          ]
+        }
+      }
+    ],
+    "variables": [
+      {
+        "name": "severity",
+        "label": "Severity",
+        "type": "dropdown",
+        "default": "*",
+        "values": ["*", "critical", "high", "medium", "low"]
+      },
+      {
+        "name": "status",
+        "label": "Status",
+        "type": "dropdown",
+        "default": "*",
+        "values": ["*", "firing", "resolved", "acknowledged"]
+      }
+    ],
+    "layout": {
+      "type": "grid",
+      "columns": 12,
+      "gap": [16, 16]
+    },
+    "refresh": "15s",
+    "time": {
+      "from": "now-1h",
+      "to": "now"
+    }
+  }
+}