feat: add available data services metric

Author: JuArce

crates/batcher/src/lib.rs

@@ -1911,6 +1911,9 @@ impl Batcher {
             warn!("Failed to upload batch to secondary S3");
         }
 
+        // Update metrics with number of available data services
+        self.metrics.available_data_services.set(successful_urls.len() as i64);
+        
         // If no uploads succeeded, return error
         if successful_urls.is_empty() {
             error!("Failed to upload batch to both S3 buckets");

crates/batcher/src/metrics.rs

@@ -27,6 +27,7 @@ pub struct BatcherMetrics {
     pub cancel_create_new_task_duration: IntGauge,
     pub batcher_gas_cost_create_task_total: GenericCounter<AtomicF64>,
     pub batcher_gas_cost_cancel_task_total: GenericCounter<AtomicF64>,
+    pub available_data_services: IntGauge,
 }
 
 impl BatcherMetrics {
@@ -79,6 +80,10 @@ impl BatcherMetrics {
                 "batcher_gas_cost_cancel_task_total",
                 "Batcher Gas Cost Cancel Task Total"
             ))?;
+        let available_data_services = register_int_gauge!(opts!(
+            "available_data_services",
+            "Number of available data services (0-2)"
+        ))?;
 
         registry.register(Box::new(open_connections.clone()))?;
         registry.register(Box::new(received_proofs.clone()))?;
@@ -96,6 +101,7 @@ impl BatcherMetrics {
         registry.register(Box::new(cancel_create_new_task_duration.clone()))?;
         registry.register(Box::new(batcher_gas_cost_create_task_total.clone()))?;
         registry.register(Box::new(batcher_gas_cost_cancel_task_total.clone()))?;
+        registry.register(Box::new(available_data_services.clone()))?;
 
         let metrics_route = warp::path!("metrics")
             .and(warp::any().map(move || registry.clone()))
@@ -124,6 +130,7 @@ impl BatcherMetrics {
             cancel_create_new_task_duration,
             batcher_gas_cost_create_task_total,
             batcher_gas_cost_cancel_task_total,
+            available_data_services,
         })
     }
 

grafana/provisioning/dashboards/aligned/aggregator_batcher.json

@@ -18,7 +18,7 @@
   "editable": true,
   "fiscalYearStartMonth": 0,
   "graphTooltip": 0,
-  "id": 2,
+  "id": 4,
   "links": [],
   "liveNow": false,
   "panels": [
@@ -1786,6 +1786,84 @@
       "title": "Total Batcher Restarts",
       "type": "stat"
     },
+    {
+      "datasource": {
+        "type": "prometheus",
+        "uid": "prometheus"
+      },
+      "description": "",
+      "fieldConfig": {
+        "defaults": {
+          "color": {
+            "mode": "thresholds"
+          },
+          "mappings": [],
+          "max": 2,
+          "min": 0,
+          "thresholds": {
+            "mode": "absolute",
+            "steps": [
+              {
+                "color": "green",
+                "value": null
+              },
+              {
+                "color": "red",
+                "value": 0
+              },
+              {
+                "color": "orange",
+                "value": 1
+              },
+              {
+                "color": "green",
+                "value": 2
+              }
+            ]
+          },
+          "unit": "none"
+        },
+        "overrides": []
+      },
+      "gridPos": {
+        "h": 5,
+        "w": 3,
+        "x": 17,
+        "y": 25
+      },
+      "id": 64,
+      "options": {
+        "colorMode": "value",
+        "graphMode": "area",
+        "justifyMode": "auto",
+        "orientation": "auto",
+        "reduceOptions": {
+          "calcs": [
+            "lastNotNull"
+          ],
+          "fields": "",
+          "values": false
+        },
+        "textMode": "auto"
+      },
+      "pluginVersion": "10.1.10",
+      "targets": [
+        {
+          "datasource": {
+            "type": "prometheus",
+            "uid": "prometheus"
+          },
+          "editorMode": "code",
+          "expr": "available_data_services{job=\"aligned-batcher\"}",
+          "instant": false,
+          "legendFormat": "__auto",
+          "range": true,
+          "refId": "A"
+        }
+      ],
+      "title": "Available Data Services",
+      "type": "stat"
+    },
     {
       "datasource": {
         "type": "prometheus",
@@ -3929,7 +4007,8 @@
             "mode": "absolute",
             "steps": [
               {
-                "color": "green"
+                "color": "green",
+                "value": null
               }
             ]
           },
@@ -4024,7 +4103,8 @@
             "mode": "absolute",
             "steps": [
               {
-                "color": "green"
+                "color": "green",
+                "value": null
               },
               {
                 "color": "red",
@@ -4120,7 +4200,8 @@
             "mode": "absolute",
             "steps": [
               {
-                "color": "green"
+                "color": "green",
+                "value": null
               },
               {
                 "color": "red",
@@ -4251,7 +4332,8 @@
             "mode": "absolute",
             "steps": [
               {
-                "color": "green"
+                "color": "green",
+                "value": null
               },
               {
                 "color": "red",
@@ -4301,21 +4383,21 @@
       "type": "timeseries"
     }
   ],
-  "refresh": "",
+  "refresh": "5s",
   "schemaVersion": 38,
   "style": "dark",
   "tags": [],
   "templating": {
     "list": []
   },
   "time": {
-    "from": "now-1h",
+    "from": "now-15m",
     "to": "now"
   },
   "timepicker": {},
   "timezone": "browser",
   "title": "System Data",
   "uid": "aggregator",
-  "version": 6,
+  "version": 18,
   "weekStart": ""
-}
+}