fix(metrics): unify partial/full Squads webhook failure into single metric

Author: hash-envy

jito-bell/src/lib.rs

@@ -274,19 +274,24 @@ impl JitoBellHandler {
 
         if errors.is_empty() {
             Ok(())
-        } else if errors.len() < webhook_urls.len() {
-            self.epoch_metrics
-                .increment_squads_partial_webhook_failure();
-            warn!(
-                "dispatch_slack: partial webhook failure ({}/{} failed) for Squads notification",
-                errors.len(),
-                webhook_urls.len()
-            );
-            Ok(())
         } else {
-            Err(JitoBellError::Notification(
-                "All Squads Slack webhooks failed".to_string(),
-            ))
+            self.epoch_metrics.increment_squads_webhook_failure();
+            if errors.len() < webhook_urls.len() {
+                warn!(
+                    "dispatch_slack: partial webhook failure ({}/{} failed) for Squads notification",
+                    errors.len(),
+                    webhook_urls.len()
+                );
+                Ok(())
+            } else {
+                warn!(
+                    "dispatch_slack: full webhook failure ({} failed) for Squads notification",
+                    webhook_urls.len()
+                );
+                Err(JitoBellError::Notification(
+                    "All Squads Slack webhooks failed".to_string(),
+                ))
+            }
         }
     }
 

jito-bell/src/metrics.rs

@@ -19,8 +19,9 @@ pub(crate) struct SquadsMetrics {
     pub(crate) no_config: u64,
     /// dispatch_slack called but no webhook URLs resolved from the destination list.
     pub(crate) no_webhook: u64,
-    /// At least one webhook failed while at least one succeeded in the same dispatch.
-    pub(crate) partial_webhook_failure: u64,
+    /// A Squads webhook dispatch had at least one delivery failure.
+    /// This is the single alert surface for both partial and total webhook failure.
+    pub(crate) webhook_failure: u64,
     /// Total individual webhook HTTP/transport errors across all dispatches.
     pub(crate) webhook_errors: u64,
 }
@@ -116,9 +117,9 @@ impl EpochMetrics {
         self.emit_live_metric("jito-bell-squads-no-webhook", 1);
     }
 
-    pub fn increment_squads_partial_webhook_failure(&mut self) {
-        self.squads.partial_webhook_failure += 1;
-        self.emit_live_metric("jito-bell-squads-partial-webhook-failure", 1);
+    pub fn increment_squads_webhook_failure(&mut self) {
+        self.squads.webhook_failure += 1;
+        self.emit_live_metric("jito-bell-squads-webhook-failure", 1);
     }
 
     pub fn increment_squads_webhook_errors(&mut self) {