fix: MetricsCollectorService does not track APPROVED or retry-pending batches

2026-06-27 20:58:57 +08:00
parent 903cd54a51
commit 5db60f46eb
3 changed files with 47 additions and 4 deletions
@@ -39,7 +39,8 @@ public class MetricsCollectorService : BackgroundService
        BatchStatus.Verified,
        BatchStatus.AwaitingClinicalApproval,
        BatchStatus.Approved,
-        BatchStatus.Promoted
+        BatchStatus.Promoted,
+        BatchStatus.Cancelled
    };

    public MetricsCollectorService(
@@ -121,11 +122,41 @@ public class MetricsCollectorService : BackgroundService
            DiagnosticsMetrics.QueueAgeSeconds.Set(0);
        }

+        // --- Promotion retry gauges ---
+        var pendingRetries = await db.PromotionAttempts
+            .AsNoTracking()
+            .CountAsync(a => !a.Succeeded && a.NextRetryAt != null, ct);
+        DiagnosticsMetrics.PromotionPendingRetries.Set(pendingRetries);
+
+        var exhaustedCount = await db.DigitizationBatches
+            .AsNoTracking()
+            .Where(b => b.Status == BatchStatus.Approved)
+            .Where(b => db.PromotionAttempts
+                .Any(a => a.BatchId == b.Id && !a.Succeeded && a.NextRetryAt == null))
+            .CountAsync(ct);
+        DiagnosticsMetrics.PromotionExhaustedTotal.Set(exhaustedCount);
+
+        // --- Approval queue age: oldest APPROVED batch ---
+        var oldestApprovedUpdatedAt = await db.DigitizationBatches
+            .AsNoTracking()
+            .Where(b => b.Status == BatchStatus.Approved)
+            .OrderBy(b => b.UpdatedAt)
+            .Select(b => (DateTimeOffset?)b.UpdatedAt)
+            .FirstOrDefaultAsync(ct);
+
+        DiagnosticsMetrics.ApprovalQueueAgeSeconds.Set(
+            oldestApprovedUpdatedAt.HasValue
+                ? (DateTimeOffset.UtcNow - oldestApprovedUpdatedAt.Value).TotalSeconds
+                : 0);
+
        _logger.LogDebug(
-            "Metrics collected: {StatusCount} status groups, queue age {QueueAge}s",
+            "Metrics collected: {StatusCount} status groups, queue age {QueueAge}s, " +
+            "pending retries {PendingRetries}, exhausted {Exhausted}",
            statusCounts.Count,
            oldestPendingUpdatedAt.HasValue
                ? (DateTimeOffset.UtcNow - oldestPendingUpdatedAt.Value).TotalSeconds
-                : 0);
+                : 0,
+            pendingRetries,
+            exhaustedCount);
    }
 }