fix

apps/api/internal/processing/pipeline.go

@@ -548,11 +548,13 @@ func (p *Pipeline) RetryJob(ctx context.Context, companyID, id uuid.UUID) (Job,
 		firstStep = progress[0].Step
 	}
 
-	itemStatuses := []string{"failed", "cancelled"}
+	// Include "processing": ProcessJob can fail after claim (hydrate/etc.) and leave
+	// rows stuck in processing — RetryJob must reclaim them or the job never drains.
+	itemStatuses := []string{"failed", "cancelled", "processing"}
 	resetProcessed := false
 	if job.Status == "completed" {
 		// Completed retries should rerun the whole job, not immediately no-op.
-		itemStatuses = []string{"processed", "failed", "cancelled"}
+		itemStatuses = []string{"processed", "failed", "cancelled", "processing"}
 		resetProcessed = true
 	}
 
@@ -677,6 +679,7 @@ func (p *Pipeline) ProcessJob(ctx context.Context, jobID uuid.UUID) error {
 		items, err := p.loadPendingItems(ctx, jobID, batch)
 		if err != nil {
 			flushProgress(true)
+			_ = p.reclaimOrphanedProcessingItems(ctx, jobID)
 			return err
 		}
 		if len(items) == 0 {
@@ -687,24 +690,34 @@ func (p *Pipeline) ProcessJob(ctx context.Context, jobID uuid.UUID) error {
 			if err := p.Pool.QueryRow(ctx, `
 				SELECT COUNT(*) FROM processing_job_products
 				WHERE job_id = $1 AND processed_product_id IS NULL
-				  AND status IN ('pending', 'processing')`, jobID).Scan(&open); err != nil {
+					AND status IN ('pending', 'processing')`, jobID).Scan(&open); err != nil {
 				flushProgress(true)
+				_ = p.reclaimOrphanedProcessingItems(ctx, jobID)
 				return fmt.Errorf("processing: count open items job=%s: %w", jobID, err)
 			}
 			if open > 0 {
-				if _, err := p.Pool.Exec(ctx, `
+				ct, err := p.Pool.Exec(ctx, `
 					UPDATE processing_job_products
 					SET status = 'pending', error = NULL, updated_at = now()
-					WHERE job_id = $1 AND status = 'processing' AND processed_product_id IS NULL`, jobID); err != nil {
+					WHERE job_id = $1 AND status = 'processing' AND processed_product_id IS NULL`, jobID)
+				if err != nil {
 					flushProgress(true)
+					_ = p.reclaimOrphanedProcessingItems(ctx, jobID)
 					return fmt.Errorf("processing: reclaim fresh processing job=%s: %w", jobID, err)
 				}
+				// Fail closed: pending-only open that claim skipped must not spin forever.
+				if ct.RowsAffected() == 0 {
+					flushProgress(true)
+					_ = p.reclaimOrphanedProcessingItems(ctx, jobID)
+					return fmt.Errorf("processing: open items unclaimable job=%s open=%d", jobID, open)
+				}
 				continue
 			}
 			break
 		}
 		if err := p.hydrateJobItems(ctx, companyID, items); err != nil {
 			flushProgress(true)
+			_ = p.reclaimOrphanedProcessingItems(ctx, jobID)
 			return fmt.Errorf("processing: hydrate batch job=%s: %w", jobID, err)
 		}
 		creditsStop := false