diff --git a/crates/zclaw-saas/src/scheduler.rs b/crates/zclaw-saas/src/scheduler.rs
index 17184eb..cbb4201 100644
--- a/crates/zclaw-saas/src/scheduler.rs
+++ b/crates/zclaw-saas/src/scheduler.rs
@@ -82,6 +82,7 @@ pub fn start_scheduler(config: &SchedulerConfig, _db: PgPool, dispatcher: Worker
 pub fn start_db_cleanup_tasks(db: PgPool) {
     let db_devices = db.clone();
     let db_key_pool = db.clone();
+    let db_relay = db.clone();
 
     // 每 24 小时清理不活跃设备
     tokio::spawn(async move {
@@ -128,6 +129,28 @@ pub fn start_db_cleanup_tasks(db: PgPool) {
             }
         }
     });
+
+    // 每 5 分钟清理超时的 relay_tasks（status=processing 且 updated_at 超过 10 分钟）
+    tokio::spawn(async move {
+        let mut interval = tokio::time::interval(Duration::from_secs(300));
+        loop {
+            interval.tick().await;
+            match sqlx::query(
+                "UPDATE relay_tasks SET status = 'failed', error_message = 'timeout: upstream not responding', completed_at = NOW() \
+                 WHERE status = 'processing' AND updated_at < NOW() - INTERVAL '10 minutes'"
+            )
+            .execute(&db_relay)
+            .await
+            {
+                Ok(result) => {
+                    if result.rows_affected() > 0 {
+                        tracing::warn!("Cleaned up {} timed-out relay tasks (>10m processing)", result.rows_affected());
+                    }
+                }
+                Err(e) => tracing::error!("Relay task timeout cleanup failed: {}", e),
+            }
+        }
+    });
 }
 
 /// 用户任务调度器