auto-sync: 2026-05-23 00:40:00

src/daemon/spawner.py

@@ -558,12 +558,33 @@ curl -X POST http://{api_host}:{api_port}/api/projects/{project_id}/tasks/{task_
     async def _do_retry(self, session_id, agent_id, task_id, on_complete,
                         db_path, retry_field="retry_count"):
         """续杯：用同一 session_id 再 spawn 一次"""
-        retry_counts = self._get_retry_counts(db_path, task_id)
-        count = retry_counts.get(retry_field, 0) + 1
-
-        # 更新计数器并写回最新 attempt 的 metadata
-        retry_counts[retry_field] = count
-        self._update_retry_counts(db_path, task_id, retry_counts)
+        # 直接读写 tasks 表的 retry_count（广播场景下所有 Agent 共享同一 tasks 记录）
+        # task_attempts metadata 的 retry_count 不可靠（多 Agent 互相覆盖）
+        if retry_field == "retry_count" and db_path and task_id:
+            try:
+                conn = get_connection(db_path)
+                try:
+                    conn.execute("BEGIN IMMEDIATE")
+                    conn.execute(
+                        "UPDATE tasks SET retry_count = COALESCE(retry_count, 0) + 1 WHERE id=?",
+                        (task_id,),
+                    )
+                    conn.commit()
+                    row = conn.execute(
+                        "SELECT retry_count FROM tasks WHERE id=?", (task_id,)
+                    ).fetchone()
+                    count = row["retry_count"] if row else 1
+                finally:
+                    conn.close()
+            except Exception:
+                logger.exception("Failed to update retry_count for task %s", task_id)
+                count = 1
+        else:
+            # 非 retry_count 的计数器（connect/api/lock）仍用 task_attempts metadata
+            retry_counts = self._get_retry_counts(db_path, task_id)
+            count = retry_counts.get(retry_field, 0) + 1
+            retry_counts[retry_field] = count
+            self._update_retry_counts(db_path, task_id, retry_counts)
 
         if count >= self.max_retries:
             logger.error("Agent %s max retries (session=%s, %s=%d)",