临时提交

Utils/ThreadManager.py

@@ -67,13 +67,6 @@ class ThreadManager:
 
     @classmethod
     def batch_stop(cls, udids: List[str]) -> Tuple[int, List[str], str]:
-        """
-        批量停止任务——瞬间下发停止信号，仍统计失败结果。
-        返回格式与原接口 100% 兼容：
-        (code, fail_list, msg)
-            code=200   全部成功
-            code=1001  部分/全部失败
-        """
         if not udids:
             return 200, [], "无设备需要停止"
 
@@ -83,48 +76,53 @@ class ThreadManager:
         with cls._lock:
             for udid in udids:
                 task = cls._tasks.get(udid)
-                if not task or not task.get("running"):
-                    # fail_list.append(f"设备{udid}停止失败：当前设备没有执行相关任务")
-                    continue
-                task["event"].set()  # 下发停止信号
+                if task and task.get("running"):
+                    task["event"].set()
 
-        # ---------- 2. 并发等 0.2 s 收尾 ----------
-        def _wait_and_clean(udid: str) -> Tuple[int, str]:
-            with cls._lock:
-                task = cls._tasks.get(udid)
-                if not task:
-                    return 400, "任务记录已丢失"
-                thread = task["thread"]
-            # 第一次等 3 秒，让“分片睡眠”有机会退出
-            thread.join(timeout=3)
-            # 如果还活，再补 2 秒
-            if thread.is_alive():
-                thread.join(timeout=2)
-            # 最终仍活，记录日志但不硬杀，避免僵尸
-            with cls._lock:
-                cls._tasks.pop(udid, None)
-            if thread.is_alive():
-                LogManager.warning(f"[batch_stop] 线程 5s 未退出，已清理记录但线程仍跑 {udid}")
-                return 201, "已下发停止，线程超长任务未立即结束"
-            return 200, "已停止"
+        # ---------- 2. 单设备重试 5 次 ----------
+        def _wait_and_clean_with_retry(udid: str) -> Tuple[int, List[str], str]:
+            """
+            内部重试 5 次，最终返回格式与 batch_stop 完全一致：
+            (code, fail_list, msg)   fail_list 空表示成功
+            """
+            for attempt in range(1, 6):
+                with cls._lock:
+                    task = cls._tasks.get(udid)
+                    if not task:
+                        return 200, [], "任务已不存在"  # 成功
+                    thread = task["thread"]
 
+                thread.join(timeout=5)
+                still_alive = thread.is_alive()
+
+                # 最后一次重试才清理记录
+                if attempt == 5:
+                    with cls._lock:
+                        cls._tasks.pop(udid, None)
+
+                if not still_alive:
+                    return 200, [], "已停止"
+
+                LogManager.warning(f"[batch_stop] {udid} 第{attempt}/5 次停止未立即结束，重试", udid)
+
+            # 5 次都失败
+            LogManager.error(f"[batch_stop] {udid} 停止失败（5 次重试后线程仍活）", udid)
+            return 1001, [udid], "部分设备停止失败"
+
+        # ---------- 3. 并发执行 ----------
         with ThreadPoolExecutor(max_workers=min(32, len(udids))) as executor:
-            future_map = {executor.submit(_wait_and_clean, udid): udid for udid in udids}
+            future_map = {executor.submit(_wait_and_clean_with_retry, udid): udid for udid in udids}
             for future in as_completed(future_map):
                 udid = future_map[future]
-                try:
-                    code, reason = future.result()
-                    if code != 200:
-                        fail_list.append(f"设备{udid}停止失败：{reason}")
-                except Exception as exc:
-                    fail_list.append(f"设备{udid}停止异常：{exc}")
+                code, sub_fail_list, sub_msg = future.result()  # ← 现在解包正确
+                if code != 200:
+                    fail_list.extend(sub_fail_list)  # 收集失败 udid
 
-        # ---------- 3. 返回兼容格式 ----------
+        # ---------- 4. 返回兼容格式 ----------
         if fail_list:
             return 1001, fail_list, "部分设备停止失败"
         return 200, [], "全部设备停止成功"
 
-
     @classmethod
     def is_task_running(cls, udid: str) -> bool:
         """