Update app.py

app.py

@@ -544,61 +544,160 @@ async def on_disconnect():
 @bot.event
 async def on_resumed():
     logger.info("✅ [Gateway] Session resumed successfully.")
-
 # ==============================================================================
-# 12. 启动入口 — 关键修复区域
+# 12. 启动入口 — DNS 就绪探测 + 安全重试
 # ==============================================================================
+import socket
+
+def wait_for_network(host="discord.com", port=443, timeout=120):
+    """
+    在容器启动后等待 DNS 解析就绪。
+    HuggingFace Spaces 容器冷启动时网络栈可能延迟 10-30 秒。
+    """
+    start = time.time()
+    attempt = 0
+    while time.time() - start < timeout:
+        attempt += 1
+        try:
+            socket.setdefaulttimeout(10)
+            socket.getaddrinfo(host, port)
+            logger.info(f"✅ [Network] DNS resolved {host} successfully (attempt {attempt})")
+            return True
+        except socket.gaierror as e:
+            logger.warning(f"⏳ [Network] DNS not ready (attempt {attempt}): {e}")
+            time.sleep(5)
+        except Exception as e:
+            logger.warning(f"⏳ [Network] Check failed (attempt {attempt}): {e}")
+            time.sleep(5)
+    logger.error(f"❌ [Network] DNS resolution failed after {timeout}s")
+    return False
+
+
 def run_bot():
     """
-    [核心修复] bot.run() 内部调用 asyncio.run()，会注册信号处理器，
-    但信号处理器只能在主线程注册。在子线程中必须：
-      1. 手动创建事件循环 asyncio.new_event_loop()
-      2. 使用 bot.start()（协程版本）而非 bot.run()
+    [核心修复]
+    1. bot.start() 替代 bot.run()：避免子线程信号处理器崩溃
+    2. DNS 就绪探测：等待 HuggingFace 容器网络栈初始化完成
+    3. 安全重试：DNS 暂时失败时自动等待重连
     """
     if not DISCORD_TOKEN:
-        logger.critical("❌ [FATAL] DISCORD_TOKEN is missing! Set it in HuggingFace Secrets.")
+        logger.critical("❌ [FATAL] DISCORD_TOKEN is missing!")
         return
 
-    logger.info("🚀 [Bot Thread] Initializing Discord connection...")
+    # ====== 第一步：等待网络就绪 ======
+    logger.info("🌐 [Bot Thread] Waiting for network/DNS to be ready...")
+    if not wait_for_network():
+        logger.critical("❌ [FATAL] Network never became available. Bot cannot start.")
+        return
 
-    # ====== 关键：为子线程创建专属事件循环 ======
-    loop = asyncio.new_event_loop()
-    asyncio.set_event_loop(loop)
+    # ====== 第二步：带重试的连接 ======
+    max_retries = 5
+    for attempt in range(1, max_retries + 1):
+        logger.info(f"🚀 [Bot Thread] Connection attempt {attempt}/{max_retries}...")
+
+        loop = asyncio.new_event_loop()
+        asyncio.set_event_loop(loop)
 
-    try:
-        # bot.start() 是纯协程，不会触发信号注册
-        loop.run_until_complete(bot.start(DISCORD_TOKEN, reconnect=True))
-    except discord.LoginFailure:
-        logger.critical(
-            "❌ [FATAL] DISCORD_TOKEN is invalid!\n"
-            "Go to https://discord.com/developers/applications → Bot → Reset Token"
-        )
-    except discord.PrivilegedIntentsRequired:
-        logger.critical(
-            "❌ [FATAL] Privileged Intents not enabled!\n"
-            "Go to Discord Developer Portal → Bot tab →\n"
-            "Enable 'SERVER MEMBERS INTENT' and 'MESSAGE CONTENT INTENT'"
-        )
-    except Exception as e:
-        logger.error(f"❌ [Bot Fatal Error] {e}")
-        traceback.print_exc()
-    finally:
-        logger.warning("🔴 [Bot Thread] Event loop ending, cleaning up...")
         try:
-            # 清理所有未完成的任务
-            pending = asyncio.all_tasks(loop)
-            for task in pending:
-                task.cancel()
-            # 等待取消完成
-            if pending:
-                loop.run_until_complete(asyncio.gather(*pending, return_exceptions=True))
-            if not bot.is_closed():
-                loop.run_until_complete(bot.close())
-        except Exception as cleanup_err:
-            logger.error(f"Cleanup error: {cleanup_err}")
+            loop.run_until_complete(bot.start(DISCORD_TOKEN, reconnect=True))
+            # 如果 bot.start() 正常返回（比如被手动 close），跳出循环
+            break
+
+        except discord.LoginFailure:
+            logger.critical("❌ [FATAL] DISCORD_TOKEN is invalid! Check your token.")
+            break  # Token 错误不重试
+
+        except discord.PrivilegedIntentsRequired:
+            logger.critical(
+                "❌ [FATAL] Privileged Intents not enabled!\n"
+                "→ Discord Developer Portal → Bot tab\n"
+                "→ Enable 'SERVER MEMBERS INTENT' and 'MESSAGE CONTENT INTENT'"
+            )
+            break  # 权限错误不重试
+
+        except (OSError, ConnectionError, asyncio.TimeoutError) as e:
+            # 网络类错误：值得重试
+            logger.error(f"⚠️ [Bot] Network error on attempt {attempt}: {e}")
+
+        except Exception as e:
+            logger.error(f"❌ [Bot] Unexpected error on attempt {attempt}: {e}")
+            traceback.print_exc()
+
         finally:
-            loop.close()
-        logger.error("🔴 [Bot Thread] Discord bot has stopped.")
+            # 清理本次循环的事件循环
+            try:
+                pending = asyncio.all_tasks(loop)
+                for task in pending:
+                    task.cancel()
+                if pending:
+                    loop.run_until_complete(asyncio.gather(*pending, return_exceptions=True))
+                if not bot.is_closed():
+                    loop.run_until_complete(bot.close())
+            except Exception:
+                pass
+            finally:
+                loop.close()
+
+        # 重试等待（递增延迟）
+        if attempt < max_retries:
+            wait_time = min(30 * attempt, 120)
+            logger.info(f"⏳ [Bot] Waiting {wait_time}s before retry...")
+            time.sleep(wait_time)
+
+            # 重试前再次确认 DNS
+            if not wait_for_network(timeout=60):
+                logger.error("❌ [Bot] Network still unavailable, retrying anyway...")
+
+            # 重建 bot 实例（因为 close() 后无法复用）
+            _rebuild_bot()
+
+    logger.error("🔴 [Bot Thread] Bot has stopped after all attempts.")
+
+
+def _rebuild_bot():
+    """
+    重建 bot 实例并重新绑定所有命令和事件。
+    bot.close() 后内部 session 被销毁，必须创建新实例。
+    """
+    global bot
+    old_commands = {cmd.name: cmd for cmd in bot.commands}
+    old_listeners = {}
+
+    # 保存已注册的事件监听器
+    for event_name in ['on_ready', 'on_disconnect', 'on_resumed']:
+        listener = getattr(bot, event_name, None)
+        if listener:
+            old_listeners[event_name] = listener
+
+    # 创建新实例
+    bot = commands.Bot(command_prefix='!', intents=intents)
+
+    # 恢复命令
+    for name, cmd in old_commands.items():
+        try:
+            bot.add_command(cmd)
+        except Exception:
+            pass
+
+    # 重新注册事件
+    @bot.event
+    async def on_ready():
+        if not auto_backup.is_running():
+            auto_backup.start()
+        await bot.change_presence(
+            activity=discord.Activity(type=discord.ActivityType.watching, name="HuggingFace Node")
+        )
+        logger.info(f"✅ [Online] BK-GTA V62.9-Titan Live. Session: {SESSION_ID}")
+
+    @bot.event
+    async def on_disconnect():
+        logger.warning("⚠️ [Gateway] Disconnected. discord.py will auto-reconnect...")
+
+    @bot.event
+    async def on_resumed():
+        logger.info("✅ [Gateway] Session resumed successfully.")
+
+    logger.info("🔄 [Bot] Instance rebuilt with all commands restored.")
 
 
 if __name__ == "__main__":
@@ -615,10 +714,7 @@ if __name__ == "__main__":
     bot_thread.start()
     logger.info("📡 [Main] Bot thread launched.")
 
-    # 2. 给 bot 线程一点启动时间
-    time.sleep(3)
-
-    # 3. 启动 Gradio 前台哨兵 (占据主线程)
+    # 2. 启动 Gradio 前台哨兵
     logger.info("📡 [Main] Starting Gradio on port 7860...")
     iface = gr.Interface(
         fn=get_status,