修复转图运行时重复启动与高频误重建问题

变更项：\n1. 修复 md2img 运行时并发启动竞争，新增启动中标记，避免同名线程被重复拉起。\n2. 优化浏览器心跳探测策略：由高频激进重建改为断连连续判定后再重建，降低误判。\n3. 新增截图进行中保护，截图期间心跳跳过探测，避免与业务并发导致误重建。\n4. 为浏览器重建增加 reason 日志字段，便于线上追踪重建触发原因。\n5. 保留截图后断连自愈能力，但改为更稳健的触发路径，减少无意义重建。

utils/markdown_to_image.py

@@ -523,6 +523,10 @@ class _PersistentBrowser:
         self._owner_loop_id: Optional[int] = None
         # 保活心跳任务：定期探测浏览器连通性，异常时自动重建。
         self._heartbeat_task: Optional[asyncio.Task] = None
+        # 心跳断连计数：避免单次抖动就触发重建。
+        self._disconnect_streak = 0
+        # 截图进行中标记：心跳期间若业务在跑，跳过本轮探测以避免误判。
+        self._capture_in_progress = False
 
     async def _launch_browser(self):
         if self._playwright is None:
@@ -581,7 +585,7 @@ class _PersistentBrowser:
             self._ensure_heartbeat_task()
             return self._browser
 
-    async def restart_browser(self):
+    async def restart_browser(self, reason: str = "unknown"):
         async with self._lock:
             if self._browser:
                 try:
@@ -591,10 +595,11 @@ class _PersistentBrowser:
                 self._browser = None
             self._browser = await self._launch_browser()
             self._owner_loop_id = id(asyncio.get_running_loop())
+            self._disconnect_streak = 0
             browser_pid = getattr(getattr(self._browser, "process", None), "pid", None)
             logger.info(
                 f"[md2img] 常驻浏览器已重建: source={self._last_launch_source}, "
-                f"loop={self._owner_loop_id}, pid={browser_pid}"
+                f"loop={self._owner_loop_id}, pid={browser_pid}, reason={reason}"
             )
             self._ensure_heartbeat_task()
             return self._browser
@@ -613,13 +618,24 @@ class _PersistentBrowser:
         """周期性探测浏览器可用性，断连后自动重建。"""
         while True:
             try:
-                await asyncio.sleep(10)
+                await asyncio.sleep(20)
                 # 没有浏览器实例时只保持心跳存活，不主动创建，避免空闲时不必要消耗。
                 if not self._browser:
+                    self._disconnect_streak = 0
                     continue
-                if not await self._is_browser_alive(self._browser, timeout_seconds=2.0):
-                    logger.warning("[md2img] 心跳探测发现浏览器已断连，准备自动重建")
-                    await self.restart_browser()
+                if self._capture_in_progress:
+                    # 截图期间跳过探测，避免与业务并发导致误判。
+                    continue
+                if self._browser and self._browser.is_connected():
+                    self._disconnect_streak = 0
+                    continue
+
+                self._disconnect_streak += 1
+                if self._disconnect_streak >= 3:
+                    logger.warning(
+                        f"[md2img] 心跳探测连续{self._disconnect_streak}次发现浏览器断连，准备自动重建"
+                    )
+                    await self.restart_browser(reason="heartbeat_disconnected")
             except asyncio.CancelledError:
                 raise
             except Exception as e:
@@ -635,8 +651,10 @@ class _PersistentBrowser:
         browser = await self.ensure_browser()
 
         async def _capture_with_browser(active_browser):
-            context = await active_browser.new_context(viewport={"width": 780, "height": 960}, device_scale_factor=1.2)
+            self._capture_in_progress = True
+            context = None
             try:
+                context = await active_browser.new_context(viewport={"width": 780, "height": 960}, device_scale_factor=1.2)
                 page = await context.new_page()
                 logger.debug("Set page content")
                 await page.set_content(html_content, wait_until='domcontentloaded', timeout=15000)
@@ -649,9 +667,11 @@ class _PersistentBrowser:
                     raise RuntimeError(f"截图失败，输出文件不存在: {output_image}")
             finally:
                 try:
-                    await context.close()
+                    if context:
+                        await context.close()
                 except Exception:
                     pass
+                self._capture_in_progress = False
 
         try:
             await _capture_with_browser(browser)
@@ -659,11 +679,11 @@ class _PersistentBrowser:
             # 在部分系统环境中，浏览器可能在任务完成后迅速断连，这里主动重建保证“常驻”语义。
             if not await self._is_browser_alive(browser, timeout_seconds=2.0):
                 logger.warning("[md2img] 截图后浏览器已断连，立即执行自动重建")
-                await self.restart_browser()
+                await self.restart_browser(reason="post_capture_disconnected")
         except Exception as e:
             # 首次失败后重建一次浏览器再重试，提升抗偶发故障能力。
             logger.warning(f"[md2img] 常驻浏览器截图失败，准备重建后重试: {e}")
-            browser = await self.restart_browser()
+            browser = await self.restart_browser(reason="capture_exception_retry")
             await _capture_with_browser(browser)
 
 
@@ -685,6 +705,8 @@ class _Md2ImgRuntime:
         self._loop: Optional[asyncio.AbstractEventLoop] = None
         self._lock = threading.Lock()
         self._ready = threading.Event()
+        # 启动中标记：避免并发调用 ensure_started 时重复创建线程。
+        self._starting = False
 
     @property
     def loop(self) -> Optional[asyncio.AbstractEventLoop]:
@@ -706,11 +728,21 @@ class _Md2ImgRuntime:
         with self._lock:
             if self._thread and self._thread.is_alive() and self._loop and self._loop.is_running():
                 return
-            self._ready.clear()
-            self._thread = threading.Thread(target=self._thread_main, name="md2img-runtime", daemon=True)
-            self._thread.start()
-            if not self._ready.wait(timeout=10):
-                raise RuntimeError("md2img 专用运行时启动超时")
+            if self._starting:
+                # 已有其他调用在启动中，当前线程等待启动完成即可。
+                pass
+            else:
+                self._starting = True
+                self._ready.clear()
+                self._thread = threading.Thread(target=self._thread_main, name="md2img-runtime", daemon=True)
+                self._thread.start()
+        # 注意：等待动作放到锁外，避免阻塞其他读取逻辑。
+        if not self._ready.wait(timeout=10):
+            with self._lock:
+                self._starting = False
+            raise RuntimeError("md2img 专用运行时启动超时")
+        with self._lock:
+            self._starting = False
 
     def submit(self, coro) -> ConcurrentFuture:
         """向专用运行时提交协程任务。"""