celery异步恢复任务独立队列|扩增celery子进程连接池上限配置

video-gen-api/app/tasks/generation_poll_tasks.py

@@ -194,7 +194,8 @@ async def _run(task_id: str):
             await remove_poll_active(task.id)
             return
 
-        if _deadline_expired(task):
+        final_poll_before_timeout = _deadline_expired(task)
+        if final_poll_before_timeout and not (task.seedance_task_id or task.provider_task_id):
             await _mark_timeout(db, task, message="任务轮询超时")
             return
 
@@ -202,6 +203,14 @@ async def _run(task_id: str):
             await _mark_failed(db, task, message="缺少外部任务ID")
             return
 
+        if final_poll_before_timeout:
+            await log_task_event(
+                task,
+                event_type="FINAL_POLL_BEFORE_TIMEOUT",
+                message="任务已到 deadline，执行最后一次供应商查询后再判定超时",
+                detail={"deadline_at": task.deadline_at, "stage": task.pipeline_stage},
+            )
+
         # 标记本次正在轮询，并登记 poll lease。
         # 如果 worker 在供应商接口调用过程中退出，启动恢复会在 lease 过期后重新投递。
         task.pipeline_stage = "polling"
@@ -274,6 +283,16 @@ async def _run(task_id: str):
                 )
                 return
 
+            if final_poll_before_timeout:
+                await log_task_event(
+                    task,
+                    event_type="FINAL_POLL_BEFORE_TIMEOUT_PENDING",
+                    message=f"最终查询后供应商仍未完成，按超时处理。status={status}",
+                    detail=poll_result,
+                )
+                await _mark_timeout(db, task, message="任务轮询超时")
+                return
+
             # 供应商仍在 pending / running 时，把阶段从 polling 改回 waiting_remote。
             # 同时登记下一次 poll active，Celery countdown 丢失时可由恢复任务拉起。
             task.pipeline_stage = "waiting_remote"
@@ -309,6 +328,15 @@ async def _run(task_id: str):
                 await remove_poll_active(task_id)
                 return
 
+            if final_poll_before_timeout:
+                await log_task_event(
+                    task,
+                    event_type="FINAL_POLL_BEFORE_TIMEOUT_ERROR",
+                    message=str(exc),
+                )
+                await _mark_timeout(db, task, message="任务轮询超时")
+                return
+
             task.retry_count = (task.retry_count or 0) + 1
 
             if task.retry_count > settings.CHATAPI_ASYNC_MAX_RETRIES:
@@ -339,7 +367,13 @@ async def _run(task_id: str):
 if celery_app:
     @celery_app.task(name="generation.poll_generation_task", bind=True, max_retries=3, default_retry_delay=30)
     def poll_generation_task(self, task_id: str):
-        return run_async(_run(task_id))
+        try:
+            return run_async(_run(task_id))
+        except Exception as exc:
+            # 只重试基础设施异常；供应商失败/业务失败已在 _run 内处理。
+            retries = int(getattr(self.request, "retries", 0) or 0) + 1
+            countdown = int(settings.CHATAPI_ASYNC_RETRY_BACKOFF_SECONDS or 30) * max(1, retries)
+            raise self.retry(exc=exc, countdown=countdown)
 else:
     class _DisabledTask:
         def delay(self, *args, **kwargs):