解决极速成片问题

core/backend/apps/projects/services/quick_create.py

@@ -423,15 +423,23 @@ def _advance_script(job: QuickCreateJob) -> int | None:
     if not metadata.get("script_started"):
         from apps.projects.tasks import run_quick_script_task
 
+        from apps.common.celery_health import worker_consumes_queue
+
         metadata["script_started"] = True
         metadata["script_started_at"] = timezone.now().isoformat()
+        listens = worker_consumes_queue("airshelf.quick")
+        if listens is False:
+            metadata["script_local"] = True
         _save_job(
             job,
             metadata=metadata,
             progress=28,
             message="正在根据商品名称与图片生成带货脚本",
         )
-        run_quick_script_task.apply_async(args=[str(job.id)], queue="airshelf.quick")
+        if listens is False:
+            _run_quick_script_in_thread(str(job.id))
+        else:
+            run_quick_script_task.apply_async(args=[str(job.id)], queue="airshelf.quick")
         return SCRIPT_POLL_SECONDS
 
     latest = _latest_script_task(job.project)
@@ -946,12 +954,26 @@ def _run_quick_script_in_thread(job_id: str) -> None:
 
 
 def _enqueue_advance(job: QuickCreateJob) -> None:
+    from apps.common.celery_health import worker_consumes_queue
     from apps.projects.tasks import advance_quick_create_task
 
     job_id = str(job.id)
+    if worker_consumes_queue("airshelf.quick") is False:
+        _advance_without_quick_queue(job_id)
+        return
     try:
         advance_quick_create_task.apply_async(args=[job_id], queue="airshelf.quick")
     except Exception:  # noqa: BLE001 — 队列不可用时就地推进一步
+        _advance_without_quick_queue(job_id)
+
+
+def _advance_without_quick_queue(job_id: str) -> None:
+    """部署后的 worker 若只听 celery 队列，airshelf.quick 里的任务会永远没人拿。"""
+    delay = advance_quick_create(job_id)
+    job = QuickCreateJob.objects.filter(id=job_id).first()
+    if job is None or delay is None:
+        return
+    if job.phase == QuickCreateJob.Phase.SCRIPT and not (job.metadata or {}).get("script_started"):
         advance_quick_create(job_id)
 
 
@@ -1028,6 +1050,13 @@ def recover_quick_create(job: QuickCreateJob) -> None:
         return
     if _is_finished(job):
         return
+    if job.status == QuickCreateJob.Status.QUEUED or job.phase == QuickCreateJob.Phase.PRODUCT:
+        if timezone.now() - job.updated_at > STALE_AFTER:
+            try:
+                _advance_without_quick_queue(str(job.id))
+            except Exception:  # noqa: BLE001 — 恢复失败不能把进度接口打成 500
+                logger.exception("quick create recover advance failed for job %s", job.id)
+            return
     job_id = str(job.id)
     metadata = dict(job.metadata or {})
     if job.phase == QuickCreateJob.Phase.SCRIPT and metadata.get("script_started"):