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"):

core/backend/apps/projects/test_quick_create.py

@@ -48,7 +48,7 @@ class QuickCreateApiTests(TestCase):
 
     @patch("apps.projects.services.quick_create.get_quick_script_model", return_value=object())
     @patch("apps.projects.views.get_default_model")
-    @patch("apps.projects.views.advance_quick_create_task.apply_async")
+    @patch("apps.projects.tasks.advance_quick_create_task.apply_async")
     @patch("apps.projects.views.require_worker_task")
     @patch("apps.projects.views._store_uploaded_asset")
     def test_submit_creates_product_project_and_persistent_job(self, store_asset, require_worker_task, enqueue, get_model, get_quick_model):
@@ -106,7 +106,7 @@ class QuickCreateApiTests(TestCase):
 
     @patch("apps.projects.services.quick_create.get_quick_script_model", return_value=object())
     @patch("apps.projects.views.get_default_model", return_value=object())
-    @patch("apps.projects.views.advance_quick_create_task.apply_async")
+    @patch("apps.projects.tasks.advance_quick_create_task.apply_async")
     @patch("apps.projects.views.require_worker_task")
     @patch("apps.projects.views._store_uploaded_asset")
     def test_submit_reuses_source_product_images(self, store_asset, require_worker_task, enqueue, get_model, get_quick_model):
@@ -447,6 +447,32 @@ class QuickCreateCoordinatorTests(TestCase):
         self.assertEqual(self.job.status, QuickCreateJob.Status.RUNNING)
         run_local.assert_called_once_with(str(self.job.id))
 
+    def test_recover_starts_stale_queued_job_without_waiting_for_quick_queue(self):
+        QuickCreateJob.objects.filter(id=self.job.id).update(
+            status=QuickCreateJob.Status.QUEUED,
+            phase=QuickCreateJob.Phase.PRODUCT,
+            message="等待开始极速成片",
+            updated_at=timezone.now() - timedelta(seconds=30),
+        )
+        self.job.refresh_from_db()
+        with patch("apps.projects.services.quick_create._advance_without_quick_queue") as advance:
+            recover_quick_create(self.job)
+        advance.assert_called_once_with(str(self.job.id))
+
+    @patch("apps.common.celery_health.worker_consumes_queue", return_value=False)
+    @patch("apps.projects.services.quick_create._run_quick_script_in_thread")
+    def test_advance_script_runs_locally_when_quick_queue_has_no_consumer(self, run_local, _listens):
+        self.job.status = QuickCreateJob.Status.RUNNING
+        self.job.phase = QuickCreateJob.Phase.SCRIPT
+        self.job.save(update_fields=["status", "phase", "updated_at"])
+
+        delay = advance_quick_create(str(self.job.id))
+        self.job.refresh_from_db()
+        self.assertEqual(delay, 5)
+        self.assertTrue(self.job.metadata.get("script_started"))
+        self.assertTrue(self.job.metadata.get("script_local"))
+        run_local.assert_called_once_with(str(self.job.id))
+
     @patch("apps.projects.services.quick_create._run_quick_script_in_thread")
     def test_recover_does_not_steal_script_when_model_call_is_inflight(self, run_local):
         self._script_task(AITask.Status.SUBMITTED, key="script-inflight-1")

core/backend/apps/projects/views.py

@@ -90,7 +90,7 @@ from .services.pipeline import (
 )
 from .services.script_import import ScriptFileError, extract_script_text
 from .services.templates import build_template_fields, coerce_persona, coerce_template_combo, render_outline_text
-from .tasks import advance_quick_create_task, poll_video_segment_task
+from .tasks import poll_video_segment_task
 
 logger = logging.getLogger(__name__)
 
@@ -661,7 +661,9 @@ class ProjectViewSet(TeamScopedViewSetMixin, ModelViewSet):
             )
 
         try:
-            advance_quick_create_task.apply_async(args=[str(job.id)], queue="airshelf.quick")
+            from .services.quick_create import _enqueue_advance
+
+            _enqueue_advance(job)
         except Exception as exc:  # noqa: BLE001 — broker 极小窗口失败也必须给任务落终态
             from .services.quick_create import fail_quick_create