refactor: 切换 Celery broker 为火山引擎 Redis + 僵尸任务自动恢复

- Redis 从阿里云切换到火山引擎（同区域低延迟）
- delay() 失败改为 warning 日志 + 重试一次（不再静默吞异常）
- 新增 recover_stuck_tasks 定时任务，每10分钟扫描卡住的任务重新派发
- 轮询时每次 touch updated_at，防止活跃任务被误判为僵尸
- Celery worker 启用内嵌 Beat 调度器（-B 参数）

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

backend/apps/generation/tasks.py

@@ -79,10 +79,9 @@ def poll_video_task(self, record_id):
             continue  # retry on next interval
 
         if new_status in ('queued', 'processing'):
-            # Still running, update status and continue
-            if record.status != new_status:
+            # Still running, update status and touch updated_at
             record.status = new_status
-                record.save(update_fields=['status'])
+            record.save(update_fields=['status', 'updated_at'])
             continue
 
         # Terminal state reached — process result
@@ -135,6 +134,33 @@ def _handle_completed(record, ark_resp):
         _release_freeze(record)
 
 
+@shared_task(ignore_result=True)
+def recover_stuck_tasks():
+    """定时扫描卡在 processing/queued 超过 10 分钟的任务，重新派发轮询。"""
+    from datetime import timedelta
+    from django.utils import timezone
+    from apps.generation.models import GenerationRecord
+
+    cutoff = timezone.now() - timedelta(minutes=10)
+    stuck_records = GenerationRecord.objects.filter(
+        status__in=('queued', 'processing'),
+        ark_task_id__isnull=False,
+        updated_at__lt=cutoff,  # updated_at 超过 10 分钟没更新，说明没有 worker 在轮询
+    ).exclude(ark_task_id='')
+
+    count = 0
+    for record in stuck_records:
+        logger.warning('recover_stuck_tasks: re-dispatching record=%s ark=%s', record.id, record.ark_task_id)
+        try:
+            poll_video_task.delay(record.id)
+            count += 1
+        except Exception:
+            logger.error('recover_stuck_tasks: failed to dispatch record=%s', record.id)
+
+    if count:
+        logger.info('recover_stuck_tasks: re-dispatched %d stuck tasks', count)
+
+
 def _handle_failed(record, ark_resp):
     """Process a failed task: record error and release frozen amount."""
     from utils.airdrama_client import ERROR_MESSAGES

backend/apps/generation/views.py

@@ -372,12 +372,19 @@ def video_generate_view(request):
             record.ark_task_id = ark_task_id
             record.status = 'processing'
             record.save(update_fields=['ark_task_id', 'status'])
-            # 触发后端异步轮询（连不上 Redis 时静默降级，前端轮询兜底）
+            # 触发后端异步轮询
             try:
                 from apps.generation.tasks import poll_video_task
                 poll_video_task.delay(record.id)
             except Exception:
-                logger.debug('Celery not available, falling back to frontend polling')
+                logger.warning('Celery dispatch failed for record %s, retrying once...', record.id)
+                import time
+                time.sleep(1)
+                try:
+                    from apps.generation.tasks import poll_video_task as _poll
+                    _poll.delay(record.id)
+                except Exception:
+                    logger.error('Celery dispatch failed twice for record %s, relying on recovery task', record.id)
         except Exception as e:
             logger.exception('AirDrama API create task failed')
             record.status = 'failed'

backend/config/settings.py

@@ -173,12 +173,18 @@ CSRF_TRUSTED_ORIGINS = [o for o in CORS_ALLOWED_ORIGINS if o.startswith('https:/
 # ──────────────────────────────────────────────
 # Celery (async task queue)
 # ──────────────────────────────────────────────
-CELERY_BROKER_URL = os.environ.get('REDIS_URL', 'redis://:vAhRnAA6VMco@r-7xvat0vez5clwbzk5vpd.redis.rds.aliyuncs.com:6379/8')
+CELERY_BROKER_URL = os.environ.get('REDIS_URL', 'redis://:vAhRnAA6VMco@redis-cngzyc2r77ka16g7a.redis.ivolces.com:6379/0')
 CELERY_RESULT_BACKEND = CELERY_BROKER_URL
 CELERY_ACCEPT_CONTENT = ['json']
 CELERY_TASK_SERIALIZER = 'json'
 CELERY_RESULT_SERIALIZER = 'json'
 CELERY_TIMEZONE = 'Asia/Shanghai'
+CELERY_BEAT_SCHEDULE = {
+    'recover-stuck-tasks': {
+        'task': 'apps.generation.tasks.recover_stuck_tasks',
+        'schedule': 600,  # 每 10 分钟
+    },
+}
 
 LANGUAGE_CODE = 'zh-hans'
 TIME_ZONE = 'Asia/Shanghai'

k8s/backend-deployment.yaml

@@ -56,7 +56,7 @@ spec:
           value: "3306"
         # Redis (Celery broker)
         - name: REDIS_URL
-          value: "redis://:vAhRnAA6VMco@r-7xvat0vez5clwbzk5vpd.redis.rds.aliyuncs.com:6379/8"
+          value: "redis://:vAhRnAA6VMco@redis-cngzyc2r77ka16g7a.redis.ivolces.com:6379/0"
         # CORS
         - name: CORS_ALLOWED_ORIGINS
           value: "https://airflow-studio.airlabs.art"

k8s/celery-deployment.yaml

@@ -20,7 +20,7 @@ spec:
       - name: celery-worker
         image: ${CI_REGISTRY_IMAGE}/video-backend:latest
         imagePullPolicy: Always
-        command: ["celery", "-A", "config", "worker", "--loglevel=info", "--concurrency=4"]
+        command: ["celery", "-A", "config", "worker", "--loglevel=info", "--concurrency=4", "-B"]
         env:
         - name: USE_MYSQL
           value: "true"
@@ -35,7 +35,7 @@ spec:
               key: DJANGO_SECRET_KEY
         # Redis
         - name: REDIS_URL
-          value: "redis://:vAhRnAA6VMco@r-7xvat0vez5clwbzk5vpd.redis.rds.aliyuncs.com:6379/8"
+          value: "redis://:vAhRnAA6VMco@redis-cngzyc2r77ka16g7a.redis.ivolces.com:6379/0"
         # Database (Aliyun RDS)
         - name: DB_HOST
           valueFrom: