共享数据库的两个Django项目django-celery-beat周期任务冲突解决咨询

两个共享同一PostgreSQL数据库的Django项目，均已部署Celery和django-celery-beat应用。当前出现问题：A项目的Celery Worker会收到B项目的周期任务，判定为未注册任务并标记为REJECTED。

推测原因是两个项目的django-celery-beat共享了数据库中的同一张调度表——两个项目为容器化部署，各自拥有独立Docker网络、不同端口的Redis服务器，且已配置专属默认队列名称。

设想每个项目需拥有独立的django-celery-beat实例，是否可通过AppConfig实现，再通过migrate生成专属表？预期配置如下：

项目A的settings.py：

INSTALLED_APPS = [
    ...
    'django_celery_beat_project_a',
    ...
]

项目B的settings.py：

INSTALLED_APPS = [
    ...
    'django_celery_beat_project_b',
    ...
]

附错误日志：

project-a-celery_worker-1  | 2022-09-09T11:35:15.974463294Z Received unregistered task of type 'trading.tasks.bulk_trade'.
project-a-celery_worker-1  | 2022-09-09T11:35:15.974621504Z The message has been ignored and discarded.
project-a-celery_worker-1  | 2022-09-09T11:35:15.974655749Z 
project-a-celery_worker-1  | 2022-09-09T11:35:15.974677303Z Did you remember to import the module containing this task?
project-a-celery_worker-1  | 2022-09-09T11:35:15.974699464Z Or maybe you're using relative imports?
project-a-celery_worker-1  | 2022-09-09T11:35:15.974721029Z 
project-a-celery_worker-1  | 2022-09-09T11:35:15.974741804Z Please see
project-a-celery_worker-1  | 2022-09-09T11:35:15.974782954Z http://docs.celeryq.org/en/latest/internals/protocol.html
project-a-celery_worker-1  | 2022-09-09T11:35:15.974815238Z for more information.
project-a-celery_worker-1  | 2022-09-09T11:35:15.974839132Z 
project-a-celery_worker-1  | 2022-09-09T11:35:15.974897495Z The full contents of the message body was:
project-a-celery_worker-1  | 2022-09-09T11:35:15.974938229Z b'[[], {}, {"callbacks": null, "errbacks": null, "chain": null, "chord": null}]' (77b)
project-a-celery_worker-1  | 2022-09-09T11:35:15.974982482Z 
project-a-celery_worker-1  | 2022-09-09T11:35:15.975032349Z The full contents of the message headers:
project-a-celery_worker-1  | 2022-09-09T11:35:15.975090058Z {'lang': 'py', 'task': 'trading.tasks.bulk_trade', 'id': 'd86620c2-04ab-4b59-bb2e-77685389f8d5', 'shadow': None, 'eta': None, 'expires': None, 'group': None, 'group_index': None, 'retries': 0, 'timelimit': [None, None], 'root_id': 'd86620c2-04ab-4b59-bb2e-77685389f8d5', 'parent_id': None, 'argsrepr': '()', 'kwargsrepr': '{}', 'origin': 'gen9@03298c28290e', 'ignore_result': False}
project-a-celery_worker-1  | 2022-09-09T11:35:15.975125751Z 
project-a-celery_worker-1  | 2022-09-09T11:35:15.975194550Z The delivery info for this task is:
project-a-celery_worker-1  | 2022-09-09T11:35:15.975257110Z {'exchange': '', 'routing_key': 'project-a_queue_1'}
project-a-celery_worker-1  | 2022-09-09T11:35:15.981241271Z  exc_info=(<class 'KeyError'>, KeyError('trading.tasks.bulk_trade'), <traceback object at 0x7fe49ca1b8c0>)
project-a-celery_worker-1  | 2022-09-09T11:35:15.996976839Z [error    ] task_not_found                              [django_structlog.celery.receivers] message=<Message object at 0x7fe49cb6e680 with details {'state': 'REJECTED', 'content_type': 'application/json', 'delivery_tag': '43925569-3d7a-4272-9629-2e08581f6579', 'body_length': 77, 'properties': {'correlation_id': 'd86620c2-04ab-4b59-bb2e-77685389f8d5'}, 'delivery_info': {'exchange': '', 'routing_key': 'project-a_queue_1'}}>

方案1：自定义AppConfig实现独立表（你的思路落地） #

该方案通过复制django-celery-beat源码并修改，为每个项目生成专属表：

• 复制django_celery_beat源码到项目目录，分别重命名为django_celery_beat_project_a（项目A）和django_celery_beat_project_b（项目B）
• 打开每个副本的models.py，给所有模型的Meta类添加db_table属性，指定专属表名，示例：

  class PeriodicTask(models.Model):
      # 原有字段保留
      class Meta:
          db_table = 'project_a_periodic_task'
          unique_together = ('name',)
  

• 修改每个副本的apps.py，更新AppConfig的名称：

  from django.apps import AppConfig
  
  class DjangoCeleryBeatProjectAConfig(AppConfig):
      name = 'django_celery_beat_project_a'
      verbose_name = 'Django Celery Beat (Project A)'
  

• 在项目的settings.py中，将原INSTALLED_APPS里的django_celery_beat替换为自定义应用名
• 分别在两个项目中执行python manage.py makemigrations和python manage.py migrate，生成专属表

方案2：使用原生表前缀配置（推荐） #

django-celery-beat原生支持通过配置项给所有表添加前缀，无需复制源码，更简洁易维护：

• 项目A的settings.py添加：

  CELERY_BEAT_DB_TABLE_PREFIX = 'project_a_'
  

• 项目B的settings.py添加：

  CELERY_BEAT_DB_TABLE_PREFIX = 'project_b_'
  

• 分别在两个项目中执行python manage.py migrate，系统会自动生成带前缀的独立表（如project_a_periodictask、project_b_crontabschedule等）

方案3：队列隔离+Worker专属监听（补充） #

结合队列配置，确保周期任务只发送到对应项目的队列，Worker只监听自身队列：

• 在每个项目的settings.py中配置任务路由：

  CELERY_TASK_ROUTES = {
      'trading.tasks.*': {'queue': 'project_a_queue_1'},  # 项目A配置
      # 项目B对应配置为自己的队列
  }
  

• 启动Worker时明确指定监听的队列：

  # 项目A启动命令
  celery -A project_a worker -l info -Q project_a_queue_1
  # 项目B启动命令
  celery -A project_b worker -l info -Q project_b_queue_1
  

该方案可与方案2结合，彻底隔离两个项目的周期任务调度流程。

内容的提问来源于stack exchange，提问作者Florent