在Airflow的2.0版本中,可能会出现这样的问题,即任务一直处于排队状态,尽管在池中有空闲的槽位。这通常是由于Airflow的一个特性,即它将任务在调度程序周期内分配给池,而不管它们当前是否有可用的槽位。因此,我们可以通过以下代码示例来解决此问题:
1.首先,应该创建一个名为task_instance_pool.py的文件,并将其放置在您的$AIRFLOW_HOME/plugins文件夹中。
2.在task_instance_pool.py文件中,我们将要编写以下代码示例:
from airflow.utils.db import provide_session from airflow.models import TaskInstance, Pool
@provide_session def allocate_pool_slot_if_available(ti:TaskInstance, session=None)-> None: pool_name = ti.pool if pool_name is None: return
pool = session.query(Pool).filter(Pool.pool == pool_name).first()
if pool is None:
return
used_slots = session.query(TaskInstance).filter(TaskInstance.pool == pool_name, TaskInstance.state!= 'success').count()
if used_slots > pool.slots:
ti.state = None
return
ti.state = 'queued'
3.在您的dag文件中,您可以在operator之前添加以下代码:
from airflow.plugins_manager import AirflowPlugin from task_instance_pool import allocate_pool_slot_if_available
class MyPlugin(AirflowPlugin): name = "my_plugin" operators = [] sensors = [] hooks = [] executors = [] macros = [] admin_views = [] flask_blueprints = [] menu_links = [] appbuilder_views = [] appbuilder_menu_items = [] global_operator_extra_links = [] operator_extra_links = [] register_blueprint_on_plugin_load = True
4.最后,在您的任务中,您可以将我们编写的代码示例应用到您的任务中:
from airflow.decorators import dag, task from airflow.utils.dates import days_ago
default_args = {