我已经在SO上看到了this和this的问题并做出了相应的更改。但是,我的依赖DAG仍然处于戳状态。以下是我的主DAG:
from airflow import DAG
from airflow.operators.jdbc_operator import JdbcOperator
from datetime import datetime
from airflow.operators.bash_operator import BashOperator
today = datetime.today()
default_args = {
'depends_on_past': False,
'retries': 0,
'start_date': datetime(today.year, today.month, today.day),
'schedule_interval': '@once'
}
dag = DAG('call-procedure-and-bash', default_args=default_args)
call_procedure = JdbcOperator(
task_id='call_procedure',
jdbc_conn_id='airflow_db2',
sql='CALL AIRFLOW.TEST_INSERT (20)',
dag=dag
)
call_procedure
以下是我的依赖DAG:
from airflow import DAG
from airflow.operators.jdbc_operator import JdbcOperator
from datetime import datetime, timedelta
from airflow.sensors.external_task_sensor import ExternalTaskSensor
today = datetime.today()
default_args = {
'depends_on_past': False,
'retries': 0,
'start_date': datetime(today.year, today.month, today.day),
'schedule_interval': '@once'
}
dag = DAG('external-dag-upstream', default_args=default_args)
task_sensor = ExternalTaskSensor(
task_id='link_upstream',
external_dag_id='call-procedure-and-bash',
external_task_id='call_procedure',
execution_delta=timedelta(minutes=-2),
dag=dag
)
count_rows = JdbcOperator(
task_id='count_rows',
jdbc_conn_id='airflow_db2',
sql='SELECT COUNT(*) FROM AIRFLOW.TEST',
dag=dag
)
count_rows.set_upstream(task_sensor)
以下是主DAG执行后依赖DAG的日志:
[2019-01-10 11:43:52,951] {{external_task_sensor.py:91}} INFO - Poking for call-procedure-and-bash.call_procedure on 2019-01-10T11:45:47.893735+00:00 ...
[2019-01-10 11:44:52,955] {{external_task_sensor.py:91}} INFO - Poking for call-procedure-and-bash.call_procedure on 2019-01-10T11:45:47.893735+00:00 ...
[2019-01-10 11:45:52,961] {{external_task_sensor.py:91}} INFO - Poking for call-procedure-and-bash.call_procedure on 2019-01-10T11:45:47.893735+00:00 ...
[2019-01-10 11:46:52,949] {{external_task_sensor.py:91}} INFO - Poking for call-procedure-and-bash.call_procedure on 2019-01-10T11:45:47.893735+00:00 ...
[2019-01-10 11:47:52,928] {{external_task_sensor.py:91}} INFO - Poking for call-procedure-and-bash.call_procedure on 2019-01-10T11:45:47.893735+00:00 ...
[2019-01-10 11:48:52,928] {{external_task_sensor.py:91}} INFO - Poking for call-procedure-and-bash.call_procedure on 2019-01-10T11:45:47.893735+00:00 ...
[2019-01-10 11:49:52,905] {{external_task_sensor.py:91}} INFO - Poking for call-procedure-and-bash.call_procedure on 2019-01-10T11:45:47.893735+00:00 ...
以下是主DAG执行的日志:
[2019-01-10 11:45:20,215] {{jdbc_operator.py:56}} INFO - Executing: CALL AIRFLOW.TEST_INSERT (20)
[2019-01-10 11:45:21,477] {{logging_mixin.py:95}} INFO - [2019-01-10 11:45:21,476] {{dbapi_hook.py:166}} INFO - CALL AIRFLOW.TEST_INSERT (20)
[2019-01-10 11:45:24,139] {{logging_mixin.py:95}} INFO - [2019-01-10 11:45:24,137] {{jobs.py:2627}} INFO - Task exited with return code 0
我的假设是,如果主机运行正常,Airflow应该触发相关的DAG?我试过玩execution_delta
,但这似乎不起作用。
此外,两个DAG的schedule_interval
和start_date
相同,所以不要认为这会造成任何麻烦。
我错过了什么吗?
可能你应该使用正时间delta:https://airflow.readthedocs.io/en/stable/_modules/airflow/sensors/external_task_sensor.html,因为当减去执行增量时,它最终将寻找在其自身后2分钟运行的任务。
但是,delta实际上不是一个范围,TI必须在日期时间列表中具有匹配的Dag ID,任务ID,成功结果以及执行日期。当你将execution_delta
作为delta时,它是一个日期时间列表,它取当前执行日期并减去timedelta。
这可能取决于您要么删除timedelta以便两个执行日期匹配,传感器将等到另一个任务成功,或者您的开始日期和计划间隔基本上设置为今天,而@once
的执行日期不是可预测的相互锁定。您可以尝试设置说datetime(2019,1,10)
和0 1 * * *
让它们每天凌晨1点运行(再次没有execution_delta
)。
确保两个DAG同时启动,并且不要手动启动任一DAG。