Apache Airflow-即使关键任务失败，DAG也会注册为成功

时间：2018-08-06 13:04:17

标签： python etl pipeline airflow

我是Apache Airflow的新手，我想编写DAG将一些数据从源数据库中的一组表移动到目标数据库中的一组表。我正在尝试设计DAG，以便有人可以为新的源表->目标表过程简单地编写create table和insert into SQL脚本并将其放入文件夹中。然后，在下一次DAG运行时，DAG将从文件夹中拾取脚本并运行新任务。我将DAG设置为：

source_data_check_task_1 (Check Operator or ValueCheckOperator)
source_data_check_task_2 (Check Operator or ValueCheckOperator, Trigger on ALL_SUCCESS)
source_data_check_task_3 (Check Operator or ValueCheckOperator, Trigger on ALL_SUCCESS)

source_data_check_task_1 >> source_data_check_task_2 >> source_data_check_task_3

for tbl_name in tbl_name_list:
    tbl_exists_check (Check Operator, trigger on ALL_SUCCESS): check if `new_tbl` exists in database by querying `information_schema`
        tbl_create_task (SQL Operator, trigger on ALL_FAILED): run the `create table` SQL script
    tbl_insert_task (SQL Operator ,trigger on ONE_SUCCESS): run the `insert into` SQL script

    source_data_check_task_3 >> tbl_exists_check
    tbl_exists_check >> tbl_create_task
    tbl_exists_check >> tbl_insert_task
    tbl_create_task >> tbl_insert)task

此设置遇到两个问题：（1）如果任何数据质量检查任务失败，tbl_create_task仍会启动，因为它在ALL_FAILED上触发；（2）不管是哪个任务失败，DAG显示运行是SUCCESS。如果tbl_exists_check失败了就很好了，因为它应该至少失败一次，但是如果某些关键任务失败了（比如任何数据质量检查任务），那不是理想的选择。

是否可以通过其他方式来设置DAG以解决这些问题？

下面的实际代码：

from airflow import DAG
from airflow.operators.postgres_operator import PostgresOperator
from airflow.operators.check_operator import ValueCheckOperator, CheckOperator
from airflow.operators.bash_operator import BashOperator
from airflow.models import Variable
from datetime import datetime, timedelta
from airflow.utils.trigger_rule import TriggerRule

sql_path = Variable.get('sql_path')

default_args = {
    'owner': 'enmyj',
    'depends_on_past':True,
    'email_on_failure': False,
    'email_on_retry': False,
    'retries': 0
}

dag = DAG(
    'test', 
    default_args=default_args, 
    schedule_interval=None,
    template_searchpath=sql_path
)

# check number of weeks in bill pay (made up example)
check_one = CheckOperator(
    task_id='check_one',
    conn_id='conn_name',
    sql="""select count(distinct field) from dbo.table having count(distinct field) >= 4 """,
    dag=dag
)

check_two = CheckOperator(
    task_id='check_two',
    conn_id='conn_name',
    sql="""select count(distinct field) from dbo.table having count(distinct field) <= 100""",
    dag=dag
)

check_one >> check_two

ls = ['foo','bar','baz','quz','apple']
for tbl_name in ls:
    exists = CheckOperator(
        task_id='tbl_exists_{}'.format(tbl_name),
        conn_id='conn_name',
        sql =""" select count(*) from information_schema.tables where table_schema = 'test' and table_name = '{}' """.format(tbl_name),
        trigger_rule=TriggerRule.ALL_SUCCESS,
        depends_on_past=True,
        dag = dag
    )

    create = PostgresOperator(
        task_id='tbl_create_{}'.format(tbl_name),
        postgres_conn_id='conn_name',
        database='triforcedb',
        sql = 'create table test.{} (like dbo.source)'.format(tbl_name), # will be read from SQL file
        trigger_rule=TriggerRule.ONE_FAILED,
        depends_on_past=True,
        dag = dag
    )

    insert = PostgresOperator(
        task_id='tbl_insert_{}'.format(tbl_name),
        postgres_conn_id='conn_name',
        database='triforcedb',
        sql = 'insert into test.{} (select * from dbo.source limit 10)'.format(tbl_name), # will be read from SQL file
        trigger_rule=TriggerRule.ONE_SUCCESS,
        depends_on_past=True,
        dag = dag
    )

    check_two >> exists
    exists >> create
    create >> insert
    exists >> insert

2 个答案:

答案 0 :(得分：3)

您有一个完美的用例，可以利用BranchPythonOperator进行检查，以查看表是否存在，然后在插入表之前继续创建表，而不必担心TRIGGER_RULES并通过UI使您的DAG逻辑更加清晰。

答案 1 :(得分：0)

下面是我最终得到的代码。此解决方案解决了上述两个问题： 1.如果上游任务失败，则不会触发tbl_create任务 2.如果任何FAILED任务失败，则DAG注册为check。 我觉得这个解决方案似乎有些混乱，很乐意提出改进建议或使它更加“流畅”的方法

from airflow.models import DAG
from airflow.models import Variable
from airflow.operators.postgres_operator import PostgresOperator
from airflow.operators.check_operator import ValueCheckOperator, CheckOperator
from airflow.operators.python_operator import BranchPythonOperator
from airflow.operators.dummy_operator import DummyOperator
from airflow.utils.trigger_rule import TriggerRule
from datetime import datetime, timedelta
from airflow.hooks.postgres_hook import PostgresHook

sql_path = Variable.get('sql_path')

default_args = {
    'owner': 'enmyj',
    'depends_on_past':False,
    'email_on_failure': False,
    'email_on_retry': False,
    'retries': 0
}

dag = DAG(
    'test', 
    default_args=default_args, 
    schedule_interval=None,
    template_searchpath=sql_path
)

# check number of weeks in bill pay (made up example)
check_one = CheckOperator(
    task_id='check_one',
    conn_id='conn_id',
    sql="""select count(distinct field) from dbo.table having count(distinct field) >= 4 """,
    dag=dag
)

def check_two_func():
    p = Hook('conn_id')
    sql="""select count(distinct field) from dbo.table having count(distinct field) <= 100"""
    count = p.get_records(sql)[0][0]
    if count == 0: 
        return 'dummy_fail'
    else:
        return 'dummy_success'


check_two = BranchPythonOperator(
    task_id = 'check_two',
    python_callable = check_two_func,
    dag=dag
)

dummy_fail = DummyOperator(task_id='dummy_fail',dag=dag)
dummy_success = DummyOperator(task_id='dummy_success',dag=dag)
join = DummyOperator(task_id='join',dag=dag)

check_one >> check_two
check_two >> dummy_fail
check_two >> dummy_success

ls = ['foo','bar','baz','quz','apple']
for tbl_name in ls:
    def has_table(tbl_name=tbl_name):
        p = PostgresHook('conn_id')
        sql =""" select count(*) from information_schema.tables where table_schema = 'test' and table_name = '{}' """.format(tbl_name)
        count = p.get_records(sql)[0][0] #unpack the list/tuple

        # If the query didn't return rows, branch to create table
        # otherwise, branch to dummy
        if count == 0:
            return 'tbl_create_{}'.format(tbl_name)
        else:
            return 'dummy_{}'.format(tbl_name) 

    exists = BranchPythonOperator(
        task_id='tbl_exists_{}'.format(tbl_name),
        python_callable=has_table,
        depends_on_past=False,
        dag=dag
    )

    create = PostgresOperator(
        task_id='tbl_create_{}'.format(tbl_name),
        postgres_conn_id='conn_id',
        database='database_name',
        sql = 'create table test.{} (like dbo.source)'.format(tbl_name), # will be read from SQL file
        dag = dag
    )

    insert = PostgresOperator(
        task_id='tbl_insert_{}'.format(tbl_name),
        postgres_conn_id='conn_id',
        database='database_name',
        sql = 'insert into test.{} (select * from dbo.source limit 10)'.format(tbl_name), # will be read from SQL file
        trigger_rule=TriggerRule.ONE_SUCCESS,
        dag = dag
    )

    dummy_success >> exists
    exists >> create >> insert 
    exists >> dummy >> insert
    insert >> join
相关问题
最新问题