Apache airflow: setting catchup to False is not working

Question

I have a DAG created on Apache airflow. It seems scheduler is configured to run it from June on 2015 (By the way. I do not know why, but it is a new DAG y created and I didn't backfill it, I only backfilled other dag with different DAG ID with these date intervals, and the scheduler took those dates and bakfilled my new dag. I'm starting to work with airflow).

(Update: I realized DAG is backfilled because start date is set on the DAG default config, although this does not explain the behaviour I expose below)

I'm trying to stop the scheduler to run all the DAG executions from that date. airflow backfill --mark_success tutorial2 -s '2015-06-01' -e '2019-02-27' command is giving me database errors (see below), so I'm trying to set catchup to False.

sqlalchemy.exc.OperationalError: (sqlite3.OperationalError) no such table: job [SQL: 'INSERT INTO job (dag_id, state, job_type, start_date, end_date, latest_heartbeat, executor_class, hostname, unixname) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)'] [parameters: ('tutorial2', 'running', 'BackfillJob', '2019-02-27 10:52:37.281716', None, '2019-02-27 10:52:37.281733', 'SequentialExecutor', '08b6eb432df9', 'airflow')] (Background on this error at: http://sqlalche.me/e/e3q8)

So I'm using another approach. What I've tried:

Setting catchup_by_default = False in airflow.cfg and restarting the whole docker container.
Setting catchup = False on my pyhton DAG file and launching the file with python again.

What I'm seeing on the web UI:

DAG's executions are being launched starting at June 2015: Catchup is set to False on DAG's configuration:

So I don't understand why those DAG's executions are being launched.

Thank you

DAG code:

"""
Code that goes along with the Airflow tutorial located at:
https://github.com/apache/airflow/blob/master/airflow/example_dags/tutorial.py
"""
from airflow import DAG
from airflow.operators.bash_operator import BashOperator
from datetime import datetime, timedelta


default_args = {
    'owner': 'airflow',
    'depends_on_past': False,
    'start_date': datetime(2015, 6, 1),
    'email': ['[email protected]'],
    'email_on_failure': False,
    'email_on_retry': False,
    'retries': 1,
    'retry_delay': timedelta(minutes=5),
    'catchup' : False,
    # 'queue': 'bash_queue',
    # 'pool': 'backfill',
    # 'priority_weight': 10,
    # 'end_date': datetime(2016, 1, 1),
}

dag = DAG(
    'tutorial2', default_args=default_args, schedule_interval='* * * * *')

# t1, t2 and t3 are examples of tasks created by instantiating operators
t1 = BashOperator(
    task_id='print_date',
    bash_command='date',
    dag=dag)

t2 = BashOperator(
    task_id='sleep',
    bash_command='sleep 5',
    retries=3,
    dag=dag)

templated_command = """
    {% for i in range(5) %}
        echo "{{ ds }}"
        echo "{{ macros.ds_add(ds, 7)}}"
        echo "{{ params.my_param }}"
    {% endfor %}
"""

t3 = BashOperator(
    task_id='templated',
    bash_command=templated_command,
    params={'my_param': 'Parameter I passed in'},
    dag=dag)

t2.set_upstream(t1)
t3.set_upstream(t1)

default_args are passed to the tasks, so if you changed it afterwards, i suggest you change the dag version as well. Suffix it with _v2 and see if that works. — bosnjak
When you say changing the version, do you mean creating a new dag with a new dag id (suffixed with _v2) to reload default args? // By the way, I guess catchup argument should be only used by the scheduler, not by the tasks, right? — dhalfageme
I've created a new DAG with catchup = False in default args, I'm not updating default args in any place, and it still behaves the same way, DAG runs are scheduled starting at 2015-06-01 without backfilling the DAG, I'm very confused with this behaviour — dhalfageme
Yes, I meant to create a new DAG. catchup=False should work as expected. What Airflow version are you using? — bosnjak

santon santon · Accepted Answer · 2019-03-02T01:07:32

I think you actually need to specify the catchup at the dag level, not pass it in through default_args. (The latter doesn't really make sense anyway, since those are the default args for the tasks. You couldn't have some tasks catch up and others not.)

Try this:

dag = DAG(
    'tutorial2', default_args=default_args, schedule_interval='* * * * *', catchup=False)

Apache airflow: setting catchup to False is not working

1 Answers