Problem with start date and scheduled date in Apache Airflow

Question

I am working with Apache Airflow and I have a problem with the scheduled day and the starting day.

I want a DAG to run every day at 8:00 AM UTC. So, I did:

default_args = {
        'owner': 'airflow',
        'depends_on_past': False,
        'start_date': datetime(2020, 12, 7, 10, 0,0),
        'email': ['[email protected]'],
        'email_on_failure': True,
        'email_on_retry': False,
        'retries': 1,
        'retry_delay': timedelta(hours=5)
    }
# Never run
dag = DAG(dag_id='id', default_args=default_args, schedule_interval='0 8 * * *',catchup=True)

The day I upload the DAG was 2020-12-07 and I wanted to run it on 2020-12-08 at 08:00:00.

I set the start_date at 2020-12-07 at 10:00:00 to avoid running it at 2020-12-07 at 08:00:00 and only trigger it the next day, but it didn't work.

Then I modified the starting day:

default_args = {
        'owner': 'airflow',
        'depends_on_past': False,
        'start_date': datetime(2020, 12, 7, 7, 59,0),
        'email': ['[email protected]'],
        'email_on_failure': True,
        'email_on_retry': False,
        'retries': 1,
        'retry_delay': timedelta(hours=5)
    }
# Never run
dag = DAG(dag_id='etl-ca-cpke-spark_dev_databricks', default_args=default_args, schedule_interval='0 8 * * *',catchup=True)

Now the start date is 1 minute before the DAG should run, and indeed, because the catchup is set to True, the DAG has been triggered for 2020-12-07 at 08:00:00, but it has not being triggered for 2020-12-08 at 08:00:00.

Why?

Prosimian

Answer 1

Airflow schedules tasks at the end of the interval (See documentation reference)

Meaning that when you do:

start_date: datetime(2020, 12, 7, 8, 0,0)
schedule_interval: '0 8 * * *'

The first run will kick in at 2020-12-08 at 08:00+- (depends on resources)

This run's execution_date will be: 2020-12-07 08:00

The next run will kick in at 2020-12-09 at 08:00

This run's execution_date of 2020-12-08 08:00.

Since today is 2020-12-08 the next run didn't kick in because it's not the end of the interval yet.

Phenolic

Then if I want to start running 2020-12-08 at 08:00 and after that run every day at 08:00 I should set the start date at start_date: 2020-12-06 08:00? I don't understand why if start date is 2020-12-07 08:00, it doesn't run on 2020-12-07 08:00 and 2020-12-08 08:00. – Prosimian

You are interpreting start_date like cronjob where you specify when and it's starting on that date. This is not how Airflow works. Airflow takes the start_date + interval when this period END it start the run. The logic behind this is when you write ETL you want to run at the end of interval over the window -> today you want to process yesterday data. – Phenolic

Then if I want to modify my dag to start being scheduled every day at 08:00:00 from tomorrow on, what start date I should set? – Prosimian

Go with the same logic. The run kicks in at the end of interval so start_date = datetime(2020, 12, 8, 8, 0,0) and interval of 0 8 * * * will end at 2020-12-09 08:00 and that is when the first run will kick in. Note that this run will have execution_date of 2020-12-08 08:00 – Phenolic