Reputation: 2973
In a catchup setting, I observed that the execution date send to executor is correctly aligned, but when I turn off catchup setting like this:
import airflow
from airflow import DAG
from airflow.operators.bash_operator import BashOperator
from airflow.operators.dummy_operator import DummyOperator
from datetime import datetime, timedelta
default_args = {
'owner': 'airflow',
'depends_on_past': True,
'start_date': airflow.utils.dates.days_ago(5),
'schedule_interval': '@daily'
}
dag = DAG('xiang-01', catchup=False, default_args=default_args)
task = BashOperator(
task_id='task',
bash_command='echo "{{ task_instance_key_str }} {{ ts }}" && sleep 10',
dag=dag)
The execution date is not aligned, for example, the rendered task is:
echo "xiang-01__task__20180909 2018-09-09T22:33:17.961926+00:00" && sleep 10
According to the document, it should be aligned: https://airflow.apache.org/scheduler.html#backfill-and-catchup
So what am I missing?
Update:
To be more precise, since my start date is set via days_ago(5)
, which is set to the midnight of 5 days ago, start from 00:00:00. What I expected is, the execution date shall also be aligned by midnight, something like 2018-09-09T00:00:00
, but what I got is a time 2018-09-09T22:33:17.961926+00:00
, which looks like aligned by the time when I un-paused this DAG.
Upvotes: 0
Views: 512
Reputation: 2973
I figured it out, the following modification works:
import airflow
from airflow import DAG
from airflow.operators.bash_operator import BashOperator
from airflow.operators.dummy_operator import DummyOperator
from datetime import datetime, timedelta
default_args = {
'owner': 'airflow',
'depends_on_past': True,
'start_date': airflow.utils.dates.days_ago(5),
#'schedule_interval': '@daily'
}
dag = DAG('xiang-02', catchup=False, default_args=default_args, schedule_interval='@daily')
task = BashOperator(
task_id='task',
bash_command='echo "{{ task_instance_key_str }} {{ ts }}" && sleep 10',
dag=dag)
The catch is, looks like schedule_interval
is a DAG argument now, if I use it inside DAG construction, it works correctly now.
Upvotes: 1