I'm using airflow.operators.sensors.ExternalTaskSensor to make one Dag wait for another.
dag = DAG(
'dag2',
default_args={
'owner': 'Me',
'depends_on_past': False,
'start_date': start_datetime,
'email': ['me#example.com'],
'email_on_failure': True,
'email_on_retry': False,
'retries': 2,
'retry_delay': timedelta(minutes=10),
},
template_searchpath="%s/me/resources/" % DAGS_FOLDER,
schedule_interval="{} {} * * *".format(minute, hour),
max_active_runs=1
)
wait_for_dag1 = ExternalTaskSensor(
task_id='wait_for_dag1',
external_dag_id='dag1',
external_task_id='dag1_task1',
dag=dag
)
If something seriously wrong happens with upstream Dag and it fails to complete during the given time period, I want upstream Dag (ExternalTaskSensor operator) crash as well, instead of hanging forever.
How can I add a timeout to ExternalTaskSensor?
I'm looking into documentation, but it does not seem to have a timeout parameter or something similar. What should I do?
https://airflow.readthedocs.io/en/stable/_modules/airflow/sensors/external_task_sensor.html
The ExternalTaskSensor does take a timeout argument in seconds. It inherits the argument from BaseSensorOperator (https://airflow.apache.org/docs/apache-airflow/stable/_api/airflow/sensors/base/index.html). If you pass it timeout=60 on instantiation, it will fail after 60 seconds.
Related
I'm trying to add a cross dag dependency using ExternalTaskSensor but haven't been able to get it to work. Dag A has schedule_interval=None as it doesn't have a fixed schedule and is triggered externally by a file creation event. Dag B should execute once Dag A has completed. Here is code for dag_a and dag_b.
DAG A
default_args = {
'depends_on_past': False,
'start_date': datetime.today()-timedelta(1),
'email_on_failure': True,
'email_on_retry': False,
'queue': 'default'
}
dag = DAG(
'dag_a', default_args=default_args, schedule_interval=None)
dag_a = AWSBatchOperator(
task_id='dag_a',
job_name='dag_a',
job_definition='dag_a',
job_queue='MyAWSJobQueue',
max_retries=10,
aws_conn_id='aws_conn',
region_name='us-east-1',
dag=dag,
parameters={},
overrides={})
DAG B
default_args = {
'depends_on_past': False,
'start_date': datetime.today()-timedelta(1),
'email_on_failure': True,
'email_on_retry': False,
'queue': 'default'
}
dag = DAG(
'dag_b', default_args=default_args, schedule_interval=None)
dag_b = AWSBatchOperator(
task_id='dag_b',
job_name='dag_b',
job_definition='dag_b',
job_queue='MyAWSJobQueue',
max_retries=10,
aws_conn_id='aws_conn',
region_name='us-east-1',
dag=dag,
parameters={},
overrides={})
wait_for_dag_a = ExternalTaskSensor(
task_id='wait_for_irr',
external_dag_id='dag_a',
external_task_id=None,
execution_delta = timedelta(hours=1),
dag=dag,
timeout = 300)
dag_b.set_upstream(wait_for_dag_a)
I set both dags with schedule_interval=None and same start_date. I even added execution_delta = timedelta(hours=1) for dag_b, but dag_b hasn't triggered so far, though dag_a is complete. Any help is appreciated.
I have tried using TriggerDagRunOperator which works, but is not suitable for my use case since dag_b will eventually be dependent on multiple parent dags.
I've met similar problem before, so there are two things need to check, first I cannot see any time delta between DAG A and DAG B, both use the default arg so you should not give the waiting task a execution_delta, and for the airflow trigger, somehow it cannot detect the DAG finish sign if there are multiple parents DAGs, so I've tried give a value to external_task_id, like 'dag_a-done' instead of the default 'None', and that works. One more thing to mention is the task_id normally should not contain underscore.
The link is the source code of external sensor:
https://airflow.apache.org/docs/stable/_modules/airflow/sensors/external_task_sensor.html
Also an article describes how the ExternalTaskSensors works:
https://medium.com/#fninsiima/sensing-the-completion-of-external-airflow-tasks-827344d03142
I did a DAG's with the following configuration:
default_args = {
'owner': 'airflow',
'depends_on_past': False,
'start_date': airflow.utils.dates.days_ago(0, 0, minute=1),
'email': ['francisco.salazar.12#sansano.usm.cl'],
'email_on_failure': False,
'email_on_retry': False,
'max_active_runs': 1,
'retries': 1,
'retry_delay': timedelta(minutes=1),
'provide_context': True
}
dag = DAG(
'terralink_environmetal_darksky',
default_args=default_args,
description='Extract Data from Darksky API',
catchup=False,
schedule_interval='31 * * * *',
)
The issue is that scheduler works correctly and execute DAG run at every hour that I defined in schedule_inverval (in minute 31 of every hour) BUT in midnight or the last execution of the day (scheduled at 00:31:00 for the next day) the DAG execution is not triggered.
I think that is a problem based on start_date but I don't know yet how to define this parameter in order to avoid the problem.
Airflow recommends to state a fixed startstart_date for your DAG. start_date is mainly for the purpose to specify when do you want your DAG to start running for the very first time. schedule_interval will be the most relevant one after the start_date did its purpose or (if you do not need to backfill or reset your dag).
I have a DAG configured like this:
AIRFLOW_DEFAULT_ARGS = {
'owner': 'airflow',
'depends_on_past': False,
'email_on_failure': False,
'email_on_retry': False,
'retries': 1,
'retry_delay': timedelta(minutes=5),
'dagrun_timeout': timedelta(hours=5)
}
DAILY_RUNNER = DAG(
'daily_runner',
max_active_runs=1,
start_date=datetime(2019, 1, 1),
schedule_interval="0 17 * * *",
default_args=AIRFLOW_DEFAULT_ARGS)
My current understanding is that retries says that a task will be retried once before failing for good. Is there a way to set a similar limit for the number of times a DAG gets retried? If I have a dag in the running state, I want to be able to set it to failed from within the UI once and have it stop rerunning.
Currently, there is no way to set retry at dag level.
Please refer the below answer for retrying a set of tasks/whole-dag in case of failures.
Can a failed Airflow DAG Task Retry with changed parameter
I want the tasks in the DAG to all finish before the 1st task of the next run gets executed.
I have max_active_runs = 1, but this still happens.
default_args = {
'depends_on_past': True,
'wait_for_downstream': True,
'max_active_runs': 1,
'start_date': datetime(2018, 03, 04),
'owner': 't.n',
'email': ['t.n#example.com'],
'email_on_failure': True,
'email_on_retry': False,
'retries': 3,
'retry_delay': timedelta(minutes=4)
}
dag = DAG('example', default_args=default_args, schedule_interval = schedule_interval)
(All of my tasks are dependent on the previous task. Airflow version is 1.8.0)
Thank you
I changed to put max_active_runs as an argument of DAG() instead of in default_arguments, and it worked.
Thanks SimonD for giving me the idea, though not directly pointing to it in your answer.
You've put the 'max_active_runs': 1 into the default_args parameter and not into the correct spot.
max_active_runs is a constructor argument for a DAG and should not be put into the default_args dictionary.
Here is an example DAG that shows where you need to move it to:
dag_args = {
'owner': 'Owner',
# 'max_active_runs': 1, # <--- Here is where you had it.
'depends_on_past': False,
'start_date': datetime(2018, 01, 1, 12, 00),
'email_on_failure': False
}
sched = timedelta(hours=1)
dag = DAG(
job_id,
default_args=dag_args,
schedule_interval=sched,
max_active_runs=1 # <---- Here is where it is supposed to be
)
If the tasks that your dag is running are actually sub-dags then you may need to pass max_active_runs into the subdags too but not 100% sure on this.
You can use xcoms to do it. First take 2 python operators as 'start' and 'end' to the DAG. Set the flow as:
start ---> ALL TASKS ----> end
'end' will always push a variable
last_success = context['execution_date'] to xcom (xcom_push). (Requires provide_context = True in the PythonOperators).
And 'start' will always check xcom (xcom_pull) to see whether there exists a last_success variable with value equal to the previous DagRun's execution_date or to the DAG's start_date (to let the process start).
Followed this answer
Actually you should use DAG_CONCURRENCY=1 as environment var. Worked for me.
I configured my DAG like this:
default_args = {
'owner': 'Aviv',
'depends_on_past': False,
'start_date': datetime(2017, 1, 1),
'email': ['aviv#oron.com'],
'email_on_failure': False,
'email_on_retry': False,
'retries': 0,
'retry_delay': timedelta(minutes=1)
}
dag = DAG(
'MyDAG'
, schedule_interval=timedelta(minutes=3)
, default_args=default_args
, catchup=False
)
and for some reason, when i un-pause the DAG, its being executed twice immediatly.
Any idea why? And is there any rule i can apply to tell this DAG to never run more than once in the same time?
You can specify max_active_runs like this:
dag = airflow.DAG(
'customer_staging',
schedule_interval="#daily",
dagrun_timeout=timedelta(minutes=60),
template_searchpath=tmpl_search_path,
default_args=args,
max_active_runs=1)
I've never seen it happening, are you sure that those runs are not backfills, see: https://stackoverflow.com/a/47953439/9132848
I think its because you have missed the scheduled time and airflow is backfilling it automatically when you ON it again. You can disable this by
catchup_by_default = False in the airflow.cfg.