Manual success of a mid-pipeline task triggers its downstream even though an upstream sensor is still deferred (question about none_failed behavior)

[darenpang]

Apache Airflow version

Other Airflow 2 version (please specify below)

If "Other Airflow 2 version" selected, which one?

2.10.1

What happened?

I have a DAG with three tasks in sequence:

wait_for_3am (a DateTimeSensorAsync), which waits until 3:00 AM and thus remains in deferred state until the target time.
empty_task_1 (an EmptyOperator), which depends on wait_for_3am.
empty_task_2 (another EmptyOperator), which depends on empty_task_1.
All tasks inherit the default trigger rule of none_failed. According to the documentation for none_failed, all upstream tasks must have “not failed or upstream_failed” (i.e. they should have succeeded or been skipped) in order to trigger the downstream tasks.

However, if I manually mark empty_task_1 as success while wait_for_3am is still in the deferred state (i.e., it has neither failed nor succeeded yet), then empty_task_2 immediately runs. From a scheduler standpoint, empty_task_2 sees empty_task_1 = success and therefore proceeds under none_failed, even though wait_for_3am has not actually completed.

What you think should happen instead?

I expected that if an upstream sensor (wait_for_3am) is still in a non-terminal (deferred) state, marking empty_task_1 as success would not trigger empty_task_2 immediately .

How to reproduce

Try this DAG, mark empty_task_1 as success .

from datetime import datetime, timedelta
from airflow import DAG
from airflow.utils.trigger_rule import TriggerRule
from airflow.sensors.date_time import DateTimeSensorAsync
from airflow.operators.empty import EmptyOperator

default_args = {
    'owner': 'airflow',
    "trigger_rule": TriggerRule.NONE_FAILED,
}

with DAG(
    'simple_dag_with_datetime_sensor',
    default_args=default_args,
    description='A simple DAG with DateTimeSensorAsync and two EmptyOperators',
    schedule_interval=timedelta(days=1),
    start_date=datetime(2024, 12, 30),
    catchup=False,
    max_active_runs=1,
) as dag:

    wait_for_3am = DateTimeSensorAsync(
        task_id='wait_for_3am',
        target_time=datetime.combine(datetime.today(), datetime.min.time()) + timedelta(hours=3),
    )

    empty_task_1 = EmptyOperator(
        task_id='empty_task_1',
    )

    empty_task_2 = EmptyOperator(
        task_id='empty_task_2',
    )

    wait_for_3am >> empty_task_1 >> empty_task_2

Operating System

RHEL8

Versions of Apache Airflow Providers

No response

Deployment

Virtualenv installation

Deployment details

No response

Anything else?

No response

Are you willing to submit PR?

• Yes I am willing to submit a PR!

Code of Conduct

• I agree to follow this project's Code of Conduct

[RNHTTR]

Assigned you @darenpang

[darenpang]

@RNHTTR
Would you please confirm the intended behavior?

[darenpang]

@RNHTTR
I’ve found this in the document.

By default, Airflow will wait for all upstream (direct parents) tasks for a task to be successful before it runs that task.

So Airflow does not recursively check grandparent tasks. It only looks to see if all direct parents are not failed. So this behavior is expected and consistent with the documentation.
If you confirm this , maybe this issue should be closed . Thank you .

[potiuk]

Correct. That's expected.