Deferrable mode of TriggerDagRunOperator stays stuck if used with `reset_dag_run`

[gvergnolle]

Apache Airflow version

3.1.1

If "Other Airflow 2/3 version" selected, which one?

No response

What happened?

When using the TriggerDagRunOperator with a given trigger_run_id, wait_for_completion, and reset_dag_run set to True, the operator remains stuck in a deferred state forever once I restarted the task.

What you think should happen instead?

Instead, once the DAG run has been reset and completed, the trigger should mark the state as success.

How to reproduce

Add two dags, one that triggers, the other is triggered:

DAG 1:

with DAG(
    dag_id="test1",
    schedule=None,
    start_date=pendulum.datetime(2020, 1, 1),
    tags=[],
    default_args={},
    max_active_runs=1,
    is_paused_upon_creation=False,
) as dag:
    TriggerDagRunOperator(
        task_id="trigger_test2_dag",
        trigger_dag_id="test2",
        trigger_run_id="a_dag_run_id",  # important
        wait_for_completion=True,  # important
        deferrable=True,  # important
        reset_dag_run=True,  # important
        retries=0,
    )

DAG 2 (does nothing really):

with DAG(
    dag_id="test2",
    schedule=None,
    start_date=pendulum.datetime(2020, 1, 1),
    tags=[],
    default_args={},
    max_active_runs=1,
    is_paused_upon_creation=False,
) as dag:

    @task()
    def hello_world():
        print("Hello, World!")

    hello_world()

Then:

• Manually trigger test1
• test2 will be scheduled, then successes. Seconds later, test1 task succeeds.
• Clear the same task in test1 so it re-triggers test2 with the same dag_run_id.
• test2 DAG will be reset, and completes without issues.
• The only test1 task is now stuck in deferred state forever

Operating System

Debian GNU/Linux 12 (bookworm)

Versions of Apache Airflow Providers

apache-airflow-providers-standard==1.9.0

Deployment

Docker-Compose

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

[mykola-shyshov]

Hello. I will try to reproduce. And investigate the issue.

[mykola-shyshov]

I was able to reproduce it. Seems like I found root issue. Every start after clearing happens with new logical date. But DAG 2 keeps the original logical date. Due to this mismatch it can not link completion to the first DAG.
I need to think about good solution.