ed-sparkes opened a new issue, #27913: URL: https://github.com/apache/airflow/issues/27913
### Apache Airflow version Other Airflow 2 version (please specify below) ### What happened Definitely present in Airflow 2.4.2 https://airflow.apache.org/api/v1/dags/{dag_id}/clearTaskInstances is supposed to return a list of cleared task instances and it used to work as expected but some version between 2.3.4 and 2.4.2 the behaviour changed. in this file `https://github.com/apache/airflow/blob/main/airflow/api_connexion/endpoints/task_instance_endpoint.py` it appears that something is mutating `task_instances` if not dry run. If i run dry run the task_instances are returned, as soon as a i run non dry_run the response is always an empty list. ``` task_instances = dag.clear(dry_run=True, dag_bag=get_airflow_app().dag_bag, task_ids=task_ids, **data) if not dry_run: clear_task_instances( task_instances.all(), session, dag=dag, dag_run_state=DagRunState.QUEUED if reset_dag_runs else False, ) return task_instance_reference_collection_schema.dump( TaskInstanceReferenceCollection(task_instances=task_instances.all()) ) ``` I looked into `https://github.com/apache/airflow/blob/main/airflow/models/taskinstance.py` where the clear_task_instances function lives, i compared the code between 2.3.4 and 2.4.2 and the only difference i thought that may cause it is the addition of a ` session.flush()` at the end of the method. Any thoughts on this would be welcome as we cant upgrade and resolve some security dependencies until this is resolved as we rely on the response from this api. If the `session.flush()` is the culprit, i presume that was added for some other reason to resolve some other issue? ### What you think should happen instead It should return the cleared tasks as per the documentation ### How to reproduce Detail in the What happened section above ### Operating System linux ### Versions of Apache Airflow Providers _No response_ ### Deployment Official Apache Airflow Helm Chart ### 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 - [X] I agree to follow this project's [Code of Conduct](https://github.com/apache/airflow/blob/main/CODE_OF_CONDUCT.md) -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
