Better way of finding successful dag duration

[ayush-san]

In the last query of get_dag_duration_info(), why are we filtering on the basis of TaskInstance? This results in an explosion of the final result with duplicate rows.

dag_duration = get_dag_duration_info() 
len(dag_duration)  # 960640
set_dag_duration = set() 
for dag in dag_duration: 
    set_dag_duration.add((dag.dag_id, dag.start_date, dag.end_date)) 
len(set_dag_duration) # 38

Instead, we should be adding the filter condition in dag_start_dt_query subquery and remove it from the last query.

dag_start_dt_query = (
    session.query(
        max_execution_dt_query.c.dag_id,
        max_execution_dt_query.c.max_execution_dt.label(
            "execution_date"
        ),
        func.min(TaskInstance.start_date).label("start_date"),
    )
    .join(
        TaskInstance,
        and_(
            TaskInstance.dag_id == max_execution_dt_query.c.dag_id,
            (
                TaskInstance.execution_date
                == max_execution_dt_query.c.max_execution_dt
            ),
        ),
    )
    .filter(
        TaskInstance.start_date.isnot(None),
        TaskInstance.end_date.isnot(None),
    )
    .group_by(
        max_execution_dt_query.c.dag_id,
        max_execution_dt_query.c.max_execution_dt,
    )
    .subquery()
)

return (
    session.query(
        dag_start_dt_query.c.dag_id,
        dag_start_dt_query.c.start_date,
        DagRun.end_date,
    )
    .join(
        DagRun,
        and_(
            DagRun.dag_id == dag_start_dt_query.c.dag_id,
            DagRun.execution_date
            == dag_start_dt_query.c.execution_date,
        ),
    )
    .all()
)

[astleychen]

The fix is a must if you'd like to use this Airflow plugin, otherwise, you can just fork yours.