7

Suppose a following situation:

[c1, c2, c3] >> child_task

where all c1, c2, c3 and child_task are operators and have task_id equal to id1, id2, id3 and child_id respectively.

Task child_task is also a PythonOperator with provide_context=True and python_callable=dummy_func

def dummy_func(**context):
    #...

Is it possible to retrieve all parents' ids inside the dummy_func(perhaps by browsing the dag somehow using the context)?

Expected result in this case would be a list ['id1', 'id2', 'id3'].

y2k-shubham
  • 10,183
  • 11
  • 55
  • 131
blahblah
  • 1,161
  • 2
  • 14
  • 30

2 Answers2

12

The upstream_task_ids and downstream_task_ids properties of BaseOperator are meant just for this purpose.

from typing import List
..
parent_task_ids: List[str] = my_task.upstream_task_ids
child_task_ids: List[str] = my_task_downstream_task_ids

Do note however that with this property, you only get immediate (upstream / downstream) neighbour(s) of a task. In order to get all ancestor or descendent tasks, you can quickly cook-up the good old graph theory approach such as this BFS-like implementation

from typing import List, Set
from queue import Queue
from airflow.models import BaseOperator

def get_ancestor_tasks(my_task: BaseOperator) -> List[BaseOperator]:
    ancestor_task_ids: Set[str] = set()
    tasks_queue: Queue = Queue()
    # determine parent tasks to begin BFS
    for task in my_task.upstream_list:
        tasks_queue.put(item=task)
    # perform BFS
    while not tasks_queue.empty():
        task: BaseOperator = tasks_queue.get()
        ancestor_task_ids.add(element=task.task_id)
        for _task in task.upstream_list:
            tasks_queue.put(item=_task)
    # Convert task_ids to actual tasks
    ancestor_tasks: List[BaseOperator] = [task for task in my_task.dag.tasks if task.task_id in ancestor_task_ids]
    return ancestor_tasks

Above snippet is NOT tested, but I'm sure you can take inspiration from it


References

Anirudh Bagri
  • 2,346
  • 1
  • 21
  • 33
y2k-shubham
  • 10,183
  • 11
  • 55
  • 131
-1

A modern approach with Airflow 2

from airflow.decorators import dag, task

...

@task
def task1():
    return "x" 

@task
def task2():
    return "y"

@task
def print_task_ids(x, y, **context):
    dag_run = context["dag_run"]
    tis = dag_run.get_task_instances()

    for ti in tis:
        print(ti.task_id)


@dag(**dag_args)
def show_task_ids():
    print_task_ids(task1(), task2())
    

dag = show_task_ids()
Matias Lopez
  • 94
  • 1
  • 7
  • your answer does not event use `x`, `y` parameters that are crucial for getting correct answer. – sol25 Jan 11 '23 at 11:39