Conditional Tasks using PythonOperator and BranchPythonOperator

前端 未结 1 653
清歌不尽
清歌不尽 2021-01-29 11:44

Hi Guys am new to airflow and python. I need to run the tasks based on the value of a variable in the input json. If the value of the variable \'insurance\' is

1条回答
  •  清酒与你
    2021-01-29 12:06

    Primary problem in your code

    1. The dag-definition-file is continuously parsed by Airflow in background and the generated DAGs & tasks are picked by scheduler. The way your file wires tasks together creates several problems

      • all 6 tasks (task1 .. task6) are ALWAYS created (and hence they will always run, irrespective of insurance_flag); just their inter-task dependency is set in accordance with insurance_flag

      • the correct way instead is to put both task instantiation (creation of PythonOperator taskn object) as well as task wiring within that if .. else block. That ways, the unnecessary tasks won't be created (and hence won't run)

    2. While the point 1. above alone should be enough to fix your code, may i offer you a suggestion for improvement: having a Variable being read in dag definition file means a SQL query being fired by Airflow's SQLAlchemy ORM very frequently in background (every cycle of continuously parsing dag-definition file)

      • this not just unnecessarily overloads your SQLAlchemy backend meta-db, but also slows down parser (in extreme case can lead of DagBag timeout if parsing starts taking too long)
      • instead you can leverage that BranchPythonOperator in right way to move that Variable reading on runtime (when DAG / tasks will be actually run) rather than Dag generation time (when dag-file is parsed by Airflow and DAG is generated on webserver); here is the code for that (and you should do away with that if-else block completely)
      """ branch 1 """
      task1 >> task2 >> task3
      """ branch 2 """
      task4 >> task5 >> task6
      
      def branch_decider(**kwargs):
          my_var_dict = Variable.get('my_var_name', deserialize_json=True)
          # decide which branch to take based on insurance flag
          if my_var_dict['car']['json']['insurance']:
              return 'task1'
          else:
              return 'task4'
      
      branch_task = BranchPythonOperator(task_id='branch_task',
                                         dag=dag,
                                         python_callable=branch_decider)
      

    Other (minor) problems in your code

    • Missing mandatory dag argument from task instantiations

         task1 = BashOperator(
           task_id='task1',
           bash_command='echo 1',
           dag=dag
         )
      
    • a dagling PythonOperator with a callable which json.dumps Variable that is solving no purpose (unless i misunderstood you code / intent here, remove it completely)

      PythonOperator(
          task_id='sample_task',
          python_callable=sample_fun,
          op_kwargs={
              json: '{{ dag_run.car.json}}'
          },
          provide_context=True,
          dag=dag
      )
      
      
      def sample_fun(json, **kwargs):
          insurance_flag = json.dumps(json)['insurance']
      

    UPDATE-1

    Responding to queries raised over comments

    We have used Variable.get( my_ var_ name). What is this my_ var_ name

    Variables have a key & value, my_var_name is the key of variable (see the Key column in following screenshot from Airflow UI)


    If condition satisfies return 'task1', 'task2', 'task3' else 'task4', 'task5', 'task6'. Can we add more than 1 tasks in return

    No you can't. (you don't have to)

    BranchPythonOperator requires that it's python_callable should return the task_id of first task of the branch only

    • 1st branch: task1, task2, task3, first task's task_id = task1
    • 2nd branch: task4, task5, task6, first task's task_id = task4

    Furthermore do understand that since the above two sets of tasks have already been wired together, so they will be naturally executed after one-another in that sequence (otherwise what would be the point of wiring them anyways?)

    task1 >> task2 >> task3
    

    Check out these links (in addition to links already inlined in answer above)

    • (official repo) example_branch_python_dop_operator_3.py
    • (official repo) example_branch_operator_3.py
    • AirflowPythonBranchOperator examples (note the incorrect name of Operator here)

    0 讨论(0)
提交回复
热议问题