Airflow - How to pass xcom variable into Python function

问题

I need to reference a variable that's returned by a BashOperator. I may be doing this wrong so please forgive me. In my task_archive_s3_file, I need to get the filename from get_s3_file. The task simply prints {{ ti.xcom_pull(task_ids=submit_file_to_spark) }} as a string instead of the value.

If I use the bash_command, the value prints correctly.

get_s3_file = PythonOperator(
    task_id='get_s3_file',
    python_callable=obj.func_get_s3_file,
    trigger_rule=TriggerRule.ALL_SUCCESS,
    dag=dag)

submit_file_to_spark = BashOperator(
    task_id='submit_file_to_spark',
    bash_command="echo 'hello world'",
    trigger_rule="all_done",
    xcom_push=True,
    dag=dag)

task_archive_s3_file = PythonOperator(
    task_id='archive_s3_file',
#    bash_command="echo {{ ti.xcom_pull(task_ids='submit_file_to_spark') }}",
    python_callable=obj.func_archive_s3_file,
    params={'s3_path_filename': "{{ ti.xcom_pull(task_ids=submit_file_to_spark) }}" },
    dag=dag)

get_s3_file >> submit_file_to_spark >> task_archive_s3_file

回答1:

Templates like {{ ti.xcom_pull(...) }} can only be used inside of parameters that support templates or they won't be rendered prior to execution. See the template_fields and template_ext attributes of the PythonOperator and BashOperator.

So templates_dict is what you use to pass templates to your python operator:

def func_archive_s3_file(**context):
    archive(context['templates_dict']['s3_path_filename'])

task_archive_s3_file = PythonOperator(
    task_id='archive_s3_file',
    dag=dag,
    python_callable=obj.func_archive_s3_file,
    provide_context=True,  # must pass this because templates_dict gets passed via context
    templates_dict={'s3_path_filename': "{{ ti.xcom_pull(task_ids='submit_file_to_spark') }}" })

However in the case of fetching an XCom value, another alternative is just using the TaskInstance object made available to you via context:

def func_archive_s3_file(**context):
    archive(context['ti'].xcom_pull(task_ids='submit_file_to_spark'))

task_archive_s3_file = PythonOperator(
    task_id='archive_s3_file',
    dag=dag,
    python_callable=obj.func_archive_s3_file,
    provide_context=True,

回答2:

Upvoted both the question and the answer, but I think that this can be made a little more clear for those users who just want to pass small data objects between PythonOperator tasks in their DAGs. Referencing this question and this XCom example got me to the following solution. Super simple:

from airflow.models import DAG
from airflow.operators.python_operator import PythonOperator

DAG = DAG(
  dag_id='example_dag',
  start_date=datetime.now(),
  schedule_interval='@once'
)

def push_function(**kwargs):
    ls = ['a', 'b', 'c']
    return ls

push_task = PythonOperator(
    task_id='push_task', 
    python_callable=push_function,
    provide_context=True,
    dag=DAG)

def pull_function(**kwargs):
    ti = kwargs['ti']
    ls = ti.xcom_pull(task_ids='push_task')
    print(ls)

pull_task = PythonOperator(
    task_id='pull_task', 
    python_callable=pull_function,
    provide_context=True,
    dag=DAG)

push_task >> pull_task

I'm not sure why this works, but it does. A few questions for the community:

What's happening with ti here? How is that built in to **kwargs?
Is provide_context=True necessary for both functions?

Any edits to make this answer clearer are very welcome!

回答3:

Used the same code and modified params like Startdate etc.

import airflow
from datetime import datetime, timedelta
from airflow.models import DAG
from airflow.operators.python_operator import PythonOperator

args = {
    'owner': 'Airflow',
    'start_date': airflow.utils.dates.days_ago(2),
}

DAG = DAG(
  dag_id='simple_xcom',
  default_args=args,
#  start_date=datetime(2019, 04, 21),
  schedule_interval="@daily",
  #schedule_interval=timedelta(1)
)

def push_function(**context):
    msg='the_message'
    print("message to push: '%s'" % msg)
    task_instance = context['task_instance']
    task_instance.xcom_push(key="the_message", value=msg)

push_task = PythonOperator(
    task_id='push_task', 
    python_callable=push_function,
    provide_context=True,
    dag=DAG)

def pull_function(**kwargs):
    ti = kwargs['ti']
    msg = ti.xcom_pull(task_ids='push_task',key='the_message')
    print("received message: '%s'" % msg)

pull_task = PythonOperator(`enter code here`
    task_id='pull_task', 
    python_callable=pull_function,
    provide_context=True,
    dag=DAG)

push_task >> pull_task

来源：https://stackoverflow.com/questions/46059161/airflow-how-to-pass-xcom-variable-into-python-function

标签

airflow