Group tasks inside DAGs

Cloud Composer 3 | Cloud Composer 2 | Cloud Composer 1

This page describes how you can group tasks in your Airflow pipelines using the following design patterns:

• Grouping tasks in the DAG graph.
• Triggering children DAGs from a parent DAG.
• Grouping tasks with the TaskGroup operator.

Important: Airflow provides SubDAGs to address repeating tasks. Despite being a common design pattern for grouping tasks together, SubDAGs often cause performance and functional issues, and is deprecated in Airflow. We recommend to avoid using SubDAGs to group tasks together in your workflow and prefer one of the alternative approaches described in this page.

Group tasks in the DAG graph

To group tasks in certain phases of your pipeline, you can use relationships between the tasks in your DAG file.

Consider the following example:

Figure 1. Tasks can be grouped together in an Airflow DAG (click to enlarge)

In this workflow, tasks op-1 and op-2 run together after the initial task start. You can achieve this by grouping tasks together with the statement start >> [task_1, task_2].

The following example provides a complete implementation of this DAG:

Airflow 2

from airflow import DAG
from airflow.operators.bash import BashOperator
from airflow.operators.dummy import DummyOperator
from airflow.utils.dates import days_ago

DAG_NAME = "all_tasks_in_one_dag"

args = {"owner": "airflow", "start_date": days_ago(1), "schedule_interval": "@once"}

with DAG(dag_id=DAG_NAME, default_args=args) as dag:
    start = DummyOperator(task_id="start")

    task_1 = BashOperator(task_id="op-1", bash_command=":", dag=dag)

    task_2 = BashOperator(task_id="op-2", bash_command=":", dag=dag)

    some_other_task = DummyOperator(task_id="some-other-task")

    task_3 = BashOperator(task_id="op-3", bash_command=":", dag=dag)

    task_4 = BashOperator(task_id="op-4", bash_command=":", dag=dag)

    end = DummyOperator(task_id="end")

    start >> [task_1, task_2] >> some_other_task >> [task_3, task_4] >> end

Airflow 1

from airflow import DAG
from airflow.operators.bash_operator import BashOperator
from airflow.operators.dummy_operator import DummyOperator
from airflow.utils.dates import days_ago

DAG_NAME = "all_tasks_in_one_dag"

args = {"owner": "airflow", "start_date": days_ago(1), "schedule_interval": "@once"}

with DAG(dag_id=DAG_NAME, default_args=args) as dag:
    start = DummyOperator(task_id="start")

    task_1 = BashOperator(task_id="op-1", bash_command=":", dag=dag)

    task_2 = BashOperator(task_id="op-2", bash_command=":", dag=dag)

    some_other_task = DummyOperator(task_id="some-other-task")

    task_3 = BashOperator(task_id="op-3", bash_command=":", dag=dag)

    task_4 = BashOperator(task_id="op-4", bash_command=":", dag=dag)

    end = DummyOperator(task_id="end")

    start >> [task_1, task_2] >> some_other_task >> [task_3, task_4] >> end

Trigger children DAGs from a parent DAG

You can trigger one DAG from another DAG with the TriggerDagRunOperator operator.

Consider the following example:

Figure 2. DAGs can be triggered from within a DAG with the TriggerDagRunOperator (click to enlarge)

In this workflow, the blocks dag_1 and dag_2 represent a series of tasks that are grouped together in a separate DAG in the Cloud Composer environment.

The implementation of this workflow requires two separate DAG files. The controlling DAG file looks like the following:

Airflow 2

from airflow import DAG
from airflow.operators.dummy import DummyOperator
from airflow.operators.trigger_dagrun import TriggerDagRunOperator
from airflow.utils.dates import days_ago


with DAG(
    dag_id="controller_dag_to_trigger_other_dags",
    default_args={"owner": "airflow"},
    start_date=days_ago(1),
    schedule_interval="@once",
) as dag:
    start = DummyOperator(task_id="start")

    trigger_1 = TriggerDagRunOperator(
        task_id="dag_1",
        trigger_dag_id="dag-to-trigger",  # Ensure this equals the dag_id of the DAG to trigger
        conf={"message": "Hello World"},
    )
    trigger_2 = TriggerDagRunOperator(
        task_id="dag_2",
        trigger_dag_id="dag-to-trigger",  # Ensure this equals the dag_id of the DAG to trigger
        conf={"message": "Hello World"},
    )

    some_other_task = DummyOperator(task_id="some-other-task")

    end = DummyOperator(task_id="end")

    start >> trigger_1 >> some_other_task >> trigger_2 >> end

Airflow 1

from airflow import DAG
from airflow.operators.dagrun_operator import TriggerDagRunOperator
from airflow.operators.dummy_operator import DummyOperator
from airflow.utils.dates import days_ago


with DAG(
    dag_id="controller_dag_to_trigger_other_dags",
    default_args={"owner": "airflow"},
    start_date=days_ago(1),
    schedule_interval="@once",
) as dag:
    start = DummyOperator(task_id="start")

    trigger_1 = TriggerDagRunOperator(
        task_id="dag_1",
        trigger_dag_id="dag-to-trigger",  # Ensure this equals the dag_id of the DAG to trigger
        conf={"message": "Hello World"},
    )
    trigger_2 = TriggerDagRunOperator(
        task_id="dag_2",
        trigger_dag_id="dag-to-trigger",  # Ensure this equals the dag_id of the DAG to trigger
        conf={"message": "Hello World"},
    )

    some_other_task = DummyOperator(task_id="some-other-task")

    end = DummyOperator(task_id="end")

    start >> trigger_1 >> some_other_task >> trigger_2 >> end

Note: The value for trigger_dag_id inside TriggerDagRunOperator must match the dag_id value of the DAG you want to trigger.

The implementation of the child DAG, which is triggered by the controlling DAG, looks like the following:

Airflow 2

from airflow import DAG
from airflow.operators.dummy import DummyOperator
from airflow.utils.dates import days_ago

DAG_NAME = "dag-to-trigger"

args = {"owner": "airflow", "start_date": days_ago(1), "schedule_interval": "None"}

with DAG(dag_id=DAG_NAME, default_args=args) as dag:
    dag_task = DummyOperator(task_id="dag-task")

Airflow 1

from airflow import DAG
from airflow.operators.dummy_operator import DummyOperator
from airflow.utils.dates import days_ago


DAG_NAME = "dag-to-trigger"

args = {"owner": "airflow", "start_date": days_ago(1), "schedule_interval": "None"}

with DAG(dag_id=DAG_NAME, default_args=args) as dag:
    dag_task = DummyOperator(task_id="dag-task")

You must upload both DAG files in your Cloud Composer environment for the DAG to work.

Grouping tasks with the TaskGroup operator

This approach works only in Airflow 2.

You can use the TaskGroup operator to group tasks together in your DAG. Tasks defined within a TaskGroup block are still part of the main DAG.

Consider the following example:

Figure 3. Tasks can be visually grouped together in the UI with the TaskGroup operator (click to enlarge)

The tasks op-1 and op-2 are grouped together in a block with ID taskgroup_1. An implementation of this workflow looks like the following code:

from airflow.models.dag import DAG
from airflow.operators.bash import BashOperator
from airflow.operators.dummy import DummyOperator
from airflow.utils.dates import days_ago
from airflow.utils.task_group import TaskGroup

with DAG(dag_id="taskgroup_example", start_date=days_ago(1)) as dag:
    start = DummyOperator(task_id="start")

    with TaskGroup("taskgroup_1", tooltip="task group #1") as section_1:
        task_1 = BashOperator(task_id="op-1", bash_command=":")
        task_2 = BashOperator(task_id="op-2", bash_command=":")

    with TaskGroup("taskgroup_2", tooltip="task group #2") as section_2:
        task_3 = BashOperator(task_id="op-3", bash_command=":")
        task_4 = BashOperator(task_id="op-4", bash_command=":")

    some_other_task = DummyOperator(task_id="some-other-task")

    end = DummyOperator(task_id="end")

    start >> section_1 >> some_other_task >> section_2 >> end

What's next

• Write DAGs
• Test DAGs