Core Concepts: DAG, Task, Run, Instance

Definition vs execution

A DAG is the definition: tasks and their dependencies plus a schedule. A task is one unit of work, created from an operator (a template such as BashOperator) or the @task decorator. Each time the DAG is scheduled, Airflow creates a DAG run; every task in it becomes a task instance with its own state (queued, running, success, failed, skipped, up_for_retry…). A DAG run is identified by its logical date (called the data interval in newer terms), which says which slice of time the run is about, not necessarily when it executes.

A first DAG

Save in the dags/ folder. This uses Airflow 3 imports (airflow.sdk); in Airflow 2 the equivalent imports are from airflow import DAG and from airflow.operators.bash import BashOperator. {{ ds }} is the logical date as YYYY-MM-DD. I ran it with airflow dags test hello_airflow 2026-10-01 on Airflow 3.0.6 and say_hello printed the line below.

from datetime import datetime

from airflow.sdk import DAG
from airflow.providers.standard.operators.bash import BashOperator

with DAG(
    dag_id="hello_airflow",
    start_date=datetime(2026, 9, 1),
    schedule="@daily",
    catchup=False,
    tags=["demo"],
) as dag:
    say_hello = BashOperator(task_id="say_hello", bash_command="echo Hello from {{ ds }}")
    say_bye = BashOperator(task_id="say_bye", bash_command="echo Bye")

    say_hello >> say_bye

Output:

Hello from 2026-10-01

Quick check: What is a task instance?

• The DAG file itself
• One task within one specific DAG run, with its own state
• The metadata database
• A worker machine