Teams often come to Dagster from Airflow. The concepts map, but not one to one.
- An Airflow DAG of tasks maps to a Dagster job, or, better, to assets when each task produces a table or file. The asset graph then comes from data dependencies instead of task order.
- XCom passes small values between tasks; Dagster passes values through IO managers or, for tables, has downstream assets read the table with
deps.
- Connections and variables become resources with
EnvVar.
- Airflow schedules become schedules or automation conditions; Airflow sensors become sensors.
catchup and backfills become partitions and Dagster backfills.
Airflow 3 added data-aware scheduling with assets too, so the difference today is emphasis: Airflow is task-first with the widest operator ecosystem; Dagster is asset-first with lineage, partitions and checks built in. Migrate incrementally: observe the Airflow DAGs from Dagster first, then move one DAG at a time, keeping both running until the outputs match.
The Airflow tasks load_orders and build_revenue become the tables they write, connected by what each one reads.