Azure 6 's offering is Fabric's Apache Airflow 129 job, the successor to Data Factory's Workflow Orchestration Manager. It still runs Airflow 2.10.5 and consumes 5 or 10 Fabric capacity units while its pool is up. For BookNest's single daily DAG, about 26 seconds of task time a day, the options compare like this:
| Option | You still operate | Fixed cost a month | Airflow |
|---|---|---|---|
| Docker Compose 514 (Airflow in Docker Compose) | Everything | A host with 2 GB free | 3.3.2 |
| Amazon MWAA 24 mw1.small | DAGs, packages, IAM | $364.56 plus workers | 3.3.1 |
| Google Gen 3 | DAGs, packages, IAM | DCU-hours at $0.06 | 3.3.1 |
| Astro 17,728 Developer | DAGs, Runtime image | $260.40 plus worker time | 3.3.2 |
| Fabric Airflow job | DAGs, packages | 5-10 capacity units | 2.10.5 |
| Dagster 177,056 + Solo, Hybrid | Code and agent | About $22 | n/a |
Managed Airflow bills for an always-on scheduler, so one small DAG costs about as much as fifty. What the money buys is upgrades, a backed-up metadata database, autoscaling, monitoring and single sign-on. Ask who is on call when the stack's one owner is away; run the orchestrator in the cloud that holds your data, so tasks use IAM roles instead of long-lived keys; check how far behind Apache each service runs; and keep DAGs portable, since MWAA Serverless YAML and Fabric items do not move elsewhere.
BookNest stays self-hosted for now: one DAG, a database on its own server, and a team that knows the stack. A second team or a promised deadline (SLAs and Deadline Alerts) would change that. Because the DAGs, the dbt 37,942 project and booknest_pipeline.py are plain Airflow and Python, moving to MWAA or Astro means copying files, not rewriting.