At a glance
| Source connectors | No built-in extraction; provider operators and hooks call external systems |
| Destination connectors | Same: tasks write through providers, SQL operators or other tools |
| Batch or streaming | Batch and event-triggered workflows; not a streaming engine |
| Change data capture | Not built in; orchestrate a CDC or ingestion tool |
| Transformations | Runs SQL, Python, Spark or dbt tasks you define |
| Orchestration | Core purpose: scheduling, dependencies, retries, backfills and monitoring |
| Deployment | Self-hosted (pip, Docker, Helm chart) or managed (MWAA, Google, Astronomer, Fabric) |
| Licensing | Apache License 2.0; managed services billed by their providers |
Apache Airflow is the default choice for many data teams that need to schedule and monitor work across several systems, helped by its provider packages and managed versions from AWS, Google and Astronomer. It does not move data on its own, and self-hosting means running a scheduler, API server and metadata database. Airflow 2 is end of life, so plan any new work on Airflow 3.
Overview
Apache Airflow is an Apache Software Foundation platform to programmatically author, schedule and monitor workflows. A workflow is a DAG (directed acyclic graph) of tasks defined in Python; the scheduler decides what runs and when, workers run the tasks, and a web UI shows runs, logs and history. Provider packages add operators and hooks for clouds, databases, warehouses and tools such as dbt, Airbyte and Fivetran. Airflow orchestrates; the data movement happens in whatever each task calls.
Airflow 3.0 was released on 22 April 2025 with DAG versioning, assets and event-driven scheduling, a Task Execution Interface and a new UI. Airflow 3.3.0 (6 July 2026) added stateful tasks and a Task SDK for Java and Go, and Airflow 3.4.0 (6 November 2026) added bounded task loops and a Dag importer that accepts Dags written in TypeScript, Go, Java or YAML. Airflow 3.4.0 supports Python 3.11 to 3.14, with PostgreSQL 14 to 18 or MySQL 8.0 and 8.4 as the metadata database.
Key features
- Workflows defined as Python DAGs, with schedules, retries, dependencies and backfills
- Assets and event-driven scheduling, so downstream DAGs can run when data is updated
- DAG versioning and a React-based UI introduced in Airflow 3
- Large catalogue of provider packages for AWS, Google Cloud, Azure, databases, dbt, Airbyte and Fivetran
- Task SDK for Java and Go (3.3) and a Dag importer for TypeScript, Go, Java and YAML Dags (3.4)
- Official Helm chart for Kubernetes deployments
- Managed versions from AWS, Google Cloud, Astronomer and Microsoft Fabric
Pros and cons
Pros
- Free and open source under the Apache License 2.0, governed by the Apache Software Foundation
- Broad provider ecosystem covering clouds, warehouses and data tools
- More managed hosting choices than other open source orchestrators
- Airflow 3 added assets, DAG versioning and event-driven scheduling
Cons
- Self-hosting means operating a scheduler, API server, DAG processor, workers and a metadata database
- An orchestrator only; it needs other tools or code to extract and load data
- Airflow 2 is end of life and Airflow 3 changed or removed several Airflow 2 features, so upgrades need work
- Managed services lag the open source release: MWAA and Google listed 3.3.1 when 3.4.0 was current
Pricing
Apache Airflow is free and open source under the Apache License 2.0; self-hosting costs are your own infrastructure. Managed services are sold separately by other vendors: Amazon MWAA, Google's Managed Service for Apache Airflow (formerly Cloud Composer), Astronomer Astro and Apache Airflow jobs in Microsoft Fabric, each with its own usage-based pricing. We have not quoted their rates here; check each provider's pricing page.
Pricing checked on the vendor's official page on 7 November 2026. Prices change; confirm before buying.
Who it is best for
Data teams that need one scheduler and monitor for pipelines spanning several tools and clouds, and that write Python.
Who should look elsewhere
- Teams that only need to copy data into a warehouse: an ingestion tool alone may be enough, see Apache Airflow vs Airbyte.
- Teams that want an asset-first model or a lighter Python framework: see Apache Airflow vs Dagster and Apache Airflow vs Prefect.
Verdict
Apache Airflow remains a safe, widely supported orchestrator, with the broadest managed hosting and provider ecosystem of the open source options. Airflow 3 closed some gaps with assets and event-driven scheduling, and 3.4 widens authoring beyond Python. Budget for the operations work if you self-host, and for the Airflow 2 to 3 migration if you still run Airflow 2.
Frequently asked questions
Is Airflow 2 still supported?
No. The Airflow version life cycle page lists Airflow 2 as end of life since 22 April 2026, after limited maintenance from 22 October 2025. Some managed services still list Airflow 2.11 builds, but the Apache project no longer maintains it.
What is the latest version of Apache Airflow?
Airflow 3.4.0, released on 6 November 2026, according to the Airflow announcements page checked on 7 November 2026.
Is Airflow an ETL tool?
Not on its own. Airflow schedules and monitors tasks; the extraction, loading and transformation happen in the systems and tools those tasks call. See Apache Airflow vs Fivetran.
Sources
- Apache Airflow announcements (release history)
- Apache Airflow supported versions and life cycle
- Apache Airflow prerequisites
- Apache Airflow blog (3.3 and 3.4 release posts)
- Amazon MWAA supported Airflow versions
- Google Managed Service for Apache Airflow versions
Checked November 2026.
How we research reviews and what star ratings mean: our editorial method.