To more strongly embrace the success and growing customer preference for OSS solutions, Cloud Composer is evolving to become Managed Service for Apache Airflow. This name change provides improved customer understanding of our portfolio while reinforcing our commitment to being the most open cloud ecosystem.

Managed Service for Apache Airflow versioning overview

This page describes Managed Service for Apache Airflow versioning and lists differences between major Managed Airflow versions.

Major versions of Managed Airflow

Managed Airflow has the following major versions:

Managed Airflow (Gen 3). This version has simplified network setup and hides infrastructure components including the environment's cluster and environment's dependencies on other services.
Managed Airflow (Gen 2). In this version, the environment's cluster scales automatically to meet the resource demands.

Important: On September 15, 2026, versions 2.0.x of Managed Airflow (Gen 2) will reach their planned end of life, and you won't be able to use them. Versions 2.1.x and later aren't affected by this change. We recommend upgrading your environment to the latest Managed Airflow (Gen 2) version or migrating to Managed Airflow (Gen 3).
Managed Airflow (Legacy Gen 1). Has manual environment scaling, the infrastructure is deployed to your projects and networks.

Important: On March 25, 2024, Managed Airflow (Legacy Gen 1) entered its post-maintenance mode. Google will not release any further updates to Managed Airflow (Legacy Gen 1), including new versions of Airflow, bugfixes, and security updates. On September 15, 2026, all Managed Airflow (Legacy Gen 1) environments will reach their planned end of life, and you won't be able to use them. We recommend migration to Managed Airflow (Gen 3).

Comparison of Managed Airflow versions

The following table lists major differences between Managed Airflow (Legacy Gen 1), Managed Airflow (Gen 2), and Managed Airflow (Gen 3).

Feature	Managed Airflow (Legacy Gen 1)	Managed Airflow (Gen 2)	Managed Airflow (Gen 3)
Image version	`composer-1.b.c-airflow-x.y.z`	`composer-2.b.c-airflow-x.y.z`	`composer-3-airflow-x.y.z-build.t`
Airflow versions	Airflow 1.10.* and Airflow 2	Airflow 2	Airflow 2, Airflow 3
Python versions	3.8.12	3.11.8 (3.11.5 and 3.8.12 in earlier versions)	3.11.8
Environment's cluster	Standard mode VPC-native or Routes-based Google Kubernetes Engine cluster	Autopilot mode VPC-native Google Kubernetes Engine cluster	Environment's cluster is not deployed into your project.
Horizontal scaling	Can adjust the number of nodes in the environment's cluster. This changes the number of Airflow workers. Can adjust the number of Airflow schedulers.	Automatically scaling number of Airflow workers, based on demand. Can set and change upper and lower limits for the number of workers. Can adjust the number of Airflow schedulers and triggerers	Automatically scaling number of Airflow workers, based on demand. Can set and change upper and lower limits for the number of workers. Can adjust the number of Airflow schedulers, triggerers, and DAG processors
Vertical scaling	Can set machine types for cluster nodes, Airflow web server and database when creating an environment. Can change machine types for Airflow web server and database.	Can set and change the workloads configuration: CPU, memory, and storage parameters for Airflow workers, schedulers, web server, and the environment size.	Can set and change the workloads configuration: CPU, memory, and storage parameters for Airflow workers, schedulers, web server, DAG processors, and the environment size.
Pricing model	Managed Airflow (Legacy Gen 1) pricing model	Managed Airflow (Gen 2) pricing model	Managed Airflow (Gen 3) pricing model
Access control	Managed Airflow (Legacy Gen 1) roles	Managed Airflow roles. Uses Workload Identity Federation for GKE and requires additional permissions for the Managed Airflow Service Agent account.	Managed Airflow roles. Uses Workload Identity Federation for GKE. Extra permissions are not required
Access through Workforce Identity Federation	Not supported	Supported	Supported
Highly Resilient environments	Not supported	Supported	Supported
Database retention policy	Not supported	Not yet available	Supported
Private IP networking	VPC peerings	Private Service Connect	Simplified network setup. Can be changed between Public and Private IP in an existing environment.
VPC Service Controls	Supported	Supported	Supported
Airflow executors	Celery Executor	Celery Executor	CeleryKubernetes Executor
Airflow triggerer (deferrable operators)	Not supported	Supported	Supported
Custom service account for KubernetesPodOperator	Supported	Supported	Not supported
CMEK	Supported	Supported	Supported
Web server access controls	Supported	Supported	Supported
Restarting web server	Supported	Supported	Supported
Web server plugins	Supported in Airflow 1 environments with disabled DAG serialization.	Supported	Supported, plugins can be disabled and enabled on demand.
Scheduler plugins	Not supported	Supported	Not supported
Listeners	Not supported	Supported	Only task-level listeners
TimeTables	Only Built-In	Custom and Built-In	Only Built-In
Scheduler custom PyPI packages	Not supported	Supported	Not supported
Installing custom plugins	Supported	Supported	Supported
Maintenance operations	All tasks can be impacted.	Tasks that take less than 55 minutes to execute are not impacted.	Tasks that take less than 24 hours to execute are not impacted.
Storing Snapshots	Supported for Airflow 2.x and 1.10.15 versions	Supported	Supported
Loading Snapshots	Not supported	Supported	Supported
Scheduled Snapshots	Not supported	Supported	Supported
Custom environment's bucket	Not supported	Supported	Supported
Saving Airflow task logs only in Cloud Logging	Not supported	Supported	Supported
Data lineage integration	Not supported	Supported	Supported
Composer Local Development CLI tool	Not supported	Supported	Supported

Comparison of Airflow versions

Managed Airflow (Gen 3) provides environments with Airflow 2 and Airflow 3.

Support for Managed Airflow (Gen 3) features

The following table summarizes Managed Airflow (Gen 3) features that aren't yet supported in Airflow 3.

Feature	Status	Description
Upgrades to Airflow 3 (snapshots)	Not yet supported in Airflow 3	Side-by-side migration of existing Managed Airflow (Gen 3) environments from Airflow 2 to Airflow 3 using snapshots.
Upgrades to Airflow 3 (in-place)	Not yet supported in Airflow 3	In-place upgrades of existing Managed Airflow (Gen 3) environments from Airflow 2 to Airflow 3 by specifying a version with Airflow 3 during an upgrade.

Known issues in Airflow 3

The following known issues have been identified in Airflow 3:

Warning messages in the logs of Airflow workloads. These messages can be safely ignored.
The airflowctl utility can't be easily used from headless environments such as Airflow workers in Managed Airflow or Docker containers. This is a limitation of the utility. For more information about this limitation, see Issue 58230 in the Airflow repository on GitHub.

Support for Airflow 3 features

The following table lists Airflow 3 features and their support:

Feature	Status
Airflow 3.0 UI	Supported
DAG versioning	Supported
Backfills (from Airflow UI)	Supported
Event-driven scheduling	Supported
Assets	Supported
Inference execution and hyper-parameter tuning	Supported
airflowctl command-line tool	Supported
DAG bundles (other than LocalDagBundle)	Not yet supported in Airflow 3
Edge Executor and tasks in other programming languages	Not yet supported in Airflow 3

Managed Service for Apache Airflow images

To run Apache Airflow, Managed Airflow builds Docker images that bundle Airflow releases with other common binaries and Python libraries.

Managed Airflow images include Airflow modifications that are specific to Managed Airflow and unsuitable for the upstream Airflow codebase. Each Managed Airflow image contains preinstalled packages along with Managed Airflow-specific Airflow changes.