Skip to main content

Astro Runtime architecture

Astro Runtime is a production ready, data orchestration tool based on Apache Airflow that is distributed as a Docker image and is required by all Astronomer products. It is intended to provide organizations with improved functionality, reliability, efficiency, and performance.

Deploying Astro Runtime is a requirement if your organization is using Astro. Astro Runtime includes the following features:

  • Timely support for new patch, minor, and major versions of Apache Airflow. This includes bug fixes that have not been released by the open source project but are backported to Astro Runtime and available to users earlier.
  • Exclusive features to enrich the task execution experience, including smart task concurrency defaults and high availability configurations.
  • The astronomer-providers package. This package is an open source collection of Apache Airflow providers and modules maintained by Astronomer. It includes deferrable versions of popular operators such as ExternalTaskSensor, DatabricksRunNowOperator, and SnowflakeOperator. See Astronomer Providers documentation
  • The openlineage-airflow package. OpenLineage standardizes the definition of data lineage, the metadata that forms lineage metadata, and how data lineage metadata is collected from external systems. This package enables data lineage on Astro. See OpenLineage and Airflow.
  • A custom logging module that ensures Airflow task logs are reliably available to the Astro data plane.
  • A custom security manager that enforces user roles and permissions as defined by Astro. See User permissions.
  • A custom Airflow UI that includes links to Astronomer resources and exposes the currently running Docker image tag in the footer of all UI pages.

For more information about the features that are available in Astro Runtime releases, see the Astro Runtime release notes.

Runtime versioning

Astro Runtime versions are released regularly and use semantic versioning. Astronomer ships major, minor, and patch releases of Astro Runtime in the format of major.minor.patch.

  • Major versions are released for significant feature additions. This includes new major or minor versions of Apache Airflow, as well as API or DAG specification changes that are not backward compatible.
  • Minor versions are released for functional changes. This includes API or DAG specification changes that are backward compatible, which might include new minor versions of astronomer-providers and openlineage-airflow.
  • Patch versions are released for bug and security fixes that resolve unwanted behavior. This includes new patch versions of Apache Airflow, astronomer-providers, and openlineage-airflow.

Every version of Astro Runtime correlates to an Apache Airflow version. All Deployments on Astro must run only one version of Astro Runtime, but you can run different versions of Astro Runtime on different Deployments within a given cluster or Workspace. See Create a Deployment.

For a list of supported Astro Runtime versions and more information on the Astro Runtime maintenance policy, see Astro Runtime versioning and lifecycle policy.

Astro Runtime and Apache Airflow parity

This table lists Astro Runtime releases and their associated Apache Airflow versions.

Astro RuntimeApache Airflow version
42.2
52.3
62.4
72.5

For version compatibility information, see the Runtime release notes.

Default environment variables

The following table lists the default Runtime environment variables. You can override the values of these environment variables to meet the unique requirements of your organization. For information about the global environment variables set on the Astro data plane, see Global environment variables. To edit default environment variable values, see Set environment variables on Astro.

Environment VariableDescriptionValue
AIRFLOW__SCHEDULER__DAG_DIR_LIST_INTERVALThe time in seconds that Airflow waits before re-scanning the dags directory for new files. Note that this environment variable is set for all Deployments regardless of Runtime version.30
AIRFLOW__CELERY__STALLED_TASK_TIMEOUTThe maximum time in seconds that tasks running with the Celery executor can remain in a queued state before they are automatically rescheduled.600
AIRFLOW_CORE_PARALLELISMThe maximum number of task instances that can run concurrently for each scheduler in your Deployment.[number-of-running-workers-for-all-worker-queues] * [max-tasks-per-worker]
AIRFLOW__KUBERNETES_EXECUTOR__WORKER_PODS_CREATION_BATCH_SIZEThe number of worker Pods that can be created each time the scheduler parses DAGs. This setting limits the number of tasks that can be scheduled at one time.16

Provider packages

All Astro Runtime images have the following open source provider packages pre-installed:

Provider package versioning

If an Astro Runtime release includes changes to an installed version of a provider package that is maintained by Astronomer (astronomer-providers or openlineage-airflow), the version change is documented in the Astro Runtime release notes.

To determine the version of any provider package installed in your current Astro Runtime image, run:

docker run --rm <runtime-image> pip freeze | grep <provider>

Python versioning

Astro Runtime supports Python 3.9. This is the only version of Python that Astro Runtime supports. If your data pipelines require an unsupported Python version and you're running Astro Runtime 6.0 (based on Airflow 2.4) or later, Astronomer recommends that you use the ExternalPythonOperator. See ExternalPythonOperator.

If you're currently using the KubernetesPodOperator or the PythonVirtualenvOperator in your DAGs, you can continue to use them to create virtual or isolated environments that can run tasks with different versions of Python.

Executors

In Airflow, the executor is responsible for determining how and where a task is completed.

In all local environments created with the Astro CLI, Astro Runtime runs the Local executor. On Astro, Astro Runtime exclusively supports the Celery executor. Support for the Kubernetes executor is currently in Private Preview.

Soon, Astronomer will provide a new executor with intelligent worker packing, task-level resource requests, improved logging, and Kubernetes-like task isolation.

Distribution

Astro Runtime is distributed as a Debian-based Docker image. Runtime Docker images have the following format:

  • quay.io/astronomer/astro-runtime:<version>
  • quay.io/astronomer/astro-runtime:<version>-base

An Astro Runtime image must be specified in the Dockerfile of your Astro project. Astronomer recommends using non-base images, which incorporate ONBUILD commands that copy and scaffold your Astro project directory so you can more easily pass those files to the containers running each core Airflow component. A base Astro Runtime image is recommended for complex use cases that require additional customization, such as installing Python packages from private sources.

For a list of all Astro Runtime Docker images, see Quay.io.

System distribution

The following table lists the operating systems and architectures supported by each Astro Runtime version. If you're using a Mac computer with an M1 chip, Astronomer recommends using Astro Runtime 6.0.4 or later.

Astro RuntimeOperating System (OS)Architecture
4Debian 11.3 (bullseye)AMD64
5Debian 11.3 (bullseye)AMD64
6Debian 11.3 (bullseye)AMD64 and ARM64
7Debian 11.3 (bullseye)AMD64 and ARM64

Astro Runtime 6.0.4 and later images are multi-arch and support AMD64 and ARM64 processor architectures for local development. Docker automatically uses the correct processor architecture based on the computer you are using.