Manage Airflow executors on Astro

The Airflow executor determines which worker resources run your scheduled tasks.

On Astro, every Deployment requires an executor and you can change the executor at any time. After you choose an executor for an Astro Deployment, you can configure your dags and Deployment resources to maximize the executor’s efficiency and performance. Use the information provided in this topic to learn how to configure the Astro, Celery, and Kubernetes executors on Astro.

To learn more about executors in Airflow, see Airflow executors.

Choose an executor

The difference between executors is based on how tasks are distributed across worker resources. The executor you choose affects the infrastructure cost of a Deployment and how efficiently your tasks execute. Astro currently supports three executors:

• Astro executor
• Celery executor
• Kubernetes executor

Read the following topics to learn about the benefits and limitations of each executor. For information about how to change the executor of an existing Deployment, see Update the Deployment executor.

Astro executor

The Astro executor is the default for all new Airflow 3.x Deployments. The Astro executor consists of agents (workers) that pull work from an API server. The API server manages the agent lifecycle and controls task assignment logic, enabling more efficient workload distribution and scaling. This design differs from the Celery executor, where workers fetch tasks directly from a queue, and the Kubernetes executor, where the scheduler launches pods for each task.

By centrally managing both agent scaling and task assignment, the Astro executor offers increased reliability, improved performance, and cost efficiency compared to other Airflow execution models.

The Astro executor is a good fit for your Deployment if:

• You need more control over agent scaling and task distribution.
• You want a balance between dynamic scaling and efficient task assignment.
• Your workload requires high reliability and performance.
• Cost optimization is a priority for your environment.

Celery executor

The Celery executor uses a group of workers, each of which can run multiple tasks at a time. Astronomer uses worker autoscaling logic to determine how many workers run on your Deployment at a given time.

The Celery executor is a good option for most use cases. Specifically, the Celery executor is a good fit for your Deployment if:

• You’re just getting started with Airflow.
• You want to use different worker types based on the type of task you’re running. See Configure worker queues.
• You have many short-running tasks.
• Your tasks require the shortest startup latency (often milliseconds).
• You don’t require task or dependency isolation.
• You want to direct tasks to run in worker queues with varying resources. This is useful if you want to run a subset of tasks with priority for SLA reasons, or if your tasks have varied resource requirements.

If you find that some tasks consume the resources of other tasks and cause them to fail, Astronomer recommends implementing worker queues or moving to the Kubernetes executor.

See Manage the Celery executor to learn more about how to configure the Celery executor.

Kubernetes executor

The Kubernetes executor runs each task in an individual Kubernetes Pod that’s defined either in your task or Deployment configuration. When a task completes, its Pod is terminated and the resources are returned to your cluster. On Astro, the infrastructure required to run the Kubernetes executor is built into every Deployment and is managed by Astronomer.

You can specify the configuration of a task’s Pod, including CPU and memory, as part of your dag definition using the Kubernetes Python Client and the pod_override arg. Any task without a pod_override runs in a default Pod as configured on your Deployment.

The Kubernetes executor is a good option for some use cases. Specifically, the Kubernetes executor is a good fit for your Deployment if:

• Your tasks are compute-intensive or you are processing large volumes of data within the task. Kubernetes executor tasks run separately in a dedicated Pod per task.
• Your tasks can tolerate some startup latency (often seconds)
• Your tasks require task or dependency isolation.
• You have had issues running certain tasks reliably with the Celery executor.

If you’re running a high task volume or cannot tolerate startup latency, Astronomer recommends the Celery executor. To learn more about using the Kubernetes executor, see Manage the Kubernetes executor.