Description / Context

The DAG file processor runs at regular intervals to parse and prepare python files ready for execution.

Airflow 2.x

This can be started automatically by default as a subprocess as part of the command airflow scheduler, or run as a stand-alone process by setting AIRFLOW__SCHEDULER__STANDALONE_DAG_PROCESSOR

Airflow 3.x

Always runs as a standalone process, to be started with airflow dag-processor.

Problem

As of 25.7.0 and Airflow 3.x, the Airflow operator runs the dag-processor as a standalone process, but as part of the scheduler role (and statefulset). This has a number of drawbacks:

• if the dag-processor crashes it is not detected and does not trigger a re-start automatically
• resources can not be set independently for the scheduler and the dag-processor
• logging levels can not be set independently for the scheduler and the dag-processor

Proposal

Introduce a new optional role to go alongside the existing ones:

• webservers
• schedulers
• celeryExecutors / kubernetesExecutors
• dagProcessors

This new role will encapsulate a command that runs airflow dag-processor.

The CRD would look something like this:

---
apiVersion: airflow.stackable.tech/v1alpha1
kind: AirflowCluster
metadata:
  name: airflow
spec:
  ...
  webservers:
    ...
  celeryExecutors:
    ...
  schedulers:
    ...
  dagProcessors:
    roleGroups:
      default:
        replicas: 1

Note the following:

• airflow dag-processor cannot be started as part of the scheduler in Airflow 3.x
  • so it will be either (still) included as a specific command in the scheduler pod, or moved to its own pod, depending on whether a dagProcessors role has been specified
• airflow dag-processor should default to being a scheduler subprocess in Airflow 2.x
  • so users can include a dagProcessors role, which will cause AIRFLOW__SCHEDULER__STANDALONE_DAG_PROCESSOR to be set to True