Skip to main content

Install Remote Execution Agents in a restricted kubernetes namespace

Airflow 3 only
This feature is only available for Airflow 3.x Deployments.

You can install the Remote Execution Agent in a Kubernetes namespace with restricted pod security standards. Your organization might have different security standards for infrastructure supporting internal-only sandboxes compared to production environments.

Kubernetes Pod Security Standards define different security levels for Pods:

  • Privileged: No restrictions (least secure)
  • Baseline: Prevents known privilege escalations
  • Restricted: Highly-constrained settings following security best practices (most secure)

The Restricted profile enforces the following limitations:

  • Runs containers as non-root users
  • Prevents privilege escalation
  • Drops all Linux capabilities
  • Uses read-only root filesystems when possible
  • Requires a runtime default seccomp profile

However, because of these limitations, you need to complete the following additional Remote Execution Agent configuration set up.

Step 1: Create a restricted namespace

Create a Namespace in your Kubernetes manifest with the following restricted Pod security standards:

apiVersion: v1
kind: Namespace
metadata:
name: astro-agent-restricted
labels:
pod-security.kubernetes.io/enforce: restricted
pod-security.kubernetes.io/audit: restricted
pod-security.kubernetes.io/warn: restricted

Step 2: Configure Global Security Settings

Modify your Agent's values.yaml file to set global security context settings that apply to all Agent components' Pods and containers:

# Global pod security context for all components
podSecurityContext:
seccompProfile:
type: RuntimeDefault
runAsUser: 50000
fsGroup: 50000
runAsNonRoot: true

# Global container security context for all components
containerSecurityContext:
allowPrivilegeEscalation: false
readOnlyRootFilesystem: true
capabilities:
drop:
- ALL

Step 3: Configure component-specific settings

When using the Agent in a restricted namespace, you must configure volume mounts because:

  1. The container security context sets readOnlyRootFilesystem: true
  2. These directories need write access during runtime
  3. Using emptyDir volumes provides isolated, writable storage that meets security requirements

Worker configuration

workers:
- name: default-worker
# Other worker settings...

# Required volumes for filesystem access
volumes:
- name: tmp
emptyDir: {}
- name: task-logs
emptyDir: {}

volumeMounts:
- name: tmp
mountPath: /tmp # This folder is used by the Astro Agent to maintain the socket file
- name: task-logs
mountPath: /usr/local/airflow/logs # Or the configured folder that holds the task logs

DAG Processor configuration

dagProcessor:
# Other dag processor settings...

# Required volumes for filesystem access
volumes:
- name: tmp
emptyDir: {}

volumeMounts:
- name: tmp
mountPath: /tmp # This folder is used by the Astro Agent to maintain the socket file

Triggerer configuration

triggerer:
# Other triggerer settings...

# Required volumes for filesystem access
volumes:
- name: tmp
emptyDir: {}
- name: task-logs
emptyDir: {}

volumeMounts:
- name: tmp
mountPath: /tmp # This folder is used by the Astro Agent to maintain the socket file
- name: task-logs
mountPath: /usr/local/airflow/logs # Or the configured folder that holds the task logs

Step 4: (Optional) Add logging sidecar configuration

loggingSidecar:
# Other logging sidecar settings...

securityContext:
allowPrivilegeEscalation: false
readOnlyRootFilesystem: true
capabilities:
drop:
- ALL
volumeMounts:
- name: task-logs
mountPath: /etc/airflow/logs # Or the configured folder that holds the task logs

Was this page helpful?