High availability configuration in Apache Airflow - Commands & Configuration

Introduction

High availability configuration in Airflow ensures that the workflow system keeps running smoothly even if one part fails. It helps avoid downtime by having backup components ready to take over automatically.

When you want your Airflow scheduler to keep working even if one scheduler instance crashes

When you need your Airflow webserver to be accessible without interruption during maintenance

When you want to prevent losing running tasks if a worker node goes down

When you run Airflow in a production environment with many users and critical workflows

When you want to scale Airflow components across multiple machines for reliability

Config File - airflow.cfg

airflow.cfg

[core]
executor = CeleryExecutor
sql_alchemy_conn = postgresql+psycopg2://airflow:airflow@postgres/airflow
load_examples = False

[webserver]
web_server_port = 8080
web_server_worker_timeout = 120

[scheduler]
max_threads = 2
scheduler_heartbeat_sec = 5
num_runs = -1

[celery]
broker_url = redis://redis:6379/0
result_backend = db+postgresql://airflow:airflow@postgres/airflow
worker_concurrency = 4

[logging]
remote_logging = False

[metrics]
enable = True

[database]
sql_alchemy_pool_size = 5
sql_alchemy_max_overflow = 10

[secrets]
backend = airflow.providers.hashicorp.secrets.vault.VaultBackend

[ha]
enable = True

This airflow.cfg file configures Airflow for high availability:

• executor = CeleryExecutor: Runs tasks distributed across multiple worker nodes.
• sql_alchemy_conn: Connects to a PostgreSQL database shared by all components.
• broker_url and result_backend: Use Redis and PostgreSQL for messaging and task results.
• scheduler settings: Allow multiple scheduler threads and continuous running.
• ha.enable = True: Enables high availability features in Airflow.

This setup allows multiple schedulers and workers to run simultaneously, so if one fails, others continue working.

Commands

Initializes the Airflow metadata database. This sets up the tables Airflow needs to track workflows and tasks.

Terminal

airflow db init

Expected Output

INFO  [alembic.runtime.migration] Context impl PostgresqlImpl.

INFO  [alembic.runtime.migration] Will assume transactional DDL.

INFO  [alembic.runtime.migration] Running upgrade head

INFO  [alembic.runtime.migration] Upgrade done.

Creates an admin user to log into the Airflow web interface.

Terminal

airflow users create --username admin --firstname Admin --lastname User --role Admin --email [e-mail] --password admin123

Expected Output

Created user admin

→

--username - Set the login username

→

--role - Assign user role for permissions

Starts the Airflow scheduler in the background. Multiple schedulers can run for high availability.

Terminal

airflow scheduler &

Expected Output

No output (command runs silently)

Starts a Celery worker in the background to execute tasks. Multiple workers improve reliability and scalability.

Terminal

airflow celery worker &

Expected Output

No output (command runs silently)

Starts the Airflow webserver on port 8080 in the background, providing the user interface.

Terminal

airflow webserver -p 8080 &

Expected Output

[2024-06-01 12:00:00,000] {webserver} INFO - Starting web server on port 8080

→

-p - Specify the port number

Checks the health of the Airflow webserver and scheduler to confirm high availability is active.

Terminal

curl -s http://localhost:8080/health | jq .

Expected Output

{

  "metadatabase": {

    "status": "healthy"

  },

  "scheduler": {

    "status": "healthy"

  },

  "webserver": {

    "status": "healthy"

  }

}

Key Concept

If you remember nothing else from this pattern, remember: running multiple schedulers and workers with a shared database and message broker keeps Airflow running smoothly even if one component fails.

Common Mistakes

Using the SequentialExecutor instead of CeleryExecutor for high availability

SequentialExecutor runs tasks one at a time on a single machine, so it cannot handle multiple workers or schedulers.

Set executor = CeleryExecutor in airflow.cfg to enable distributed task execution.

Not configuring a shared database and message broker for all Airflow components

Without a shared backend, schedulers and workers cannot coordinate, causing failures and inconsistent state.

Use a central PostgreSQL database and Redis broker configured in airflow.cfg for all nodes.

Starting only one scheduler or worker process

Single processes create a single point of failure, defeating high availability.

Run multiple scheduler and worker processes on different machines or containers.

Summary

Initialize the Airflow database to prepare metadata storage.

Create an admin user to access the Airflow web interface.

Start multiple schedulers and Celery workers to enable high availability.

Run the Airflow webserver to provide the user interface.

Verify the health of Airflow components to confirm high availability is active.