Counting 1,361 Big Data & Machine Learning Frameworks, Toolsets, and Examples...
Suggestion? Feedback? Tweet @stkim1

Airflow Scheduler Failover Controller

Project Purpose

The purpose of this project is to create a failover controller that will control which scheduler is up and running to allow HA across an Airflow cluster.


We had attempted to setup a Highly Available Airflow Cluster where we had 2 machines with all the normal Airflow Daemons (web server, scheduler, workers, etc.) running on them. Each of the instances would share a MySQL instances as its MetaStore and share a RabbitMQ Queue for its Queueing Services (since we were using CeleryExecutors).

What we noticed after a month running the Cluster is that the schedulers would occasionally push a duplicate task instance to the RabbitMQ Queue. Therefore the Airflow executors would execute the same task instance twice. This caused a lot of data inconsistency issues.

This is what motivated us to search for an alternative to our initial approach to build a highly available Airflow Cluster. It lead to us creating this module.

How it Works

The Airflow Scheduler Failover Controller (ASFC) is a mechanism that ensures that only one Scheduler instance is running in an Airflow Cluster at a time. This way you don't come across the issues we described in the "Motivation" section above.

You will first need to startup the ASFC on each of the instances you want the scheduler to be running on. When you start up multiple instances of the ASFC one of them takes on the Active state and the other takes on a Standby state. There is a heart beat mechanism setup to track if the Active ASFC is still active. If the Active ASFC misses multiple heart beats, the Standby ASFC becomes active.

The Active ASFC will poll every 10 seconds to see if the scheduler is running on the desired node. If it is not, the ASFC will try to restart the daemon. If the scheduler daemons still doesn't startup, the daemon is started on another node in the cluster.


Local Development

In case you want to do development work on the project

  1. Clone the project from GitHub to your local machine

    git clone
  2. Run pip install

    pip install -e .
  3. You're done!

    • After, you will be able to run the project through the CLI Interface (See bellow for more details), be able to make any changes to the project you just brought down and have those changes be immediately reflected in the CLI


  1. Select which version of the code you want to install and use this value as the {BRANCH_OR_TAG} placeholder. Current Options:

    • master
      • latest code in master
      • BRANCH
    • development
      • code in development branch - experimental fixes and features
      • BRANCH
    • v1.0.0
      • version 1.0.0 of the code
      • RELEASE
    • v1.0.1
      • version 1.0.1 of the code
      • RELEASE
  2. Run pip install

    pip install git+git://{BRANCH_OR_TAG}

CLI Interface

usage: scheduler_failover_controller [-h]

positional arguments:
                        sub-command help
    version             Prints out the version of the Scheduler Failover
    init                Initialize Configurations to allow Scheduler Failover
                        Controller to run
    test_connection     Tests if you can connect to all the necessary machines
                        listed in 'scheduler_nodes_in_cluster' config
                        Checks if the Scheduler is running on the machines you
                        have listed in 'scheduler_nodes_in_cluster' config
    clear_metadata      Clear the Metadata in Metastore
    metadata            Get the Metadata from Metastore
    send_test_email     Send a Test Email
    get_current_host    Get the Current Hostname
    start               Start the Airflow Scheduler Failover Controller

optional arguments:
  -h, --help            show this help message and exit


  1. Run pip uninstall

    pip uninstall scheduler_failover_controller
  2. If you ran the installation for development, follow these steps:

    a. Get the bin file location and use this value as the {BIN_CLI_FILE_PATH} placeholder

    which scheduler_failover_controller

    b. Remove the bin file


Startup/Status/Shutdown Instructions



Startup in the foreground
scheduler_failover_controller start
Startup as a background process
nohup scheduler_failover_controller start > /dev/null &


ps -eaf | grep scheduler_failover_controller


for pid in `ps -ef | grep "scheduler_failover_controller" | awk '{print $2}'` ; do kill -9 $pid ; done


The Airflow Scheduler Failover Controller also allows you to control the process with systemd. The systemd files can be found under ${PROJECT_HOME}/scripts/systemd. Within this directory is a file which describes how to deploy the systemd files. Bellow illustrates how to run the process after deploying the files.

Note: Run as Root


sudo systemctl start scheduler_failover_controller


sudo systemctl status scheduler_failover_controller


sudo systemctl stop scheduler_failover_controller


sudo systemctl restart scheduler_failover_controller

Getting Started

This is a step by step set of instructions you can take to get up and running with the scheduler_failover_controller

  1. Install the ASFC on all the desired machines

    • See the above section entitled "Installation"
  2. Run the following CLI command to get the default configurations setup in airflow.cfg

    scheduler_failover_controller init
  3. Update the default configurations that were added to the bottom of the airflow.cfg file under the [scheduler_failover] section

    a. Main ones include updating: scheduler_nodes_in_cluster, alert_to_email

    • For scheduler_nodes_in_cluster property, it is recommended that you use the value printed from the following command:

      scheduler_failover_controller get_current_host

    • So, if this command printed out 'ip-10-0-0-98', include this as the value for that particular host.

  4. Enable all the machines to be able to ssh to each of the other machines with the user you're running airflow as

    a. Create a public and private key SSH key on all of the machines you want to act as schedulers. You can follow these instructions:

    b. Add the public key content to the ~/.ssh/authorized_keys file on all the other machines

  5. Run the following CLI command to test the connection to all the machines that will act as Schedulers

    scheduler_failover_controller test_connection
  6. Startup the following Airflow Daemons

    a. webserver

    nohup airflow webserver $* >> ~/airflow/logs/webserver.logs &

    b. workers (If you're using the CeleryExecutor)

    nohup airflow worker $* >> ~/airflow/logs/celery.logs &
  7. Startup the Airflow Scheduler Failover Controller on each node you would like acting as the Scheduler Failover Controller (ONE AT A TIME).

    • See the above section entitled "Startup/Status/Shutdown Instructions"
  8. View the logs to ensure things are running correctly

    • Location of the logs can be determined by the 'logging_dir' configuration entry in the airflow.cfg

    • Note: Logs are set by default to rotate at midnight and only keep 7 days worth of backups. This can be overridden in the configuration file.

  9. View the metadata to ensure things are being set correctly

    scheduler_failover_controller metadata

Recommended Steps for a Better Deployment

Above describes a quickstart approach. However, if you're looking for a better long term approach for using the Airflow Scheduler Failover Controller then you can follow the bellow steps.

Setup Systemd for Airflow

Airflow provides scripts to help you control the airflow daemons through the systemctl command. It is recommended that you setup the airflow-scheduler, at least, for systemd.

Go to and follow the instructions in the README file to get it setup.

Update airflow.cfg configs to use the Systemd

airflow_scheduler_start_command = sudo systemctl restart airflow-scheduler

airflow_scheduler_stop_command = sudo systemctl stop airflow-scheduler

Setup the Airflow Scheduler Failover Controller to use Systemd

Follow the instructions in ${PROJECT_HOME}/scripts/systemd/ to set it up.

Latest Releases
 Dec. 13 2016
 Nov. 28 2016
 Nov. 14 2016