Airflow, Big Data, Python

Installing and Configuring Apache Airflow

Apache Airflow is a platform to programmatically author, schedule and monitor workflows – it supports integration with 3rd party platforms so that you, our developer and user community, can adapt it to your needs and stack.

Additional Documentation:


Install Documentation:

GitHub Repo:

Preparing the Environment

Install all needed system dependencies


  1. SSH onto target machine (s) where you want to install Airflow
  2. Login as Root
  3. Install Required Libraries
    #Run upgrade
    apt-get update
    apt-get install unzip
    #Build Essentials - GCC Compiler
    apt-get install build-essential
    #Python Development
    apt-get install python-dev
    apt-get install libsasl2-dev
    apt-get install python-pandas
  4. Check Python Version
    1. Run the command:
      python -V
    2. If the version comes back as “Python 2.7.X” you can skip the rest of this step
    3. Install Python 2.7.X
      cd /opt
      sudo wget --no-check-certificate
      tar xf Python-2.7.6.tar.xz
      cd Python-2.7.6
      ./configure --prefix=/usr/local
      make && make altinstall
      ls -ltr /usr/local/bin/python*
      vi ~/.bashrc
      #add this line alias python='/usr/local/bin/python2.7'
      source ~/.bashrc
  5. Install PIP
    1. Run Install
      cd /tmp/
      unzip setuptools-*.zip
      cd setuptools-*
      easy_install pip
    2. Verify Installation
       which pip
      # Should print out the path to the pip command
    3. If you come across an issue where while using pip bellow, its still referring to python2.6, you can follow these instructions
      1. Replace the binaries in the /usr/bin/ directory with the ones that were just installed
        cd /usr/bin/
        #Backup old binaries
        mv pip pip-BACKUP
        mv pip2 pip2-BACKUP
        mv pip2.6 pip2.6-BACKUP
        #Setup symlinks to the new version of pip that was installed
        ln -s /usr/local/bin/pip pip
        ln -s /usr/local/bin/pip2 pip2
        ln -s /usr/local/bin/pip2.7 pip2.7

Troubleshooting installation on Ubuntu:

  • If you later get the error “error trying to exec ‘as’: execvp: No such file or directory” while trying to install airflow with PIP
    • Install the following:
      apt-get install binutils
      apt-get install gcc
      apt-get install build-essential
      pip install pandas
    • Retry installation
    • If the problem persists, uninstall the packages listed above and reinstall. Then rerun.


  1. SSH onto target machine(s) where you want to install Airflow
  2. Login as Root
  3. Install Required Libraries
    yum groupinstall "Development tools"
    yum install zlib-devel bzip2-devel openssl-devel ncurses-devel sqlite-devel python-devel wget cyrus-sasl-devel.x86_64
  4. Check Python Version
    1. Run the command:
      python -V
    2. If the version comes back as “Python 2.7.X” you can skip the rest of this step
    3. Install Python 2.7.X
      cd /opt
      sudo wget --no-check-certificate
      tar xf Python-2.7.6.tar.xz
      cd Python-2.7.6
      ./configure --prefix=/usr/local
      make && make altinstall
      ls -ltr /usr/local/bin/python*
      vi ~/.bashrc
      #add this line alias python='/usr/local/bin/python2.7'
      source ~/.bashrc
  5. Install PIP
    1. Run Install
      cd /tmp/
      cd setuptools-X.X
      easy_install pip
    2. Verify Installation
       which pip
      #Should print out "/usr/local/bin/pip"

Troubleshooting on CentOS:

  • If you get an error saying ImportError: No module named extern while Installing PIP with easy_install
    1. Reinstall python-setuptools:
      yum reinstall python-setuptools
    2. Retry installation

Install Airflow

Login as Root and run:

pip install airflow==1.7.0
pip install airflow[hive]==1.7.0
pip install airflow[celery]==1.7.0

Update: Common Issue with Celery

Recently there were some updates to the dependencies of Airflow where if you were to install the airflow[celery] dependency for Airflow 1.7.x, pip would install celery version 4.0.2. This version of celery is incompatible with Airflow 1.7.x. This would result in various types of errors including messages saying that the CeleryExecutor can’t be loaded or that tasks are not getting executed as they should.

To get around this issue, install an older version of celery using pip:

pip install celery==3.1.17

Install RabbitMQ

If you intend to use RabbitMQ as a message broker you will need to install RabbitMQ.If you don’t intend to, you can skip this step. For production it is recommended that you use CeleryExecutors which requires a message broker such as RabbitMQ.


Follow these steps: Install RabbitMQ

Recovering from a RabbitMQ Node Failure

If you’ve opted to setup RabbitMQ to run on as a cluster, and one of those cluster nodes fails, you can follow these steps to recover on airflow:

  1. Bring the RabbitMQ node and daemon back up
  2. Navigate to the RabbitMQ Management UI
  3. Click on Queues
  4. Delete the “Default” queue
  5. Restart Airflow Scheduler service

Install MySQL Dependencies

If you intend to use MySQL as an DB repo you will need to install some MySQL dependencies. If you don’t intend to, you can skip this step.

Install MySQL Dependencies on Ubuntu

  1. Install MySQL Dependencies
     apt-get install python-dev libmysqlclient-dev
     pip install MySQL-python

Install MySQL Dependencies on CentOS

  1. Install MySQL Dependencies
    yum install -y mysql-devel python-devel python-setuptools
    pip install MySQL-python

Configuring Airflow

Its recommended to use RabbitMQ.

Apache Airflow needs a home, ~/airflow is the default, but you can lay foundation somewhere else if you prefer (OPTIONAL)

export AIRFLOW_HOME=~/airflow

Run the following as the desired user (who ever you want executing the Airflow jobs) to setup the airflow directories and default configs

airflow initdb
#note: When you run this the first time, it will generate a sqlite file (airflow.db) in the AIRFLOW_HOME directory for the Airflow Metastore. If you don't intend to use sqlite as the Metastore then you can remove this file.

Make the following changes to the {AIRFLOW_HOME}/airflow.cfg file

  1. Change the Executor to CeleryExecutor (Recommended for production)
    executor = CeleryExecutor
  2. Point SQL Alchemy to MySQL (if using MySQL)
    sql_alchemy_conn = mysql://{USERNAME}:{PASSWORD}@{MYSQL_HOST}:3306/airflow
  3. Set dags are paused on startup. This is a good idea to avoid unwanted runs of the workflow. (Recommended)
    # Are DAGs paused by default at creation
    dags_are_paused_at_creation = True
  4. Don’t load examples
    load_examples = False
  5. Set the Broker URL (If you’re using CeleryExecutors)
    1. If you’re using RabbitMQ:
      broker_url = amqp://guest:guest@{RABBITMQ_HOST}:5672/
    2. If you’re using AWS SQS:
      broker_url = sqs://{ACCESS_KEY_ID}:{SECRET_KEY}@
      # Note: You will also need to install boto:
      $ pip install -U boto
  6. Point Celery to MySQL (if using MySQL)
    celery_result_backend = db+mysql://{USERNAME}:{PASSWORD}@{MYSQL_HOST}:3306/airflow
  7. Set the default_queue name used by CeleryExecutors (Optional: Primarily for if you have a preference of the default queue name or plan on using the same broker for multiple airflow instances)
    # Default queue that tasks get assigned to and that worker listen on.
    default_queue = {YOUR_QUEUE_NAME_HERE}
  8. Setup MySQL (if using MySQL)
    1. Login to the mysql machine
    2. Create the airflow database if it doesn’t exist
      CREATE DATABASE airflow CHARACTER SET utf8 COLLATE utf8_unicode_ci;
    3. Grant access
      grant all on airflow.* TO ‘USERNAME'@'%' IDENTIFIED BY ‘{password}';
  9. Run initdb to setup the database tables
    airflow initdb
  10. Create needed directories
    mkdir dags
    mkdir logs

Configuring Airflow – Advanced (Optional)

Email Alerting

Allow Email alerting for if a task or job fails.

  1. Edit the {AIRFLOW_HOME}/airflow.cfg file
  2. Set the properties
    1. Properties
      • SMTP_HOST - Host of the SMTP Server
      • SMTP_TLS - Whether to use TLS when connecting to the SMTP Server
      • SMTP_USE_SSL - Whether to use SSL when connecting to the SMTP Server
      • STMP_USER - Username for connecting to SMTP Server
      • SMTP_PORT - Port to use for SMTP Server
      • SMTP_PASSWORD - Password associated with the user thats used to connect to SMTP Server
      • SMTP_EMAIL_FROM - Email to send Alert Emails as
    2. Example
      email_backend = airflow.utils.send_email_smtp
      # If you want airflow to send emails on retries, failure, and you want to
      # the airflow.utils.send_email function, you have to configure an smtp
      # server here
      smtp_host = {SMTP_HOST}
      smtp_starttls = {SMTP_TLS: True or False}
      smtp_ssl = {SMTP_USE_SSL: True or False}
      smtp_user = {STMP_USER}
      smtp_port = {SMTP_PORT}
      smtp_password = {SMTP_PASSWORD}
      smtp_mail_from = {SMTP_EMAIL_FROM}

Password Authentication

To enable password authentication for the web app.

Follow these instructions:

Controlling Airflow Services

By default you have to use the Airflow Command line Tool to startup the services. You can use the bellow commands to startup the processes in the background and dump the output to log files.

Starting Services

  1. Start Web Server
    nohup airflow webserver $* >> ~/airflow/logs/webserver.logs &
  2. Start Celery Workers
    nohup airflow worker $* >> ~/airflow/logs/worker.logs &
  3. Start Scheduler
    nohup airflow scheduler >> ~/airflow/logs/scheduler.logs &
  4. Navigate to the Airflow UI
  5. Start Flower (Optional)
    • Flower is a web UI built on top of Celery, to monitor your workers.
    nohup airflow flower >> ~/airflow/logs/flower.logs &
  6. Navigate to the Flower UI (Optional)

Stopping Services

Search for the service and run the kill command:

# Get the PID of the service you want to stop
ps -eaf | grep airflow
# Kill the process
kill -9 {PID}

Setting up Systemd to Run Airflow

Deploy Systemd Scripts

  1. Login as Root
  2. Get the zipped up Airflow
    cd /tmp/
    #Example: "wget"
  3. Unzip the file
    unzip {AIRFLOW_VERSION}.zip
    # This will output extract the contents into: incubator-airflow-{AIRFLOW_VERSION}
  4. Distribute the Systemd files
    cd incubator-airflow-{AIRFLOW_VERSION}/scripts/systemd/
    # Update the contents of the airflow file.
    # Set the AIRFLOW_HOME if its anything other then the default
    vi airflow
    # Copy the airflow property file to the target location
    cp airflow /etc/sysconfig/
    # Update the contents of the airflow-*.service files
    # Set the User and Group values to the user and group you want the airflow service to run as
    vi airflow-*.service
    # Copy the airflow services files to the target location
    cp airflow-*.service /etc/systemd/system/

How to Use Systemd

# Starting up the Service
service airflow-webserver start

# Stopping the Service
service airflow-webserver stop
# Restarting the Service
service airflow-webserver restart
# Checking the Status of the Service
service airflow-webserver status
# Viewing the Logs
journalctl -u airflow-webserver -e
Celery Worker
# Starting up the Service
service airflow-worker start

# Stopping the Service
service airflow-worker stop
# Restarting the Service
service airflow-worker restart
# Checking the Status of the Service
service airflow-worker status
# Viewing the Logs
journalctl -u airflow-worker -e
# Starting up the Service
service airflow-scheduler start

# Stopping the Service
service airflow-scheduler stop
# Restarting the Service
service airflow-scheduler restart
# Checking the Status of the Service
service airflow-scheduler status
# Viewing the Logs
journalctl -u airflow-scheduler -e
Flower (Optional)
# Starting up the Service
service airflow-flower start

# Stopping the Service
service airflow-flower stop
# Restarting the Service
service airflow-flower restart
# Checking the Status of the Service
service airflow-flower status
# Viewing the Logs
journalctl -u airflow-flower -e

Setting up Airflow Services to Run on Machine Startup

chkconfig airflow-webserver on
Celery Worker
chkconfig airflow-worker on
chkconfig airflow-scheduler on
Flower (Optional)
chkconfig airflow-flower on

Testing Airflow

Example Dags

High Level Testing

Note: You will need to deploy the dag.

airflow test tutorial print_date 2016-03-30

#[2016-03-30 18:39:46,621] {} INFO - Output:
#[2016-03-30 18:39:46,623] {} INFO - Wed Mar 30 18:39:46 UTC 2016

Running a Sample Airflow DAG

Assume the following code is in the dag at {AIRFLOW_HOME}/dags/

from airflow import DAG
from airflow.operators import DummyOperator
from datetime import datetime, timedelta

default_args = {
    'owner': 'airflow',
    'start_date': - timedelta(seconds=10),
    'retries': 0

dag = DAG('sample', default_args=default_args, - timedelta(seconds=10))

op = DummyOperator(task_id='dummy', dag=dag)
Verify the DAG is Available

Verify that the DAG you deployed is available in the list of DAGs

airflow list_dags

The output should list the ‘sample’ DAG

Running a Test

Let’s test by running the actual task instances on a specific date. The date specified in this context is an execution_date, which simulates the scheduler running your task or dag at a specific date + time:

airflow test sample dummy 2016-03-30

Heres how to run a particular task. Note: It might fail if the dependent tasks are not run successfully.

airflow run sample dummy 2016-04-22T00:00:00 --local
Trigger DAG

Trigger a DAG run

airflow trigger_dag sample

Backfill will respect your dependencies, emit logs into files and talk to the database to record status. If you do have a webserver up, you’ll be able to track the progress. airflow webserver will start a web server if you are interested in tracking the progress visually as your backfill progresses.

airflow backfill sample -s 2016-08-21

Helpful Operations

Getting Airflow Version

airflow version

Find Airflow Site-Packages Installation Location

Sometimes it might be helpful to find the source code so you can perform some other operations to help customize the experience in Airflow. This is how you can find the location of where the airflow source code is installed:

  1. Start up a Python CLI
  2. Run the following code to find where the airflow source code is installed
    import site
    import os
    SITE_PACKAGES = site.getsitepackages()
    print "All Site Packages: " + str(SITE_PACKAGES)
    for site_package in SITE_PACKAGES:
    	test_path = site_package + "/airflow"
    	if os.path.exists(test_path):
    		AIRFLOW_INSTALL_DIR = test_path
    print "Site Page Containing Airflow: " + str(AIRFLOW_INSTALL_DIR)

Usual Site Package Paths:

  • Centos
    • /usr/lib/python2.7/site-packages

Change Alert Email Subject

By default, the Airflow Alert Emails are always sent with the subject like: Airflow alert: <TaskInstance: [DAG_NAME].[TASK_ID] [DATE] [failed]>. If you would like to change this to provide more information as to which Airflow cluster you’re working with you can follow these steps.

Note: It requires a very small modification of the Airflow Source Code.

  1. Go to the Airflow Site-Packages Installation Location
    1. Example Path: /usr/lib/python2.7/site-packages/airflow
  2. Edit the file
  3. Search for the text “Airflow alert: ”
    1. Using nano
      1. Open the file
      2. Hit CTRL+w
      3. Type in “Airflow alert” and hit enter
  4. Modify this string to whatever you would like.
    1. Original value ‘title = “Airflow alert: {self}”.format(**locals())”‘ will produce ‘Airflow alert: <TaskInstance: [DAG_NAME].[TASK_ID] [DATE] [failed]>’
    2. An updated value like ‘title = “Test Updated Airflow alert: {self}”.format(**locals())”‘ will produce ‘Test Updated Airflow alert: <TaskInstance: [DAG_NAME].[TASK_ID] [DATE] [failed]>’

Set Logging Level

If you want to get more information in the logs (debug) or log less information (warn) you can follow these steps to set the logging level

Note: It requires a very small modification of the Airflow Source Code.

  1. Go to the Airflow Site-Packages Installation Location of airflow
  2. Edit the file
  3. Set the LOGGING_LEVEL variable to your desired value
    1. debug → logging.DEBUG
    2. info → logging.INFO
    3. warn → logging.WARN
  4. Restart the Airflow Services


  1. Pingback: Setting up an Apache Airflow Cluster - Home

  2. Dan

    Hey Robert!

    Amazing article I really really appreciate you writing this out. I have an issue that I was hoping you might be able to help me out with though, related to mysql set up. I followed your instructions but I am getting the following error

    sqlalchemy.exc.OperationalError: (_mysql_exceptions.OperationalError) (1071, ‘Specified key was too long; max key length is 1000 bytes’) [SQL: u’\nCREATE TABLE sla_miss (\n\ttask_id VARCHAR(250) NOT NULL, \n\tdag_id VARCHAR(250) NOT NULL, \n\texecution_date DATETIME NOT NULL, \n\temail_sent BOOL, \n\ttimestamp DATETIME, \n\tdescription TEXT, \n\tPRIMARY KEY (task_id, dag_id, execution_date), \n\tCHECK (email_sent IN (0, 1))\n)\n\n’]

    I am hoping that there might be some way I can change the CHARACTER SET and or the sqlalchemy connection to handle this max_key_length issue?

    1. Dan

      Actually I think I was able to get this to work, just changed my create database to be ascii instead of utf-8 and added that to my sqlalchemy engine

      Created this question too

      Thanks again for the post I really enjoyed it!

  3. Pingback: Upgrading Apache Airflow Versions - Home

  4. JER

    Now you can set log level from airflow.cfg :
    logging_level = debug|info|warning|error|…

    It’s a really recent commit, so it’s not for production for now but soon 😉

    1. Robert Sanders Post author

      True this will hopefully be supported soon and we won’t have to make this kind source code change. 🙂

  5. Kom

    I’ve installed airflow using the root user, but cannot access airflow using the other users. Any suggestions on enabling airflow for the other users so that they don’t have to have root access to use airflow?

    1. Aditi

      Hey ,did you get answer to this question . Your help will be appreciated!

  6. Roger Kelly

    How do you start a DAG at startup time that should always run? I see CRON type schedules and “None” for manual start, but nothing that is auto startup? Do I have to do a “airflow trigger_dag” as part of my airflow service startup?

    1. Robert Sanders Post author

      You can setup a DAG to startup (by this i mean unpaused and ready to be scheduled) on creation by modifying the airflow.cfg file. Specifically by setting the “dags_are_paused_at_creation” variable to “False”. This will however do this for all the DAGs that you create, so be careful.

      After that, whenever you restart Airflow services, the DAG will retain its state (paused or unpaused). So if you restart Airflow, the scheduler will check to see if any DAG Runs have been missed based off the last time it ran and the current time and trigger DAG Runs as needed.

  7. Sergio

    I have seen several other docs referring to the usage of python virtual environment to install airflow, which is not the case with yours, can you explain the reason behind your decision?
    thanks and regards,

  8. Sue

    Hi, Robert,

    Love the way you wrote this technical article, it is very clear. Could you please provide some instructions on how to set up distributed airflow configuration and how to execute python programs or shell programs remotely from airflow?
    We have many existing cron jobs running on many servers, and we would like to set up airflow to manage all those crons in one place. Please help.
    Thank you very much!

  9. Satya

    Well composed precise information on Airflow. Much appreciated. Thank you so much.

  10. sravan

    I have successfully configured rabbitmq and mysql. I set celery executor. i am trying to execute example_xcom dag. push task is working fine. it is inserting data to xcom table. i checked in xcom table. but the puller is not working. xcom_pull method not able to pull the values of task id. could you please guide me.

  11. Meir

    any idea how to configure the airflow’s mysql connection to work with ssl?

  12. Glenn Sampson

    Great article, helped my get a production level environment up and running with RabbitMQ, Celery & MySql.

    Can i ask about the scheduler.logs and this command “nohup airflow scheduler >> ~/airflow/logs/scheduler.logs”

    If we set a scheduler_heartbeat_sec = 5 after 24 hour this file is 1GB.

    Is this the best method? Is there a way to reduce this file?

  13. Pingback: How to Create a Workflow in Apache Airflow to Track Disease Outbreaks in India – Devops Dojo

  14. RockStar

    Which versions of MySQL and AIrflow have you used?

  15. Maries

    I am adding new views to and also modifying I am making modifications in my local system. Please let me know how to deploy the changes to Airflow running in the localhost.

  16. Nikhil Bhalerao

    HI All,

    could you please help me how to install Apache Airflow cloudera,



    Hi, I am new to this community.

    Please provide more details to setup this tool in our organization.

    Thanks in advance for your reply

  18. Pingback: Apache Airflow Installation on Ubuntu 18.04 & 18.10 – Ryan Merlin

  19. Jeff

    Why would you install airflow via pip as root? Isn’t that bad practice

  20. Pingback: Airflow SequentialExecutor Installation manual and basic commands | Big Data Demytified

Leave a Reply

Your email address will not be published. Required fields are marked *