Distributed airflow - local

Use docker to run a distributed airflow on my own machine


  • python 2.7-slim (docker pull python:2.7-slim)

  • redis (docker pull redis)

  • postgres (docker pull postgres)

Start up the docker containers

  • Redis

$ docker run -p 6379:6379 --name some-redis -d redis redis-server --appendonly yes
  • Postgres

$ docker run -p 5432:5432 --name some-postgres -e POSTGRES_PASSWORD=mysecretpassword -d postgres
  • Master node ( webserver and scheduler)

$ docker run -it -p 8080:8080 --name webserver --link some-redis:redis --link some-postgres:postgres -d python:2.7-slim
  • Woker

$ docker run -it -p 5555:5555 -p 8793:8793 --name worker --link some-redis:redis --link some-postgres:postgres -d python:2.7-slim

Airflow on master and worker

Repeat the steps for both 2 containers

Access the container docker exec -it webserver bash

  • Install airflow and necessary configurations

$ apt-get update
$ apt-get install -y --no-install-recommends git make g++ libssl-dev libffi-dev
$ pip install apache-airflow==1.9.0
$ pip install celery[redis]
$ pip install psycopg2
$ apt-get install vim lsof -y
# Add AIRFLOW_HOME environment to .bash_rc
$ vim ~/.bashrc
export AIRFLOW_HOME=/home/airflow
$ source ~/.bashrc
  • Config airflow.cfg inside the webserver and worker node

$ airflow initdb
$ cd /home/airflow
$ vim airflow.cfg
# Edit airflow.cfg to our requirements
executor = CeleryExecutor
sql_alchemy_conn = postgresql+psycopg2://postgres:[email protected]$POSTGRES_PORT_5432_TCP_ADDR/postgres
broker_url = redis://
result_backend = redis://
  • Run airflow worker command to make sure that the scheduler and worker both point to same redis url. The result for both should look like:

  • Simple dags for testing. Put the dag in both master and worker machine

from airflow import DAG
from airflow.operators.bash_operator import BashOperator
from datetime import datetime, timedelta
default_args = {
'owner': 'airflow',
'depends_on_past': False,
'start_date': datetime(2015, 6, 1),
'email': ['[email protected]'],
'email_on_failure': False,
'email_on_retry': False,
'retries': 1,
'retry_delay': timedelta(minutes=5)
dag = DAG('test_airflow', default_args=default_args, schedule_interval=timedelta(days=1))
t1 = BashOperator(
bash_command='echo "Hello from Task 1"',
t2 = BashOperator(
bash_command='echo "World from Task 2"',
  • Start up the airflow and check the result

# On master machine
$ airflow start webserver
$ airflow scheduler
# On worker machine
$ airflow worker