airflow-local

• Source code - Github
• Author - Gavin Noronha - [email protected]

About

This project provides a Ubuntu (16.04) Vagrant Virtual Machine (VM) with Airflow, a data workflow management system from Airbnb.

There are Ansible scripts that automatically install the software when the VM is started.

1. Run Airflow

Connect to the VM

1. To start the virtual machine(VM) type

   vagrant up
   

2. Setup private keys

   copy-private-keys.bat
   

3. Connect to the VM

   vagrant ssh airflow-master
   

4. Setup private keys

   /vagrant/scripts/setup-private-keys.sh
   

Initialize Airflow

1. Initialize Airflow home

   export AIRFLOW_HOME=~/airflow
   

2. Setup the sqlite database

   airflow initdb
   

3. Change to the Airflow directory

   cd $AIRFLOW_HOME
   

4. Reduce the load on the Airflow system by setting values in airflow.cfg

   parallelism = 4
   dag_concurrency = 2
   celeryd_concurrency = 4
   

5. Start the web server

   airflow webserver -p 8080
   

6. Open a web browser to the UI at http://192.168.33.10:8080

Run a task

1. List DAGS

   airflow list_dags
   

2. List tasks for example_bash_operator DAG

   airflow list_tasks example_bash_operator
   

3. List tasks for example_bash_operator in a tree view

   airflow list_tasks example_bash_operator -t
   

4. Run the runme_0 task on the example_bash_operator DAG today

   airflow run example_bash_operator runme_0 `date +%Y-%m-%d`
   

5. Backfill a DAG

   export START_DATE=$(date -d "-2 days" "+%Y-%m-%d")
   airflow backfill -s $START_DATE example_bash_operator
   

6. Clear the history of DAG runs

   airflow clear example_bash_operator
   

Add a new task

1. Go to the Airflow config directory

   cd ~/airflow
   

2. Set the airflow dags directory in airflow.cfg by change the line:

   dags_folder = /vagrant/airflow/dags
   

3. Remove example dags

   load_examples = False
   

4. Restart the web server

   airflow webserver -p 8080
   

Run a task

1. Run the dynamic_dags task

   airflow list_dags
   

2. Run the dag

   airflow trigger_dag dynamic_dags
   

3. Run the scheduler to actually run the dag

   airflow scheduler
   

Disable logging

1. Change to the airflow directory

   cd /vagrant/airflow
   

2. Set airflow environment

   source set_airflow_env.sh
   

3. Run airflow without any logging messages

Setup airflow dags directory

1. Edit file ~/airflow/airflow.cfg

2. Set the following:

   dags_folder = /vagrant/airflow/dags
   load_examples = False
   

3. Start the webserver & scheduler by running the following

   tmuxp load /vagrant/scripts/tmux-webserver-scheduler.yaml
   

Setup Airflow in Pseudo-distributed mode using Local Executor

Follow the instructions here

Setup Airflow in distributed mode using Celery Executor

Follow the instructions here

2. Run RabbitMQ

Start RabbitMQ

1. Start the RabbitMQ in a Docker container

   export RMQ_IMG=rabbitmq:3.6.10-management
   docker run -d --rm --hostname airflow-rmq \
       --name airflow-rmq -p 192.168.33.10:15672:15672 -p 5672:5672 $RMQ_IMG
   

2. Display the list of running Docker instances

   docker ps
   

3. Go to the RabbitMQ dashboard at http://192.168.33.10:15672/

4. Login using guest/guest

Stop RabbitMQ

1. Connect to the RabbitMQ Docker container

   export RMQ=$(docker ps -aq --filter name=airflow-rmq)
   

2. List queues

   docker exec -ti $RMQ rabbitmqctl list_queues
   

3. Stop RabbitMQ

   docker stop $RMQ
   

3. Connect to RabbitMQ using Python

Connect to RabbitMQ only using Python (no Celery)

The RabbitMQ web site demonstrates how to connect using Python and the Pika library.

1. List queues

   docker exec -ti $RMQ rabbitmqctl list_queues
   

2. Send a message to a RabbitMQ queue called hello

   python rmq-send.py
   

3. Receive a message from RabbitMQ queue called hello

   python rmq-receive.py
   

4. List queues displaying the hello queue

   docker exec -ti $RMQ rabbitmqctl list_queues
   

5. Stop the app

   docker exec -ti $RMQ rabbitmqctl stop_app
   

6. Start the app

   docker exec -ti $RMQ rabbitmqctl start_app
   

7. List queues and the hello queue is not displayed

   docker exec -ti $RMQ rabbitmqctl list_queues
   

Connect to RabbitMQ using Celery

1. Start the Celery worker

   export PYTHONPATH=/vagrant/scripts
   celery -A tasks worker --loglevel=info
   

2. Call the task

   export PYTHONPATH=/vagrant/scripts
   python -c "from tasks import add; add.delay(2, 3)"
   

4. Run Postgres

Start Postgres

1. Start the Postgres in the Docker container with the name

   export PG_IMG=postgres:9.6.3
   export PGPASSWORD=airflow_pg_pass
   docker run -d --rm --name airflow-pg -p 0.0.0.0:5432:5432 \
       -e POSTGRES_PASSWORD=$PGPASSWORD $PG_IMG
   export PG=$(docker ps -aq --filter name=airflow-pg)
   

Stop Postgres

1. List the Docker container

   docker ps --filter id=$PG
   

2. Stop Postgres

   docker stop $PG
   

5. Connect to Postgres using Psycopg2 and SQLAlchemy

Start the Postres database before running these steps

1. Connect to the database using psql and create the database test

   docker exec -ti $PG psql -U postgres -c "create database test"
   

2. Create table test in database test only using Psycopg2

   export PGHOST=localhost
   python /vagrant/scripts/pg-psycopg2.py
   

3. Connect to database test using Psycopg2 and SQLAlchemy

   python pg-sqlalchemy-read.py
   

4. Connect to the postgres database again

   docker exec -ti $PG psql -U postgres
   

5. List the databases

   \l
   

6. Connect to the test database

   \c test
   

7. List objects in the test database

   \d
   

8. Select all rows from the test database

   select id, num, data from test;
   

9. Quit the psql utility

   \q
   

10. Drop database test

    docker exec -ti $PG psql -U postgres -c "drop database test"
    

6. Setup Airflow with Postgres

1. Using the LocalExecutor

1. Change the executor in ~/airflow.cfg file to the following values

   executor = LocalExecutor
   

2. Change the sql_alchemy_conn in ~/airflow.cfg file to the following values

   # Change the meta db configuration
   sql_alchemy_conn = postgresql+psycopg2://postgres:airflow_pg_pass@localhost/test
   

2. Using the CeleryExecutor with Redis

1. Change the executor in ~/airflow.cfg file to the following values

   executor = CeleryExecutor
   

2. Set the following two values in ~/airflow.cfg file

   broker_url = redis://localhost:6379/0
   celery_result_backend = redis://localhost:6379/0
   

3. Initialize the database with Airflow data

1. Initialze the test database as the Airflow database

   airflow initdb
   

4. Start airflow daemons with Tmux

```
tmuxp load /vagrant/scripts/tmux-airflow-daemons.yaml
```

5. View the Airflow web server at http://192.168.33.10:8080

6. View the Airflow flower server at http://192.168.33.10:5555

7. Setup Airflow worker machine

1. Copy Airflow configuration

   rsync -zvh airflow/airflow*.cfg worker:~/airflow/
   

2. Initialize Airflow home export AIRFLOW_HOME=~/airflow

3. Run airflow worker airflow worker

4. Stop redis sudo systemctl stop redis

5. Modify Redis configuration sudo vim /etc/redis/redis.conf

6. Change bind line to the following bind 0.0.0.0

7. Start Redis sudo systemctl start redis

8. Start airflow daemons with supervisord

1. Create postgres Docker container
2. Create test database
3. Run airflow init
4. Run sudo supervisord
5. Need to setup logs in airflow-scheduler.conf, airflow-webserver.conf, airflow-worker.conf

9. Setup netdata for monitoring

1. Edit netdata configuration

   vim /opt/netdata/etc/netdata/netdata.conf
   

2. Show status of netdata

   sudo systemctl status netdata
   

3. Start netdata

   sudo systemctl start netdata
   

4. View the netdata at http://192.168.33.10:19999/

5. Stop netdata

   sudo systemctl stop netdata
   

6. Netdata stores data in memory and updates every second. To store hours of data without using up memory add the following

   [global]
   update every = 10
   

Documentation

1. Main documentation

   • https://pythonhosted.org/airflow/

2. Videos on Airflow

   • Best practices with Airflow
   • A Practical Introduction to Airflow

3. Slides

   • http://www.slideshare.net/walterliu7/airflow-a-data-flow-engine
   • http://www.slideshare.net/Hadoop_Summit/airflow-an-open-source-platform-to-author-and-monitor-data-pipelines

4. Airflow reviews

   • http://bytepawn.com/airflow.html
   • https://www.pandastrike.com/posts/20150914-airflow

5. Airflow tips and tricks

   • https://medium.com/handy-tech/airflow-tips-tricks-and-pitfalls-9ba53fba14eb#.i2hu0syug
   • Airflow with Postgres + RabbitMQ
   • Three tips on using Celery
   • Building a Data Pipeline with Airflow
   • https://databricks.com/blog/2016/12/08/integrating-apache-airflow-databricks-building-etl-pipelines-apache-spark.html
   • http://site.clairvoyantsoft.com/installing-and-configuring-apache-airflow/
   • https://gtoonstra.github.io/etl-with-airflow/principles.html
   • https://cwiki.apache.org/confluence/display/AIRFLOW/Common+Pitfalls
   • http://michal.karzynski.pl/blog/2017/03/19/developing-workflows-with-apache-airflow/

Requirements

The following software is needed to get the software from github and run Vagrant to set up the Python development environment. The Git environment also provides an SSH client for Windows.

• Oracle VM VirtualBox
• Vagrant version 1.9 or higher
• Git