Skip to content

joa-rodrigues/airflow

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

8 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Airflow

Introduction

This project is part of an example data pipeline, presented as part of a talk at ACM's [Applicative 2016 Conference] (http://applicative.acm.org/speakers/ypodeswa.html). Slides are available [here] (https://docs.google.com/presentation/d/1hX_fPTu92YBIny6LwvUfyF597YT6Bu0F7TgLr6focGk/edit?usp=sharing), which describe the data pipeline. This pipeline is made of 3 projects, all meant to be stitched together:

  • An event loading job, which reads JSON events from S3, and loads them into different database tables based on the class of event (organization payments, generic organization events, generic user events, and unknown events)
  • A job that calculates organization statistics, including key stats like how much each organization is paying, how active the users in the org are, etc. These stats could be used by an Account Manager to monitor the health of an organization. It depends on the output of the event loading job
  • This project, an implementation of Airbnb's Airflow system, which acts as a communication and orchestration layer. It runs the jobs, making sure the the event loading job runs before the organization statistics job, and also handles things like job retries, job concurrency levels, and monitoring/alerting on failures

Note that this is meant to be somewhat of a skeleton pipeline - fork it and use the code as a starting point, tweaking it for your own needs, with your own business logic.

Airflow

Airflow expresses relationships between jobs as a Directed Acyclic Graph. It lets you set dependencies for jobs, so they only run when their dependencies complete successfully. It also lets you define retry logic for jobs, monitor job completion/errors, view job runs in a web UI, and more. Full docs here.

Configuration/Setup

Change the default production inventory for the playbook (playbook/inventories/production) to whichever host you want to deploy Airflow to. Update playbook/vars/airflow-dev.yml and playbook/vars/airflow-prod.yml with your choice of credentials/settings (mysql users, fernet keys, aws credentials that can be used to run Lambda jobs, etc.).

Running the App in Dev

vagrant up, then visit 192.168.33.11 in your browser to see the Airflow web interface.

Airflow consists of 3 Python services: a scheduler, a set of workers, and a web app. The scheduler determines what tasks airflow should perform when (i.e. what to monitor), the workers actually perform the tasks, and the web server gives you a web interface where you can view the statuses of all your jobs.

The logs for these services are located at:

$AIRFLOW_HOME/airflow-worker.log
$AIRFLOW_HOME/airflow-scheduler.log
$AIRFLOW_HOME/airflow-webserver.log

And you can start/stop/restart any of them with:

$ sudo service airflow-worker {start|stop|restart}
$ sudo service airflow-scheduler {start|stop|restart}
$ sudo service airflow-webserver {start|stop|restart}

You can also start/stop services with the dev-runner.sh script, run ./runner.sh -h for usage.

The DAG definitions can be found in the workflows dir.

Running the App in Prod

Run the playbook against the prod inventory:

$ ansible-playbook main.yml -i inventories/production

Deploy the airflow dir via your favourite means to $AIRFLOW_HOME on the prod server. For a quick MVP, if you don't want to use a more formal build/deploy tool, you can just tar and scp the dir up to the server. Restart Airflow services, and the jobs should run.

About

Workflow manager from Airbnb

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 54.1%
  • Shell 45.9%