Airflow Single-Engineer Group

1. You are here:
2. Engineering
3. Incubation Engineering Department
4. Airflow Single-Engineer Group

On this page

• About the Airflow SEG
  • Latest video
  • Previous 5 videos
• Apache Airflow
• Typical workflow of developing DAGs
• Common challenges
• GitLab integration

About the Airflow SEG

Latest video

Previous 5 videos

Date	Tl;DW;	Video
2023-01-12	DAG overview page is now pretty	https://youtu.be/E3_YGF7Wr2k
2023-01-05	Developed the first Airflow page with an overview of Dags	https://youtu.be/oFs4OsHZfRw
2022-12-21	First video that started this SEG	https://youtu.be/Jrjp6_rdDo4

Apache Airflow

Airflow is the de facto tool for data teams to schedule and execute ELT pipelines, Machine Learning pipelines, DevOps tasks and really any task that requires scheduling. Its cronjob turned up to 11.

According to Airflow themselves:

Airflow is a platform created by the community to programmatically author, schedule and monitor workflows

Source: https://airflow.apache.org

A workflow is also called a Directed Acyclic Graph (DAG) in Airflow, a DAG contains tasks which utilize operators.

Typical workflow of developing DAGs

A typical development workflow looks like:

1. User locally creates or updates a DAG
2. Push code to Git
3. Pipeline deploys the DAG to a non-production Airflow instance
4. User visits the Airflow webserver to inspect and run the DAG
5. If the pipeline succeeds the DAG will via merge requests make its way to production

Common challenges

Below are some common challenges related to Airflow, in no particular order:

1. Airflow is single-tenant. In development, users are overwriting each others DAGs if they deploy to the same instance
2. Airflow is quite difficult to set up properly for a production environment
3. Developing DAGs is often very iterative
4. It's difficult to spot bugs in a DAG during code review without actually deploying it to an Airflow instance

GitLab integration

Below are a few of the initial options to integrate GitLab and Airflow:

1. Integration of the DAG overview into GitLab
2. GitLab as authentication provider for Airflow
3. Using GitLab runners as compute for Airflow
4. Using preview apps to create an instance of Airflow per MR to ease the code review process
5. Provision an Airflow instance directly from GitLab