Scaling On-Prem Airflow With 2,000 DAGs at Numberly with Sébastien Crocquevieille

Scaling 2,000+ data pipelines isn’t easy. But with the right tools and a self-hosted mindset, it becomes achievable.

In this episode, Sébastien Crocquevieille, Data Engineer at Numberly, unpacks how the team scaled their on-prem Airflow setup using open-source tooling and Kubernetes. We explore orchestration strategies, UI-driven stakeholder access and Airflow’s evolving features.

Key Takeaways:

00:00 Introduction.

02:13 Overview of the company’s operations and global presence.

04:00 The tech stack and structure of the data engineering team.

04:24 Running nearly 2,000 DAGs in production using Airflow.

05:42 How Airflow’s UI empowers stakeholders to self-serve and troubleshoot.

07:05 Details on the Kubernetes-based Airflow setup using Helm charts.

09:31 Transition from GitSync to NFS for DAG syncing due to performance issues.

14:11 Making every team member Airflow-literate through local installation.

17:56 Using custom libraries and plugins to extend Airflow functionality.

Resources Mentioned:

Sébastien Crocquevieille

https://www.linkedin.com/in/scroc/

Numberly | LinkedIn

https://www.linkedin.com/company/numberly/

Numberly | Website

https://numberly.com/

Apache Airflow

https://airflow.apache.org/

Grafana

https://grafana.com/

Apache Kafka

https://kafka.apache.org/

Helm Chart for Apache Airflow

https://airflow.apache.org/docs/helm-chart/stable/index.html

Kubernetes

https://kubernetes.io/

GitLab

https://about.gitlab.com/

KubernetesPodOperator – Airflow

https://airflow.apache.org/docs/apache-airflow-providers-cncf-kubernetes/stable/operators.html

Beyond Analytics Conference

https://astronomer.io/beyond/dataflowcast

Than