talk-data.com

Observability in data workflows often stops at logs and metrics, leaving data lineage as a blind spot. At Booking, we set out to change that by treating lineage as a core observability layer. In this talk, I'll walk through how we integrated lineage tracking into our Airflow ecosystem, what metadata we capture, and how we surface it to users in a meaningful way. I'll also share how lineage data helps us debug failures, detect unexpected changes, and ensure compliance. You'll leave with a practical view of what it takes to make lineage not just visible, but actionable.

Running Airflow at scale for thousands of workflows across multiple teams introduces challenges around standardization, governance, and isolation. At Booking, we've built a multi-tenant Airflow platform that serves over 4,000 workflows using a custom DSL defined in workflow.yaml files. In this talk, I'll show how we use automated DAG generation to bring structure to complexity, how we achieved horizontal scalability by decoupling orchestration from execution, and how reusable step templates help us enforce governance--without sacrificing workflow isolation. You'll leave with a blueprint for taming Airflow at scale.

Traditionally, managing the lifecycle of data in workflows at Booking involved ad hoc tracking and custom logic to handle data changes. With the adoption of data assets, we now have a standardized way to represent, version, and evolve data over time. In this talk, I'll introduce how data assets are implemented at Booking, how versioning is handled under the hood, and how our workflows are built to consume and respond to these evolving assets. I'll close with a real-world example that shows how data assets help us ensure consistency and traceability in a complex production workflow.