creating-openlineage-extractors

To capture OpenLineage lineage for unsupported Airflow operators, you create custom extractors or operator lineage methods that map operator properties to OpenLineage datasets, then register them via the AIRFLOW__OPENLINEAGE__EXTRACTORS configuration.

You need a custom OpenLineage extractor when third-party operators lack built-in support, when inputs and outputs are determined at runtime, or when you need richer lineage details like failure-aware metadata that inlets and outlets cannot express.

You register a custom OpenLineage extractor in Airflow by setting the AIRFLOW__OPENLINEAGE__EXTRACTORS configuration to the import path of your custom extractor class, enabling Airflow to emit lineage events for the targeted operator.

Yes, you can optionally enrich OpenLineage extractors with facets to support advanced needs like output statistics and column-level lineage, providing deeper metadata about the datasets involved in your Airflow tasks.

You safely implement OpenLineage extraction by following operational guardrails for approach selection and registration, ensuring that custom extractor hooks handle lineage capture without disrupting the actual Airflow task execution.

OpenLineage works with third-party provider operators by implementing custom extractors that define operator hooks, allowing you to emit input and output dataset lineage for operators you cannot directly modify.