Datahub Airflow Plugin

See the DataHub Airflow docs for details.

Version Compatibility

The plugin supports Apache Airflow versions 2.7+ and 3.1+.

Note on Airflow 3.0.x: Airflow 3.0.6 pins pydantic==2.11.7, which contains a bug that prevents the DataHub plugin from importing correctly. This issue is resolved in Airflow 3.1.0+ which uses pydantic>=2.11.8. If you must use Airflow 3.0.6, you can manually upgrade pydantic to >=2.11.8, though this may conflict with Airflow's dependency constraints. We recommend upgrading to Airflow 3.1.0 or later.

Related issue: https://github.com/pydantic/pydantic/issues/10963

Installation

The installation command varies depending on your Airflow version due to different OpenLineage dependencies.

For Airflow 2.x (2.7+)

pip install 'acryl-datahub-airflow-plugin[airflow2]'

This installs the plugin with Legacy OpenLineage (openlineage-airflow>=1.2.0), which is required for Airflow 2.x lineage extraction.

Alternative: Using Native OpenLineage Provider on Airflow 2.7+

If your Airflow 2.7+ environment rejects the Legacy OpenLineage package (e.g., due to dependency conflicts), you can use the native OpenLineage provider instead:

# Install the native Airflow provider first
pip install 'apache-airflow-providers-openlineage>=1.0.0'

# Then install the DataHub plugin without OpenLineage extras
pip install acryl-datahub-airflow-plugin

The plugin will automatically detect and use apache-airflow-providers-openlineage when available, providing the same functionality.

For Airflow 3.x (3.1+)

pip install 'acryl-datahub-airflow-plugin[airflow3]'

This installs the plugin with apache-airflow-providers-openlineage>=1.0.0, which is the native OpenLineage provider for Airflow 3.x.

Note: If using Airflow 3.0.x (3.0.6 specifically), you'll need to manually upgrade pydantic:

pip install 'acryl-datahub-airflow-plugin[airflow3]' 'pydantic>=2.11.8'

We recommend using Airflow 3.1.0+ which resolves this issue. See the Version Compatibility section above for details.

What Gets Installed

Base Installation (No Extras)

When you install without any extras:

pip install acryl-datahub-airflow-plugin

You get:

• acryl-datahub[sql-parser,datahub-rest] - DataHub SDK with SQL parsing and REST emitter
• pydantic>=2.4.0 - Required for data validation
• apache-airflow>=2.5.0,<4.0.0 - Airflow itself
• No OpenLineage package - You'll need to provide your own or use one of the extras below

With [airflow2] Extra

pip install 'acryl-datahub-airflow-plugin[airflow2]'

Adds:

• openlineage-airflow>=1.2.0 - Standalone OpenLineage package for Airflow 2.x

With [airflow3] Extra

pip install 'acryl-datahub-airflow-plugin[airflow3]'

Adds:

• apache-airflow-providers-openlineage>=1.0.0 - Native OpenLineage provider for Airflow 3.x

Additional Extras

You can combine multiple extras if needed:

# For Airflow 3.x with Kafka emitter support
pip install 'acryl-datahub-airflow-plugin[airflow3,datahub-kafka]'

# For Airflow 2.x with file emitter support
pip install 'acryl-datahub-airflow-plugin[airflow2,datahub-file]'

Available extras:

• airflow2: OpenLineage support for Airflow 2.x (adds openlineage-airflow>=1.2.0)
• airflow3: OpenLineage support for Airflow 3.x (adds apache-airflow-providers-openlineage>=1.0.0)
• datahub-kafka: Kafka-based metadata emission (adds acryl-datahub[datahub-kafka])
• datahub-file: File-based metadata emission (adds acryl-datahub[sync-file-emitter]) - useful for testing

Why Different Extras?

Airflow 2.x and 3.x have different OpenLineage integrations:

• Airflow 2.x (2.5-2.6) typically uses Legacy OpenLineage (openlineage-airflow package)
• Airflow 2.x (2.7+) can use either Legacy OpenLineage or native OpenLineage Provider (apache-airflow-providers-openlineage)
• Airflow 3.x uses native OpenLineage Provider (apache-airflow-providers-openlineage)

The plugin automatically detects which OpenLineage variant is installed and uses it accordingly. This means:

1. With extras ([airflow2] or [airflow3]): The appropriate OpenLineage dependency is installed automatically
2. Without extras: You provide your own OpenLineage installation, and the plugin auto-detects it

This flexibility allows you to adapt to different Airflow environments and dependency constraints.

Configuration

The plugin can be configured via airflow.cfg under the [datahub] section. Below are the key configuration options:

Extractor Patching (OpenLineage Enhancements)

When enable_extractors=True (default), the DataHub plugin enhances OpenLineage extractors to provide better lineage. You can fine-tune these enhancements:

[datahub]
# Enable/disable all OpenLineage extractors
enable_extractors = True  # Default: True

# Fine-grained control over DataHub's OpenLineage enhancements

# --- SQL Parsing Configuration ---

# Enable multi-statement SQL parsing (resolves temp tables, merges lineage)
enable_multi_statement_sql_parsing = False  # Default: False

# --- Patches (work with both Legacy OpenLineage and OpenLineage Provider) ---

# Patch SqlExtractor to use DataHub's advanced SQL parser (enables column-level lineage)
patch_sql_parser = True  # Default: True

# Patch SnowflakeExtractor to fix default schema detection
patch_snowflake_schema = True  # Default: True

# --- Custom Extractors (only apply to Legacy OpenLineage) ---

# Use DataHub's custom AthenaOperatorExtractor (better Athena lineage)
extract_athena_operator = True  # Default: True

# Use DataHub's custom BigQueryInsertJobOperatorExtractor (handles BQ job configuration)
extract_bigquery_insert_job_operator = True  # Default: True

Multi-Statement SQL Parsing:

When enable_multi_statement_sql_parsing=True, if a task executes multiple SQL statements (e.g., CREATE TEMP TABLE ...; INSERT ... FROM temp_table;), DataHub parses all statements together and resolves temporary table dependencies within that task. By default (False), only the first statement is parsed.

How it works:

Patches (apply to both Legacy OpenLineage and OpenLineage Provider):

• Apply monkey-patching to OpenLineage extractor/operator classes at runtime
• Work on both Airflow 2.x and Airflow 3.x
• When patch_sql_parser=True:
  • Airflow 2: Patches SqlExtractor.extract() method
  • Airflow 3: Patches SQLParser.generate_openlineage_metadata_from_sql() method
  • Provides: More accurate lineage extraction, column-level lineage (CLL), better SQL dialect support
• When patch_snowflake_schema=True:
  • Airflow 2: Patches SnowflakeExtractor.default_schema property
  • Airflow 3: Currently not needed (handled by Airflow's native support)
  • Fixes Snowflake schema detection issues

Custom Extractors/Operator Patches:

• Register DataHub's custom implementations for specific operators
• Work on both Airflow 2.x and Airflow 3.x
• extract_athena_operator:
  • Airflow 2 (Legacy OpenLineage only): Registers AthenaOperatorExtractor
  • Airflow 3: Patches AthenaOperator.get_openlineage_facets_on_complete()
  • Uses DataHub's SQL parser for better Athena lineage
• extract_bigquery_insert_job_operator:
  • Airflow 2 (Legacy OpenLineage only): Registers BigQueryInsertJobOperatorExtractor
  • Airflow 3: Patches BigQueryInsertJobOperator.get_openlineage_facets_on_complete()
  • Handles BigQuery job configuration and destination tables

Example use cases:

Disable DataHub's SQL parser to use OpenLineage's native parsing:

[datahub]
enable_extractors = True
patch_sql_parser = False  # Use OpenLineage's native SQL parser
patch_snowflake_schema = True  # Still fix Snowflake schema detection

Disable custom Athena extractor (only relevant for Legacy OpenLineage):

[datahub]
enable_extractors = True
extract_athena_operator = False  # Use OpenLineage's default Athena extractor

Other Configuration Options

For a complete list of configuration options, see the DataHub Airflow documentation.

Developing

See the developing docs.