Tableau Ingestion Fails with Invalid SchemaField URN Error – DataHub Customer Support Portal

Area: Ingestion Issues

Sub-Area: Tableau Connector — Column-Level Lineage

Issue

When running a Tableau ingestion recipe that includes column-level lineage (CLL) extraction, the pipeline fails because the connector generates invalid schemaField URNs. This occurs specifically when Tableau workbooks or datasources use Custom SQL as a data source and the SQL references tables whose upstream URN cannot be fully resolved — for example, cross-database references, aliased tables, or tables from a source not yet ingested into DataHub. The connector's get_upstream_fields_from_custom_sql() method passes an unresolved or None table reference to make_schema_field_urn(), which enforces strict URN validation and raises an assertion error. The resulting malformed URN — containing an empty column name portion — is rejected by the DataHub backend, causing the entire ingestion run to fail.

Error Messages

Unable to emit metadata to DataHub GMS: Failed to validate record with class com.linkedin.dataset.UpstreamLineage: ERROR :: /fineGrainedLineages/15/upstreams/0 :: "Provided urn urn:li:schemaField:(urn:li:dataset:(urn:li:dataPlatform:<platform>,<dataset-path>,PROD),)" is invalid
invalid schemaField URN

You Might Be Asking

Why does my Tableau ingestion fail with an invalid URN error even though my recipe worked before?
What causes a schemaField URN to have a trailing comma and an empty field name?
How can I unblock Tableau ingestion while a permanent fix is applied?
Is this a bug in the Tableau connector or a configuration problem in my recipe?

Solution

The recommended workaround is to disable column-level lineage extraction for Custom SQL datasources in your Tableau recipe. This unblocks ingestion while preserving table-level lineage and all other metadata extraction.

Identify the failure point. Review your ingestion logs for lines containing invalid schemaField URN or Failed to validate record with class com.linkedin.dataset.UpstreamLineage. The malformed URN will have a trailing comma and an empty field name, for example:
```
urn:li:schemaField:(urn:li:dataset:(urn:li:dataPlatform:<platform>,<dataset-path>,PROD),)
```
The empty string after the final comma is the missing column name — this confirms the Custom SQL column-level lineage parsing path is the root cause.
Disable column-level lineage for Custom SQL in your recipe. Add or update the following flag in the source.config section of your Tableau ingestion recipe YAML:
```
source:
  type: tableau
  config:
    # ... your existing connection settings ...
    extract_column_level_lineage: false
```
Setting extract_column_level_lineage to false prevents the connector from attempting to resolve column references from Custom SQL datasources, eliminating the code path that produces the invalid URN.
Re-run the ingestion recipe. After saving the updated recipe, trigger the ingestion run again. Table-level lineage, dataset metadata, and all other Tableau entities will continue to be extracted normally.
(Optional) Narrow the scope with table pattern filters. If you need column-level lineage for some datasources but not others, you can use tableau_project_pattern or workbook_pattern to exclude only the workbooks or projects that contain the problematic Custom SQL datasources:
```
source:
  type: tableau
  config:
    # ... your existing connection settings ...
    workbook_pattern:
      deny:
        - ".*<problematic-workbook-name>.*"
```
Upgrade to a patched connector version. A fix has been merged in the DataHub open-source repository (PR #18301) that adds null and validity checks before calling make_schema_field_urn(), causing invalid column references to be skipped with a warning log instead of crashing the pipeline. Upgrade to a connector version that includes this fix to restore column-level lineage extraction safely. Confirm the target version with your DataHub administrator or support contact.

Additional Notes

Root cause detail: The failure occurs in tableau.py inside the get_upstream_fields_from_custom_sql() method. When the SQL parser cannot fully resolve a column_ref.table value (returning None, a raw table name, or a partial path instead of a valid URN), the result is passed directly to make_schema_field_urn() in mce_builder.py, which enforces a strict assertion. Defensive null checks present elsewhere in the same file were missing from this specific code path.
Affected scenarios: Custom SQL datasources that reference cross-database tables, aliased tables, or upstream tables not yet ingested into DataHub are most likely to trigger this bug.
Table-level lineage is unaffected: Disabling extract_column_level_lineage does not affect table-level upstream lineage, dataset schema extraction, ownership, tags, or any other Tableau metadata.
Version context: This issue has been observed on DataHub connector versions in the v1.3.x range. The permanent fix is included in the patch referenced above; consult the DataHub release notes to confirm when it becomes available in your deployment channel.
URN format reference: A valid schemaField URN must follow the pattern urn:li:schemaField:(<datasetUrn>,<fieldPath>). A URN ending in a bare comma — urn:li:schemaField:(...,) — indicates an empty fieldPath and will always be rejected by the DataHub GMS validation layer.

Issue

Error Messages

You Might Be Asking

Solution

Additional Notes

Related Documentation