What causes FEATURE_STORE_TRAINING_SET_ERROR?

A feature table referenced in the training set definition was deleted or renamed. The join key column used to look up features does not exist in the observation data. The feature table has not been published or is empty. Unity Catalog permissions are missing on the feature table for the service principal running the job

How do I fix FEATURE_STORE_TRAINING_SET_ERROR?

Step 1: Verify the feature table exists in the Databricks Feature Store UI and that its name matches the one in the training set definition.. Step 2: Confirm the join key column is present and has the same data type in both the feature table and the observation data.. Step 3: Grant SELECT permission on the feature table to the service principal or user running the job.. Step 4: If the feature table is empty, check that the feature engineering pipeline ran successfully before the training job.. Step 5: Validate the training set definition with fs.get_table() before calling fs.create_training_set().

High severityconfiguration

Power BI Refresh Error:
FEATURE_STORE_TRAINING_SET_ERROR

What does this error mean?

Databricks Feature Store failed to create or read a training set, typically because a referenced feature table does not exist or the join key is missing.

Common causes

1A feature table referenced in the training set definition was deleted or renamed
2The join key column used to look up features does not exist in the observation data
3The feature table has not been published or is empty
4Unity Catalog permissions are missing on the feature table for the service principal running the job

How to fix it

1Step 1: Verify the feature table exists in the Databricks Feature Store UI and that its name matches the one in the training set definition.
2Step 2: Confirm the join key column is present and has the same data type in both the feature table and the observation data.
3Step 3: Grant SELECT permission on the feature table to the service principal or user running the job.
4Step 4: If the feature table is empty, check that the feature engineering pipeline ran successfully before the training job.
5Step 5: Validate the training set definition with fs.get_table() before calling fs.create_training_set().

Frequently asked questions

Do feature tables need to be published before they can be used in training sets?

Yes — feature tables must be written to the Feature Store using fs.write_table() before they can be joined in a training set.

Can I use Feature Store with Unity Catalog?

Yes — Unity Catalog is the recommended backend for Databricks Feature Store, using three-part names (catalog.schema.table). Ensure the required Unity Catalog privileges are granted.