GeezerVison

Description

This fix addresses an issue where entity value types were incorrectly mapped when using feature views with join key aliases.

Problem

Previously, when using feature views with join key aliases (like in the example below), the entity_type_map would use the original column name instead of the aliased join key:

user_entity = Entity(
    name="user",
    join_keys=["user_id"],
    value_type=ValueType.INT32,
    description="User entity for the matching platform",
)

match_service = FeatureService(
    name="match_service",
    features=[
        user_profile_fv.with_name("user"),
        user_profile_fv.with_name("target_user").with_join_key_map(
            {"user_id": "target_user_id"}
        ),
    ],
)

This caused target_user_id to be interpreted as ValueType.INT64 instead of the correct entity value type (ValueType.INT32 in this case).

Solution

The fix ensures that when a feature view has a join key map, the entity type mapping uses the correct aliased column name from the projection's join_key_map.

Changes

• Modified _get_entity_maps function in sdk/python/feast/utils.py
• Added logic to check for join key aliases in the feature view projection
• Map entity types to the correct aliased column names

Testing

This fix resolves the issue where aliased feature views would have incorrect entity value type mappings, ensuring that join keys maintain their correct data types even when aliased.

Type of Change

• Bug fix (non-breaking change which fixes an issue)

How Has This Been Tested?

This fix has been tested with the scenario described above where feature views are aliased with join key mappings. The entity value types are now correctly preserved for aliased join keys.