microsoft/semantic-kernel · error · VectorStoreInitializationException

Distance function {field.distance_function} is not supported

Error message

Distance function {field.distance_function} is not supported. Supported distance functions are: {list(DISTANCE_FUNCTION_MAP.keys())}

What it means

Raised by _create_vector_field (VectorStoreInitializationException) when a vector field's distance_function is not one MongoDB Atlas supports. The supported set is fixed in DISTANCE_FUNCTION_MAP: EUCLIDEAN_DISTANCE, COSINE_SIMILARITY, DOT_PROD, and DEFAULT (mapped to euclidean). Any other DistanceFunction value (or a value added in a newer SK release) is rejected while building the Atlas vector search index definition.

Source

Thrown at python/semantic_kernel/connectors/mongodb.py:102

    env_prefix: ClassVar[str] = "MONGODB_ATLAS_"

    connection_string: SecretStr
    database_name: str = DEFAULT_DB_NAME
    index_name: str = DEFAULT_SEARCH_INDEX_NAME


def _create_vector_field(field: VectorStoreField) -> dict:
    """Create a vector field.

    Args:
        field (VectorStoreRecordVectorField): The vector field.

    Returns:
        dict: The vector field.
    """
    if field.distance_function not in DISTANCE_FUNCTION_MAP:
        raise VectorStoreInitializationException(
            f"Distance function {field.distance_function} is not supported. "
            f"Supported distance functions are: {list(DISTANCE_FUNCTION_MAP.keys())}"
        )
    return {
        "type": "vector",
        "numDimensions": field.dimensions,
        "path": field.storage_name or field.name,
        "similarity": DISTANCE_FUNCTION_MAP[field.distance_function],
    }


def _create_index_definitions(
    record_definition: VectorStoreCollectionDefinition, index_name: str
) -> list[SearchIndexModel]:
    """Create the index definitions."""
    indexes = []
    if record_definition.vector_fields:
        vector_fields = [_create_vector_field(field) for field in record_definition.vector_fields]

View on GitHub (pinned to c028a0c7dc)

Solutions

Use one of the supported distance functions: COSINE_SIMILARITY, DOT_PROD, EUCLIDEAN_DISTANCE, or DEFAULT.
Remove an explicit distance_function so the default applies (DEFAULT -> euclidean).
Upgrade semantic-kernel so the connector map and your enum version align.

Example fix

// before
vector=VectorStoreRecordVectorField(distance_function=DistanceFunction.MANHATTAN)
// after
vector=VectorStoreRecordVectorField(distance_function=DistanceFunction.COSINE_SIMILARITY)

Defensive patterns

Strategy: validation

Validate before calling

from semantic_kernel.connectors.mongodb import DISTANCE_FUNCTION_MAP
from semantic_kernel.data.vector import DistanceFunction

if field.distance_function not in DISTANCE_FUNCTION_MAP:
    raise ValueError(f'use one of {list(DISTANCE_FUNCTION_MAP)}')
_create_vector_field(field)

Type guard

from semantic_kernel.connectors.mongodb import DISTANCE_FUNCTION_MAP

def is_supported_distance(fn) -> bool:
    return fn in DISTANCE_FUNCTION_MAP

Try / catch

from semantic_kernel.exceptions import VectorStoreInitializationException
try:
    _create_vector_field(field)
except VectorStoreInitializationException as e:
    field.distance_function = DistanceFunction.COSINE_SIMILARITY
    _create_vector_field(field)

Prevention

Restrict vector field annotations to COSINE_SIMILARITY, DOT_PROD, EUCLIDEAN_DISTANCE, or DEFAULT.
Keep semantic-kernel version aligned between model definitions and the mongodb connector.
Add a startup assertion that all vector fields use a supported distance function.

When it happens

Trigger: Annotating a VectorStoreRecordVectorField with distance_function=DistanceFunction.MANHATTAN (or HAMMING/COSINE_DISTANCE/etc.), or a None/unset distance function, then creating/registering the MongoDB collection which calls _create_index_definitions -> _create_vector_field.

Common situations: Copying a model from another connector that uses a distance function MongoDB doesn't map; version skew where the DistanceFunction enum grew but this connector map didn't; explicitly setting an unsupported metric.

Related errors

AI-assisted analysis of microsoft/semantic-kernel@c028a0c7dc (2026-08-13). Data as JSON: /api/errors/090e9d7c9e89046e. Report an issue: GitHub.