跳到主要内容
版本：v3.0.x

add_function_field()

This operation adds a function-backed field to an existing collection. It is a high-level convenience wrapper that commits the schema change — the new output field plus the function definition — together with the index meta bound to the new output field. The server creates the index meta atomically with the schema change, so backfill compaction is never blocked on a missing vector index.

index_params must contain exactly one entry with an explicit index_type for the output field. The server rejects vector output fields without bound index params.

Request syntax​

Python
client.add_function_field(
    collection_name: str,
    field_schema: FieldSchema,
    func: Function,
    index_params: IndexParams,
    timeout: Optional[float] = None
)

PARAMETERS:

  • collection_name (str) -

    [REQUIRED]

    The name of the collection to modify.

  • field_schema (FieldSchema) -

    [REQUIRED]

    The schema of the new output field produced by the function. For example, a SPARSE_FLOAT_VECTOR field for BM25 or a BINARY_VECTOR field for MinHash.

  • func (Function) -

    [REQUIRED]

    The function definition that generates the output field. For example, a FunctionType.BM25 or FunctionType.MINHASH function.

    The supported combinations are FunctionType.BM25 with DataType.SPARSE_FLOAT_VECTOR and FunctionType.MINHASH with DataType.BINARY_VECTOR.

  • index_params (IndexParams) -

    [REQUIRED]

    The index definition bound to the new output field. It must contain exactly one entry with an explicit index_type (for example, SPARSE_INVERTED_INDEX/BM25 or MINHASH_LSH/MHJACCARD).

  • timeout (float | None) -

    The timeout in seconds for the schema-change RPC. Setting this to None indicates that this operation timeouts when any response arrives or any error occurs.

RETURN TYPE:

NoneType

RETURNS:

None

EXCEPTIONS:

  • ParamError

    Raised when a required parameter is missing or invalid.

  • MilvusException

    This exception will be raised when the schema change fails.

Example​

Python
from pymilvus import MilvusClient, DataType, FieldSchema, CollectionSchema, Function, FunctionType

client = MilvusClient(
    uri="http://localhost:19530",
    token="root:Milvus"
)

# The target collection must exist first. The BM25 input field must have
# enable_analyzer set to True, and the schema must contain a vector field.
client.create_collection(
    collection_name="my_collection",
    schema=CollectionSchema(
        fields=[
            FieldSchema("id", DataType.INT64, is_primary=True),
            FieldSchema("text", DataType.VARCHAR, max_length=512, enable_analyzer=True),
            FieldSchema("dense", DataType.FLOAT_VECTOR, dim=3),
        ],
    ),
)

# Define the BM25 function and its output sparse vector field
bm25_function = Function(
    name="bm25",
    function_type=FunctionType.BM25,
    input_field_names=["text"],
    output_field_names=["sparse_vector"],
)

sparse_field = FieldSchema(
    name="sparse_vector",
    dtype=DataType.SPARSE_FLOAT_VECTOR,
)

# The index params bound to the output field
index_params = client.prepare_index_params()
index_params.add_index(
    field_name="sparse_vector",
    index_type="SPARSE_INVERTED_INDEX",
    metric_type="BM25",
)

client.add_function_field(
    collection_name="my_collection",
    field_schema=sparse_field,
    func=bm25_function,
    index_params=index_params,
)