DataType

This Enum specifies the data type of a collection field. Pass a DataType value when constructing a FieldSchema or calling FieldSchema::WithDataType().

enum class DataType {
    UNKNOWN = 0,
    BOOL = 1,
    INT8 = 2,
    INT16 = 3,
    INT32 = 4,
    INT64 = 5,
    FLOAT = 10,
    DOUBLE = 11,
    VARCHAR = 21,
    ARRAY = 22,
    JSON = 23,
    GEOMETRY = 24,
    TIMESTAMPTZ = 26,
    BINARY_VECTOR = 100,
    FLOAT_VECTOR = 101,
    FLOAT16_VECTOR = 102,
    BFLOAT16_VECTOR = 103,
    SPARSE_FLOAT_VECTOR = 104,
    INT8_VECTOR = 105,
    STRUCT = 201,
};

VALUES:

Scalar types:

• BOOL (1) - Boolean value (true / false).

• INT8 (2) - 8-bit signed integer (−128 to 127).

• INT16 (3) - 16-bit signed integer.

• INT32 (4) - 32-bit signed integer.

• INT64 (5) - 64-bit signed integer. The only scalar type supported as a primary key.

• FLOAT (10) - 32-bit single-precision floating-point number.

• DOUBLE (11) - 64-bit double-precision floating-point number.

• VARCHAR (21) - Variable-length UTF-8 string. Requires WithMaxLength() (max 65535 bytes).

• ARRAY (22) - Array of scalar elements of a single type. Requires WithElementType() and WithMaxCapacity().

• JSON (23) - Unstructured JSON document. Supports dynamic filtering on any nested key path.

• GEOMETRY (24) - Geometric/spatial data stored in Well-Known Binary (WKB) format.

• TIMESTAMPTZ (26) - Timestamp with timezone (RFC 3339 string).

Vector types:

• BINARY_VECTOR (100) - Bit-packed binary vector. Dimension must be a multiple of 8. Requires WithDimension(). Typically paired with MetricType::HAMMING or MetricType::JACCARD.

• FLOAT_VECTOR (101) - 32-bit float dense vector. Requires WithDimension(). The most common vector type.

• FLOAT16_VECTOR (102) - 16-bit half-precision (FP16) float vector. Requires WithDimension(). Uses half the memory of FLOAT_VECTOR with minimal recall loss.

• BFLOAT16_VECTOR (103) - Brain Float 16 (BF16) vector. Requires WithDimension(). Better numeric range than FP16; popular for ML model outputs.

• SPARSE_FLOAT_VECTOR (104) - Sparse float vector where most dimensions are zero. No fixed dimension. Used for keyword search with MetricType::BM25.

• INT8_VECTOR (105) - INT8 quantized dense vector. Requires WithDimension(). Smallest memory footprint among dense vector types.

Multi-vector type:

• STRUCT (201) - Multi-vector struct field containing several named sub-vectors. Used with StructFieldSchema.

Internal:

• UNKNOWN (0) - Uninitialized or unrecognized type. Do not use directly.

Example

#include "milvus/MilvusClientV2.h"
#include <milvus/MilvusClientV2.h>
#include <milvus/types/DataType.h>
using namespace milvus;

CollectionSchemaPtr schema = std::make_shared<CollectionSchema>();

// Scalar fields
schema->AddField(FieldSchema("id",   DataType::INT64,   "primary key").WithPrimaryKey(true));
schema->AddField(FieldSchema("name", DataType::VARCHAR, "user name").WithMaxLength(200));
schema->AddField(FieldSchema("age",  DataType::INT8,    "user age"));
schema->AddField(FieldSchema("tags", DataType::ARRAY,   "tag list")
                    .WithElementType(DataType::VARCHAR).WithMaxCapacity(10));
schema->AddField(FieldSchema("meta", DataType::JSON,    "extra metadata"));

// Vector field
schema->AddField(FieldSchema("vec", DataType::FLOAT_VECTOR, "embedding").WithDimension(128));

auto client = MilvusClientV2::Create();
client->Connect(ConnectParam("http://localhost:19530").WithToken("root:Milvus"));
client->CreateCollection(
    CreateCollectionRequest()
        .WithCollectionName("my_collection")
        .WithCollectionSchema(schema));