milvus insert api的数据结构源码分析-编程知识

milvus insert api的数据结构源码分析

insert api的数据结构

一个完整的insert例子:

import numpy as np
from pymilvus import (connections,FieldSchema, CollectionSchema, DataType,Collection,
)num_entities, dim = 10, 3print("start connecting to Milvus")
connections.connect("default", host="192.168.230.71", port="19530")fields = [FieldSchema(name="pk", dtype=DataType.INT64, is_primary=True, auto_id=True),FieldSchema(name="book_id", dtype=DataType.INT64),FieldSchema(name="embeddings", dtype=DataType.FLOAT_VECTOR, dim=dim)
]schema = CollectionSchema(fields, "hello_milvus is the simplest demo to introduce the APIs")print("Create collection `hello_milvus`")
hello_milvus = Collection("hello_milvus", schema, consistency_level="Eventually",shards_num=1)print("Start inserting entities")
rng = np.random.default_rng(seed=19530)
entities = [[i for i in range(num_entities)],  # field book_idrng.random((num_entities, dim)),    # field embeddings
]insert_result = hello_milvus.insert(entities)hello_milvus.flush()

InsertRequest数据结构:

type InsertRequest struct {Base                 *commonpb.MsgBaseDbName               stringCollectionName       stringPartitionName        stringFieldsData           []*schemapb.FieldDataHashKeys             []uint32NumRows              uint32XXX_NoUnkeyedLiteral struct{}XXX_unrecognized     []byteXXX_sizecache        int32
}

FieldsData是一个数组，如果insert有3列，则数组长度为3，按照插入顺序。

FieldData数据结构:

type FieldData struct {Type      DataType FieldName string   // Types that are valid to be assigned to Field:////	*FieldData_Scalars//	*FieldData_VectorsField                isFieldData_FieldFieldId              int64IsDynamic            boolXXX_NoUnkeyedLiteral struct{}XXX_unrecognized     []byteXXX_sizecache        int32
}

isFieldData_Field是一个接口:

type isFieldData_Field interface {isFieldData_Field()
}

它有2个实现:FieldData_Scalars和FieldData_Vectors。

type FieldData_Scalars struct {Scalars *ScalarField
}type FieldData_Vectors struct {Vectors *VectorField
}

FieldData_Scalars存储标量数据，FieldData_Vectors存储向量数据。

ScalarField数据结构:

type ScalarField struct {// Types that are valid to be assigned to Data:////	*ScalarField_BoolData//	*ScalarField_IntData//	*ScalarField_LongData//	*ScalarField_FloatData//	*ScalarField_DoubleData//	*ScalarField_StringData//	*ScalarField_BytesData//	*ScalarField_ArrayData//	*ScalarField_JsonDataData                 isScalarField_DataXXX_NoUnkeyedLiteral struct{}XXX_unrecognized     []byteXXX_sizecache        int32
}

isScalarField_Data是一个接口。

type isScalarField_Data interface {isScalarField_Data()
}

isScalarField_Data的实现有9个：

ScalarField_BoolData
ScalarField_IntData
ScalarField_LongData
ScalarField_FloatData
ScalarField_DoubleData
ScalarField_StringData
ScalarField_BytesData
ScalarField_ArrayData
ScalarField_JsonData

以ScalarField_LongData为例:

type ScalarField_LongData struct {LongData *LongArray
}type LongArray struct {Data                 []int64XXX_NoUnkeyedLiteral struct{}XXX_unrecognized     []byteXXX_sizecache        int32
}

VectorField数据结构:

type VectorField struct {Dim int64// Types that are valid to be assigned to Data:////	*VectorField_FloatVector//	*VectorField_BinaryVector//	*VectorField_Float16VectorData                 isVectorField_DataXXX_NoUnkeyedLiteral struct{}XXX_unrecognized     []byteXXX_sizecache        int32
}

isVectorField_Data是一个接口。

type isVectorField_Data interface {isVectorField_Data()
}

isVectorField_Data有3种实现:

VectorField_FloatVector
VectorField_BinaryVector
VectorField_Float16Vector

以VectorField_FloatVector为例:

type VectorField_FloatVector struct {FloatVector *FloatArray
}type FloatArray struct {Data                 []float32XXX_NoUnkeyedLiteral struct{}XXX_unrecognized     []byteXXX_sizecache        int32
}

案例

向hello_milvus插入10个3维向量。

num_entities, dim = 10, 3
rng = np.random.default_rng(seed=19530)
entities = [[i for i in range(num_entities)],rng.random((num_entities, dim)), 
]
insert_result = hello_milvus.insert(entities)