milvus insert api的数据结构源码分析

原创

melodyshu

发布于 2024-02-18 17:07:49

1130

发布于 2024-02-18 17:07:49

文章被收录于专栏：milvus数据库milvus数据库

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, 3

print("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_id
    rng.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.MsgBase
	DbName               string
	CollectionName       string
	PartitionName        string
	FieldsData           []*schemapb.FieldData
	HashKeys             []uint32
	NumRows              uint32
	XXX_NoUnkeyedLiteral struct{}
	XXX_unrecognized     []byte
	XXX_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_Vectors
	Field                isFieldData_Field
	FieldId              int64
	IsDynamic            bool
	XXX_NoUnkeyedLiteral struct{}
	XXX_unrecognized     []byte
	XXX_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_JsonData
	Data                 isScalarField_Data
	XXX_NoUnkeyedLiteral struct{}
	XXX_unrecognized     []byte
	XXX_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                 []int64
	XXX_NoUnkeyedLiteral struct{}
	XXX_unrecognized     []byte
	XXX_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_Float16Vector
	Data                 isVectorField_Data
	XXX_NoUnkeyedLiteral struct{}
	XXX_unrecognized     []byte
	XXX_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                 []float32
	XXX_NoUnkeyedLiteral struct{}
	XXX_unrecognized     []byte
	XXX_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)

FloatVector是一个长度为30的float32数组，插入的是10个3维向量，1个向量是3个float32，在这里展开了。

原创声明：本文系作者授权腾讯云开发者社区发表，未经许可，不得转载。

如有侵权，请联系 cloudcommunity@tencent.com 删除。

向量数据库

原创声明：本文系作者授权腾讯云开发者社区发表，未经许可，不得转载。

如有侵权，请联系 cloudcommunity@tencent.com 删除。

向量数据库

#milvus

登录后参与评论

0 条评论

热度