關於 Milvus
開始使用
概念
使用者指南
- 資料庫
- 收藏集
- 模式與資料欄位
- 插入與刪除
- 索引
- 搜尋
- 嵌入與重新排名
- 儲存優化
資料匯入
AI 工具
管理指南
工具
整合
教學
常見問題
API Reference

Home
Docs
使用者指南
搜尋
查詢

查詢

除了 ANN 搜尋，Milvus 也支援透過查詢來過濾元資料。本頁介紹如何使用 Query、Get 和 QueryIterators 來執行元資料篩選。

如果您在集合建立後動態新增欄位，包含這些欄位的查詢將會回傳定義的預設值，或對於沒有明確設定值的實體回傳 NULL。如需詳細資訊，請參閱新增欄位到現有的集合。

概觀

一個集合可以儲存各種類型的標量欄位。你可以讓 Milvus 基於一個或多個標量欄位篩選實體。Milvus 提供三種類型的查詢：Query、Get 和 QueryIterator。下表比較了這三種查詢類型。

	獲取	查詢	查詢迭代器
適用場景	要尋找持有指定主鍵的實體。	尋找符合自訂篩選條件的所有或指定數量的實體	在分頁查詢中尋找符合自訂篩選條件的所有實體。
篩選方法	透過主索引鍵	透過篩選表達式。	透過篩選表達式。
必須參數	集合名稱主鍵	集合名稱篩選表達式	集合名稱篩選表達式每次查詢要返回的實體數量
可選參數	分區名稱輸出欄位	分區名稱要返回的實體數量輸出欄位	分區名稱要返回的實體總數輸出欄位
返回值	回傳指定集合或分割區中持有指定主索引鍵的實體。	傳回指定集合或分割區中符合自訂篩選條件的所有或指定數量的實體。	透過分頁查詢傳回指定集合或分割區中符合自訂篩選條件的所有實體。

關於元資料篩選的更多資訊，請參閱 .NET Framework 2.0。

使用 Get

當您需要根據主鍵尋找實體時，您可以使用Get方法。以下的程式碼範例假設在您的集合中有三個欄位名為id,vector, 和color 。

[
        {"id": 0, "vector": [0.3580376395471989, -0.6023495712049978, 0.18414012509913835, -0.26286205330961354, 0.9029438446296592], "color": "pink_8682"},
        {"id": 1, "vector": [0.19886812562848388, 0.06023560599112088, 0.6976963061752597, 0.2614474506242501, 0.838729485096104], "color": "red_7025"},
        {"id": 2, "vector": [0.43742130801983836, -0.5597502546264526, 0.6457887650909682, 0.7894058910881185, 0.20785793220625592], "color": "orange_6781"},
        {"id": 3, "vector": [0.3172005263489739, 0.9719044792798428, -0.36981146090600725, -0.4860894583077995, 0.95791889146345], "color": "pink_9298"},
        {"id": 4, "vector": [0.4452349528804562, -0.8757026943054742, 0.8220779437047674, 0.46406290649483184, 0.30337481143159106], "color": "red_4794"},
        {"id": 5, "vector": [0.985825131989184, -0.8144651566660419, 0.6299267002202009, 0.1206906911183383, -0.1446277761879955], "color": "yellow_4222"},
        {"id": 6, "vector": [0.8371977790571115, -0.015764369584852833, -0.31062937026679327, -0.562666951622192, -0.8984947637863987], "color": "red_9392"},
        {"id": 7, "vector": [-0.33445148015177995, -0.2567135004164067, 0.8987539745369246, 0.9402995886420709, 0.5378064918413052], "color": "grey_8510"},
        {"id": 8, "vector": [0.39524717779832685, 0.4000257286739164, -0.5890507376891594, -0.8650502298996872, -0.6140360785406336], "color": "white_9381"},
        {"id": 9, "vector": [0.5718280481994695, 0.24070317428066512, -0.3737913482606834, -0.06726932177492717, -0.6980531615588608], "color": "purple_4976"},
]

您可以依據實體的 ID 取得實體，如下所示。

Python Java Go NodeJS cURL

from pymilvus import MilvusClient

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

res = client.get(
    collection_name="my_collection",
    ids=[0, 1, 2],
    output_fields=["vector", "color"]
)

print(res)

import io.milvus.v2.client.ConnectConfig;
import io.milvus.v2.client.MilvusClientV2;
import io.milvus.v2.service.vector.request.GetReq
import io.milvus.v2.service.vector.request.GetResp
import io.milvus.v2.service.vector.response.QueryResp;
import java.util.*;

MilvusClientV2 client = new MilvusClientV2(ConnectConfig.builder()
        .uri("http://localhost:19530")
        .token("root:Milvus")
        .build());
        
GetReq getReq = GetReq.builder()
        .collectionName("my_collection")
        .ids(Arrays.asList(0, 1, 2))
        .outputFields(Arrays.asList("vector", "color"))
        .build();

GetResp getResp = client.get(getReq);

List<QueryResp.QueryResult> results = getResp.getGetResults();
for (QueryResp.QueryResult result : results) {
    System.out.println(result.getEntity());
}

// Output
// {color=pink_8682, vector=[0.35803765, -0.6023496, 0.18414013, -0.26286206, 0.90294385], id=0}
// {color=red_7025, vector=[0.19886813, 0.060235605, 0.6976963, 0.26144746, 0.8387295], id=1}
// {color=orange_6781, vector=[0.43742132, -0.55975026, 0.6457888, 0.7894059, 0.20785794], id=2}

import (
    "context"
    "fmt"

    "github.com/milvus-io/milvus/client/v2/column"
    "github.com/milvus-io/milvus/client/v2/entity"
    "github.com/milvus-io/milvus/client/v2/milvusclient"
)

ctx, cancel := context.WithCancel(context.Background())
defer cancel()

milvusAddr := "localhost:19530"
client, err := milvusclient.New(ctx, &milvusclient.ClientConfig{
    Address: milvusAddr,
})
if err != nil {
    fmt.Println(err.Error())
    // handle error
}
defer client.Close(ctx)

resultSet, err := client.Get(ctx, milvusclient.NewQueryOption("my_collection").
    WithConsistencyLevel(entity.ClStrong).
    WithIDs(column.NewColumnInt64("id", []int64{0, 1, 2})).
    WithOutputFields("vector", "color"))
if err != nil {
    fmt.Println(err.Error())
    // handle error
}

fmt.Println("id: ", resultSet.GetColumn("id").FieldData().GetScalars())
fmt.Println("vector: ", resultSet.GetColumn("vector").FieldData().GetVectors())
fmt.Println("color: ", resultSet.GetColumn("color").FieldData().GetScalars())

import { MilvusClient, DataType } from "@zilliz/milvus2-sdk-node";

const address = "http://localhost:19530";
const token = "root:Milvus";
const client = new MilvusClient({address, token});

const res = client.get({
    collection_name="my_collection",
    ids=[0,1,2],
    output_fields=["vector", "color"]
})

export CLUSTER_ENDPOINT="http://localhost:19530"
export TOKEN="root:Milvus"

curl --request POST \
--url "${CLUSTER_ENDPOINT}/v2/vectordb/entities/get" \
--header "Authorization: Bearer ${TOKEN}" \
--header "Content-Type: application/json" \
--header "Request-Timeout: 10" \
-d '{
    "collectionName": "my_collection",
    "id": [0, 1, 2],
    "outputFields": ["vector", "color"]
}'

# {"code":0,"cost":0,"data":[{"color":"pink_8682","id":0,"vector":[0.35803765,-0.6023496,0.18414013,-0.26286206,0.90294385]},{"color":"red_7025","id":1,"vector":[0.19886813,0.060235605,0.6976963,0.26144746,0.8387295]},{"color":"orange_6781","id":2,"vector":[0.43742132,-0.55975026,0.6457888,0.7894059,0.20785794]}]}

使用查詢

當您需要透過自訂過濾條件來尋找實體時，請使用Query方法。下面的程式碼範例假設有三個欄位分別命名為id,vector, 和color ，並傳回以red 開始持有color 值的指定數量的實體。

Python Java Go NodeJS cURL

from pymilvus import MilvusClient

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

res = client.query(
    collection_name="my_collection",
    filter="color like \"red%\"",
    output_fields=["vector", "color"],
    limit=3
)

import io.milvus.v2.service.vector.request.QueryReq
import io.milvus.v2.service.vector.request.QueryResp

QueryReq queryReq = QueryReq.builder()
        .collectionName("my_collection")
        .filter("color like \"red%\"")
        .outputFields(Arrays.asList("vector", "color"))
        .limit(3)
        .build();

QueryResp queryResp = client.query(queryReq);

List<QueryResp.QueryResult> results = queryResp.getQueryResults();
for (QueryResp.QueryResult result : results) {
    System.out.println(result.getEntity());
}

// Output
// {color=red_7025, vector=[0.19886813, 0.060235605, 0.6976963, 0.26144746, 0.8387295], id=1}
// {color=red_4794, vector=[0.44523495, -0.8757027, 0.82207793, 0.4640629, 0.3033748], id=4}
// {color=red_9392, vector=[0.8371978, -0.015764369, -0.31062937, -0.56266695, -0.8984948], id=6}

resultSet, err := client.Query(ctx, milvusclient.NewQueryOption("my_collection").
    WithFilter("color like \"red%\"").
    WithOutputFields("vector", "color"))
if err != nil {
    fmt.Println(err.Error())
    // handle error
}

fmt.Println("id: ", resultSet.GetColumn("id").FieldData().GetScalars())
fmt.Println("vector: ", resultSet.GetColumn("vector").FieldData().GetVectors())
fmt.Println("color: ", resultSet.GetColumn("color").FieldData().GetScalars())

import { MilvusClient, DataType } from "@zilliz/milvus2-sdk-node";

const address = "http://localhost:19530";
const token = "root:Milvus";
const client = new MilvusClient({address, token});

const res = client.query({
    collection_name="my_collection",
    filter='color like "red%"',
    output_fields=["vector", "color"],
    limit(3)
})

export CLUSTER_ENDPOINT="http://localhost:19530"
export TOKEN="root:Milvus"

curl --request POST \
--url "${CLUSTER_ENDPOINT}/v2/vectordb/entities/query" \
--header "Authorization: Bearer ${TOKEN}" \
--header "Content-Type: application/json" \
--header "Request-Timeout: 10" \
-d '{
    "collectionName": "my_collection",
    "filter": "color like \"red%\"",
    "limit": 3,
    "outputFields": ["vector", "color"]
}'
#{"code":0,"cost":0,"data":[{"color":"red_7025","id":1,"vector":[0.19886813,0.060235605,0.6976963,0.26144746,0.8387295]},{"color":"red_4794","id":4,"vector":[0.44523495,-0.8757027,0.82207793,0.4640629,0.3033748]},{"color":"red_9392","id":6,"vector":[0.8371978,-0.015764369,-0.31062937,-0.56266695,-0.8984948]}]}

使用查詢迭代器

當您需要透過分頁查詢以自訂過濾條件尋找實體時，請建立一個QueryIterator並使用其next()方法遍歷所有實體，以找出符合過濾條件的實體。以下程式碼範例假設有三個欄位，分別命名為id,vector, 和color ，並返回所有持有color 值的實體，從red 開始。

Python Java Go NodeJS cURL

from pymilvus import connections, Collection

connections.connect(
    uri="http://localhost:19530",
    token="root:Milvus"
)

collection = Collection("my_collection")

iterator = collection.query_iterator(
    batch_size=10,
    expr="color like \"red%\"",
    output_fields=["color"]
)

results = []

while True:
    result = iterator.next()
    if not result:
        iterator.close()
        break

    print(result)
    results += result

import io.milvus.orm.iterator.QueryIterator;
import io.milvus.response.QueryResultsWrapper;
import io.milvus.v2.common.ConsistencyLevel;
import io.milvus.v2.service.vector.request.QueryIteratorReq;

QueryIteratorReq req = QueryIteratorReq.builder()
        .collectionName("my_collection")
        .expr("color like \"red%\"")
        .batchSize(50L)
        .outputFields(Collections.singletonList("color"))
        .consistencyLevel(ConsistencyLevel.BOUNDED)
        .build();
QueryIterator queryIterator = client.queryIterator(req);

while (true) {
    List<QueryResultsWrapper.RowRecord> res = queryIterator.next();
    if (res.isEmpty()) {
        queryIterator.close();
        break;
    }

    for (QueryResultsWrapper.RowRecord record : res) {
        System.out.println(record);
    }
}

// Output
// [color:red_7025, id:1]
// [color:red_4794, id:4]
// [color:red_9392, id:6]

// go

import { MilvusClient, DataType } from "@zilliz/milvus2-sdk-node";

const iterator = await milvusClient.queryIterator({
  collection_name: 'my_collection',
  batchSize: 10,
  expr: 'color like "red%"',
  output_fields: ['color'],
});

const results = [];
for await (const value of iterator) {
  results.push(...value);
  page += 1;
}

# Not available

分區中的查詢

您也可以透過在 Get、Query 或 QueryIterator 請求中包含分區名稱，在一個或多個分區中執行查詢。以下程式碼範例假設集合中有一個名為PartitionA的分割區。

Python Java Go NodeJS cURL

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

res = client.get(
    collection_name="my_collection",
    partitionNames=["partitionA"],
    ids=[10, 11, 12],
    output_fields=["vector", "color"]
)

from pymilvus import MilvusClient

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

res = client.query(
    collection_name="my_collection",
    partitionNames=["partitionA"],
    filter="color like \"red%\"",
    output_fields=["vector", "color"],
    limit=3
)

# Use QueryIterator
from pymilvus import connections, Collection

connections.connect(
    uri="http://localhost:19530",
    token="root:Milvus"
)

collection = Collection("my_collection")

iterator = collection.query_iterator(
    partition_names=["partitionA"],
    batch_size=10,
    expr="color like \"red%\"",
    output_fields=["color"]
)

results = []

while True:
    result = iterator.next()
    if not result:
        iterator.close()
        break

    print(result)
    results += result

GetReq getReq = GetReq.builder()
        .collectionName("my_collection")
        .partitionName("partitionA")
        .ids(Arrays.asList(10, 11, 12))
        .outputFields(Collections.singletonList("color"))
        .build();

GetResp getResp = client.get(getReq);

QueryReq queryReq = QueryReq.builder()
        .collectionName("my_collection")
        .partitionNames(Collections.singletonList("partitionA"))
        .filter("color like \"red%\"")
        .outputFields(Collections.singletonList("color"))
        .limit(3)
        .build();

QueryResp getResp = client.query(queryReq);

QueryIteratorReq req = QueryIteratorReq.builder()
        .collectionName("my_collection")
        .partitionNames(Collections.singletonList("partitionA"))
        .expr("color like \"red%\"")
        .batchSize(50L)
        .outputFields(Collections.singletonList("color"))
        .consistencyLevel(ConsistencyLevel.BOUNDED)
        .build();
QueryIterator queryIterator = client.queryIterator(req);

resultSet, err := client.Get(ctx, milvusclient.NewQueryOption("my_collection").
    WithPartitions("partitionA").
    WithIDs(column.NewColumnInt64("id", []int64{10, 11, 12})).
    WithOutputFields("vector", "color"))
if err != nil {
    fmt.Println(err.Error())
    // handle error
}

fmt.Println("id: ", resultSet.GetColumn("id").FieldData().GetScalars())
fmt.Println("vector: ", resultSet.GetColumn("vector").FieldData().GetVectors())
fmt.Println("color: ", resultSet.GetColumn("color").FieldData().GetScalars())

resultSet, err := client.Query(ctx, milvusclient.NewQueryOption("my_collection").
    WithPartitions("partitionA").
    WithFilter("color like \"red%\"").
    WithOutputFields("vector", "color"))
if err != nil {
    fmt.Println(err.Error())
    // handle error
}

fmt.Println("id: ", resultSet.GetColumn("id").FieldData().GetScalars())
fmt.Println("vector: ", resultSet.GetColumn("vector").FieldData().GetVectors())
fmt.Println("color: ", resultSet.GetColumn("color").FieldData().GetScalars())

import { MilvusClient, DataType } from "@zilliz/milvus2-sdk-node";

const address = "http://localhost:19530";
const token = "root:Milvus";
const client = new MilvusClient({address, token});

// Use get
var res = client.query({
    collection_name="my_collection",
    partition_names=["partitionA"],
    filter='color like "red%"',
    output_fields=["vector", "color"],
    limit(3)
})

// Use query
res = client.query({
    collection_name="my_collection",
    partition_names=["partitionA"],
    filter="color like \"red%\"",
    output_fields=["vector", "color"],
    limit(3)
})

// Use queryiterator
const iterator = await milvusClient.queryIterator({
  collection_name: 'my_collection',
  partition_names: ['partitionA'],
  batchSize: 10,
  expr: 'color like "red%"',
  output_fields: ['vector', 'color'],
});

const results = [];
for await (const value of iterator) {
  results.push(...value);
  page += 1;
}

export CLUSTER_ENDPOINT="http://localhost:19530"
export TOKEN="root:Milvus"

# Use get
curl --request POST \
--url "${CLUSTER_ENDPOINT}/v2/vectordb/entities/get" \
--header "Authorization: Bearer ${TOKEN}" \
--header "Content-Type: application/json" \
--header "Request-Timeout: 10" \
-d '{
    "collectionName": "my_collection",
    "partitionNames": ["partitionA"],
    "id": [0, 1, 2],
    "outputFields": ["vector", "color"]
}'

# Use query
curl --request POST \
--url "${CLUSTER_ENDPOINT}/v2/vectordb/entities/get" \
--header "Authorization: Bearer ${TOKEN}" \
--header "Content-Type: application/json" \
--header "Request-Timeout: 10" \
-d '{
    "collectionName": "my_collection",
    "partitionNames": ["partitionA"],
    "filter": "color like \"red%\"",
    "limit": 3,
    "outputFields": ["vector", "color"],
    "id": [0, 1, 2]
}'

使用查詢隨機抽樣

若要從資料集中抽取有代表性的資料子集，以進行資料探索或開發測試，請使用RANDOM_SAMPLE(sampling_factor) 表達式，其中sampling_factor 是介於 0 和 1 之間的浮動值，代表要抽樣的資料百分比。

有關詳細用法、進階範例和最佳實務，請參閱隨機抽樣。

Python Java Go NodeJS cURL

from pymilvus import MilvusClient

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

# Sample 1% of the entire collection
res = client.query(
    collection_name="my_collection",
    filter="RANDOM_SAMPLE(0.01)",
    output_fields=["vector", "color"]
)

print(f"Sampled {len(res)} entities from collection")

# Combine with other filters - first filter, then sample
res = client.query(
    collection_name="my_collection", 
    filter="color like \"red%\" AND RANDOM_SAMPLE(0.005)",
    output_fields=["vector", "color"],
    limit=10
)

print(f"Found {len(res)} red items in sample")

import io.milvus.v2.client.ConnectConfig;
import io.milvus.v2.client.MilvusClientV2;
import io.milvus.v2.service.vector.request.GetReq
import io.milvus.v2.service.vector.request.GetResp
import io.milvus.v2.service.vector.request.QueryReq
import io.milvus.v2.service.vector.request.QueryResp
import java.util.*;

MilvusClientV2 client = new MilvusClientV2(ConnectConfig.builder()
        .uri("http://localhost:19530")
        .token("root:Milvus")
        .build());

QueryReq queryReq = QueryReq.builder()
        .collectionName("my_collection")
        .filter("RANDOM_SAMPLE(0.01)")
        .outputFields(Arrays.asList("vector", "color"))
        .build();

QueryResp getResp = client.query(queryReq);
for (QueryResp.QueryResult result : getResp.getQueryResults()) {
    System.out.println(result.getEntity());
}

queryReq = QueryReq.builder()
        .collectionName("my_collection")
        .filter("color like \"red%\" AND RANDOM_SAMPLE(0.005)")
        .outputFields(Arrays.asList("vector", "color"))
        .limit(10)
        .build();

getResp = client.query(queryReq);
for (QueryResp.QueryResult result : getResp.getQueryResults()) {
    System.out.println(result.getEntity());
}

import (
    "context"
    "fmt"

    "github.com/milvus-io/milvus/client/v2/column"
    "github.com/milvus-io/milvus/client/v2/entity"
    "github.com/milvus-io/milvus/client/v2/milvusclient"
)

ctx, cancel := context.WithCancel(context.Background())
defer cancel()

milvusAddr := "localhost:19530"
client, err := milvusclient.New(ctx, &milvusclient.ClientConfig{
    Address: milvusAddr,
})
if err != nil {
    return err
}

resultSet, err := client.Query(ctx, milvusclient.NewQueryOption("my_collection").
    WithFilter("RANDOM_SAMPLE(0.01)").
    WithOutputFields("vector", "color"))
if err != nil {
    return err
}

resultSet, err = client.Query(ctx, milvusclient.NewQueryOption("my_collection").
    WithFilter("color like \"red%\" AND RANDOM_SAMPLE(0.005)").
    WithLimit(10).
    WithOutputFields("vector", "color"))
if err != nil {
    return err
}

// node

# restful

暫時設定查詢的時區

如果您的集合有TIMESTAMPTZ 欄位，您可以透過在查詢呼叫中設定timezone 參數，暫時覆寫資料庫或集合的單次操作預設時區。這可以控制TIMESTAMPTZ 值在操作過程中的顯示和比較方式。

timezone 的值必須是有效的IANA 時區識別碼（例如，亞洲/上海、美國/芝加哥或UTC）。有關如何使用TIMESTAMPTZ 欄位的詳細資訊，請參閱TIMESTAMPTZ 欄位。

以下範例說明如何為查詢作業暫時設定時區：

Python Java NodeJS Go cURL

# Query data and display the tsz field converted to "America/Havana"
results = client.query(
    collection_name,
    filter="id <= 10",
    output_fields=["id", "tsz", "vec"],
    limit=2,
    timezone="America/Havana",
)

// java

// js

// go

# restful

免費嘗試托管的 Milvus

Zilliz Cloud 無縫接入，由 Milvus 提供動力，速度提升 10 倍。

開始使用

反饋

這個頁面有幫助嗎？

查詢

概觀

使用 Get

使用查詢

使用查詢迭代器

分區中的查詢

使用查詢隨機抽樣

暫時設定查詢的時區

目錄

免費嘗試托管的 Milvus

反饋