向量数据库实战：Milvus入门与Java集成
<h2>前言</h2><p>向量数据库是 RAG 系统的核心组件，负责存储和检索高维向量数据。Milvus 是最流行的开源向量数据库，支持十亿级向量检索。本文带你快速入门 Milvus，并完成 Java 集成。</p>
<h2>一、Milvus 核心概念</h2>
<ul>
<li><strong>Collection</strong>：向量数据的集合，类似数据库中的表</li>
<li><strong>Partition</strong>：集合的分区，提升查询性能</li>
<li><strong>Vector Field</strong>：存储向量数据的字段</li>
<li><strong>Index</strong>：向量索引，决定检索性能</li>
<li><strong>Metric Type</strong>：相似度度量方式（IP欧式距离 / L2欧式距离）</li>
</ul>
<h2>二、快速部署 Milvus（Docker）</h2>
# 下载 docker-compose.yml
wget https://github.com/milvus-io/milvus/releases/download/v2.3.3/milvus-standalone-docker-compose.yml -O docker-compose.yml
# 启动 Milvus
docker-compose up -d
# 验证服务
docker ps | grep milvus
<h2>三、Java 客户端配置</h2>
<dependency>
<groupId>io.milvus</groupId>
milvus-sdk-java</artifactId>
<version>2.3.4</version>
</dependency>
import io.milvus.client.MilvusClient;
import io.milvus.param.ConnectParam;
import io.milvus.param.collection.FieldType;
import io.milvus.param.collection.CreateCollectionParam;
public class MilvusConfig {
private static final String HOST = "localhost";
private static final int PORT = 19530;
public static MilvusClient createClient() {
ConnectParam param = ConnectParam.newBuilder()
.withHost(HOST)
.withPort(PORT)
.build();
return new MilvusClient(param);
}
}
<h2>四、创建 Collection</h2>
public void createCollection(MilvusClient client) {
// 定义字段
FieldType idField = FieldType.newBuilder()
.withName("id")
.withDataType(DataType.Int64)
.withPrimaryKey(true)
.build();
FieldType vectorField = FieldType.newBuilder()
.withName("embedding")
.withDataType(DataType.FloatVector)
.withDimension(1536)// OpenAI text-embedding-3-small 默认维度
.build();
FieldType textField = FieldType.newBuilder()
.withName("text")
.withDataType(DataType.VarChar)
.withMaxLength(65535)
.build();
CreateCollectionParam param = CreateCollectionParam.newBuilder()
.withCollectionName("documents")
.withDescription("Document embeddings collection")
.addFieldType(idField)
.addFieldType(vectorField)
.addFieldType(textField)
.build();
client.createCollection(param);
System.out.println("Collection created: documents");
}
<h2>五、插入向量数据</h2>
public void insertVectors(MilvusClient client, List<Float[]> vectors, List<String> texts) {
List<List<Float>> vectorList = vectors.stream()
.map(v -> Arrays.asList(v))
.collect(Collectors.toList());
List<InsertParam.Field> fields = Arrays.asList(
new InsertParam.Field("id", Arrays.asList(1L, 2L, 3L)),
new InsertParam.Field("embedding", vectorList),
new InsertParam.Field("text", texts)
);
InsertParam param = InsertParam.newBuilder()
.withCollectionName("documents")
.withFields(fields)
.build();
MutationResult result = client.insert(param);
System.out.println("Inserted " + result.getSuccIndexCount() + " vectors");
// 加载Collection到内存
client.loadCollection("documents");
}
<h2>六、向量检索</h2>
public void search(MilvusClient client, float[] queryVector, int topK) {
SearchParam param = SearchParam.newBuilder()
.withCollectionName("documents")
.withVectorFieldName("embedding")
.withTopK(topK)
.withVectors(Arrays.asList(Arrays.asList(queryVector)))
.withMetricType(MetricType.IP)// 内积相似度
.build();
SearchResult result = client.search(param);
for (SearchResult.ScoredVector sv : result.getRowRecords()) {
System.out.println("ID: " + sv.getVector().get("id")
+ ", Score: " + sv.getScore()
+ ", Text: " + sv.getVector().get("text"));
}
}
<h2>七、性能优化建议</h2>
<ol>
<li><strong>选择合适的索引</strong>：HNSW 适合高精度场景，IVF_FLAT 适合大规模数据</li>
<li><strong>Partition 优化</strong>：按时间或类别分区，减少检索范围</li>
<li><strong>批量操作</strong>：使用批量插入和批量检索，减少网络开销</li>
<li><strong>内存管理</strong>：确保 Collection 加载到内存，避免磁盘 I/O</li>
</ol>
<h2>总结</h2>
<p>Milvus 是构建生产级 RAG 系统的最佳选择。本文涵盖了从部署到 Java 集成的完整流程，下期我们来聊：<strong>如何用 Milvus + LangChain4j 构建完整的知识库问答系统</strong>。</p>
<p>觉得有帮助请点赞收藏！有问题欢迎评论区交流 <br>来源：程序园用户自行投稿发布，如果侵权，请联系站长删除<br>免责声明：如果侵犯了您的权益，请联系站长，我们会及时删除侵权内容，谢谢合作！
页:
[1]