Spring Boot 3 + Hibernate 6中如何支持PostgreSQL的pgvector扩展？

问题原因 #

你遇到的PSQLException是因为自定义Converter仅做了字符串与向量的转换，但Hibernate仍将其以varchar类型发送给数据库，与PostgreSQL的vector原生类型不匹配，导致类型转换错误。

解决方案 #

1. 添加依赖 #

引入pgvector Java客户端和适配的PostgreSQL JDBC驱动：

<!-- Maven依赖 -->
<dependency>
    <groupId>com.github.pgvector</groupId>
    <artifactId>pgvector-java</artifactId>
    <version>0.10.0</version>
</dependency>
<dependency>
    <groupId>org.postgresql</groupId>
    <artifactId>postgresql</artifactId>
    <version>42.6.0</version>
</dependency>

2. 自定义Hibernate向量类型 #

创建适配pgvector的Hibernate基础类型，让框架正确识别数据库的vector类型：

import com.github.pgvector.Vector;
import com.github.pgvector.VectorTypeDescriptor;
import org.hibernate.type.AbstractSingleColumnStandardBasicType;
import org.hibernate.type.descriptor.sql.VarcharTypeDescriptor;

public class PgVectorType extends AbstractSingleColumnStandardBasicType<Vector> {
    public static final PgVectorType INSTANCE = new PgVectorType();

    public PgVectorType() {
        super(VarcharTypeDescriptor.INSTANCE, VectorTypeDescriptor.INSTANCE);
    }

    @Override
    public String getName() {
        return "vector";
    }
}

3. 配置JPA实体 #

在实体类中注册自定义类型，并映射向量字段：

import com.github.pgvector.Vector;
import jakarta.persistence.*;
import org.hibernate.annotations.Type;
import org.hibernate.annotations.TypeDef;

@Entity
@Table(name = "items")
@TypeDef(name = "vector", typeClass = PgVectorType.class) // 注册自定义类型
public class Item {
    @Id
    @GeneratedValue(strategy = GenerationType.IDENTITY)
    private Long id;

    @Column(columnDefinition = "vector(5)")
    @Type(type = "vector") // 使用自定义类型映射字段
    private Vector embedding;

    // Getter & Setter
    public Long getId() { return id; }
    public void setId(Long id) { this.id = id; }
    public Vector getEmbedding() { return embedding; }
    public void setEmbedding(Vector embedding) { this.embedding = embedding; }
}

4. 注册JDBC类型映射（可选但推荐） #

在Spring配置类中注册pgvector的JDBC类型映射，确保驱动能正确处理vector类型：

import com.github.pgvector.Vector;
import jakarta.annotation.PostConstruct;
import org.springframework.beans.factory.annotation.Value;
import org.springframework.stereotype.Component;

import java.sql.Connection;
import java.sql.DriverManager;
import java.sql.SQLException;

@Component
public class PgVectorConfig {
    @Value("${spring.datasource.url}")
    private String dbUrl;
    @Value("${spring.datasource.username}")
    private String dbUser;
    @Value("${spring.datasource.password}")
    private String dbPassword;

    @PostConstruct
    public void registerVectorType() throws SQLException {
        try (Connection conn = DriverManager.getConnection(dbUrl, dbUser, dbPassword)) {
            conn.getTypeMap().put("vector", Vector.class);
        }
    }
}

5. 业务代码示例 #

直接使用pgvector的Vector类创建向量并保存：

import com.github.pgvector.Vector;
import org.springframework.stereotype.Service;
import jakarta.transaction.Transactional;

@Service
public class ItemService {
    private final ItemRepository itemRepository;

    public ItemService(ItemRepository itemRepository) {
        this.itemRepository = itemRepository;
    }

    @Transactional
    public void saveWithEmbedding() {
        float[] embeddingValues = {1.2f, 3.4f, 5.6f, 7.8f, 9.0f};
        Vector embedding = new Vector(embeddingValues);
        
        Item item = new Item();
        item.setEmbedding(embedding);
        itemRepository.save(item);
    }
}

内容的提问来源于stack exchange，提问作者K. Meyer