JPA saveAll 在 OneToMany 上的 INSERT INTO 之后强制 UPDATE 调用

Question

上下文

我们的PUT 端点本质上是通过首先在所有实体上调用DELETE，然后调用INSERT INTO 来执行UPSERT。

但是，在调试时，我看到 JPA 也发出了一些 UPDATE 调用，出于性能原因，我试图避免这种情况。我希望只有 2 个INSERT 调用：一个用于Parent，一个用于所有Child。

这似乎只发生在 @OneToMany 实体上。

如果有帮助，我们使用 EclipseLink 作为我们的 JPA 提供程序，并激活 batch-writing。该应用程序是使用 Spring Data 构建的。

型号

@Entity
@Table
@Data // org.lombok
@IdClass(ParentPrimaryKey.class)
public class Parent {

    @Id
    @Column(name = "product", updatable = false)
    private String product;

    @Id
    @Column(name = "source", updatable = false)
    private String source;

    @Column
    private String whatever;


    @OneToMany(cascade = {CascadeType.ALL}, orphanRemoval = true)
    @JoinColumns({
            @JoinColumn(name = "product", referencedColumnName = "product"),
            @JoinColumn(name = "source", referencedColumnName = "source")
    })
    private List<Child> details;
}

@Entity
@Table
@Data // org.lombok
@IdClass(ChildPrimaryKey.class)
public class Child {

    @Id
    @Column(name = "product", updatable = false)
    private String product;

    @Id
    @Column(name = "source", updatable = false)
    private String source;

    @Id
    @Column(name = "extra", updatable = false)
    private String extra;

    @Column
    private String foo;

    @Column
    private String bar;
}

关联的当前日志

INSERT INTO PARENT (product, source, whatever) VALUES (?, ?, ?)
    bind => [A, A, A]
    bind => [B, B, A]

INSERT INTO CHILD (product, source, extra, foo, bar) VALUES (?, ?, ?, ?, ?)
    bind => [A, A, 1, B, B]
    bind => [A, A, 2, C, C]
    bind => [A, A, 3, D, D]
    bind => [B, B, 1, B, B]
    bind => [B, B, 2, C, C]
    bind => [B, B, 3, D, D]

UPDATE CHILD SET product = ?, source = ? WHERE ((product = ?) AND (source = ?) AND (extra = ?))
    bind => [A, A, A, A, 1]
    bind => [A, A, A, A, 2]
    bind => [A, A, A, A, 3]
    bind => [B, B, B, B, 1]
    bind => [B, B, B, B, 2]
    bind => [B, B, B, B, 3]

尝试修复

基于this answer，我尝试添加nullable=false：

public class Child {

    @Id
    @Column(name = "product", updatable = false, nullable = false)
    private String product;

    @Id
    @Column(name = "source", updatable = false, nullable = false)
    private String source;

    @Id
    @Column(name = "extra", updatable = false, nullable = false)
    private String extra;

    // the rest is identical...
}

结果日志

可以看出，这确实删除了UPDATE 调用，但它增加了将向我们的数据库发出的总查询量，因此意味着性能实际上会更差（它将从总共3 个请求，到 2N）：

INSERT INTO PARENT (product, source, whatever) VALUES (?, ?, ?)
    bind => [A, A, A]
INSERT INTO CHILD (product, source, extra, foo, bar) VALUES (?, ?, ?, ?, ?)
    bind => [A, A, 1, B, B]
    bind => [A, A, 2, C, C]
    bind => [A, A, 3, D, D]


INSERT INTO PARENT (product, source, whatever) VALUES (?, ?, ?)
    bind => [B, B, A]
INSERT INTO CHILD (product, source, extra, foo, bar) VALUES (?, ?, ?, ?, ?)
    bind => [B, B, 1, B, B]
    bind => [B, B, 2, C, C]
    bind => [B, B, 3, D, D]

Answer 1

两种可能更适合您的选择。

第一个选项

保持其他一切不变：

@Entity
@Table
@Data // org.lombok
@IdClass(ParentPrimaryKey.class)
public class Parent {

    @Id
    @Column(name = "product", updatable = false)
    private String product;

    @Id
    @Column(name = "source", updatable = false)
    private String source;

    @Column
    private String whatever;


    @OneToMany(cascade = {CascadeType.ALL}, orphanRemoval = true)
    @JoinColumns({
            @JoinColumn(name = "product", referencedColumnName = "product", insertable = false, updatable = false),
            @JoinColumn(name = "source", referencedColumnName = "source", insertable = false, updatable = false)
    })
    private List<Child> details;
}

这将确保 Child 表中的外键仅由它们各自的 @Id 映射控制。

第二个选项

或者，您可以反过来，让 JPA 使用所谓的派生 ID 完全为您设置子实体实例中的值：

@Entity
@IdClass(ChildPrimaryKey.class)
public class Child {
    @Id
    @JoinColumns({
            @JoinColumn(name = "product", referencedColumnName = "product"),
            @JoinColumn(name = "source", referencedColumnName = "source")
    })
    Parent parent;

    @Id
    @Column(name = "extra")
    private String extra;

    // the rest is identical...
}

public class ChildPrimaryKey {
    ParentPrimaryKey parent;
    String extra;
}

这将确保根据父实例的值在子表中设置源和产品列。然后，Parent 的详细信息映射可以删除连接列定义，并声明它是由 Child 的父关系映射的：

@OneToMany(mappedBy = "parent",cascade = {CascadeType.ALL}, orphanRemoval = true)
private List<Child> details;