Hibernate's PostgreSQL dialect isn't very bright. It doesn't know about your per-SERIAL sequences, and is assuming there's a global database-wide sequence called "hibernate_sequence" that it can use.
(UPDATE: It appears that newer Hibernate versions may use the default per-table sequences when GenerationType.IDENTITY
is specified. Test your version and use this instead of the below if it works for you.)
You need to change your mappings to explicitly specify each sequence. It's annoying, repetitive, and pointless.
@Entity
@Table(name = "JUDGEMENTS")
public class Judgement implements Serializable, Cloneable {
private static final long serialVersionUID = -7049957706738879274L;
@Id
@GeneratedValue(strategy = GenerationType.SEQUENCE, generator="judgements_id_seq")
@SequenceGenerator(name="judgements_id_seq", sequenceName="judgements_id_seq", allocationSize=1)
@Column(name = "JUD_ID")
private Long _judId;
...
The allocationSize=1
is quite important. If you omit it, Hibernate will blindly assume that the sequence is defined with INCREMENT 50
so when it gets a value from a sequence it can use that value and the 49 values below it as unique generated keys. If your database sequences increment by 1 - the default - then this will result in unique violations as Hibernate tries to re-use existing keys.
Note that getting one key at a time will result in an additional round trip per insert. As far as I can tell Hibernate isn't capable of using INSERT ... RETURNING
to efficiently return generated keys, nor can it apparently use the JDBC generated keys interface. If you tell it to use a sequence, it'll call nextval
to get the value then insert
that explicitly, resulting in two round trips. To reduce the cost of that, you can set a greater increment on key sequences with lots of inserts , remembering to set it on the mapping and the underlying database sequence. That'll cause Hibernate to call nextval
less frequently and cache blocks of keys to hand out as it goes.
I'm sure you can see from the above that I don't agree with the Hibernate design choices made here, at least from the perspective of using it with PostgreSQL. They should be using getGeneratedKeys
or using INSERT ... RETURNING
with DEFAULT
for the key, letting the database take care of this without Hibernate having to trouble its self over the names of the sequences or explicit access to them.
BTW, if you're using Hibernate with Pg you'll possibly also want an oplock trigger for Pg to allow Hibernate's optimistic locking to interact safely with normal database locking. Without it or something like it your Hibernate updates will tend to clobber changes made via other regular SQL clients. Ask me how I know.
与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…