Hibernate sets map-key to null during 'update' - sql

Question

Imagine you have
#EqualsAndHashcode(of = "id")
class AnAggregate {
private int id;
private Map<String, AValueObject> childrenByValueA;
void replaceChild(AValueObject childToReplace) {
childrenByValueA.put(childToReplace.getValueA(), childToReplace);
}
}
AValueObject should be a VO, but because of the nature of relational DB, it has a dummy ID:
#EqualsAndHashcode(exclude = "aDummyID")
class AValueObject {
private int aDummyID;
private String valueA;
private String valueB;
}
now, take a look at the part of xml mapping for AnAggregate class:
<map name="childrenByValueA" fetch="select" batch-size="666">
<key column="aggregate_id"/>
<map-key type="string" column="value_a"/>
<one-to-many class="AValueObject"/>
</map>
Now, when I create tables as follows:
create table an_aggregate (
aggregate_id bigint,
primary key (metric_id)
);
crate table a_value_object (
vo_id bigint,
aggregate_id bigint,
value_a varchar(255),
value_b varchar(255),
primary_key (vo_id);
);
alter table a_value_object add constraint a_fk foreign key (aggregate_id) references an_aggregate;
everything seems to work.
But if I declare:
value_a varchar(255) **not null**
then I have an integrity violation during the update operation.
Let's assume we have anAggregate - (1)
and aValueObject (1, 1, value_a, value_b)
and we want to replace that row with (_, 1, modified_value_a, modified_value_b)
It turns out, that hibernate tries to do stuff in the following order:
insert into a_value_object values (2, 1, modified_value_a,
modified_value_b)
update a_value_object set aggregate_id=null, value_a=null, where aggregate_id=1 and vo_id=1
delete from a_value_object where vo_id=1
and fail during the second step, because it violates the 'not null' constraint on 'value_a'.
Question(s):
How to overcome this? Why is hibernate executing stuff in such a strange order? Why does it try to null-out fields it shouldn't?