What is wrong with my update statement with a join in Oracle?

Question

I am working with an Oracle 10g Database.

I have the following two tables:

T_DEBTOR :
    - ID_DEBTOR
    - HEADER
T_ELEMENT :
    -         


        

Answer 1

Good question.

To simulate your situation, I've created sample tables:

SQL> create table t_debtor(id_debtor,header)
  2  as
  3  select 1, 'Header 1' from dual union all
  4  select 2, null from dual union all
  5  select 3, 'Header 3' from dual
  6  /

Tabel is aangemaakt.

SQL> create table t_element (id_element,id_debtor,insurer)
  2  as
  3  select 1, 1, 'to be updated' from dual union all
  4  select 2, 1, 'to be updated' from dual union all
  5  select 3, 2, 'not to be updated' from dual union all
  6  select 4, 2, 'not to be updated' from dual union all
  7  select 5, 3, 'to be updated' from dual
  8  /

Tabel is aangemaakt.

And with your current update statement, the problem becomes clear: the "not to be updated" values are set to NULL:

SQL> update
  2      T_ELEMENT elt
  3      set elt.INSURER = (
  4          select HEADER
  5              from T_DEBTOR debtor
  6              where
  7                  debtor.HEADER is not null
  8                  and debtor.ID_DEBTOR = elt.ID_DEBTOR)
  9  /

5 rijen zijn bijgewerkt.

SQL> select * from t_element
  2  /

ID_ELEMENT  ID_DEBTOR INSURER
---------- ---------- -----------------
         1          1 Header 1
         2          1 Header 1
         3          2
         4          2
         5          3 Header 3

5 rijen zijn geselecteerd.

The best way to do this update, is to update a join of both tables. There are some restrictions however:

SQL> rollback
  2  /

Rollback is voltooid.

SQL> update ( select elt.insurer
  2                , dtr.header
  3             from t_element elt
  4                , t_debtor dtr
  5            where elt.id_debtor = dtr.id_debtor
  6              and dtr.header is not null
  7         )
  8     set insurer = header
  9  /
   set insurer = header
       *
FOUT in regel 8:
.ORA-01779: cannot modify a column which maps to a non key-preserved table

With the bypass ujvc hint, we can circumvent this restriction. But it is not advisable to do so unless you know really really sure that t_debtor.id_debtor is unique.

SQL> update /*+ bypass_ujvc */
  2         ( select elt.insurer
  3                , dtr.header
  4             from t_element elt
  5                , t_debtor dtr
  6            where elt.id_debtor = dtr.id_debtor
  7              and dtr.header is not null
  8         )
  9     set insurer = header
 10  /

3 rijen zijn bijgewerkt.

SQL> select * from t_element
  2  /

ID_ELEMENT  ID_DEBTOR INSURER
---------- ---------- -----------------
         1          1 Header 1
         2          1 Header 1
         3          2 not to be updated
         4          2 not to be updated
         5          3 Header 3

5 rijen zijn geselecteerd.

It's better to just add a primary key. You'll probably have this one already in place:

SQL> rollback
  2  /

Rollback is voltooid.

SQL> alter table t_debtor add primary key (id_debtor)
  2  /

Tabel is gewijzigd.

SQL> update ( select elt.insurer
  2                , dtr.header
  3             from t_element elt
  4                , t_debtor dtr
  5            where elt.id_debtor = dtr.id_debtor
  6              and dtr.header is not null
  7         )
  8     set insurer = header
  9  /

3 rijen zijn bijgewerkt.

SQL> select * from t_element
  2  /

ID_ELEMENT  ID_DEBTOR INSURER
---------- ---------- -----------------
         1          1 Header 1
         2          1 Header 1
         3          2 not to be updated
         4          2 not to be updated
         5          3 Header 3

5 rijen zijn geselecteerd.

Regards, Rob.

Answer 2

@Rob Thanks for the /*+ bypass_ujvc */ Tip. I have a couple cases where I need to use this. I wish my DBA told be able this. There are a couple times I had to create a cursor to get around this.

Answer 3

Have you tried

update
    T_ELEMENT elt
    set elt.INSURER = (
        select HEADER
            from T_DEBTOR debtor
            where
                debtor.HEADER is not null
                and debtor.ID_DEBITEUR = elt.ID_DEBITEUR)
where not elt.ID_DEBITEUR is null;

Answer 4

You could use the SQL Case statement, to distinguish when HEADER is null and when it has a value:
http://www.tizag.com/sqlTutorial/sqlcase.php

Answer 5

since Oracle 8i (I haven't tried with the preceeding versions), you can update a join if the tables are "key-preserved" (i-e: if you're updating the child from in a parent-child relationship). Here, if id_debtor is the primary key of T_DEBTOR, you can :

UPDATE (SELECT e.insurer, d.header
          FROM t_element e, t_debtor d
         WHERE e.id_debtor = d.id_debtor
           AND d.header IS NOT NULL)
   SET insurer = HEADER;

Cheers,

--
Vincent

Answer 6

You can do this by updating the results of a select, but the tables have to be 'key preserved':

SQL> create table t_debtor ( id_debtor integer, header varchar2(10));

Table created.

SQL> create table t_element (id_element integer, id_debtor integer, insurer varchar2(10));

Table created.

SQL> insert into t_debtor values (1, 'something');

1 row created.

SQL> insert into t_debtor values (2, 'else');

1 row created.

SQL> insert into t_debtor values (3, null);

1 row created.

SQL>
SQL> insert into t_element values (1, 1, 'foo');

1 row created.

SQL> insert into t_element values (2, 2, null);

1 row created.

SQL> insert into t_element values (3, 3, 'bar');

1 row created.

SQL> commit;

Commit complete.

That creates your tables (hint - it's very useful if you can post SQL for your example!).

Now you can update the results of a select to give what you want ...

SQL> update (select e.id_element, d.header header, e.insurer insurer
        from t_debtor d, t_element e
  2          where d.id_debtor = e.id_debtor  3
  4          and d.header is not null)
  5  set insurer = header;
set insurer = header
    *
ERROR at line 5:
ORA-01779: cannot modify a column which maps to a non key-preserved table

This fails because the table is not key preserved, but a few constraints will solve this:

alter table t_element add constraint t_element_pk primary key (id_element) using index;

alter table t_debtor add constraint t_debtor_pk primary key (id_debtor) using index;

alter table t_element add constraint t_element_debtor_fk foreign key (id_debtor) references t_debtor(id_debtor);

Now the update will work, because the tables are key preserved:

SQL> update (select e.id_element, d.header header, e.insurer insurer
        from t_debtor d, t_element e
        where d.id_debtor = e.id_debtor
        and d.header is not null)
set insurer = header  2    3    4    5  ;

2 rows updated.

SQL> select * from t_element;

ID_ELEMENT  ID_DEBTOR INSURER
---------- ---------- ----------
         1          1 something
         2          2 else
         3          3 bar