Error Logging Clause & Alternatives

Question

Users want data from an external source validated, and any nonconformities logged. I had hoped to do this via constraints and the error logging clause, and cooked up an example to try things out:
SQL> alter table emp add constraint emp_ck_sal check(sal between 1 and 5000) enable;
Table altered.
SQL> select constraint_name, constraint_type from user_constraints where table_name = 'EMP';
CONSTRAINT_NAME                C
EMP_CK_SAL                     C
PK_EMP                         P
FK_DEPTNO                      R
SQL> exec dbms_errlog.create_error_log('emp', 'bad_emp');
PL/SQL procedure successfully completed.
SQL> insert into emp (empno, ename, job, mgr, hiredate, sal, deptno)
  2  select 8001, 'AZIZ', 'CLERK', 7782, sysdate, 1000, 10 from dual
  3  union
  4  select 8002, 'NG', 'CLERK', 7782, sysdate, 0, 10 from dual
  5  union
  6  select 8003, 'ASFAW', 'CLERK', 8000, sysdate, 8000, 50 from dual
  7  log errors into bad_emp
  8  reject limit unlimited;
1 row created.
SQL> select * from emp;
  EMPNO ENAME      JOB              MGR HIREDATE      SAL       COMM  DEPTNO
   7369 SMITH      CLERK           7902 17-DEC-80     800                 20
   7499 ALLEN      SALESMAN        7698 20-FEB-81    1600        300      30
   7521 WARD       SALESMAN        7698 22-FEB-81    1250        500      30
   7566 JONES      MANAGER         7839 02-APR-81    2975                 20
   7654 MARTIN     SALESMAN        7698 28-SEP-81    1250       1400      30
   7698 BLAKE      MANAGER         7839 01-MAY-81    2850                 30
   7782 CLARK      MANAGER         7839 09-JUN-81    2450                 10
   7788 SCOTT      ANALYST         7566 19-APR-87    3000                 20
   7839 KING       PRESIDENT            17-NOV-81    5000                 10
   7844 TURNER     SALESMAN        7698 08-SEP-81    1500          0      30
   7876 ADAMS      CLERK           7788 23-MAY-87    1100                 20
   7900 JAMES      CLERK           7698 03-DEC-81     950                 30
   7902 FORD       ANALYST         7566 03-DEC-81    3000                 20
   7934 MILLER     CLERK           7782 23-JAN-82    1300                 10
   8000 MOORE      MANAGER         7839 22-MAY-06    3600                 40
   8001 AZIZ       CLERK           7782 31-AUG-06    1000                 10
16 rows selected.
SQL> select ora_err_number$, ora_err_mesg$, empno empnox, ename enamex, sal salx, deptno deptnox
from bad_emp;
ORA_ERR_NUMBER$ ORA_ERR_MESG$                EMPNOX ENAMEX     SALX   DEPTNOX
           2290 ORA-02290: check constraint  8002   NG         0      10
                (SCOTT.EMP_CK_SAL) violated
           2290 ORA-02290: check constraint  8003   ASFAW      8000   50
                (SCOTT.EMP_CK_SAL) violatedGreat. "Clean! Modern! Rapid!" as Mr Mori said. However, there's a problem. Employee 8003 ASFAW actually violates 2 constraints: the reported EMP_CK_SAL check constraint for having a salary out of range, and the FK_DEPTNO foreign key constraint because dept 50 doesn't exist - which isn't reported. The users want all the violations reported at once so that only one fix operation is required. I don't think I'm going to do that via the error logging clause? Anything I've missed?
So - any ideas on how to avoid the 3000 line package of row-by-row processing with 7 levels of nested IFs I keep finding predecessors and associates have employed to do this kind of thing?

Answer 1

Hello
I would say that unless you can convince your users that they are going to have to accept that for some rows, they are going to have to correct more than one error, in more than one run, you will end up re-inventing the wheel - infact you will be creating a caterpillar track, capable of navigating pretty much any terrain roughly as quick as an actual caterpillar.
Realistically though, what percentage of data is going to require users to re-import the same row twice? Is the quality of the source data very poor?
One potential solution would be to use views on the source data that identify rows violating any given constraint:
SQL> CREATE TABLE dt_test_dept (id number primary key)
  2  /
Table created.
SQL>
SQL> insert into dt_test_dept values(1)
  2  /
1 row created.
SQL>
SQL> CREATE TABLE dt_test_emp (id number primary key, deptid number, sal number)
  2  /
Table created.
SQL>
SQL> alter table dt_test_emp add constraint emp_dept_fk foreign key (deptid) references dt_test_dept
  2  /
Table altered.
SQL> alter table dt_test_emp add constraint emp_ck_sal check(sal between 1 and 5000) enable
  2  /
Table altered.
SQL>
SQL> CREATE TABLE dt_test_emp_src(id number, deptid number, sal number)
  2  /
Table created.
SQL> insert into dt_test_emp_src VALUES(2, 100,500)
  2  /
1 row created.
SQL> insert into dt_test_emp_src VALUES(3, 1,5001)
  2  /
1 row created.
SQL> insert into dt_test_emp_src VALUES(4, 100,5001)
  2  /
1 row created.
SQL> insert into dt_test_emp_src VALUES(5, 1,5000)
  2  /
1 row created.
SQL>
SQL> CREATE VIEW dt_test_emp_bad_fk
  2  AS
  3  SELECT
  4     'No matching department' err_text,
  5     id,
  6     deptid,
  7     sal
  8  FROM
  9     dt_test_emp_src
10  WHERE
11     NOT EXISTS(     SELECT
12                             NULL
13                     FROM
14                             dt_test_dept
15                     WHERE
16                             dt_test_dept.id = dt_test_emp_src.deptid
17             )
18  /
View created.
SQL> CREATE VIEW dt_test_emp_bad_sal
  2  AS
  3  SELECT
  4     'Salary invalid' err_text,
  5     id,
  6     deptid,
  7     sal
  8  FROM
  9     dt_test_emp_src
10  WHERE
11     sal NOT BETWEEN 1 AND 5000
12  /
View created.
SQL>
SQL> CREATE VIEW dt_test_emp_bad
  2  AS
  3  SELECT
  4     *
  5  FROM
  6     dt_test_emp_bad_fk
  7  UNION ALL
  8  SELECT
  9     *
10  FROM
11     dt_test_emp_bad_sal
12  /
View created.
SQL> select * from dt_test_emp_bad
  2  /
ERR_TEXT                       ID     DEPTID        SAL
No matching department          2        100        500
No matching department          4        100       5001
Salary invalid                  3          1       5001
Salary invalid                  4        100       5001
SQL> select * from dt_test_emp_bad order by id, err_text
  2  /
ERR_TEXT                       ID     DEPTID        SAL
No matching department          2        100        500
Salary invalid                  3          1       5001
No matching department          4        100       5001
Salary invalid                  4        100       5001That way you can pre-emptively idetify the rows that will fail and not even attempt to insert them. It does of course mean one view or select statement per constraint but that's the price I guess.
HTH
David
And it should mean much less code to maintain...
Message was edited by:
david_tyler