Insert Master Detail

Question

Hi all,
I need to understand which is the fastest way for inserting rows in master-detail tables
Using a procedure. I often insert lots of details rows and sometimes master-details rows
Here there is my solution. Is there something faster?
Oracle Database 11g Enterprise Edition Release 11.1.0.7.0 - 64bit Production
With the Partitioning, OLAP, Data Mining and Real Application Testing options
CREATE TABLE MAIN_TRACING (
   MAINVAL VARCHAR2(10) NOT NULL,
   PAR1    VARCHAR2(10) NOT NULL);
CREATE TABLE TRACING_DET (
   MAINVAL   VARCHAR2(10) NOT NULL,
   SUBPAR1   VARCHAR2(10) NOT NULL);
CREATE UNIQUE INDEX MYPK ON MAIN_TRACING(MAINVAL);
ALTER TABLE MAIN_TRACING ADD ( CONSTRAINT MYPK PRIMARY KEY (MAINVAL) USING INDEX MYPK);
ALTER TABLE TRACING_DET ADD ( CONSTRAINT MYFK FOREIGN KEY (MAINVAL)  REFERENCES MAIN_TRACING (MAINVAL));
CREATE OR REPLACE PROCEDURE ADD_TRACE(V_MAINVAL VARCHAR2, V_PAR1 VARCHAR2, V_SUBPAR1 VARCHAR2)
AS
   PARENT_NOT_FOUND   EXCEPTION;
   PRAGMA EXCEPTION_INIT(PARENT_NOT_FOUND, -2291);
BEGIN
   INSERT INTO TRACING_DET(MAINVAL, SUBPAR1)
        VALUES (V_MAINVAL, V_SUBPAR1);
   COMMIT;
EXCEPTION
   WHEN PARENT_NOT_FOUND
   THEN
      INSERT INTO MAIN_TRACING(MAINVAL, PAR1)
           VALUES (V_MAINVAL, V_PAR1);
      INSERT INTO TRACING_DET(MAINVAL, SUBPAR1)
           VALUES (V_MAINVAL, V_SUBPAR1);
      COMMIT;
   WHEN OTHERS
   THEN
      DBMS_OUTPUT.PUT_LINE(' Error code    : ' || TO_CHAR(SQLCODE));
END ADD_TRACE;
EXEC ADD_TRACE('m1', 'a', 'suba1');
EXEC ADD_TRACE('m1', 'a', 'suba2');
EXEC ADD_TRACE('m1', 'a', 'suba3');
--exec ADD_TRACE('m1', 'b', 'suba3'); --IMPOSSIBLE!
EXEC ADD_TRACE('m2', 'b', 'subb1');
EXEC ADD_TRACE('m2', 'b', 'subb2');
SELECT * FROM MAIN_TRACING;
MAINVAL    PAR1
m1         a
m2         b
SELECT * FROM TRACING_DET;
MAINVAL    SUBPAR1
m1         suba1
m1         suba2
m1         suba3
m2         subb1
m2         subb2

Answer 1

The answer (as Tom Kyte is fond of stating) is "It depends" ;-)
Is this something that is called scattered over the application? Most of the time it will be called with an existing MAINVAL and just sometimes now and again a new MAINVAL will be inserted? Subsequent calls to ADD_TRACE will typically have a different MAINVAL than the previous call?
If that is the case, then your method is not a bad one. Many programmers would have first tested for existence of master and decided whether to insert it or not, which would then be wasted effort in 95% of the cases.
Your method of assuming the master exists and then handle the exception if it does not - that is quite neat. But the key is - it has to be the exception to the rule that the exception handler is invoked.
But if on the other hand you are calling ADD_TRACE pretty consecutively (sort of like your example) with first several calls with one MAINVAL, then several calls with a new MAINVAL, then several with a new MAINVAL, etc. - say a typical ETL process or loading data from perhaps a flatfile or something like that - then you would do better to do one insert of the parent and then bulk up the children in a bulk insert.
So it depends on your use case. There is not one generally best way to insert into master-detail - there will be several depending on whether you are bulk loading data to your master-detail or you have scattered calls.
But just one other thing concerning your code:
Usually it is not a good idea to issue those COMMIT statements inside your procedure. How do you know if the caller has finished with the transaction or not?
Either let the client/caller decide when the complete transaction is done.
Or you may decide that this is tracing/logging and should be logged no matter whether the calling transaction commits or rollbacks, and then you can make ADD_TRACE use an autonomous transaction.
(All of the above is just my personal opinion and you will likely find other people with different opinions ;-) )