Way to randomly update a set of values using another table as a collection of random values without using a loop? - sql

Question

I have a table with a column that contains some values. I want to replace all the existing values with random values from another table (but only the existing ones - so WHERE COL1 IS NOT NULL). There is no way to correlate the two tables. Each of the random values needs to be different (well... unless they are randomly the same, in which case it's fine).
For example:
CREATE TABLE T1 (ID NUMBER(11,0), COL1 VARCHAR2(20));
CREATE TABLE T2 (COL2 VARCHAR2(20));
INSERT INTO T1 VALUES (1, NULL);
INSERT INTO T1 VALUES (54, NULL);
INSERT INTO T1 VALUES (941, 'Some text');
INSERT INTO T1 VALUES (251, NULL);
INSERT INTO T1 VALUES (352, 'Some other text');
INSERT INTO T1 VALUES (354, NULL);
INSERT INTO T2 VALUES ('Val1');
INSERT INTO T2 VALUES ('Val2');
INSERT INTO T2 VALUES ('Val3');
INSERT INTO T2 VALUES ('Val4');
INSERT INTO T2 VALUES ('Val5');
INSERT INTO T2 VALUES ('Val6');
INSERT INTO T2 VALUES ('Val7');
I have tried a few things and have searched this site for answers. The answers I've found seem to require that there is some correlation between the two tables. A lot of the examples are for SQL Server. I've tried a few out anyway, but I can't seem to get a MERGE or CROSS APPLY approach to work (I appreciate this is most likely my failure...).
The only solution I have that actually works at the moment is the following:
BEGIN
FOR X IN (
SELECT ID FROM T1 WHERE COL1 IS NOT NULL
)
LOOP
UPDATE T1 SET COL1 = (
SELECT COL2 FROM (
SELECT COL2 FROM T2 ORDER BY SYS_GUID()
) WHERE ROWNUM = 1
)
WHERE T1.ID = X.ID;
END LOOP;
END;
/
This produces a (desired) result of:
SELECT * FROM T1;
ID COL1
---------------------------
1
54
941 Val3
251
352 Val7
354
(COL1 values are both random each time I run the loop).
... But I know there must be a set-based way to achieve this... right?

Answer 1

You could do it with a MERGE statement.
merge into t1
using ( select a1.id
, a2.col2
from ( select id
, row_number() over (order by dbms_random.value) rn
from t1
where col1 is not null ) a1
join ( select col2
, row_number() over (order by dbms_random.value) rn
from t2) a2
on a1.rn = a2.rn
) q
on ( q.id = t1.id)
when matched then
update set t1.col1 = q.col2
/
The USING query is a little unorthodox. There are two subqueries, one for each table, which generate analytic row_number() in a random order (this is tidier than using rownum. The two subqueries are joined on the random row numbers which gives a random combination of T1.ID and T2.COL_2. After that, it's a straightforward MERGE.

Answer 2

You can create a FUNCTION named GET_RANDOM_VALUE
create or replace FUNCTION GET_RANDOM_VALUE
RETURN VARCHAR2 AS
sValue VARCHAR2(100) := '?';
BEGIN
select col2 INTO sValue
from t2
order by dbms_random.value
fetch first 1 rows only;
RETURN sValue;
END GET_RANDOM_VALUE;
and the UPDATE command can be
UPDATE t1
SET col1 = GET_RANDOM_VALUE()
WHERE col1 IS NOT NULL;
Normally, I avoid using SELECT in function, but in this case, I found this solution more readable and more easy to understand.