unable to get a stable set of rows error - sql

Question

I am trying to perform a merge into a table (let's call it table1) from a table2. In the USING condition I need a third table (table3). This third table contains some IDs that I need in table1. A simplified version of my merge looks like:
MERGE INTO table1 a
USING (
SELECT ID, address
FROM table3 b
Where address IN
(
SELECT address
FROM table3
WHERE address IS NOT NULL
AND ID> 0
GROUP BY address
HAVING COUNT(*) = 1
)
) c
ON (a.address = c.address)
WHEN MATCHED THEN
UPDATE SET a.ID = c.ID
WHERE a.ID = 0
I know that the error I get is usually caused by the query in the USING clause, but theoretically this problem should be eliminated by the count(*)=1 condition.
I have duplicates in table2, but they should all get an ID from table3 or ID 0 if the address is duplicated in table3.
IDs are unique for an address, so they should be distinct.
P.S. This merge is performed automatically by a script that , so I can modify the query to add more conditions/restrictions, but I cannot change the structure [meaning I have to use these 3 tables as they are].
I hope this makes sense.
Any ideas why this still does not work for me?

Answer 1

Try this:
MERGE INTO table1 a
USING (
SELECT max(ID), address
FROM table3 b
WHERE address IS NOT NULL AND ID > 0
GROUP BY address
HAVING COUNT(*) = 1
) c
ON (a.address = c.address)
WHEN MATCHED THEN
UPDATE SET a.ID = c.ID
WHERE a.ID = 0;
you have where condition in inner query but not in outer query. If you want your original query please try:
MERGE INTO table1 a
USING (
SELECT ID, address
FROM table3 b
AND address IN
(
SELECT address
FROM table3
WHERE address IS NOT NULL
AND ID> 0
GROUP BY address
HAVING COUNT(*) = 1
)
WHERE address IS NOT NULL
AND ID> 0
) c
ON (a.address = c.address)
WHEN MATCHED THEN
UPDATE SET a.ID = c.ID
WHERE a.ID = 0

Answer 2

The issue is more than likely due to the duplicate rows from table2. Here's a simple test case demonstrating the issue:
Setup:
CREATE TABLE t1 (ID INTEGER PRIMARY KEY,
val VARCHAR2(1));
CREATE TABLE t2 (ID INTEGER,
val VARCHAR2(1));
INSERT INTO t1 (ID, val) VALUES (1, 'A');
INSERT INTO t2 (ID, val) VALUES (1, 'B');
INSERT INTO t2 (ID, val) VALUES (1, 'B');
COMMIT;
Merge that will error:
MERGE INTO t1 USING t2
ON (t1.id = t2.id)
WHEN MATCHED THEN
UPDATE SET t1.val = t2.val;
ORA-30926: unable to get a stable set of rows in the source tables
Merge that will succeed:
MERGE INTO t1 USING (SELECT DISTINCT id, val FROM t2) t2
ON (t1.id = t2.id)
WHEN MATCHED THEN
UPDATE SET t1.val = t2.val;
N.B. The second merge will still fail if you have different values returned for val for the same id; that means you will have more than one row returned for a given id, and Oracle won't know which one to use to update the target table with.
In order to make sure your merge statement will work, you will need to ensure that you will return at most 1 row per address in the source subquery.