Delete records with correlation query in Oracle

Question

I have a table with users and addresses.
I need to delete all records from the table where multiple different users are associated with the same address.
So, I've prepared the following query:

DELETE FROM MYTABLE A 
WHERE EXISTS (
    SELECT USER_ID 
    FROM MYTABLE B 
    WHERE A.ADDRESS = B.ADDRESS AND B.USER_ID > A.USER_ID
)

I found that sometimes this query hangs while most of the times it's working ok. I suspect that this hanging could be caused by locks.

Can someone pls confirm if this is the case? Generally, is this a valid way to achieve the goal in Oracle?

The oracle version is 12.1.

Thanks,

Answer 1

This is a bit long for a comment.

1) It is impossible to tell what is causing the slowness that you are seeing without you providing additional data.

2) However, when it comes to deleting duplicates, a solution that usually has good performance in Oracle is to use a correlated subquery with ANY . I would advocate for the following query:

DELETE FROM MYTABLE A
WHERE A.USER_ID > ANY (SELECT USER_ID FROM MYTABLE B WHERE A.ADDRESS = B.ADDRESS)

For this query (and for yours too), consider an index on (USER_ID, ADDRESS ) .

Answer 2

If you want to delete all rows where there are duplicates, I might suggest:

DELETE FROM MYTABLE A 
WHERE A.ADDRESS IN (SELECT A2.ADDRESS
                    FROM MYTABLE A2
                    GROUP BY A2.ADDRESS
                    HAVING COUNT(DISTINCT A2.USER_ID) > 1
                   );

The subquery should be evaluated once. An index on (ADDRESS, USER_ID) would be helpful for the query.

If you want to keep the minimum user id (which is not what you say but seems reasonable), then I would suggest:

DELETE FROM MYTABLE A 
WHERE A.USER_ID > (SELECT MIN(A2.USER_ID)
                    FROM MYTABLE A2
                    WHERE A2.ADDRESS = A.ADDRESS
                   );

For this, the same index is beneficial.