Delete on Oracle DB when joining with a piplined table function - sql

Question

I have a table called CUSTOMERS from which I want to delete all entries that are not present in VALUE_CUSTOMERS. VALUE_CUSTOMERS is a Pipelined Table Function which builds on customers.
DELETE
(
SELECT *
FROM CUSTOMERS
LEFT JOIN
(
SELECT
1 AS DELETABLE,
VC.*
FROM
(
CUSTOMER_PACKAGE.VALUE_CUSTOMERS(TRUNC(SYSDATE) - 30)) VC
)
USING
(
FIRST_NAME, LAST_NAME, DATE_OF_BIRTH
)
WHERE
DELETABLE IS NULL
)
;
When I try to execute the statement, I get the Error:
ORA-01752: cannot delete from view without exactly one key-preserved
table

Answer 1

It looks like you're example has wrong syntax (lack of table keyword) - I've tested it on Oracle 12c, so maybe it works on newer ones. Below some ideas - based on Oracle 12c.
You've got multiple options here:
Save result of CUSTOMER_PACKAGE.VALUE_CUSTOMERS to some temporary table and use it in your query
CREATE GLOBAL TEMPORARY TABLE TMP_CUSTOMERS (
FIRST_NAME <type based on CUSTOMERS.FIRST_NAME>
, LAST_NAME <type based on CUSTOMERS.LAST_NAME>
, DATE_OF_BIRTH <type based on CUSTOMERS.DATE_OF_BIRTH>
)
ON COMMIT DELETE ROWS;
Then in code:
INSERT INTO TMP_CUSTOMERS(FIRST_NAME, LAST_NAME, DATE_OF_BIRTH)
SELECT VC.FIRST_NAME, VC.LAST_NAME, VC.DATE_OF_BIRTH
FROM TABLE(CUSTOMER_PACKAGE.VALUE_CUSTOMERS(TRUNC(SYSDATE) - 30)) VC
;
-- and then:
DELETE FROM CUSTOMERS C
WHERE NOT EXISTS(
SELECT 1
FROM TMP_CUSTOMERS TMP_C
-- Be AWARE that NULLs are not handled here
-- so it's correct only if FIRST_NAME, LAST_NAME, DATE_OF_BIRTH are not nullable
WHERE C.FIRST_NAME = TMP_C.FIRST_NAME
AND C.LAST_NAME = TMP_C.LAST_NAME
AND C.DATE_OF_BIRTH = TMP_C.DATE_OF_BIRTH
)
;
-- If `CUSTOMER_PACKAGE.VALUE_CUSTOMERS` can return a lot of rows,
-- then you should create indexes on FIRST_NAME, LAST_NAME, DATE_OF_BIRTH
-- or maybe even 1 multi-column index on all of above columns.
Also, consider rewriting your query. You should insert into TMP_CUSTOMERS just a customer ID and then call a delete based on these ids.
The main risk here is that the data could be changed between these 2 operations and you should consider that issue.
You can save result in collection variable and then do a bulk delete with forall loop.
If number of rows to delete could be big, then you should extend this example using LIMIT clause. Even with limit, you still could encounter some problems - like not enough space in UNDO. So this solution is good only for small amount of data. The risk here is the same as in example above.
DECALRE
type t_tab is table of number index by pls_integer;
v_tab t_tab;
BEGIN
SELECT CUSTOMER.ID -- I hope you have some kind of Primary key there...
BULK COLLECT INTO v_tab
FROM CUSTOMERS
LEFT JOIN
(
SELECT
1 AS DELETABLE,
VC.*
FROM TABLE(CUSTOMER_PACKAGE.VALUE_CUSTOMERS(TRUNC(SYSDATE) - 30)) VC
)
USING
(
FIRST_NAME, LAST_NAME, DATE_OF_BIRTH
)
WHERE
DELETABLE IS NULL
;
FORALL idx in 1..V_TAB.COUNT()
DELETE FROM CUSTOMERS
WHERE CUSTOMERS.ID = V_TAB(idx)
;
END;
/
Do it completely different - that's preferable.
For example: logic from CUSTOMER_PACKAGE.VALUE_CUSTOMERS move to a view and based on that view create a delete statement. Remember to change CUSTOMER_PACKAGE.VALUE_CUSTOMERS to use that new view also (DRY principle).