How to update the parent field in SQL Server - sql-server

Question

My data looks like this
ID Text IsParent ParentID
-------------------------
1 A 1 NULL
2 B 0 NULL
3 C 0 NULL
4 D 0 NULL
5 E 1 NULL
6 F 0 NULL
7 G 1 NULL
8 H 0 NULL
I want to fill ParentID with the previous parentID.
Data is ordered so
ID : 2,3,4 have parentID : 1
ID : 6 has parentID : 5
ID : 8 has parentID : 7
How to do this with SQL?
I have tried with a cursor, but it is way too slow.
Here is my code:
DECLARE cur1 CURSOR FOR
SELECT ID Text IsParent ParentID
FROM x2
ORDER BY ID
OPEN cur1
FETCH NEXT FROM cur1 INTO #ID, #Text, #IsParent, #ParentID
WHILE ##FETCH_STATUS = 0
BEGIN
IF #IsParent = 1
BEGIN
SET #LastParentID = #ID
END
ELSE
BEGIN
UPDATE X2
SET ParentID = #LastParentID
WHERE ID = #ID
END
FETCH NEXT FROM cur1 INTO #ID, #Text, #IsParent, #ParentID
END;
CLOSE cur1;
DEALLOCATE cur1;

Answer 1

You can do this with APPLY. The premise is to find the parent record with the highest ID, where the ID is lower than the child record.
Example
DECLARE #x2 TABLE (ID INT NOT NULL, Text CHAR(1), IsParent BIT, ParentID INT);
INSERT #x2 (ID, Text, IsParent)
VALUES
(1, 'A', 1), (2, 'B', 0), (3, 'C', 0), (4, 'D', 0),
(5, 'E', 1), (6, 'F', 0), (7, 'G', 1), (8, 'H', 0);
UPDATE c
SET ParentID = p.ID
FROM #x2 AS c
CROSS APPLY
( SELECT TOP 1 ID
FROM #x2 AS p
WHERE p.IsParent = 1 -- Is a parent record
AND p.ID < c.ID -- ID is lower than child record
ORDER BY p.ID DESC -- Order descending to get the highest ID
) AS p
WHERE c.IsParent = 0
AND c.ParentID IS NULL;
SELECT *
FROM #x2;
OUTPUT
ID Text IsParent ParentID
---------------------------------
1 A 1 NULL
2 B 0 1
3 C 0 1
4 D 0 1
5 E 1 NULL
6 F 0 5
7 G 1 NULL
8 H 0 7

Answer 2

You can use CTE and window function to achieve that.
First we are creating continuous id(cid) using sum and secondly picking the minimum ID using the cid created in the first step and then finally updating the table where IsParent is 0.
try the following:
;WITH cte AS
(
SELECT *, sum(t.IsParent) OVER (ORDER BY id) cid
FROM #t t
),
cte2 AS
(
SELECT *, min(id) OVER (PARTITION BY cid ORDER BY id) pid
FROM cte c
)
UPDATE t
SET
t.ParentID = pid
FROM #t t
JOIN cte2 c ON c.id = t.ID
WHERE c.IsParent = 0
db<>fiddle demo.

Answer 3

The quickest way would be to insert child records with the parent id. Instead of populating the parent ids after the fact. From code you would first insert parent records and get back the newly generated parent ids. Then insert the child records with those newly generated parent ids.
Trying to maintain a query's speed like the ones suggested would just get gross over time as the data grows. Just because you can, doesn't mean you should.
Also as a side note, if you plan on having child of child records with an unknown depth. To avoid recursion I would recommend looking into having a hierarchyid data type column.