Can a CTE be used to update in passes? - sql-server-2008

Question

I inherited a table with the following structure:
rowID rn userID Data1 Data2 Data3
----- -- ------ ----- ---- ----
1 1 1 A null 123
2 2 1 B 111 null
3 1 2 C 222 333
4 2 2 D null null
5 3 2 E 111 null
6 1 3 F 333 222
The first recs (rn=1) need to be inserted, while the rest (rn <>1) need to update the insertions (sequentially). I can insert easily, using where rn = 1 and checking for absence of the userID.
My problem is that I need to now update all recs sequentially using rn <>1 so that the user table reflects the latest state. That is, the user table after UPDATEs should look like this:
rowID userID Data1 Data2 Data3
----- ------ ----- ----- -----
1 1 B 111 123
2 2 E 111 333
3 3 F 333 222
My thought was to write a CTE, where each "pass" would grab all the recs where rn=2, then rn=3, then rn=4.... until I have no more rn to process. This way, I can update in sets.
Is this possible (or should I use do-while)? if so, do I need a recursive one or a "regular" CTE?
Here is what I tried:
;with my_cte (rowID, rn, userID, Data1, Data2, Data3, val) As
(
SELECT rowID, rn, userID, Data1, Data2, Data3, val
from #MyTempTable x
where rn =1
UNION ALL
SELECT rowID, rn, userID, Data1, Data2, Data3, b.val +1
from #MyTempTable y
INNER JOIN
my_cte b
ON y.userID = b.userID
WHERE y.rn = b.val +1
)
UPDATE userTable
SET
[Data1] = COALESCE(c.Data1, [Data1])
,[Data2]= COALESCE(c.Data2, [Data2])
,[Data3]= COALESCE(c.Data3, [Data3])
From #MyTempTable c
JOIN
( SELECT user_id
FROM my_cte
WHERE rn<>1
) b
ON b.user_id = c.user_id
WHERE
EXISTS
( Select userID
from userTable q
Where q.userId = b.userId
)
I could not get this to work, and it looks like only the first row is updating. Any thoughts? I'm a noob with CTEs. More than anything I'd like to know what exactly the CTE is doing... is it even possible for the update to run in "passes"?

Answer 1

Following CTE returns your given output for your given inputs. You can use these results as a starting point for inserting the records into another table.
The gist of it is
Use a recursive CTE, starting with all rows where rn=1.
In the recursive part, pick Data1-3 from the recursive part if available, otherwise retain the exisiting value (COALESCE). The result of the CTE now is your final values + the initial values where rn=1
Add a ROW_NUMBER for each userID but ORDER DESC on the existing rn. This makes sure that the latest values get rownumber 1.
Finally select all with rownumber 1 and add another rownumber as per your final results.
SQL Statement
;WITH q AS (
SELECT rn
, UserID
, Data1
, Data2
, Data3
FROM Inherited
WHERE rn = 1
UNION ALL
SELECT i.rn
, i.UserID
, COALESCE(i.Data1, q.Data1)
, COALESCE(i.Data2, q.Data2)
, COALESCE(i.Data3, q.Data3)
FROM q
INNER JOIN Inherited i ON i.rn = q.rn+1 AND i.userID = q.userID
)
SELECT rn = ROW_NUMBER() OVER (ORDER BY userID)
, *
FROM (
SELECT UserID
, Data1
, Data2
, Data3
, rn = ROW_NUMBER() OVER (PARTITION BY UserID ORDER BY rn DESC)
FROM q
) t
WHERE rn = 1
Test Script
;WITH Inherited (rowID, rn, userID, Data1, Data2, Data3) AS (
SELECT * FROM (VALUES
(1, 1, 1, 'A', null, '123')
, (2, 2, 1, 'B', '111', null)
, (3, 1, 2, 'C', '222', '333')
, (4, 2, 2, 'D', null, null)
, (5, 3, 2, 'E', '111', null)
, (6, 1, 3, 'F', '333', '222')
) a (b, c, d, e, f, g)
)
, q AS (
SELECT rn
, UserID
, Data1
, Data2
, Data3
FROM Inherited
WHERE rn = 1
UNION ALL
SELECT i.rn
, i.UserID
, COALESCE(i.Data1, q.Data1)
, COALESCE(i.Data2, q.Data2)
, COALESCE(i.Data3, q.Data3)
FROM q
INNER JOIN Inherited i ON i.rn = q.rn+1 AND i.userID = q.userID
)
SELECT rn = ROW_NUMBER() OVER (ORDER BY userID)
, *
FROM (
SELECT UserID
, Data1
, Data2
, Data3
, rn = ROW_NUMBER() OVER (PARTITION BY UserID ORDER BY rn DESC)
FROM q
) t
WHERE rn = 1