Update table with using NEWID() function - sql

Question

CREATE TABLE Products(Id INT, Name CHAR(100), DefaultImageId INT NULL);
INSERT INTO Products (Id, Name, DefaultImageId) VALUES(1, 'A', NULL);
INSERT INTO Products (Id, Name, DefaultImageId) VALUES(2, 'A', NULL);
INSERT INTO Products (Id, Name, DefaultImageId) VALUES(3, 'A', NULL);
INSERT INTO Products (Id, Name, DefaultImageId) VALUES(4, 'A', NULL);
INSERT INTO Products (Id, Name, DefaultImageId) VALUES(5, 'A', NULL);
INSERT INTO Products (Id, Name, DefaultImageId) VALUES(1, 'B', NULL);
INSERT INTO Products (Id, Name, DefaultImageId) VALUES(2, 'B', NULL);
INSERT INTO Products (Id, Name, DefaultImageId) VALUES(3, 'B', NULL);
In general, I would update a table randomly like the following scripts.
update a
set DefaultImageId=1
from Products as a
where name = 'A'
and id in (
select top 2 id
from Products as b
where a.name = b.name
order by newid()
)
However, I get some issue. It would update more/less then 2 rows. I try execute the following scripts many times for debug.
The results are not always only two records. If I remove the order by newid(), the number of output result will be fine. It seems like the problem in newid(). How can I solve this problem? Thanks
select *
from Products as a
where name = 'A'
and id in (
select top 2 id
from Products as b
where a.name = b.name
order by newid()
)

Answer 1

You can try this:
UPDATE U SET DefaultImageId = 1
FROM(
SELECT TOP 2 * FROM dbo.Products WHERE Name = 'A' ORDER BY NEWID()
) AS U
However if you filter out the DefaultImageId=1 in inner query, it would be even better too. You should note that in this case, the inner query might produce less than 2 records.
UPDATE U SET DefaultImageId = 1
FROM(
SELECT TOP 2 * FROM dbo.Products WHERE Name = 'A' AND DefaultImageId <> 1 ORDER BY NEWID()
) AS U

Answer 2

I think you're just looking for a simple JOIN as
UPDATE A
SET DefaultImageId = 1
FROM Products A
JOIN
(
SELECT TOP 2 Id, Name
FROM Products B
ORDER BY NEWID()
) B ON A.Id = B.Id AND A.Name = B.Name;
Demo

Answer 3

SQL SERVER BUG:
I think you just hit the following SQL server bug, that is still under review since 2018.
https://feedback.azure.com/forums/908035-sql-server/suggestions/33272404-cte-containing-top-and-newid-producing-incorrect
WORK AROUND:
The only thing that seems to work, based on my stesting, is if you eliminate the outside reference from the subquery, things works as expected. You can eliminate the outside reference by doing any of the following as suggested by others here.
Option 1: Keeping the query as-is, except hard-coding the filter criteria inside the subquery instead of using an outside reference:
SELECT ...
FROM t1
WHERE
t1.id IN (SELECT TOP 2 id subquery with NO outside reference WHERE name = 'A' ORDER BY ORDER BY newid() )
AND t1.name = 'A'
....
Option 2: Using a JOIN with proper ON matching criteria instead of using an outside reference inside the subquery:
SELECT ...
FROM t1
INNER JOIN (SELECT TOP 2 id subquery with NO outside reference ORDER BY newid() ) t2 ON t1.id = t2.id and t1.name = t2.name
WHERE
t1.name = 'A'
....
OR
DETAILS:
Based on my testing, when all of the following are aligned, the SQL Engine returns an unpredictable result:
An outside table is referenced inside the subquery (or through CTE) as a filtering criteria
NEWID() function is used in the ORDER BY with a SELECT TOP statement (It does NOT happen with other non-deterministic functions like RAND() for example. Something specific to NEWID())
The SELECT TOP statement uses a number less than the number of records exists in that table
This seems to throw off the SQL engine. The result changes at every run: The number of records returned varies at every run between zero records and max number of records that would have been returned if SELECT TOP was not used.
The query plan does not make sense at all. See the screenshot here. TOP N returns more than it is supposed to and id IN () predicate returns random number of records (zero in this example). How could that be possible? I have no clue. Also, note that the id IN () predicate is applied before the left table scan, which shouldn't happen. I think the unreliable result has something to do with that. We'll have to wait until SQL Server team fixes the bug to be able write the queries the way you wrote.

Answer 4

As you are using Correlated subquery, where outer query name is referred by subquery name, you are getting different result set.
I would suggest you to use the name parameter twice to avoid this problem.
select *
from Products as a
where name='A'
and id in
(select top 2 id
from Products as b
where b.name = 'A'
order by newid())
Similarly, for update also, you change the query
update a
set DefaultImageId=1
from Products as a
where name = 'A'
and id in (
select top 2 id
from Products as b
where b.name = 'A'
order by newid()
)

Answer 5

Something like this
This query selects 2 rows at random.
with top_2_cte(id, [name]) as (
select top 2 id, [name]
from Products
where name='A'
order by newid())
select a.*
from Products a
join
top_2_cte ttc on a.Id=ttc.id
and a.[Name]=ttc.[name];
Update statement
;with top_2_cte(id, [name]) as (
select top 2 id, [name]
from Products
where name='A'
order by newid())
update a
set DefaultImageId=1
from Products a
join
top_2_cte ttc on a.Id=ttc.id
and a.[Name]=ttc.[name];