How to speed up update query in SQL Server 2008? - sql

Question

update orders
set tname = (select top 1 t.task
from task t
where prod_typ='2' and sorder_nbr = t.ORDER_NBR
order by t.strt_dt desc)
where Prod_type='2'
update orders
set tname= (select top 1 t.task
from task t
where prod_typ='1' and sorder_nbr=t.ORDER_NBR
order by t.strt_dt desc)
where Prod_type='1'
I am trying to update the tname column of orders table by the latest task from the task table
And the condition is prod_typ of orders table is 1 and sorder_nbr of orders table and order_nbr of task table are equal
My first update statement works well where the rows are 900k and for the second update rows are 400k for second update statement it takes more than one hour to run and at last I cancelled the query

Answer 1

1) You query and my query:
update orders
set tname = (select top 1 t.task
from task t
where prod_type='2' and order_nbr = t.ORDER_NBR
order by t.strt_dt desc)
where Prod_type='2';
go
update o
set tname = (select top 1 t.task
from task t
where prod_type='2' and o.order_nbr = t.ORDER_NBR
order by t.strt_dt desc)
from dbo.orders o
where Prod_type='2';
go
The actual execution plans:
As you can see, if default collation for current DB is CI (case insensitive) then following predicate order_nbr=t.ORDER_NBR force SQL Server to compare the values of t.ORDER_NBR with the values order_nbr column from the same table task t. Look at first execution plan which corresponds to first query.
To solve just this problem, I've used another alias
dbo.orders o and I've reqrite the predicate thus o.order_nbr = t.ORDER_NBR. You may see this also within second execution plan.
Depending on how many tasks are for every order_num & prod_type you could test S#1 if there are many tasks or S#2 if there is a small amount of tasks per order_num & prod_type. Again, you need to test with your data to see which solution is better.
2) Solution #1:
UPDATE o
SET tname =
COALESCE(
(SELECT TOP(1) t.task
FROM dbo.task t
WHERE t.prod_type=o.Prod_type
AND o.order_nbr = t.ORDER_NBR
ORDER BY t.strt_dt DESC), tname
)
FROM dbo.orders o
WHERE o.Prod_type IN ('1', '2');
3) Solution #2:
UPDATE o
SET tname = lt.task
FROM dbo.orders o
INNER JOIN
(
SELECT src.order_nbr, src.prod_type, src.task
FROM (
SELECT t.ORDER_NBR, t.prod_type, t.task,
ROW_NUMBER() OVER(PARTITION BY t.ORDER_NBR, t.prod_type ORDER BY t.strt_dt DESC) RowNum
FROM dbo.task t
) src
WHERE src.RowNum = 1
) lt -- last task
ON o.order_nbr = lt.ORDER_NBR AND o.prod_type = lt.prod_type
WHERE o.Prod_type IN ('1', '2');
If you have questions then feel free to ask.
4) An index on dbo.task(order_nbr, prod_type, strt_dt) include (task) should help both solutions.
5) Also you should publish the actual execution plans.

Answer 2

If the data size is large than i suggest you to use variables for updating the table, or Using CTE to update
Update a table using CTE and NEWID()
Updating record using CTE?
I hope this will help
with tname (t.task) as
(select top 1 t.task
from task t
where prod_typ='2' and order_nbr = t.ORDER_NBR
order by t.strt_dt desc )
insert into Orders(t.task)

Answer 3

Try something like this. This will update prod_type of 1 and 2 at the same time.
UPDATE orders
SET tname = t1.task
FROM orders o
CROSS APPLY (
SELECT order_nbr, prod_type, t.task, row_number() OVER (PARTITION BY order_nbr, prod_type ORDER BY strt_dt DESC) rownumber
FROM task t
WHERE o.prod_type = t.prod_type
AND o.order_nbr = t.order_nbr) t1
WHERE t1.rownumber = 1
AND o.prod_type in (1,2)

Answer 4

Using a CTE query will speed up this, because the subquery is need not be created for every row, it is pre-prepared. Here is the sqlfiddle
;with cteTaskNames as
(
select top 1 t.task
from task t
where prod_type='2' and order_nbr=t.ORDER_NBR
order by t.strt_dt desc
)
update orders
set tname = (select task from cteTaskNames)
where Prod_type='2'
go
Also,
1) Is "prod_type" an integer field or a string field?
2) If you add group by in the cte, you can do an inner join on orders and cte query to run all updates at once instead of doing each query.