What is the alternative for outer apply? - sql

Question

Recently I have added outer apply in my a query. Since then this query takes forever. One reason i know that the table it is associated with is the biggest table in the database now.
select
a.*,
b.*,
BTab.*,
BTTab.*
from
tableA a
join tableB b ON a.ID = b.UID
join *****
left join *******
....
....
....
outer apply
(SELECT TOP 1 *
FROM
biggestTable bt
WHERE
bt.id = a.id
and a.id <> 100
ORDER BY a.datetime desc) BTab
Outer apply
(SELECT TOP 1 *
FROM
biggestTable btt
WHERE
btt.id = a.id
AND btt.DateTime <> '1948-01-01 00:00:00.000'
and btt.value = 0
order by btt.datetime desc) BTTab
where
..................
.................
....................
.................
Is there any better solution than using outer apply?

Answer 1

Here's an alternative, can't say whether its better or not. You may simply need better indexes on your big table
WITH BTAB as
( SELECT TOP 1
* ,
row_nubmer() over (partition by b.id) rn
FROM
biggestTable bt
) ,
BTTab as (SELECT TOP 1
* ,
row_nubmer() over (partition by btt.id order by btt.datetime desc) rn
FROM
biggestTable btt
WHERE
AND btt.DateTime <> '1948-01-01 00:00:00.000'
and btt.value = 0
)
select
a.*,
b.*,
BTab.*,
BTTab.*
from
tableA a
join tableB b ON a.ID = b.UID
join *****
left join BTab on ON a.ID = BTab.ID
and BTAB.rn = 1
left join BTTabon ON a.ID = BTTab.ID
and BTTab.rn = 1

Answer 2

+1 for Conrad as his answer might be all you need and I reused some of his syntax.
Problem with Apply and CTE is they are evaluated for each row in the a, b join.
I would create two temporary tables. To represent the max rows and put a PK on them. The benefit is these two expensive quires are done once and the join is to a PK. Big benefit joining to a PK. I eat the overhead of #temp to get a single evaluation and PK a lot.
Create table #Btab (int ID PK, ...)
insert into #Btab
WITH BTAB as
( SELECT * ,
row_nubmer() over (partition by b.id) rn
FROM
biggestTable
where ID <> 100
)
Select * from BTAB
Where RN = 1
order by ID
Create table #Bttab (int ID PK, ...)
insert into #Bttab
WITH BTTAB as
( SELECT * ,
row_nubmer() over (partition by id order by datetime desc) rn
FROM
biggestTable
where DateTime <> '1948-01-01 00:00:00.000' and value = 0
)
Select * from BTAB
Where RN = 1
order by ID
select
a.*,
b.*,
#Btab.*,
#Bttab.*
from
tableA a
join tableB b ON a.ID = b.UID
join *****
left join *******
....
....
....
left outer outer join #Btab
on #Btab.ID = a.ID
left outer outer join #Bttab
on #Bttab.ID = a.ID
where
..................
.................
P.S. I am exploring TVP over #TEMP for this. A TVP supports a PK and has less overhead than #tmp. But I have not compared them head to head in this type of application.
Tested TVP over #TEMP and got a 1/2 second improvement (about the time it take to create and delete a temporary table).