Question

Let’s head straight to the example:

Query 1:

Select * 
Into #temp_v1 
From View1

Select * 
Into #temp_v2
From View2

select *
From #temp_v1 v1
where not exists (
    Select * From #temp_v2 where key = v1.key
)

This is much faster than

Query 2:

select *
From View1 v1
where not exists (
    Select * From View2 where key = v1.key
)

Now, obviously, I have simplified the example. View1 is a view of views and and more comparisons are needed that make it difficult to use a join.

My question is really not how I should rather write my SQL, but how come SQL Server can execute query 1 in 3 seconds, and query 2 in 10 minutes.

And more importantly is there a magic option that I can give the optimizer to create an execution plan like query 1.

Answer 1

If your view is complicated, then accessing the view will be the slowest part of the operation by far, and query 1 minimizes that.

Suppose each view has 1,000,000 rows. In query 1, you are only retrieving a row from a view 2,000,000 times, while in query 2 you will likely retrieve a row from a view 500,000,000,000 times or more. (In a best-case scenario, every v1.key will exist in view 2. But you still have to check roughly half the rows in view 2 each time to verify that).

I don’t know about execution plans in SQL server, but writing your query something like this should also be more efficient:

with keys as (
   select key from View2
)
select *
From View1 v1
where not exists (
    Select * From keys where key = keys.key
)

By getting all the keys from view 2 ahead of time, you essentially have the same efficiency of your query 1, without the temporary table step.