CODEWITHSUNDEEP
sqlsql-serverperformancecommon-table-expression
Seph
Seph

Reputation: 8693

Improve SQL CTE query performance

Is there any way to improve the performance of the following CTE query (@E and @R are tables with indexes in the actual system):

DECLARE @id bigint = 1

DECLARE @E TABLE
(
id bigint,
name varchar(50)
)

DECLARE @R TABLE
(
child_id bigint,
parent_id bigint
)

INSERT INTO @E SELECT 1, 'one'
INSERT INTO @E SELECT 2, 'two'
INSERT INTO @E SELECT 3, 'three'
INSERT INTO @E SELECT 4, 'four'
INSERT INTO @E SELECT 5, 'five'
INSERT INTO @E SELECT 6, 'six'
INSERT INTO @E SELECT 7, 'seven'

INSERT INTO @R SELECT 1, 2
INSERT INTO @R SELECT 1, 3
INSERT INTO @R SELECT 3, 4
INSERT INTO @R SELECT 5, 4
INSERT INTO @R SELECT 3, 6
INSERT INTO @R SELECT 7, 4

; WITH cte
(
child_id,
parent_id
)

AS (

SELECT * FROM @R R
WHERE R.child_id = @id

UNION ALL
SELECT R.* FROM @R R
INNER JOIN cte ON CTE.parent_id = R.child_id

)
SELECT * FROM @E E
WHERE e.id = @id
UNION ALL
SELECT P.* FROM @E E
INNER JOIN cte ON 1=1
INNER JOIN @E P ON P.id = cte.parent_id
WHERE e.id = @id
ORDER BY 1

Expected Results:

id | name
1  | one 
2  | two
3  | three
4  | four
6  | six

In real world data I will be dealing with many millions of rows in @R and about a hundred thousand rows in in @E. So I'm looking to see if there's anything I can do to squeeze a little more performance out.

Edit: just to clarify and summarize so far, there is a clustered pk index on R with child_id, parent_id and adding an index to @r.parent_id will also improve join performance.

Is there anything improve this? The bit after the CTE with the inner join 1=1 is there anything that could be improved upon here is is that about as good as it will get? Is there any other schema design I could do to get similar parent-child mapping with better performance?

Upvotes: 4

Views: 8920

Answers (2)

Seph
Seph

Reputation: 8693

As marc_s pointed out

a clustered index on [child_id, parent_id] for this table is not good enough - you should have separate indices on both (child_id) and (parent_id) to speed up JOIN performance. If you have a compound index on (child_id, parent_id) in that order, then this can be used for child_id alone - but not for parent_id alone (and it will need both) – marc_s Dec 19 '11 at 12:35

This greatly improved the performance of the query and helped me understand how CTE queries work internally.

Upvotes: 3

Oleg Dok
Oleg Dok

Reputation: 21766

For the very beginning:

DECLARE @E TABLE
(
    id BIGINT PRIMARY KEY,
    name varchar(50)
)

DECLARE @R TABLE
(
    child_id bigint,
    parent_id BIGINT,
    PRIMARY KEY(child_id, parent_id),
    UNIQUE (parent_id, child_id)
)

But, keep in mind that Sql Server is very poor in optimizing CTEs.

Upvotes: 0

Related Questions