When we should use a subquery?
Which one will faster in SQL server according to query performance?
These are common questions when we need to get data using more than one table. There is not a straightforward answer to these questions. Here we will discuss many aspects of this.
1. Use joins when we need to get data from both the tables in a SELECT statement.
2. Use subquery when we need to get data from only one table and another table is used only to check existence. For example:
Let's assume we have two tables: tblEmp and tblEmpDetail with the following schema:
CREATE TABLE tblEmp(
ntEmpID BIGINT PRIMARY KEY IDENTITY,
vcName VARCHAR(50),
ntAge INT
)
CREATE TABLE tblEmpDetail(
ntEmpDetailID BIGINT PRIMARY KEY IDENTITY,
ntEmpID BIGINT REFERENCES tblEmp,
moSalary MONEY,
dtDOJ DATETIME
)
--Inserting some recodes into it
INSERT tblEmp
SELECT
LEFT([Text],20),
severity
FROM Sys.messages
INSERT tblEmpDetail
SELECT
ntEmpID,
2000 * (ntEmpID % 10),
GETDATE() - (ntEmpID % 30)
FROM tblEmp
Inner join: By using inner join we can select fields of both the tables. For example:
SELECT
E.ntEmpID,
E.vcName,
E.vcName,
ED.moSalary,
ED.dtDOJ
FROM tblEmp E INNER JOIN tblEmpDetail ED
ON E.ntEmpID = ED.ntEmpID
Subquery: we can get the data from the only table. For example:
SELECT
E.ntEmpID,
E.vcName,
E.vcName
FROM tblEmp E
WHERE ntEmpID IN (SELECT ntEmpID
FROM tblEmpDetail)
Here we cannot include fields of table tblEmpDetail in a select list. So if we need to get the data from fields of both tables join is the only solution. It is not possible by the subquery.
When we need to get the data from only one table by checking the existence in another table then it is possible by both joins as well as a subquery. In this case which one will perform better inner join or subquery?
Here we have written the same query in four different ways:
1.
--Using inner join
SELECT E.*
FROM tblEmp E INNER JOIN tblEmpDetail ED
ON E.ntEmpID = ED.ntEmpID
2.
--Using IN clause
SELECT *
FROM tblEmp
WHERE ntEmpID IN (SELECT ntEmpID
FROM tblEmpDetail
3.
--Using EXISTS clause
SELECT *
FROM tblEmp E
WHERE EXISTS(SELECT *
FROM tblEmpDetail
WHERE ntEmpID = E.ntEmpID)
4.
--Using correlated subquery
SELECT *
FROM tblEmp E
WHERE ntEmpID = (SELECT ntEmpID
FROM tblEmpDetail
WHERE ntEmpID = E.ntEmpID)
Let's compare the actual execution plan:
Query 1, 2 and 3 have same execution cost while query 4 is costliest. Also, query 2 and 3 have an exactly same execution plan. Let me explain it one by one.
Why query 4 is costliest compare to others?
If we will observe its execution plan we will find it has index spool operator which cost is 95% and has an assert operator. It is due to equality arithmetic operator that is:
ntEmpID = (SELECT ntEmpID
FROM tblEmpDetail
WHERE ntEmpID = E.ntEmpID)
SQL server doesn't know how many rows will be returned by this subquery. If it is more than one then it will throw an error: Subquery returned more than 1 value. This is not permitted when the subquery follows =, !=, <, <=, >, >= or when the subquery is used as an expression.
So, to validate this, SQL server has followed following extra steps:
1. Index spool: It creates a copy of a table in tempdb which satisfy WHERE ntEmpID = E.ntEmpID and create an index on it.
2. Stream Aggregate: It counts the total records in tblEmpDetail for each ntEmpID something like this [Expr1006] = Count (*)
3. Assert: It checks if record count is one then it is fine otherwise throw error message that is CASE WHEN [Expr1006]>(1) THEN (0) ELSE NULL END
Note: Even top one clause will not help us. For example:
ntEmpID = (SELECT TOP(1) ntEmpID
FROM tblEmpDetail
WHERE ntEmpID = E.ntEmpID)
Conclusion: Never write subquery with equal to an operator for this scenario.
Solution: Create unique index on the field tblEmpDetail.ntEmpID. For example:
CREATE UNIQUE NONCLUSTERED INDEX NCI_ntEmpID
ON tblEmpDetail(ntEmpID)
Now if you will check the execution plan, you will find all four queries have equal execution cost (25%).
Suppose there is not any unique index on the field tblEmpDetail.ntEmpID so I am dropping it:
DROP INDEX NCI_ntEmpID ON tblEmpDetail
Now compare the execution plan of query 1, 2 and 3:
Query 1: It uses Hash Match (Inner Join)
Query 2 and 3: It uses Hash Match (Right Semi Join)
Operator Inner Join can get data from both tables while operator Right Semi Join can get the data from an only right table.
Conclusion: Inner join has more flexibility than a subquery. It can select the data from both tables as well as only from one table with same query cost just like subquery. For example:
--Inner Join 1
SELECT *
FROM tblEmp E INNER JOIN tblEmpDetail ED
ON E.ntEmpID = ED.ntEmpID
--Inner Join 2
SELECT E.*
FROM tblEmp E INNER JOIN tblEmpDetail ED
ON E.ntEmpID = ED.ntEmpID
Actual execution plan:
Definitely Inner join 1 will more costly than inner Join 2 since it is selecting more numbers of columns.
Is there any other difference between inner join and subquery except inner join get data from both tables while not by subquery?
We are going to check an inserting observation. Comparing the execution plan of the following query:
--Using inner join
SELECT TOP(10) E.*
FROM tblEmp E INNER JOIN tblEmpDetail ED
ON E.ntEmpID = ED.ntEmpID
--Using IN clause
SELECT TOP(10) *
FROM tblEmp
WHERE ntEmpID IN (SELECT ntEmpID
FROM tblEmpDetail)
Actual execution plan:
So, in this case, an inner join is performing better than a subquery. Now we are inserting few duplicates ntEmpID into the tblEmpDetail:
INSERT tblEmpDetail VALUES
(1,5000,GETDATE()),
(2,7000,GETDATE())
Comparing the execution plan of the same query:
--Using inner join
SELECT TOP(10) E.*
FROM tblEmp E INNER JOIN tblEmpDetail ED
ON E.ntEmpID = ED.ntEmpID
--Using IN clause
SELECT TOP(10) *
FROM tblEmp
WHERE ntEmpID IN (SELECT ntEmpID
FROM tblEmpDetail)
Now in this case subquery is performing better than inner join.
Why there are differences? Sometimes an inner join performs better and sometime subquery performs better. To understand this we are going to create two tables with fewer records and checking the output:
CREATE TABLE tblStu(
ntStuID BIGINT,
vcName VARCHAR(50),
ntAge INT
)
CREATE TABLE tblStuDetail(
ntStuDetailID BIGINT,
ntStuID BIGINT,
moSalary MONEY
)
Now we are inserting three records in both the tables:
INSERT tblStu VALUES
(1,'Scott', 18),
(2,'Greg', 21),
(3,'Davis', 22)
INSERT tblStuDetail VALUES
(1, 1, 5000),
(2, 2, 8000),
(3, 3, 6000)
Now, both the table has one to one relationship for ntStuID. That is one student has one detail and one details belong to one student.
Note: But SQL server doesn't know both tables has one to one relationship. To tell SQL server, we have to create a unique constraint in ntStuID of both tables. We can do it by making it unique key, Primary key or by creating a unique index.
Let's check output of following queries:
--Inner join
SELECT S.* FROM tblStu S INNER JOIN tblStuDetail SD
ON S.ntStuID = SD.ntStuID
--Subquery
SELECT * FROM tblStu S
WHERE S.ntStuID IN(SELECT SD.ntStuID
FROM tblStuDetail SD WHERE S.ntStuID = SD.ntStuID)
Inner Join:
ntStuID
|
vcName
|
ntAge
|
1
|
Scott
|
18
|
2
|
Greg
|
21
|
3
|
Davis
|
22
|
Sub query:
ntStuID
|
vcName
|
ntAge
|
1
|
Scott
|
18
|
2
|
Greg
|
21
|
3
|
Davis
|
22
|
We will get an exactly same output of both the query.
Now we are making the one to many or many to one relation between the tables. So we are going to insert few duplicate records into the table tblStuDetail:
INSERT tblStuDetail VALUES(4,2,5000),(5,2,8000)
Note: Again SQL server doesn't know both tables has one to many relationships or not. To tell SQL server, we have to create a unique constraint in ntStuID in only one table. We can do it by making it unique key, Primary key or by creating a unique index.
Let's check output of same queries:
--Inner join
SELECT S.* FROM tblStu S INNER JOIN tblStuDetail SD
ON S.ntStuID = SD.ntStuID
--Subquery
SELECT * FROM tblStu S
WHERE S.ntStuID IN(SELECT SD.ntStuID
FROM tblStuDetail SD WHERE S.ntStuID = SD.ntStuID)
Inner Join:
1
|
Scott
|
18
|
2
|
Greg
|
21
|
3
|
Davis
|
22
|
2
|
Greg
|
21
|
2
|
Greg
|
21
|
1
|
Scott
|
18
|
Sub query:
ntStuID
|
vcName
|
ntAge
|
1
|
Scott
|
18
|
2
|
Greg
|
21
|
3
|
Davis
|
22
|
Now inner join is returning more records than a subquery. Both queries have different output!! How can we compare the performance if both queries return different output? They are two different queries.
Last, now are going to make it many to many relationships between them. For this we are also going to insert few duplicate records into the table tblStu:
INSERT tblStu VALUES
(1,'Alan',19),
(2,'Sam',21)
Note: If two tables have not any primary key, unique key or unique index on column SD.ntStuID, SQL server assume both tables have many to many relations.
Let's again check output of same queries:
--Inner join
SELECT S.* FROM tblStu S INNER JOIN tblStuDetail SD
ON S.ntStuID = SD.ntStuID
--Subquery
SELECT * FROM tblStu S
WHERE S.ntStuID IN(SELECT SD.ntStuID
FROM tblStuDetail SD WHERE S.ntStuID = SD.ntStuID)
Inner Join:
ntStuID
|
vcName
|
ntAge
|
1
|
Scott
|
18
|
2
|
Greg
|
21
|
2
|
Greg
|
21
|
2
|
Greg
|
21
|
3
|
Davis
|
22
|
1
|
Alan
|
19
|
2
|
Sam
|
21
|
2
|
Sam
|
21
|
2
|
Sam
|
21
|
Sub query:
ntStuID
|
vcName
|
ntAge
|
1
|
Scott
|
18
|
2
|
Greg
|
21
|
3
|
Davis
|
22
|
1
|
Alan
|
19
|
2
|
Sam
|
21
|
Again, inner join returning more records than a subquery. Also subquery returning duplicate recodes. Both queries have different output. In this case, we cannot compare the performance between subquery and inner join since both queries have different output.
Question: If two tables have one to one relation for a key column and we want to get data from only one table in this case inner join will perform better or subquery?
Answer: As I know the performance of both the query will exactly same. Till now, I didn't find any differences if you please share with us.
Best practice: Inner join vs. subquery
1. If we need the data from both the tables we must have to choose inner join. If we need data from only one table then we can choose either subquery or inner join.
2. If two tables have one to many, many to one or many to many, subquery and inner join may have different output. So choose according to your application requirement.
3. If two tables have one to one relationship then you can choose either subquery or inner join since query optimizer will follow same execution plan for both of them. (Thanks to SQL server query optimizer)
4. Following subqueries are equivalent from query performance point of view:
a.
SELECT * FROM tblStu S
WHERE S.ntStuID IN(SELECT SD.ntStuID
FROM tblStuDetail SD WHERE S.ntStuID = SD.ntStuID)
b.
SELECT * FROM tblStu S
WHERE S.ntStuID IN(SELECT SD.ntStuID
FROM tblStuDetail SD)
c.
SELECT * FROM tblStu S
WHERE EXISTS(SELECT *
FROM tblStuDetail SD WHERE S.ntStuID = SD.ntStuID)
An SQL server query optimizer is smart enough that it will choose same execution for all.
5. Don't use subquery with equality operator unless there is not one to one relation between two tables. For example:
SELECT * FROM tblStu S
WHERE S.ntStuID = (SELECT SD.ntStuID
FROM tblStuDetail SD
WHERE S.ntStuID = SD.ntStuID)
Otherwise, it will decrease the performance or cause of an error.
6. If there is an aggregate function in subquery then there is no direct equivalent query using an inner join. For example:
SELECT * FROM tblStu S
WHERE S.ntStuID = (SELECT MAX(SD.ntStuID)
FROM tblStuDetail SD)
So you must have to use a subquery.
7. If a query is either inner join or subquery, internally is handled by any of these operators:
a. Nested loop join
b. Merge join
c. Hash join
So, technically there is no difference between inner join and subquery.
8. Consider on two inner join syntax:
--New way
SELECT S.* FROM tblStu S
INNER JOIN tblStuDetail SD
ON S.ntStuID = SD.ntStuID
--Old way
SELECT S.*
FROM tblStu S ,tblStuDetail SD
WHERE S.ntStuID = SD.ntStuID
There are not any differences except syntax. I will prefer using ON clause to write join condition instead of WHERE clause. If we will miss the ON clause:
SELECT S.*
FROM tblStu S
INNER JOIN tblStuDetail SD
We will get an error message: Incorrect syntax near 'SD'.
And we can correct it. What will happen if we miss the WHERE clause:
SELECT S.*
FROM tblStu S ,tblStuDetail SD
We get wrong output without any warning. The worst thing is if one or both table has too many records this query will be too costly and locked the both table for a long period of time. So I suggest don't write join query in the old way.
Write SQL queries in set-based approach SQL server
SQL server best practices for stored procedure
What is Adhoc query in SQL server: Cached Query Plan
Indexes best practices in SQL server: Where to start creating indexes?
SQL server query optimization tips: Tuning best practices with examples
Write SQL queries in set-based approach SQL server
SQL server best practices for stored procedure
What is Adhoc query in SQL server: Cached Query Plan
Indexes best practices in SQL server: Where to start creating indexes?
SQL server query optimization tips: Tuning best practices with examples
