CODEWITHSUNDEEP
sqlsql-serverstored-proceduresquery-optimizationrow-number
lije
lije

Reputation: 420

Performance optimization Row_Number() over partition by:

I have been trying to optimize performance of a stored procedure that has left join between a table (Table1) and a subquery. It gets slow when the records in the department table increase; I think, it is due to the ROW_NUMBER function. It works fine with lower number of records in the Department table. There is a left join between Table1 and a subquery at the end of the stored procedure.

Below is sample tables Table1, Employee and Department, involved in the query that is running slow.

CREATE TABLE [Table1]
(
    [ID] [int] NOT NULL PRIMARY KEY, 
    [Name] VARCHAR(250) NOT NULL,
    [DepartmentID] int 
) ON [PRIMARY] 

INSERT [Table1] ([ID], [Name], [DepartmentID])  
VALUES (1, N'A', 1), (2, N'D', 2),  
       (3, N'C', 3), (4, N'E', 4),
       (5, N'D', 5), (6, N'A', 6),
       (7, N'B', 7)  
GO 

CREATE TABLE [Department]
(
    [DepartmentID] [int] NOT NULL PRIMARY KEY, 
    [Name] VARCHAR(250) NOT NULL, 
) ON [PRIMARY] 

INSERT [Department] ([DepartmentID], [Name])  
VALUES (1, N'Engineering'), (2, N'Administration'),  
       (3, N'Sales'), (4, N'Marketing'), 
       (5, N'Finance') 
GO 

CREATE TABLE [Employee]
( 
    [EmployeeID] [int] NOT NULL PRIMARY KEY, 
    [FirstName] VARCHAR(250) NOT NULL, 
    [LastName] VARCHAR(250) NOT NULL, 
    [DepartmentID] [int] NOT NULL 
         REFERENCES [Department](DepartmentID), 
) ON [PRIMARY] 
GO

INSERT [Employee] ([EmployeeID], [FirstName], [LastName], [DepartmentID]) 
VALUES (1, N'Orlando', N'Gee', 1), (2, N'Keith', N'Harris', 2), 
       (3, N'Donna', N'Carreras', 3), (4, N'Janet', N'Gates', 3)

Sample query that is running slow.

SELECT 
    *
FROM
    Table1 AS t
LEFT JOIN 
    (SELECT 
         D.Name,
         E.DepartmentID,
         ROW_NUMBER() OVER (PARTITION BY E.DepartmentID
                            ORDER BY D.DepartmentID ASC,
                                     CASE WHEN D.Name IS NULL THEN 1 ELSE 0 END ASC, D.Name ASC) AS ord_index
     FROM 
         Department AS D
     INNER JOIN 
         Employee AS E ON D.DepartmentID = E.DepartmentID) AS x ON x.DepartmentID = t.DepartmentID
WHERE 
    x.ord_index = 1 OR x.ord_index IS NULL

I tried adding indexes per the suggestion of tuning advisor, but it doesn't improve the performance. I was also trying to use APPLY, but couldn't get it clear due to the ROW_NUMBER function. I have been trying to go through the related titles on the platform though.

I appreciate any guide in optimizing such query.

Upvotes: 2

Views: 3491

Answers (2)

The Impaler
The Impaler

Reputation: 48800

In the query, the section that is difficult to index is CASE WHEN D.Name IS NULL THEN 1 ELSE 0 END. This can, however, be indexed if you are willing to add an extra generated column in the Department table. For example:

alter table Department add name_flag as 
CASE WHEN D.Name IS NULL THEN 1 ELSE 0 END;

create index ix1 on Department (DepartmentID, name_flag, name);

With the the new column in place and the index, the query can be rephrased as:

SELECT * 
FROM Table1 t 
LEFT JOIN (
  SELECT D.Name, D.DepartmentID,  
    ROW_NUMBER() OVER (
      PARTITION BY D.DepartmentID  ORDER BY name_flag, D.Name
    ) AS ord_index
  FROM Department D 
  INNER JOIN Employee E ON D.DepartmentID = E.DepartmentID
) x ON x.DepartmentID = t.DepartmentID and x.ord_index = 1

Upvotes: 1

Stu
Stu

Reputation: 32609

You could refactor the query to use an outer apply which may result in a better execution plan (depending on the supporting indexes) such as :

select * from Table1 t 
outer apply (
    select D.Name,
        Row_Number() over (partition by E.DepartmentID order by D.DepartmentID asc, 
        case when D.Name is null then 1 
        else 0 end asc, D.Name asc) as ord_index 
    from Department D 
    join Employee E on D.DepartmentID = E.DepartmentID
    where E.DepartmentID = t.DepartmentID 
) x 
where x.ord_index = 1 or x.ord_index is null

See DB<>Fiddle showing better plan

Upvotes: 1

Related Questions