SQL Server存储过程对200万条记录的排序，分页和筛选

Question

I created a stored procedure that select some data from one table

table name ( t1 ) columns ( id , a , b )

CREATE PROCEDURE spGetData
    @DisplayLength INT,
    @DisplayStart INT,
    @SortCol INT,
    @SortDir NVARCHAR(10),
    @Search  NVARCHAR(255) = NULL
AS
BEGIN
    DECLARE @FirstRec INT, @LastRec INT

    SET @FirstRec = @DisplayStart ;
    SET @LastRec = @DisplayStart + @DisplayLength;

    WITH CTE_Employees AS
    (
        SELECT
            ROW_NUMBER() OVER (ORDER BY
                                   CASE WHEN (@SortCol = 0 AND @SortDir = 'asc') THEN id END asc,
                                   CASE WHEN (@SortCol = 0 AND @SortDir = 'desc') THEN id END desc,
                                   CASE WHEN (@SortCol = 1 AND @SortDir = 'asc') THEN a END asc,
                                   CASE WHEN (@SortCol = 1 AND @SortDir = 'desc') THEN a END desc,
                                   CASE WHEN (@SortCol = 2 AND @SortDir = 'asc') THEN b END asc,
                                   CASE WHEN (@SortCol = 2 AND @SortDir = 'desc') THEN b END desc) AS RowNumber, 
            COUNT(*) OVER () AS TotalCount, 
            id, a, b
        FROM
            t1
        WHERE
            (@Search IS NULL 
             OR id LIKE '%' + @Search + '%' 
             OR a LIKE '%' + @Search + '%' 
             OR b LIKE '%' + @Search + '%')
    )
    SELECT * 
    FROM CTE_Employees
    WHERE RowNumber > @FirstRec AND RowNumber <= @LastRec   
END

It takes about 20 seconds to run the following

spGetData 1000 ,0,0,'desc'

which is very slow. This procedure will be called later from an ASP.NET MVC project and the result will be displayed using jQuery datatables.

How can I improve the performance of it?

Answer 1

First off, lose CTE. This is way too simple query for CTE.

CREATE PROCEDURE spGetData
    @DisplayLength int,
    @DisplayStart int,
    @SortCol int,
    @SortDir nvarchar(10),
    @Search nvarchar(255) = NULL
AS
BEGIN
    SELECT 
        COUNT(*) OVER () AS TotalCount,
        id,
        a,
        b
    FROM t1
    WHERE 
        (@Search IS NULL OR 
         id LIKE '%'+@Search+'%' OR 
         a LIKE '%'+@Search+'%' OR 
         b LIKE '%'+@Search+'%')
    ORDER BY
        CASE 
            WHEN @SortDir = 'ASC' THEN 
                CASE @SortCol
                    WHEN 0 THEN id
                    WHEN 1 THEN a
                    WHEN 2 THEN b
                END 
            END desc,
        CASE 
            WHEN @SortDir = 'desc' THEN 
                CASE @SortCol
                    WHEN 0 THEN id
                    WHEN 1 THEN a
                    WHEN 2 THEN b
                END 
            END DESC
    OFFSET @DisplayStart ROWS
    FETCH NEXT @DisplayLength ROWS ONLY
END 

This should be faster, but one more remark. LIKE searching within middle of string ( '%'+@Search+'%' ) can't use any indexing and will always be slow - especially on 2M rows and even worse - doing that on three different columns. It simply has to do a full table scan. Adding additional conditions (which are not LIKE) would improve performance.