构建更高效的 MS SQL 函数以 return 均匀分割范围

Question

假设您有以下 SQL table:

-- create temp table
CREATE TABLE [tempNums] 
(
    id INT NOT NULL,
    somedate datetime NULL
)
GO

有一些数据（见下面的 tempSplitStringToInts 定义）：

-- with date
INSERT INTO [tempNums]
    SELECT id, GETUTCDATE()
    FROM [tempSplitStringToInts] ('1,2,3,5,10,100,101,102,103,233,1001,5002,5003,5005,5007,5010',',') 
GO

-- without date
INSERT INTO [tempNums]
    SELECT id, NULL
    FROM [tempSplitStringToInts] ('6,7,8,150,151,152,153,433,2001,2002,2003,2005,3007,10010',',') 
GO

How do you build BETTER/FASTER function that will take number of ranges, and a flag bit as input and return a table of range values?

例如，这样的方法可行，但对于非常大的 tables 来说速度很慢：

-- create range function
CREATE FUNCTION [tempFnGetIdRanges]
(
    @apps INT,
    @has_date BIT
)
RETURNS @ret TABLE
(
    RangeNum INT, 
    MinNum INT, 
    MaxNum INT
)
AS
BEGIN

    DECLARE @i INT = 0;
    DECLARE @count INT;
    DECLARE @min INT;
    DECLARE @max INT = 0;

    IF @has_date = 1
    BEGIN
        SELECT @count = COUNT(id) 
            FROM [tempNums] 
            WHERE somedate IS NOT NULL
    END
    ELSE
    BEGIN
        SELECT @count = COUNT(id) 
            FROM [tempNums] 
            WHERE somedate IS NULL

    END

    DECLARE @top INT = @count/@apps;

    WHILE @i<@apps
    BEGIN

        IF @i+1=@apps
        BEGIN
            -- on last get reminder
            SET @top = @top + @apps 
        END

        IF @has_date = 1
        BEGIN       
            SELECT @min = MIN(id), @max = MAX(id)
            FROM
            (
                SELECT TOP (@top) id 
                FROM [tempNums] 
                WHERE somedate IS NOT NULL
                    AND id > @max
                ORDER BY id
            ) XX
        END
        ELSE
        BEGIN
            SELECT @min = MIN(id), @max = MAX(id)
            FROM
            (
                SELECT TOP (@top) id 
                FROM [tempNums] 
                WHERE somedate IS NULL
                    AND id > @max
                ORDER BY id
            ) XX
        END


        INSERT INTO @ret VALUES(@i, @min, @max)

        SET @i = @i + 1;
        CONTINUE
    END

    RETURN
END
GO

因此，当您运行以下内容时：

SELECT * FROM [tempFnGetIdRanges](4, 0)
SELECT * FROM [tempFnGetIdRanges](4, 1)

第一个语句的结果：

RangeNum    MinNum  MaxNum
0           6       8
1           150     152
2           153     2001
3           2002    10010

第二条语句的结果：

RangeNum    MinNum  MaxNum
0           1       5
1           10      102
2           103     5002
3           5003    5010

拆分功能（仅供参考，但不是本题重点）：

-- create split string function
CREATE  FUNCTION [tempSplitStringToInts] ( @SourceString VARCHAR(MAX) , @delimeter VARCHAR(10))
RETURNS @IntList TABLE
   (
     id INT
   )
AS
BEGIN
IF RIGHT(@SourceString, LEN(@delimeter))<> @delimeter
    BEGIN
        SELECT @SourceString = @SourceString + @delimeter
    END

DECLARE @LocalStr VARCHAR(MAX)
DECLARE @start INT
DECLARE @end INT
SELECT @start = 1
SELECT @end =  CHARINDEX ( @delimeter , @SourceString , @start ) 

WHILE @end > 0
    BEGIN 
        SELECT @LocalStr = SUBSTRING ( @SourceString , @start , @end - @start ) 
        IF LTRIM(RTRIM(@LocalStr)) <> '' 
            BEGIN
                INSERT @IntList (id) VALUES (CAST(@LocalStr AS INT))
            END
        SELECT @start = @end + LEN(@delimeter)
        SELECT @end = CHARINDEX ( @delimeter , @SourceString , @start ) 
    END
   RETURN
END
GO

As I said this works, but it is slow for very large tables. Is there a better way to write tempFnGetIdRanges function? Something native to SQL? I am using MS SQL 2012 if that is relevant.

Answer 1

不太确定您的 GetRanges 函数试图做什么，但您绝对不需要循环。当您将 HasDate 作为 1 传递时，此函数 returns 与您的值相同。

create function GetRanges
(
    @NumGroups int
) returns table as return

    with MyGroups as
    (
        select NTILE(@NumGroups) over(order by t.id) as GroupNum
            , t.id
        from tempnums t
    )

    select GroupNum
        , MIN(id) as MinNum
        , MAX(id) as MaxNum
    from MyGroups
    group by GroupNum

--编辑--

现在我看到你发布了两组示例数据，我明白了这个问题。

您可以通过以下方式调整它以在某个日期适应 NULL 或 NOT NULL。

alter function GetRanges
(
    @NumGroups int
    , @HasDate bit
) returns table as return

    with MyGroups as
    (
        select NTILE(@NumGroups) over(order by t.id) as GroupNum
            , t.id
        from tempnums t
        where
        (
            @HasDate = 1
            AND
            t.somedate is not null
        )
        OR
        (
            @HasDate = 0
            AND
            t.somedate is null
        )
    )

    select GroupNum
        , MIN(id) as MinNum
        , MAX(id) as MaxNum
    from MyGroups
    group by GroupNum

我看到的问题是您只有 14 行有 NULL，所以不确定为什么您想要的输出是这样的。由于 NTILE 将不均匀的行分组的方式不同，使用 NTILE 会对样本数据产生略微不同的结果。

构建更高效的 MS SQL 函数以 return 均匀分割范围

Build more efficient MS SQL function to return evenly split ranges

sql-server

sql-function

sql-optimization