Question

我正在尝试准备一些数据以供第三方删除，不幸的是，他们只能批量处理2000条记录中的数据。我有100k记录，可能需要多次分割和导出这些数据，所以我想以某种方式自动化这个过程。

使用SQL Server 2008有一种相当简单的方法吗？我没有运行复杂的查询 - 距离SELECT PKID FROM Sometable ORDER BY PKID并不太远 - 尽管我可以使用光标做到这一点，但我想知道是否有更好的方法。

Answer 1

我认为您可以利用ROW_NUMBER然后使用BETWEEN来指定您喜欢的行范围。或者，如果您知道没有间隙，或者不关心间隙，则可以使用PKID

e.g。

SELECT ...
FROM
   (SELECT ... 
         ROW_NUMBER() OVER(ORDER BY PKID ) as RowNum
    FROM Sometable e
   ) t
WHERE RowNum BETWEEN @startRowIndex AND (@startRowIndex + @maximumRows) - 1

这通常用于分页结果。 4GuysFromRolla有一个很好的article on it

Answer 2

SET NOCOUNT ON;

CREATE TABLE [dbo].[SyncAudit] ( PkId INT, BatchNumber INT)

DECLARE  @batchsize INT
    ,@rowcount INT
    ,@batchcount INT
    ,@rootdir VARCHAR(2048)
    ,@saveas VARCHAR(2048)
    ,@query VARCHAR(2048)
    ,@bcpquery VARCHAR(2048)
    ,@bcpconn VARCHAR(64)
    ,@bcpdelim VARCHAR(2)

SET     @rootdir    = '\\SERVER1\SHARE1\FOLDER\'
SET     @batchsize  = 2000
SET     @bcpdelim   = '|'
SET     @bcpconn    = '-T' -- Trusted
--SET       @bcpconn    = '-U <username> -P <password>' -- SQL authentication

SELECT  @rowcount   = COUNT(1),
    @batchcount = CEILING(COUNT(1)/@batchsize) FROM <@TableName, string, 'SomeTable'>

SELECT [BatchSize] = @BatchSize, [BatchCount] = @Batchcount

INSERT INTO SyncAudit
SELECT 
 <@TableKey, string, 'PKField'>
 ,groupnum = NTILE(@batchcount) OVER ( ORDER BY <@TableKey, string, 'PKField'>)
FROM
<@TableName, string, 'SomeTable'>

WHILE (@batchcount > 0)
BEGIN

SET @saveas = @rootdir + 'batchnumber-' + cast(@batchcount as varchar) + '.txt'
SET @query = '  SELECT  [<@TableName, string, 'SomeTable'>].* 
                FROM    [' + db_name() + '].[dbo].[<@TableName, string, 'SomeTable'>] 
                JOIN    [' + db_name() + '].[dbo].[SyncAudit]   
                            ON  [<@TableName, string, 'SomeTable'>].<@TableKey, string, 'PKField'> = [SyncAudit].PkId 
                            AND [SyncAudit].BatchNumber = ' + cast(@batchcount as varchar) + ''

SET @bcpquery = 'bcp "' + replace(@query, char(10), '') + '" QUERYOUT "' + @saveas + '" -c -t^' + @bcpdelim + ' ' + @bcpconn + ' -S ' + @@servername
EXEC master..xp_cmdshell @bcpquery  

--EXEC (@query)


SET @batchcount = @batchcount -1
END


DROP TABLE [dbo].[SyncAudit] -- or leave for reference

Answer 3

您可以在@@ ROWCOUNT循环中计算范围以定位所需的行。它可能比ROW_NUMBER（）更好，它必须从头开始编号。

declare @startid int
declare @endid int

-- get one range, these are efficient as they go over the PKID key by range
select top(1) @startid = pkid from sometable order by pkid  -- 1 key visited
select top(2000) @endid = pkid from sometable order by pkid  -- 2000 keys visited
-- note: top 2000 may end up with the 514th id if that is the last one

while @@ROWCOUNT > 0
begin
    insert otherdb.dbo.backupcopy
    select * from sometable
    where pkid between @startid and @endid

    select top(1) @startid = pkid from sometable
    WHERE pkid > @endid -- binary locate
    order by pkid

    select top(2000) @endid = pkid from sometable
    WHERE pkid > @endid -- binary locate, then forward range lookup, max 2000 keys
    order by pkid
end

Answer 4

我最终使用了 cyberkiwi 和 Adam 提供的方法的组合。我不需要仅使用ROW_NUMBER，因为我在IDENTITY数据类型中使用了table列。

这是我使用的代码的编辑版本 - 它就像一个魅力。再次感谢大家的帮助！

use Testing
GO

SET NOCOUNT ON

declare
    @now datetime = GETDATE(), 
    @batchsize int = 2000, 
    @bcpTargetDir varchar(500) = '\\SomeServer\Upload\', 
    @csvQueryServer varchar(500) = '.\SQLExpress', 

    @rowcount integer, 
    @nowstring varchar(100), 
    @batch_id int, 
    @startid int, 
    @endid int, 
    @oidCSV varchar(max), 
    @csvQuery varchar(max), 
    @bcpFilename varchar(200), 
    @bcpQuery varchar(1000)

declare @tblBatchRanges table (
    batch_id integer NOT NULL IDENTITY(1,1) PRIMARY KEY, 
    oid_start integer NOT NULL, 
    oid_end integer NOT NULL, 
    csvQuery varchar(max)
)

-- Create a unique timestamp-based string, which will be used to name the exported files.
select @nowstring = CONVERT(varchar, @now, 112) + '-' + REPLACE(CONVERT(varchar, @now, 114), ':', '')


-- 
select top(1) @startid = oid from Testing..MyObjectIds order by oid
select top(@batchsize) @endid = oid from Testing..MyObjectIds order by oid
select @rowcount = @@ROWCOUNT

while (@rowcount > 0) begin
    -- Create a CSV of all object IDs in the batch, using the STUFF() function (http://goo.gl/EyE8L).
    select @csvQuery = 'select stuff((select distinct '','' + CAST(oid as varchar) from Testing..MyObjectIds where oid between ' + CAST(@startid as varchar) + ' and ' + CAST(@endid as varchar) + ' order by '','' + CAST(oid as varchar) for xml path('''')),1,1,'''')'


    -- Log the info and get the batch ID.   
    insert into @tblBatchRanges (oid_start, oid_end, csvQuery)
        values (@startid, @endid, @oidCSV, @csvQuery)

    select @batch_id = @@IDENTITY


    -- Advance @startid and @endid so that they point to the next batch
    select top(1) @startid = oid 
        from Testing..MyObjectIds
        where oid > @endid
        order by oid

    select top(@batchsize) @endid = oid 
        from Testing..MyObjectIds
        where oid > @endid
        order by oid

    select @rowcount = @@ROWCOUNT


    -- Export the current batch to a file.
    select @bcpFilename = 'MyExport-' + @nowstring + '-' + cast(@batch_id as varchar) + '.txt'
    select @bcpQuery = 'bcp "' + @csvQuery + '" QUERYOUT "' + @bcpTargetDir + @bcpFilename + '" -S ' + @csvQueryServer + ' -T -c'
    exec master..xp_cmdshell @bcpquery
end

SET NOCOUNT OFF


--Check all of the logged info.
select oid_start, oid_end, csvQuery from @tblBatchRanges

将SQL Server 2008查询分解为批处理

4 个答案: