Question

我需要从最近正在写入的表中删除旧记录。我的问题是，当其他进程尝试读取或写入表时，我的delete语句将因死锁而失败。它是一个大型存储过程的一部分，该存储过程以Transaction isolation level read uncommitted和deadlockpriority low开头。

如何转换此cte和delete语句以部分删除记录？例如，如果我可以选择在cte（x）中为x / 500条记录运行delete语句的记录数，那会很完美。

  ;with chargesToDelete(id, ciid) as (
        select c.id, ci.Id from @chargeids c
        left join xxx.dbo.chargeitems ci on ci.Charge_Id = c.id
        where ci.id is null
    )
    delete from xxx.dbo.charges
        where Id in (select id from chargesToDelete);

Answer 1

您可以在@@ROWCOUNT上使用会话变量TOP N和DELETE进行循环。

SELECT 1 -- Forces @@ROWCOUNT = 1

WHILE @@ROWCOUNT > 0
BEGIN

    delete TOP (500) I from 
        xxx.dbo.charges AS I
    where 
        exists (
            select 
                'to delete' 
            from 
                @chargeids c
                left join xxx.dbo.chargeitems ci on ci.Charge_Id = c.id
            where 
                ci.id is null AND
                c.id = I.Id)

END

如果您不想使用前哨SELECT（它将结果返回给客户端，如果有的话），则可以使用变量。

DECLARE @ForceStart BIT = 1

WHILE @@ROWCOUNT > 0 OR @ForceStart = 1
BEGIN

    SET @ForceStart = 0

    delete TOP (500) I from 
        xxx.dbo.charges AS I
    where 
        exists (
            select 
                'to delete' 
            from 
                @chargeids c
                left join xxx.dbo.chargeitems ci on ci.Charge_Id = c.id
            where 
                ci.id is null AND
                c.id = I.Id)

END

如果子查询需要花费很长时间进行处理，则可能要创建一个临时表，该表具有要删除的ID并在循环中与之联接。

请注意，如果要检查xxx.dbo.chargeitems上的记录是否不存在，则进行NOT EXISTS的速度要比LEFT JOIN和IS NULL快。

编辑：使用临时表保存ID：

-- Create temporary table
IF OBJECT_ID('tempdb..#ChargesToDelete') IS NOT NULL
    DROP TABLE #ChargesToDelete

SELECT DISTINCT
    c.id
INTO
    #ChargesToDelete
from 
    @chargeids c
    left join xxx.dbo.chargeitems ci on ci.Charge_Id = c.id
where 
    ci.id is null

CREATE CLUSTERED INDEX CI_ChargesToDelete ON #ChargesToDelete (id)


-- Delete rows in batches
SELECT 1 -- Forces @@ROWCOUNT = 1

WHILE @@ROWCOUNT > 0
BEGIN

    delete TOP (500) I from 
        xxx.dbo.charges AS I
    where 
        exists (select 'to delete' from #ChargesToDelete AS C where c.id = I.Id)

END

Answer 2

您可以尝试以下代码：

declare @deleted table (id int)
declare @amountToDelete int = 2

delete from @deleted
    delete top (@amountToDelete) from xxx.dbo.charges
    output deleted.* into @deleted
    where not exists(select 1 from xxx.dbo.chargeitems
                     where charge_id = xxx.dbo.charges.id)

while (select count(*) from @deleted) = @amountToDelete
begin
    delete from @deleted
    delete top (@amountToDelete) from xxx.dbo.charges
    output deleted.* into @deleted
    where not exists(select 1 from xxx.dbo.chargeitems
                     where charge_id = xxx.dbo.charges.id)
end

@amountToDelete表示一次要删除的记录批的大小。类型为table的另一个变量在循环运行一次时保存已删除的id。停止循环的条件是删除的记录少于应该删除的记录（这意味着要删除的最后一条记录已经被删除），这与以下条件相对应：

while (select count(*) from @deleted) = @amountToDelete

第一次删除是在循环之前（循环之外）执行的。如果一开始删除所有后退，则代码不会进入循环。

批量删除CTE中的记录

2 个答案: