由于您要删除90％的行，因此建议您将需要保留的行复制到具有相同结构ALTER TABLE ... SWITCH的新表中，然后用新表替换现有表，然后简单地删除旧表。请参阅此Microsoft Docs页面以获取语法。
一个简单的测试台，无需复制即可显示一般原理：
首先，我们将为测试创建一个数据库：
USE master;
IF (SELECT 1 FROM sys.databases d WHERE d.name = 'SwitchTest') IS NOT NULL
BEGIN
ALTER DATABASE SwitchTest SET SINGLE_USER WITH ROLLBACK IMMEDIATE;
DROP DATABASE SwitchTest;
END
CREATE DATABASE SwitchTest;
ALTER DATABASE SwitchTest SET RECOVERY FULL;
BACKUP DATABASE SwitchTest TO DISK = 'NUL:';
GO
在这里，我们创建了几个表，并带有一个触发器来将行从表“ A”移动到“ B”，从而近似设置。
USE SwitchTest;
GO
CREATE TABLE dbo.A
(
i int NOT NULL
CONSTRAINT PK_A
PRIMARY KEY CLUSTERED
IDENTITY(1,1)
, d varchar(300) NOT NULL
, rowdate datetime NOT NULL
) ON [PRIMARY]
WITH (DATA_COMPRESSION = PAGE);
CREATE TABLE dbo.B
(
i int NOT NULL
CONSTRAINT PK_B
PRIMARY KEY CLUSTERED
, d varchar(300) NOT NULL
, rowdate datetime NOT NULL
) ON [PRIMARY]
WITH (DATA_COMPRESSION = PAGE);
GO
CREATE TRIGGER t_a
ON dbo.A
AFTER INSERT, UPDATE
AS
BEGIN
SET NOCOUNT ON;
DELETE
FROM dbo.B
FROM dbo.B b
INNER JOIN deleted d ON b.i = d.i
INSERT INTO dbo.B (i, d, rowdate)
SELECT i.i
, i.d
, i.rowdate
FROM inserted i;
END
GO
在这里，我们将1,000,000行插入到“ A”中，由于触发器的缘故，这些行也将插入到“ B”中。
;WITH src AS (
SELECT i.n
FROM (VALUES (0), (1), (2), (3), (4), (5), (6), (7), (8), (9))i(n)
)
INSERT INTO dbo.A (d, rowdate)
SELECT d = CRYPT_GEN_RANDOM(300), DATEADD(SECOND, s6.n + (s5.n * 100000) + (s4.n * 10000) + (s3.n * 1000) + (s2.n * 100) + (s1.n * 10), '2017-01-01T00:00:00.000')
FROM src s1
CROSS JOIN src s2
CROSS JOIN src s3
CROSS JOIN src s4
CROSS JOIN src s5
CROSS JOIN src s6;
清除事务日志，以避免空间不足。不要在生产环境中运行此程序，因为它将事务日志数据发送到“ NUL”设备。
BACKUP LOG SwitchTest TO DISK = 'NUL:';
GO
这段代码创建了一个事务，以确保在我们迁移行时不会写入任何受影响的表：
BEGIN TRANSACTION
EXEC sys.sp_getapplock @Resource = N'TableSwitcher', @LockMode = 'Exclusive', @LockOwner = 'Transaction', @LockTimeout = '1000', @DbPrincipal = N'dbo';
BEGIN TRY
-- create a table to hold the rows we want to keep
CREATE TABLE dbo.C
(
i int NOT NULL
CONSTRAINT PK_C
PRIMARY KEY CLUSTERED
, d varchar(300) NOT NULL
, rowdate datetime NOT NULL
) ON [PRIMARY]
WITH (DATA_COMPRESSION = PAGE);
--copy the rows we want to keep into "C"
INSERT INTO dbo.C (i, d, rowdate)
SELECT b.i
, b.d
, b.rowdate
FROM dbo.B
WHERE b.rowdate >= '2017-01-11T10:00:00';
--truncate the entire "B" table
TRUNCATE TABLE dbo.B;
--"switch" table "C" into "B"
ALTER TABLE dbo.C SWITCH TO dbo.B;
--drop table "C", since we no longer need it
DROP TABLE dbo.C;
--shows the count of rows in "B" which were retained.
SELECT COUNT(1)
FROM dbo.B
WHERE b.rowdate >= '2017-01-11T10:00:00';
--look for rows in "B" that should no longer exist.
SELECT COUNT(1)
FROM dbo.B
WHERE b.rowdate < '2017-01-11T10:00:00';
--release the applock and commit the transaction
EXEC sys.sp_releaseapplock @Resource = N'TableSwitcher', @LockOwner = 'Transaction', @DbPrincipal = N'dbo';
COMMIT TRANSACTION;
END TRY
BEGIN CATCH
DECLARE @message nvarchar(1000) = ERROR_MESSAGE();
DECLARE @severity int = ERROR_SEVERITY();
DECLARE @state int = ERROR_STATE();
RAISERROR (@message, @severity, @state);
EXEC sys.sp_releaseapplock @Resource = N'TableSwitcher', @LockOwner = 'Transaction', @DbPrincipal = N'dbo';
ROLLBACK TRANSACTION;
END CATCH
GO
在sp_getapplock和sp_releaseapplock防止在同一时间运行此代码的多个实例。如果您允许通过GUI重用此代码，这将很有帮助。
（请注意，只有在每个访问资源的进程都明确实现了相同的手动资源锁定逻辑的情况下，应用程序锁定才有效-没有一种神奇的方法可以像在SQL Server自动锁定行，页面等过程中那样“锁定”表。插入/更新操作。）
现在，我们测试将行插入“ A”的过程，以确保通过触发器将行插入到“ B”中。
INSERT INTO dbo.A (d, rowdate)
VALUES ('testRow', GETDATE());
SELECT *
FROM dbo.B
WHERE B.d = 'testRow'
+ --------- + --------- + ------------------------- +
| 我 d | 行日期|
+ --------- + --------- + ------------------------- +
| 1000001 | testRow | 2018-04-13 03：49：53.343 |
+ --------- + --------- + ------------------------- +