我们需要创建一个保持时间有效性的表(即,对于给定的键,在本例中为下表中的 Md5,不会有重叠的周期)。用户需要能够设置日期EffectiveFrom
和EffectiveTo
日期,因此临时表没有用,因为它们似乎只允许系统生成日期。用例是将批量数据上传并设置有效日期范围,这需要应用于现有数据以确保没有时间段重叠。
表定义:
IF OBJECT_ID('dbo.IngestedData', 'U') IS NOT NULL
DROP TABLE IngestedData;
CREATE TABLE IngestedData
(
ID INT IDENTITY(1,1),
Md5 VARCHAR(15) NOT NULL,
EffectiveFrom DATE NOT NULL,
EffectiveTo DATE NOT NULL,
UpdateUser VARCHAR(50),
JsonData VARCHAR(MAX),
CONSTRAINT CK_IngestedData_Start_End CHECK (EffectiveFrom < EffectiveTo),
CONSTRAINT UK_IngestedData_Md5_Start_End UNIQUE(Md5, EffectiveFrom),
PRIMARY KEY (Id)
);
CREATE NONCLUSTERED INDEX AK_IngestedData_Md5
ON IngestedData (Md5);
CREATE NONCLUSTERED INDEX AK_IngestedData_EffectiveFrom
ON IngestedData (EffectiveFrom);
CREATE NONCLUSTERED INDEX AK_IngestedData_EffectiveTo
ON IngestedData (EffectiveTo);
我编写了一个适用于单行更新的 upsert 程序,如下所示:
上插程序:
CREATE PROCEDURE dbo.usp_UpsertIngestedDataRow
@Md5 VARCHAR(20),
@EffectiveFrom DateTime,
@EffectiveTo DateTime,
@UpdateUser VARCHAR(50),
@JsonData VARCHAR(MAX)
AS
BEGIN
SET NOCOUNT ON;
BEGIN TRY;
BEGIN TRANSACTION;
--Select the data that needs to be modified along with the action to be taken
WITH NewRow(ID, Md5, EffectiveFrom, EffectiveTo, UpdateUser, JsonData, [Action]) AS
(
SELECT NULL, @Md5, @EffectiveFrom, @EffectiveTo, @UpdateUser, @JsonData, 'I'
),
OverlappingRows(ID, Md5, EffectiveFrom, EffectiveTo, UpdateUser, JsonData) AS
(
SELECT
X.ID, X.Md5, X.EffectiveFrom, X.EffectiveTo, X.UpdateUser, X.JsonData
FROM
NewRow A
JOIN
IngestedData X ON (X.EffectiveFrom < A.EffectiveTo
AND X.EffectiveTo > A.EffectiveFrom)
AND A.Md5 = X.Md5
),
NewStartRows(ID, Md5, EffectiveFrom, EffectiveTo, UpdateUser, JsonData, [Action]) AS
(
SELECT
s.ID, s.Md5, s.EffectiveFrom,
(SELECT DATEADD(DAY, -1, MIN(EffectiveFrom))
FROM NewRow),
s.UpdateUser, s.JsonData, 'I'
FROM
OverlappingRows s
WHERE
EffectiveFrom < (SELECT MIN(EffectiveFrom) FROM NewRow)
),
NewEndRows(ID, Md5, EffectiveFrom, EffectiveTo, UpdateUser, JsonData, [Action]) AS
(
SELECT
s.ID, s.Md5,
(SELECT DATEADD(DAY, 1, MIN(EffectiveTo))
FROM NewRow),
s.EffectiveTo, s.UpdateUser, s.JsonData, 'I'
FROM
OverlappingRows s
WHERE
EffectiveTo > (SELECT MAX(EffectiveTo) FROM NewRow)
),
DeleteRows(ID, Md5, EffectiveFrom, EffectiveTo, UpdateUser, JsonData, [Action]) AS
(
SELECT
del.ID, del.Md5, del.EffectiveFrom, del.EffectiveTo,
del.UpdateUser, del.JsonData, 'D'
FROM
OverlappingRows del
INNER JOIN
NewRow n ON n.EffectiveFrom <= del.EffectiveFrom
AND n.EffectiveTo >= del.EffectiveTo
)
SELECT *
INTO #Temp
FROM
(SELECT * FROM NewRow
UNION
SELECT * FROM NewStartRows
UNION
SELECT * FROM NewEndRows
UNION
SELECT * FROM DeleteRows) AS Data;
--Delete any rows that are being replaced
DELETE FROM IngestedData WHERE ID IN (SELECT DISTINCT ID FROM #Temp)
--Insert the replacement
INSERT INTO IngestedData(Md5, EffectiveFrom, EffectiveTo, UpdateUser, JsonData)
SELECT Md5, EffectiveFrom, EffectiveTo, UpdateUser, JsonData
FROM #Temp
WHERE [Action] = 'I'
--Drop temp table
IF OBJECT_ID('tempdb.dbo.#Temp', 'U') IS NOT NULL
DROP TABLE #Temp
COMMIT;
END TRY
BEGIN CATCH
ROLLBACK;
THROW;
END CATCH
END
GO
即使在表中填充了 10,000,000 条记录,单个调用的性能也很好,调用大约需要 7 毫秒。问题是进行大量更新。通过游标对 35,000 条记录执行上述存储过程大约需要 5 分钟。
我尝试重写该过程以获取一个允许 DML 使用集合操作但在逻辑中迷失的表变量。任何人都可以帮助将上述逻辑转换为遵循此模式的基于集合的更新:
新的存储过程:
CREATE PROCEDURE usp_BulkUpsertIngestedData
@UpdateUser VARCHAR(15),
@NewRows DataIngestionRecord READONLY
AS
BEGIN
类型定义
CREATE TYPE DataIngestionRecord AS TABLE
(
Md5 VARCHAR(15) NOT NULL,
EffectiveFrom DATE NOT NULL,
EffectiveTo DATE NOT NULL,
JsonData VARCHAR(MAX)
)