SQL Server中Upsert的三种方式（转载）

本文介绍了SQL Server中Upsert的三种常见写法以及他们的性能比较。
SQL Server并不支持原生的Upsert语句，通常使用组合语句实现upsert功能。

假设有表table_A，各字段如下所示：

int型Id为主键。

方法1：先查询，根据查询结果判断使用insert或者update

IF EXISTS (SELECT 1 FROM table_A WHERE Id = @Id)
    BEGIN
        UPDATE dbo.table_A
        SET Value = @Value
        WHERE Id = @Id;
    END
ELSE
    BEGIN
        INSERT INTO dbo.table_A (Id, Value)
        VALUES(@Id, @Value)
    END

方法2：先更新，根据更新结果影响的条目数判断是否需要插入

UPDATE dbo.table_A
SET Value = @Value
WHERE Id = @Id;
    
IF(@@ROWCOUNT = 0)
BEGIN
    INSERT INTO dbo.table_A (Id, Value)
    VALUES(@Id, @Value)
END

方法3：使用MERGE语句，将待upsert的数据作为source，merge到目标表 

MERGE INTO table_A as T
USING (SELECT @Id AS id, @Value AS value ) AS S
ON T.Id = S.id
WHEN MATCHED THEN
    UPDATE SET T.Value = S.value
WHEN NOT MATCHED THEN
    INSERT(Id, Value) VALUES(S.id, S.value);

性能比较
在50万行数据项中随机upsert10万次

场景一：upsert数据项100%命中update

场景二：upsert数据项100%命中insert

场景三：upsert数据项Id为随机数，~50%insert，~50%update

从图中可以看出实验数据存在部分偏差，大体上这三种方法在性能上的差别非常小。对于绝大多数upsert并非关键路径的程序，方法2在可读性和执行性能上综合来讲是较优的方案。
在对性能苛求的场景，可以选用MERGE语句，以下是MERGE语句的优点：”
Faster performance. The Engine needs to parse, compile, and execute only one query instead of three (and no temporary variable to hold the key).
Neater and simpler T-SQL code (after you get proficient in MERGE).
No need for explicit BEGIN TRANSACTION/COMMIT. MERGE is a single statement and is executed in one implicit transaction.
Greater functionality. MERGE can delete rows that are not matched by source (SRC table above). For example, we can delete row 1 from A_Table because its Data column does not match Search_Col in the SRC table. There is also a way to return inserted/deleted values using the OUTPUT clause.“

关于SQL Server的Merge介绍可以查看这里

原文链接