从 Azure SQL DW 更新？

Question

我在尝试执行 UPDATE FROM 查询的 Azure SQL DW 中收到错误消息。错误是“UPDATE 和 DELETE 语句中的 FROM 子句不能包含子查询源或连接”

这只是特定于 SQL DW 吗？否则，我认为此查询没有任何问题。如果是 SQL DW 的限制，还有什么办法？

-- Permanent fact table with 5 billion rows
CREATE TABLE FactTable (Id1 INT, Id2 INT, EmailAddress NVARCHAR(100), Value1 INT)
WITH (DISTRIBUTION = HASH(EmailAddress));

-- Staging fact table with 10 million rows    
CREATE TABLE StageTable (Id1 INT, Id2 INT, EmailAddress NVARCHAR(100), Value1 INT)
WITH (DISTRIBUTION = HASH(EmailAddress), HEAP);

-- Add a secondary index that should help with joining to StageTable
CREATE NONCLUSTERED INDEX ix ON FactTable (Id1, Id2);

UPDATE fact
SET
   Value1 = CASE WHEN stage.Value1 > fact.Value1 THEN stage.Value1 ELSE fact.Value1 END
FROM FactTable AS fact
INNER JOIN StageTable AS stage ON fact.Id1 = stage.Id1 AND fact.Id2 = stage.Id2

Answer 1

根据the documentation，Azure SQL 数据仓库确实支持UPDATE，但不支持FROM 子句中的ANSI 联接。您可以使用 CTAS 来解决。简单的两表更新：

UPDATE dbo.FactTable
SET
   Value1 = CASE WHEN stage.Value1 > dbo.FactTable.Value1 THEN stage.Value1 ELSE dbo.FactTable.Value1 END
FROM dbo.StageTable AS stage
WHERE dbo.FactTable.Id1 = stage.Id1 
  AND dbo.FactTable.Id2 = stage.Id2;

更复杂的 CTAS 示例，从main UPDATE documentation page 批发复制：

-- Create an interim table
CREATE TABLE CTAS_acs
WITH (DISTRIBUTION = ROUND_ROBIN)
AS
SELECT  ISNULL(CAST([EnglishProductCategoryName] AS NVARCHAR(50)),0)    AS [EnglishProductCategoryName]
,       ISNULL(CAST([CalendarYear] AS SMALLINT),0)                      AS [CalendarYear]
,       ISNULL(CAST(SUM([SalesAmount]) AS MONEY),0)                     AS [TotalSalesAmount]
FROM    [dbo].[FactInternetSales]       AS s
JOIN    [dbo].[DimDate]                 AS d    ON s.[OrderDateKey]             = d.[DateKey]
JOIN    [dbo].[DimProduct]              AS p    ON s.[ProductKey]               = p.[ProductKey]
JOIN    [dbo].[DimProductSubCategory]   AS u    ON p.[ProductSubcategoryKey]    = u.[ProductSubcategoryKey]
JOIN    [dbo].[DimProductCategory]      AS c    ON u.[ProductCategoryKey]       = c.[ProductCategoryKey]
WHERE   [CalendarYear] = 2004
GROUP BY
        [EnglishProductCategoryName]
,       [CalendarYear]
;

-- Use an implicit join to perform the update
UPDATE  AnnualCategorySales
SET     AnnualCategorySales.TotalSalesAmount = CTAS_ACS.TotalSalesAmount
FROM    CTAS_acs
WHERE   CTAS_acs.[EnglishProductCategoryName] = AnnualCategorySales.[EnglishProductCategoryName]
AND     CTAS_acs.[CalendarYear]               = AnnualCategorySales.[CalendarYear]
;

--Drop the interim table
DROP TABLE CTAS_acs
;

Answer 2

我发现使用 ASDW（和 APS/PDW）是一种很好的做法，可以避免像瘟疫一样的批量更新。

这是一个纯粹的 CTAS 替代方案，在您更新大量行的情况下会更快。

它假设 id1 是一个相对较好的分布键，并且暂存行的数量少于事实行，使得复制可行。这种策略应该消除节点之间的数据移动。

如果您有一个非常大的临时表，则在每个表中创建一个 id1 和 id2 组合的代理列，然后按该列的哈希分布这两个表，将提供更好的性能。

create  table FactTable (
        id1 int,
        id2 int,
        value1 int)
with    (distribution = hash(id1));

create  table StageTable (
        id1 int,
        id2 int,
        value1 int)
with    (distribution = replicate);

create  table UpdatedFact 
with    (distribution = hash(id1)) 
as
select  f.id1,
        f.id2,
        case when s.id1 is not null and s.value1 > f.value1 
            then s.value1
            else f.value1
            end as value1
from    FactTable f
        left outer join StageTable s
        on s.id1 = f.id1
        and s.id2 = f.id2

truncate table FactTable;
alter table UpdatedFact switch to FactTable;
drop table UpdatedFact;

Answer 3

简化您的尝试会奏效。只需删除连接并从另一个表更新一个表。

update FactTable
set this = that
from StageTable s where s.something = FactTable.something

这是否是最好的方法取决于您的情况，但它会执行而不会引发错误。