3

总的来说,我相信“FULL OUTER JOIN 被认为是有害的”,转而使用这句话。

背景:

http://weblogs.sqlteam.com/jeffs/archive/2007/04/19/Full-Outer-Joins.aspx

但我确实有一个非常方便的特定情况:

鉴于:

CREATE VIEW Calcs(c1, c2, c3, fordate, ...other columns) AS
   /* Complicated set of equations, etc. */

和:

CREATE TABLE Overrides(c1, c2, c3, fordate)

我需要调整上面的视图以遵循这个逻辑:

  1. 对于其计算日期没有相应覆盖的任何Calcs行,选择计算值。
  2. 对于其计算日期与覆盖日期匹配的任何Calcs行,选择覆盖值。
  3. 对于在Calcs中没有对应行的任何Override行,选择覆盖值。

现在,通常我只会做一个三部分查询:

CREATE VIEW Calcs AS ... (current logic)

CREATE VIEW CalcsWithOverrides AS

   SELECT * FROM Calcs WHERE NOT EXISTS (match in Overrides)

   UNION ALL

   SELECT override.c1, override.c2, override.c3, (other non-overridden columns)
       FROM Calcs INNER JOIN Overrides

   UNION ALL

   SELECT *, (null placeholders for non-overridden columns) FROM Overrides WHERE
       NOT EXISTS (match in Calcs)

然而,这似乎比使用 OUTER JOIN 简单得多:

   SELECT
       COALESCE(Calcs.fordate, Overrides.fordate) AS fordate,
       -- Note that I am not using COALESCE() below on purpose: a null override should still override any calculated value
       CASE WHEN Overrides.fordate IS NULL THEN Calcs.c1 ELSE Overrides.c1 END AS C1,
       CASE WHEN Overrides.fordate IS NULL THEN Calcs.c2 ELSE Overrides.c2 END AS C2,
       CASE WHEN Overrides.fordate IS NULL THEN Calcs.c3 ELSE Overrides.c3 END AS C3,
       (....other columns from calcs that will be null for override-only rows...)
   FROM
       Calcs
       FULL OUTER JOIN Overrides ON Overrides.fordate = Calcs.fordate

那么,这是一种 OUTER JOIN 合理的情况,还是有比上述更好的选择?

4

1 回答 1

8

你想在这里完全加入,所以我认为它根本不被认为是有害的。这是获得解决方案集的最佳方式。我会在每种方式上运行查询执行播放以查看哪个最快,但我的猜测是完全连接会是。

请记住,在处理外连接时,where子句中的条件和子句中的条件之间存在明显差异。限制整个结果集,限制匹配的行。joinWherejoin

于 2009-07-29T19:28:12.953 回答