在具有复合PK的桌子上两次使用的地方是否足够？

Question

我有一个问题

SELECT * 
FROM table1
WHERE
documentId in
(
   --select items from a second table where a third column happens to be null
   select documentId from table2 t2 inner join table1 t1
   on t1.documentId = t2.documentId and t1.itemId = t2.ItemId
   WHERE t1.someOtherColumn is null
)
and itemId in
(
   --similar query as above, just selecting itemId now
   select itemId from table2 t2 inner join table1 t1
   on t1.documentId = t2.documentId and t1.itemId = t2.ItemId
   WHERE t1.someOtherColumn is null
)
order by 1

鉴于table1具有复合PK = documentId + itemId,这是否足以仅从table1中选择唯一的值？我担心在documentId存在且itemsId存在的情况下可能会出现这种情况,但由于它们没有一起查看,因此可以进行不正确的选择.

例如,

我们假设存在一个值

documentId = 1,itemId = 1.

我们假设没有复合键

documentId = 1,itemId = 1.

我不想要复合键

[documentId = 1和itemId = 3]

被包括.

我也不想要复合键

[documentId = 2和itemId = 1]

被包括.

如果稍后添加了复合键(现在不存在)

[documentId = 1和itemId = 1]

然后应该包括在内.

Answer 1

这样做无需使用连接或两个INs

SELECT *
FROM table1
WHERE exists
(
    SELECT 1
    FROM table2
    WHERE someOtherColumn is null
        AND table1.documentId = table2.documentId
        AND table1.itemId = table2.ItemId
)
ORDER BY 1

Answer 2

你是正确的,假设它可能导致错误的结果.你可以用这个:

SELECT table1.* 
FROM table1
WHERE someOtherColumn IS NULL
  AND (documentId, itemId ) IN
   (
   SELECT documentId, itemId 
   FROM table2 t2                  --- no need for table1 again here 
   )
ORDER BY 1

更新:我认为上述内容并不适用于SQL-Server,仅适用于Postgres(和Oracle？).无论如何,这应该适用于大多数系统:

SELECT table1.* 
FROM table1
WHERE someOtherColumn IS NULL
  AND EXISTS
   (
   SELECT documentId, itemId FROM table2 t2     
   WHERE table2.documentId = table1.documentId
     AND table2.itemId = table1.itemId
   )
ORDER BY 1