Question

我在查询具有父/子关系的行的表时遇到问题。在编写简化示例时，我意识到stackexchange模式非常相似。

因此想象一下我通过stackexchange数据资源管理器查询stackoverflow posts表。我正在尝试获取所有帖子及其相关答案的子集。

有关示例查询，请参阅http://data.stackexchange.com/stackoverflow/query/121981/a-subset-of-questions-and-associated-answers。

帖子的子集在视图中定义，该视图具有相当复杂和昂贵的查询计划。在下面的例子中，它被简化为简单地选择前两行。

第一种方式，使用联合：

with ExpensiveView as (select top 2 ID from Posts order by ID)

select Posts.*
from ExpensiveView
left outer join Posts
  ON ExpensiveView.Id = Posts.Id 

union all

select Posts.*
from ExpensiveView
left outer join Posts
  ON ExpensiveView.Id = Posts.ParentId

我非常想避免这种情况，因为ExpensiveView被评估了两次。对于上面的简化版本显然不是问题，但会导致更复杂的问题。

第二种方式，使用带条件连接子句的单个选择：

with ExpensiveView as (select top 2 ID from Posts order by ID)

select Posts.*
from ExpensiveView
left outer join Posts
  ON ExpensiveView.Id = Posts.Id or ExpensiveView.Id = Posts.ParentId

这可以避免ExpensiveView被评估两次，但会导致一个非常大的聚簇索引扫描。它似乎是在ExpensiveView中扫描每个ID的整个索引（所以2 * 14977623 =〜3,000万行）。这很慢。

两个问题

为什么第二个查询中的条件连接会导致如此大的索引扫描？

如果不对ExpensiveView进行多次评估，我有什么办法可以获得我想要的结果吗？

Answer 1

试试这个

with
ExpensiveView as (select top 2 ID from Posts order by ID),
CTE_Posts as (
    select *, NP.Id as New_Post_ID
    from Posts as P
        outer apply (select P.Id union all select P.ParentId) as NP
)
select
    P.*
from ExpensiveView as E
    left outer join CTE_Posts as P on P.New_Post_ID = E.ID

使用条件连接语句时执行多个完整索引扫描

1 个答案: