具有子链接和父链接的递归 CTE

Question

您好，我有一个存储数据映射的表，例如

Data                  Map
id | letter | type    l | r
------------------    -----
 1 |   AA   | HEAD    5 | 1
 2 |   BB   | HEAD    2 | 1
 3 |   CC   | HEAD    6 | 2
 4 |   DD   | HEAD    3 | 2
 5 |  END-1 |  END    7 | 3
 6 |  END-2 |  END    8 | 4
 7 |  END-3 |  END
 8 |  END-4 |  END

见http://sqlfiddle.com/#!3/4eccfe/5

我想从给定的来源中找出所有 END 类型的链接，例如对于 AA，我得到 END-1、END-2、END-3；对于 BB，我得到 END-1、END-2、END-3；对于 CC，我得到 END-1、END-2、END-3；对于 DD，我得到 END-4

我已经使用递归 CTE 编写了我想要的内容：

;WITH data(id, letter, type) AS (
    SELECT '1', 'AA', 'HEAD' UNION SELECT '2', 'BB', 'HEAD' UNION SELECT '3', 'CC', 'HEAD' UNION
    SELECT '4', 'DD', 'HEAD' UNION SELECT '5', 'END-1', 'END' UNION SELECT '6', 'END-2', 'END' UNION
    SELECT '7', 'END-3', 'END' UNION SELECT '8', 'END-4', 'END'
), map (l, r) AS (
    SELECT '5', '1' UNION SELECT '2', '1' UNION
    SELECT '6', '2' UNION SELECT '3', '2' UNION
    SELECT '7', '3' UNION SELECT '8', '4'
), my_list (origin, source, target, target_type, sid, tid, level) AS (
    SELECT s.letter, s.letter, t.letter, t.type, s.id, t.id, 0
    FROM data s JOIN map ON (s.id = l OR s.id = r)
    JOIN data t ON (t.id = l OR t.id = r)
    WHERE t.id <> s.id AND s.type <> 'END'
    UNION ALL
    SELECT my_list.origin, s.letter, t.letter, t.type, s.id, t.id, level + 1
    FROM data s JOIN map ON (s.id = l OR s.id = r)
    JOIN data t ON (t.id = l OR t.id = r) JOIN my_list ON s.id = my_list.tid
    WHERE t.id <> s.id AND s.type <> 'END' AND t.id <> my_list.sid
)
SELECT * FROM my_list
WHERE origin = 'BB' AND target_type = 'END'
ORDER BY level
GO

但性能不是很好（在我的真实桌子上）。然后我意识到是连接条件中的 OR 导致了问题，然后我尝试使用 UNION

my_list (origin, source, target, target_type, sid, tid, level) AS (
    SELECT s.letter, s.letter, t.letter, t.type, s.id, t.id, 0
    FROM data s JOIN map ON s.id = l JOIN data t ON t.id = r
    WHERE s.type <> 'END'
    UNION ALL
    SELECT s.letter, s.letter, t.letter, t.type, s.id, t.id, 0
    FROM data s JOIN map ON s.id = r JOIN data t ON t.id = l
    WHERE s.type <> 'END'
    UNION ALL
    SELECT my_list.origin, s.letter, t.letter, t.type, s.id, t.id, level + 1
    FROM data s JOIN map ON (s.id = l OR s.id = r)
    JOIN data t ON (t.id = l OR t.id = r) JOIN my_list ON s.id = my_list.tid
    WHERE t.id <> s.id AND s.type <> 'END' AND t.id <> my_list.sid
)

差异很大（在我的真实桌子上，时间减半）。对于上面的例子，我得到了

Table 'Worktable'. Scan count 6, logical reads 100

对

Table 'Worktable'. Scan count 5, logical reads 75

但是当我尝试对递归部分做同样的事情时，例如

my_list (origin, source, target, target_type, sid, tid, level) AS (
    SELECT s.letter, s.letter, t.letter, t.type, s.id, t.id, 0
    FROM data s JOIN map ON s.id = l JOIN data t ON t.id = r
    WHERE s.type <> 'END'
    UNION ALL
    SELECT s.letter, s.letter, t.letter, t.type, s.id, t.id, 0
    FROM data s JOIN map ON s.id = r JOIN data t ON t.id = l
    WHERE s.type <> 'END'
    UNION ALL
    SELECT my_list.origin, s.letter, t.letter, t.type, s.id, t.id, level + 1
    FROM data s JOIN map ON s.id = l
    JOIN data t ON t.id = r JOIN my_list ON s.id = my_list.tid
    WHERE s.type <> 'END' AND t.id <> my_list.sid
    UNION ALL
    SELECT my_list.origin, s.letter, t.letter, t.type, s.id, t.id, level + 1
    FROM data s JOIN map ON s.id = r
    JOIN data t ON t.id = l JOIN my_list ON s.id = my_list.tid
    WHERE s.type <> 'END' AND t.id <> my_list.sid
)

结果变慢了（在我的真实桌子上，慢了 5 倍）。

我想知道为什么它会变慢，是否还有其他方法可以摆脱 OR 来加快查询速度？数据库是 MS SQL SERVER 2008R2

谢谢

Answer 1

我可能弄错了，但你不能推动谓词：

WHERE origin = 'BB'

在 CTE 内部。即：

;WITH my_list (origin, source, target, target_type, sid, tid, level) AS (
SELECT s.letter, s.letter, t.letter, t.type, s.id, t.id, 0
FROM data s 
    JOIN map 
        ON s.id = l 
    JOIN data t 
        ON t.id = r
WHERE s.letter = 'BB'
UNION ALL
SELECT s.letter, s.letter, t.letter, t.type, s.id, t.id, 0
FROM data s 
    JOIN map 
        ON s.id = r 
    JOIN data t 
        ON t.id = l
WHERE s.letter = 'BB'
UNION ALL
SELECT my_list.origin, s.letter, t.letter, t.type, s.id, t.id, level + 1 
    FROM data s 
    JOIN map 
        ON (s.id = l OR s.id = r)
JOIN data t 
        ON (t.id = l OR t.id = r) 
    JOIN my_list 
        ON s.id = my_list.tid
WHERE t.id <> s.id 
      AND s.type <> 'END' 
      AND t.id <> my_list.sid
)
SELECT * FROM my_list
WHERE origin = 'BB' AND target_type = 'END'
ORDER BY level

这会提高性能吗？