尝试避免 T-SQL 中的相关子查询 for Netezza re-write - MAX(COALESCE()) 在聚合查询中不起作用

Question

看看这个 T-SQL 查询。它有一个嵌套查询。我试图以相同的方式使其在没有嵌套查询的情况下工作。我这样做的原因是因为我想在 Netezza 中编写类似的东西，但是 Netezza 的查询引擎不允许您在嵌套查询中引用外部查询的表。请注意，在我的虚拟数据中，我硬编码了它是树枝还是树叶，但在我的第一个查询中，我计算它是树枝还是树叶，结果是一样的。

在我的问题的最后，我有我的虚拟数据和一个简单的查询，所以你可以看到那里有什么。

SELECT
    DISTINCT F1.[PATH],  
    F1.BRANCH_OR_LEAF,  
    (
        SELECT
            COUNT(DISTINCT [FILE].ID)
        FROM FOLDER F2
            JOIN [FILE] ON F2.ID = [FILE].FOLDER_ID
        WHERE   
            F2.[PATH] LIKE (F1.[PATH] + '%')

    ) file_count,
    (
        SELECT
            CASE WHEN 
                MAX(COALESCE(F2.ID, -1)) != COALESCE(F1.ID, -1)
            THEN
                'B'
            ELSE
                'L'
            END AS BRANCH_OR_LEAF           
        FROM FOLDER F2
            JOIN [FILE] ON F2.ID = [FILE].FOLDER_ID
        WHERE   
            F2.[PATH] LIKE (F1.[PATH] + '%')

    ) branch_or_leaf
FROM
    [FOLDER] F1
ORDER BY 
    F1.[PATH]

查询聚合 MAX（COALESCE 不起作用：

SELECT
    F1.PATH AS FOLDER_PATH,
    COUNT(DISTINCT F.ID) AS FILE_COUNT,
    CASE
        WHEN COUNT(DISTINCT F.ID) > 0 THEN 'A'
        ELSE 'H'
    END,
    (
        LEN(F1.PATH) - LEN(REPLACE(F1.PATH, '/', '')) - 1
    ) AS FOLDER_LEVEL
    --,
    --CASE
    --  WHEN MAX(coalesce(F1.ID,'-1')) != coalesce(F2.ID,'-1') THEN 'B'
    --  ELSE 'L'
    --END AS BRANCH_OR_LEAF
FROM
    [FOLDER] F1
    LEFT JOIN [FOLDER] F2 ON F2.PATH LIKE (F1.PATH + '%')
    JOIN [FILE] F ON F2.ID = F.FOLDER_ID
GROUP BY
    F1.PATH

当我取消评论时，我明白了。

列 'FOLDER.ID' 在 select 列表中无效，因为它未包含在聚合函数或 GROUP BY 子句中。

这是我当前的虚拟数据的样子。

SELECT F1.ID, F1.[PATH], F1.BRANCH_OR_LEAF, F.ID, F.NAME
FROM [FOLDER] F1
JOIN [dbo].[FILE] F ON F1.ID = F.FOLDER_ID

Answer 1

我根据您的虚拟数据创建了一个 table，但我什至没有查看您的查询。我反而写了我自己的。非常非常接近，但您的虚拟数据似乎缺少路径为“/AA/BB/CC/”

的一行

您的查询应如下所示。我假设您会将 CTE 切换为子查询或临时查询 table。但我将其作为可读性的 CTE。

WITH CTE
AS
(
    SELECT F1.ID FolderID, F1.[PATH], F1.BRANCH_OR_LEAF, F.ID FileID, F.NAME
    FROM [FOLDER] F1
    JOIN [dbo].[FILE] F ON F1.ID = F.FOLDER_ID
)

SELECT A.[Path],A.Branch_or_Leaf,COUNT(DISTINCT B.FileID) file_Count
FROM CTE A
INNER JOIN  CTE B
ON A.[Path] = LEFT(B.[Path],LEN(A.[Path]))
GROUP BY A.Path,A.Branch_or_Leaf,B.Branch_Or_Leaf
ORDER BY A.[Path]

结果：

Path                      Branch_or_Leaf file_Count
------------------------- -------------- -----------
/AA/                      B              5
/AA/                      B              8
/AA/BB/                   B              4
/AA/BB/                   B              8
/AA/BB/CC/CC.1/           L              2
/AA/BB/CC/CC.2/           B              1
/AA/BB/CC/CC.2/           B              4
/AA/BB/CC/CC.2/CC.22/     L              3
/AA/BB/CC/CC.2/CC.23/     L              4
/AA/BB/CC/DD/             L              2

Answer 2

这应该可以让我在 Netezza 中完成我想要的。在实际情况下，您无法确认 LEN 函数是否足够，但由于我们已经在检查一个字符串是否包含另一个字符串，所以这应该没问题。我会使用 CHECKSUM_AGG 函数，但 Netezza 不使用它。所以这将不得不做。

create table #branch_or_leaf
(
    [folder_id] int,
    [path] nvarchar(50),  -- for reference purposes only
    [branch_or_leaf] [nvarchar](50)
)

INSERT INTO #branch_or_leaf
SELECT
    DISTINCT F1.ID, 
    F1.PATH,
    CASE 
        WHEN SUM(LEN(F2.PATH)) = LEN(F1.PATH) THEN 'L'
        ELSE 'B'
    END
FROM
    [FOLDER] F1
    LEFT JOIN [FOLDER] F2 ON F2.PATH LIKE (F1.PATH + '%') 
GROUP BY 
    F1.ID, F1.PATH

SELECT * FROM #branch_or_leaf

SELECT
    DISTINCT F1.PATH AS FOLDER_PATH,
    COUNT(DISTINCT F.ID) AS FILE_COUNT,
    CASE
        WHEN COUNT(DISTINCT F.ID) > 0 THEN 'A'
        ELSE 'H'
    END,
    (
        LEN(F1.PATH) - LEN(REPLACE(F1.PATH, '/', '')) - 1
    ) AS FOLDER_LEVEL,
    BL.branch_or_leaf AS branch_or_leaf
FROM
    [FOLDER] F1
    LEFT JOIN [FOLDER] F2 ON F2.PATH LIKE (F1.PATH + '%')
    JOIN [FILE] F ON F2.ID = F.FOLDER_ID
    JOIN #branch_or_leaf BL ON F1.ID = BL.folder_id
GROUP BY
    F1.PATH, BL.branch_or_leaf
ORDER BY
    F1.PATH

DROP TABLE #branch_or_leaf

编辑：

类似查询，但现在我们在 FROM 子句中有了嵌套查询。可能在 Netezza 中是合法的。

SELECT
    DISTINCT F1.PATH AS FOLDER_PATH,
    COUNT(DISTINCT F.ID) AS FILE_COUNT,
    CASE
        WHEN COUNT(DISTINCT F.ID) > 0 THEN 'A'
        ELSE 'H'
    END,
    (
        LEN(F1.PATH) - LEN(REPLACE(F1.PATH, '/', '')) - 1
    ) AS FOLDER_LEVEL,
    BL.BL AS branch_or_leaf
FROM
    [FOLDER] F1
    LEFT JOIN [FOLDER] F2 ON F2.PATH LIKE (F1.PATH + '%')
    JOIN [FILE] F ON F2.ID = F.FOLDER_ID
    JOIN 
    (
        SELECT
            DISTINCT F1.ID, 
            CASE 
                WHEN SUM(LEN(F2.PATH)) = LEN(F1.PATH) THEN 'L'
                ELSE 'B'
            END AS BL
        FROM
            [FOLDER] F1
            LEFT JOIN [FOLDER] F2 ON F2.PATH LIKE (F1.PATH + '%') 
        GROUP BY 
            F1.ID, F1.PATH      
    ) AS BL ON F1.ID = BL.ID
GROUP BY
    F1.PATH, BL.BL
ORDER BY
    F1.PATH

尝试避免 T-SQL 中的相关子查询 for Netezza re-write - MAX(COALESCE()) 在聚合查询中不起作用

Trying to avoid correlated subquery in T-SQL for Netezza re-write - MAX(COALESCE()) not working in aggregate query

sql-server

tsql

max

aggregate

coalesce

编辑：