查询优化：在我的情况下哪个更好 - CURSOR 或基于集合的方法？

Question

查询优化：在我的情况下哪个更好 - CURSOR 或基于集合的方法？

Leg*_*end 1 sql-server-2008 sql-server query

首先，我可以描述表的架构，但我的表本身大约是 80GB，所以我希望一些专家的眼睛可以看到可以执行的任何明显的优化。我正在进行一些基于活动的分析，以了解来自内部虚拟游戏环境的一组用户。

我有两个基表：

EventTable: Moderately big at 10GB
Activity_ID UserName ActivityStart ActivityEnd Badge

ActivityTable: Super big at 100GB
UserName ActivityTime Game1Points Game2Points

Run Code Online (Sandbox Code Playgroud)

简而言之，我正在尝试以下操作：

- For each activity in the EventTable:
   - Compute the median of Game1Points, Game2Points 
     for two cases: for all entries in the ActivityTable 
     that were present since 1 hour before the activity 
     and for all entries stored during the activity itself

Run Code Online (Sandbox Code Playgroud)

因此，我期待的决赛桌是这样的：

ResultTable:
Activity_ID Badge Game1_Before Game1_During Game2_Before Game2_During

Run Code Online (Sandbox Code Playgroud)

这显然看起来像是CURSORs的工作，所以我首先使用这种方法编写了我的查询。它正在运行，我可以看到进度，但在阅读了这么多恐怖故事之后，我决定用基于集合的方法重写我的 200 行基于 CURSOR 的方法。这非常适用于小桌子。但是，对于我这种大小的桌子，它似乎停滞不前（或者至少我不确定应该运行多长时间）。我运行Display Estimated Plan并添加了相关索引。现在我的问题是我什至不确定这个查询需要多长时间才能运行（不像CURSOR我可以打印一些消息并查看进度的方法）。

有人可以告诉我是否有更好的方法来做到这一点？我当前的查询如下。虽然查询本身看起来很长，但结构非常简单：首先，我定义了一个 CTE，它获取与特定事件关联的所有点。使用这个 CTE，我计算了两个单独的中位数。

WITH ST AS
(
    -- Fetching the required data: Get entries that exist 
    -- since 1 hour before the activity until the activity ended
    SELECT    ROW_NUMBER() OVER(ORDER BY Activity_ID) AS ID
            , Activity_ID
            , Badge
            , (CASE WHEN
                    Y.ActivityTime BETWEEN DATEADD(HOUR, -1, dbo.ROUNDTIME(X.ActivityStart) AND dbo.ROUNDTIME(X.ActivityStart)
                    THEN 'Game1ActivityBefore'
                    WHEN
                    Y.ActivityTime BETWEEN dbo.ROUNDTIME(X.ActivityStart) AND dbo.ROUNDTIME(X.ActivityFinish)
                    THEN 'Game1ActivityDuring'
              END) Game1Type
            , (CASE WHEN
                    Y.ActivityTime BETWEEN DATEADD(HOUR, -1, dbo.ROUNDTIME(X.ActivityStart)) AND dbo.ROUNDTIME(X.ActivityStart)
                    THEN 'Game2ActivityBefore'
                    WHEN
                    Y.ActivityTime BETWEEN dbo.ROUNDTIME(X.ActivityStart) AND dbo.ROUNDTIME(X.ActivityFinish)
                    THEN 'Game2ActivityDuring'
              END) Game2Type
            , CAST(Game1Points AS bigint) Game1Points
            , CAST(Game2Poins AS bigint) Game2Points
    FROM
    dbo.EventTable X
    INNER JOIN dbo.ActivityTable Y
    ON X.UserName = Y.UserName
    WHERE
    ActivityTime BETWEEN DATEADD(HOUR, -1, dbo.ROUNDTIME(X.ActivityStart))
                         AND dbo.ROUNDTIME(X.ActivityFinish)
    AND 
    (
            Badge LIKE 'GREEN%'
        OR Badge LIKE 'RED%'
    )
)
-- Eliminating any NULL values that resulted from the median computation step
SELECT Activity_ID
        , Badge
        , MAX(CAST([Game1ActivityBefore] AS DECIMAL)) AS 'Game1_Before'
        , MAX(CAST([Game1ActivityDuring] AS DECIMAL)) AS 'Game1_During'
        , MAX(CAST([Game2ActivityBefore] AS DECIMAL)) AS 'Game2_Before'
        , MAX(CAST([Game2ActivityDuring] AS DECIMAL)) AS 'Game2_During'
FROM
(
    -- Median computation for the two columns - Aggregation Step
    SELECT Activity_ID
            , Badge
            , Game1Type
            , Game2Type
            , AVG(Game1Points) Game1Median
            , AVG(Game2Points) Game2Median
    FROM
    (
            -- Median computation for the two columns - Inner step
        SELECT Activity_ID
                , Badge
                , Game1Type
                , Game2Type
                , Game1Points
                , ROW_NUMBER() OVER (
                    PARTITION BY Activity_ID, Badge, Game1Type
                    ORDER BY Game1Points ASC, ID ASC) AS Game1RowAsc
                , ROW_NUMBER() OVER (
                    PARTITION BY Activity_ID, Badge, Game1Type
                    ORDER BY Game1Points DESC, ID DESC) AS Game1RowDesc
                , Game2Points
                , ROW_NUMBER() OVER (
                    PARTITION BY Activity_ID, Badge, Game2Type
                    ORDER BY Game2Points ASC, ID ASC) AS Game2RowAsc
                , ROW_NUMBER() OVER (
                    PARTITION BY Activity_ID, Badge, Game2Type
                    ORDER BY Game2Points DESC, ID DESC) AS Game2RowDesc
        FROM ST TS
    ) X
    WHERE Game1RowAsc IN (Game1RowDesc, Game1RowDesc - 1, Game1RowDesc + 1)
    OR Game2RowAsc IN (Game2RowDesc, Game2RowDesc - 1, Game2RowDesc + 1)
    GROUP BY Activity_ID, Badge, Game1Type, Game2Type
) Y
PIVOT
(
    MAX(Game1Median)
    FOR Game1Type IN ([Game1ActivityBefore], [Game1ActivityDuring])
) AS PivotTable
PIVOT
(
    MAX(Game2Median)
    FOR Game2Type IN ([Game2ActivityBefore], [Game2ActivityDuring])
) AS PivotTable2
GROUP BY Activity_ID, Badge

Run Code Online (Sandbox Code Playgroud)

更新：在建议之后。取 1

CREATE TABLE #ST(ID INT
                    , Activity_ID INT
                    , Game1Type CHAR(13)
                    , Game2Type CHAR(13)
                    , Game1Points DECIMAL
                    , Game2Points DECIMAL
                )

INSERT INTO #ST
SELECT    ROW_NUMBER() OVER(ORDER BY Activity_ID) AS ID
        , Activity_ID
        , Badge
        , (CASE WHEN
                Y.ActivityTime BETWEEN DATEADD(HOUR, -1, dbo.ROUNDTIME(X.ActivityStart) AND dbo.ROUNDTIME(X.ActivityStart)
                THEN 'Game1ActivityBefore'
                WHEN
                Y.ActivityTime BETWEEN dbo.ROUNDTIME(X.ActivityStart) AND dbo.ROUNDTIME(X.ActivityFinish)
                THEN 'Game1ActivityDuring'
          END) Game1Type
        , (CASE WHEN
                Y.ActivityTime BETWEEN DATEADD(HOUR, -1, dbo.ROUNDTIME(X.ActivityStart)) AND dbo.ROUNDTIME(X.ActivityStart)
                THEN 'Game2ActivityBefore'
                WHEN
                Y.ActivityTime BETWEEN dbo.ROUNDTIME(X.ActivityStart) AND dbo.ROUNDTIME(X.ActivityFinish)
                THEN 'Game2ActivityDuring'
          END) Game2Type
        , CAST(Game1Points AS bigint) Game1Points
        , CAST(Game2Poins AS bigint) Game2Points
FROM
dbo.EventTable X
INNER JOIN dbo.ActivityTable Y
ON X.UserName = Y.UserName
WHERE
ActivityTime BETWEEN DATEADD(HOUR, -1, dbo.ROUNDTIME(X.ActivityStart))
                     AND dbo.ROUNDTIME(X.ActivityFinish)
AND 
(
        Badge LIKE 'GREEN%'
    OR Badge LIKE 'RED%'
)

CREATE TABLE #INTERMEDIATE (Activity_ID INT
                            , Badge VARCHAR(255)
                            , Game1Type CHAR(13)
                            , Game2Type CHAR(13)
                            , Game1Points DECIMAL
                            , Game1RowAsc INT
                            , Game1RowDesc INT
                            , Game2Points DECIMAL
                            , Game2RowAsc INT
                            , Game2RowDesc INT 
                            )

INSERT INTO #INTERMEDIATE
SELECT Activity_ID
        , Badge
        , Game1Type
        , Game2Type
        , Game1Points
        , ROW_NUMBER() OVER (
            PARTITION BY Activity_ID, Badge, Game1Type
            ORDER BY Game1Points ASC, ID ASC) AS Game1RowAsc
        , ROW_NUMBER() OVER (
            PARTITION BY Activity_ID, Badge, Game1Type
            ORDER BY Game1Points DESC, ID DESC) AS Game1RowDesc
        , Game2Points
        , ROW_NUMBER() OVER (
            PARTITION BY Activity_ID, Badge, Game2Type
            ORDER BY Game2Points ASC, ID ASC) AS Game2RowAsc
        , ROW_NUMBER() OVER (
            PARTITION BY Activity_ID, Badge, Game2Type
            ORDER BY Game2Points DESC, ID DESC) AS Game2RowDesc
FROM #ST

CREATE CLUSTERED INDEX [TT1] ON #INTERMEDIATE (Acitivity_ID
                                             , Badge)

CREATE NONCLUSTERED INDEX [TT2] ON #INTERMEDIATE (Game1RowAsc
                                                , Game1RowDesc
                                                , Game2RowAsc
                                                , Game2RowDesc)

SELECT Activity_ID
        , Badge
        , MAX(CAST([Game1ActivityBefore] AS DECIMAL)) AS 'Game1_Before'
        , MAX(CAST([Game1ActivityDuring] AS DECIMAL)) AS 'Game1_During'
        , MAX(CAST([Game2ActivityBefore] AS DECIMAL)) AS 'Game2_Before'
        , MAX(CAST([Game2ActivityDuring] AS DECIMAL)) AS 'Game2_During'
FROM
(
    SELECT Activity_ID
            , Badge
            , Game1Type
            , Game2Type
            , AVG(Game1Points) Game1Median
            , AVG(Game2Points) Game2Median
    FROM #ST
    WHERE Game1RowAsc IN (Game1RowDesc, Game1RowDesc - 1, Game1RowDesc + 1)
    OR Game2RowAsc IN (Game2RowDesc, Game2RowDesc - 1, Game2RowDesc + 1)
    GROUP BY Activity_ID, Badge, Game1Type, Game2Type
) Y
PIVOT
(
    MAX(Game1Median)
    FOR Game1Type IN ([Game1ActivityBefore], [Game1ActivityDuring])
) AS PivotTable
PIVOT
(
    MAX(Game2Median)
    FOR Game2Type IN ([Game2ActivityBefore], [Game2ActivityDuring])
) AS PivotTable2
GROUP BY Activity_ID, Badge

DROP TABLE #INTERMEDIATE
DROP TABLE #ST

Run Code Online (Sandbox Code Playgroud)

更新2：：执行计划

可能是该Display Execution Plan选项不显示正在使用的索引？

Answer 1

JNK*_*JNK 7

1 - 不要使用 CTE 来提高性能。CTE 基本上是一次性视图。它们绝对没有表现出任何性能优势。如果性能很重要并且您有足够的数据，则索引#temp表可能会更好地工作。

2-绝对使用基于集合的方法。我想不出任何CURSOR比基于集合的查询更快的场景。

我不会完全重写您的查询，但我确实看到了一些应该会产生巨大差异的事情：

将您的初始选择（最外层 CTE）和您的子查询放入一个#temp表中并为其编制索引。 现在您的过滤器是动态创建的，不可搜索！

您正在对ROW_NUMBER输出进行 4 次过滤，因此这是对每一行的表扫描。

归档时间：	13 年，11 月前
查看次数：	631 次
最近记录：	13 年，7 月前