优化SQL Server中的表假脱机执行计划

Question

我有以下sql查询并尝试使用执行计划对其进行优化。在执行计划中，它表示估计的子树成本是36.89。有几个表轴（Eager Spool）。任何人都可以帮助我优化这个查询。提前致谢。

SELECT 
    COUNT(DISTINCT bp.P_ID) AS total, 
    COUNT(DISTINCT CASE WHEN bc.Description != 'S' THEN bp.P_ID END) AS m_count, 
    COUNT(DISTINCT CASE WHEN bc.Description = 'S' THEN bp.P_ID END) AS s_count, 
    COUNT(DISTINCT CASE WHEN bc.Description IS NULL THEN bp.P_ID END) AS n_count 
FROM 
    progress_tbl AS progress 
INNER JOIN Person_tbl AS bp ON bp.P_ID = progress.person_id 
LEFT OUTER JOIN Status_tbl AS bm ON bm.MS_ID = bp.MembershipStatusID   
LEFT OUTER JOIN Membership_tbl AS m ON m.M_ID = bp.CurrentMembershipID 
LEFT OUTER JOIN Category_tbl AS bc ON bc.MC_ID = m.MembershipCategoryID 
WHERE 
    logged_when BETWEEN '2017-01-01' AND '2017-01-31' 

Answer 1

前面已经指出的那样，似乎与你查询的一些错字/复制粘贴问题。这使我们很难弄清楚发生了什么事情。

表轴可能是CASE WHEN b.description etc...结构中正在发生的事情。 MSSQL首先创建一个包含所有结果值的（内存）表，然后通过COUNT(DISTINCT ...)运算符对其进行排序和流式处理。由于工作需要在某个地方完成，我不认为你可以做些什么。

无论如何，一些言论和乱撞：

• 我猜logged_when在progress_tbl表？
• 如果是这样，你真的需要LEFT OUTER JOIN所有其他表吗？从我能告诉他们没有被使用？
• 您正在尝试计算与条件匹配的P_ID s的数量，并且您希望将这个数字拆分为具有b.Description'S'，其他值或NULL的值。
  • 为此，您可以将总数计算为m_count，s_count和n_count之和。这样可以节省您的操作，不确定它在更大的图片中有很大的帮助，但所有的位都帮助我猜测。

事情是这样的：

;WITH counts AS (

        SELECT 
         COUNT(DISTINCT CASE WHEN b.Description != 'S' THEN b_p.P_ID END) AS m_count, 
         COUNT(DISTINCT CASE WHEN b.Description = 'S' THEN b_p.P_ID END) AS s_count, 
         COUNT(DISTINCT CASE WHEN b.Description IS NULL THEN b_p.P_ID END) AS n_count 
        FROM 
         progress_tbl AS progress 
        INNER JOIN Person_tbl AS bp ON bp.P_ID = progress.person_id 
        LEFT OUTER JOIN Status_tbl AS bm ON bm.MS_ID = bp.MembershipStatusID  -- really needed? 
        LEFT OUTER JOIN Membership_tbl AS m ON m.M_ID = bp.CurrentMembershipID -- really needed? 
        LEFT OUTER JOIN Category_tbl AS bc ON bc.MC_ID = m.MembershipCategoryID -- really needed? 
        WHERE 
         logged_when BETWEEN '2017-01-01' AND '2017-01-31' -- what table does logged_when column come from???? 

    ) 

SELECT total = m_count + s_count + n_count, 
     * 
    FROM counts 

UPDATE

请注意：使用马丁·史密斯的答案/例子代码，我才明白，total不一定总和其他领域。它可能是一个给定的P_ID显示不同的description，然后可能属于不同的类别。根据您的数据，可能因此我的答案显然是错误的。

Answer 2

以下是您可以使用的技巧。

WITH T AS 
(
SELECT DISTINCT CASE 
        WHEN bc.Description != 'S' THEN 'M' 
        WHEN bc.Description = 'S' THEN 'S' 
        WHEN bc.Description IS NULL THEN 'N' 
       END AS type, 
       bp.P_ID 
FROM progress_tbl AS progress 
     INNER JOIN Person_tbl AS bp 
     ON bp.P_ID = progress.person_id 
     LEFT OUTER JOIN Status_tbl AS bm 
     ON bm.MS_ID = bp.MembershipStatusID 
     LEFT OUTER JOIN Membership_tbl AS m 
     ON m.M_ID = bp.CurrentMembershipID 
     LEFT OUTER JOIN Category_tbl AS bc 
     ON bc.MC_ID = m.MembershipCategoryID 
WHERE logged_when BETWEEN '2017-01-01' AND '2017-01-31' 
) 
SELECT COUNT(DISTINCT P_ID) AS total, 
     COUNT(CASE WHEN type= 'M' THEN P_ID END) AS m_count, 
     COUNT(CASE WHEN type= 'S' THEN P_ID END) AS s_count, 
     COUNT(CASE WHEN type= 'N' THEN P_ID END) AS n_count 
FROM T 

我会在一个更简单的例子上演示它。

假设你现有的查询是

SELECT 
    COUNT(DISTINCT number) AS total, 
    COUNT(DISTINCT CASE WHEN name != 'S' THEN number END) AS m_count, 
    COUNT(DISTINCT CASE WHEN name = 'S' THEN number END) AS s_count, 
    COUNT(DISTINCT CASE WHEN name IS NULL THEN number END) AS n_count 
FROM master..spt_values; 

可以按如下

WITH T AS 
(
SELECT DISTINCT CASE 
        WHEN name != 'S' 
        THEN 'M' 
        WHEN name = 'S' 
        THEN 'S' 
        ELSE 'N' 
       END AS type, 
       number 
FROM master..spt_values 
) 
SELECT COUNT(DISTINCT number) AS total, 
     COUNT(CASE WHEN type= 'M' THEN number END) AS m_count, 
     COUNT(CASE WHEN type= 'S' THEN number END) AS s_count, 
     COUNT(CASE WHEN type= 'N' THEN number END) AS n_count 
FROM T 

注重写作为相当便宜成本核算及计划要简单得多重写。