如何提高 Sql server 中 Distinct Query 的性能答案

【问题标题】：how to Improve Performance of Distinct Query In Sql server如何提高 Sql server 中 Distinct Query 的性能
【发布时间】：2018-09-21 06:11:45
【问题描述】：

我有一个查询，目前仅 324 条记录需要 10 秒，有什么方法可以提高此性能。Ps 我对 Sql 服务器非常陌生。

试过了：我已经在 SP 中使用了SET NOCOUNT ON，因为我读到它可以提高性能，甚至对每个表都使用了alias。请告诉我可以做些什么来提高它的性能

DECLARE @vRequestedBy VARCHAR(2000) = CASE WHEN @RequestedBy <> '' THEN @RequestedBy END,  
   @vJobType NVARCHAR(2000) = CASE WHEN @JobType <> '' THEN @JobType END;  
SELECT  distinct ts.JobID,   
  dbo.TSP_CAT_Category.Category,   
  ts.JobType,   
  dbo.TSP_TSR_JobStatus.JobStatus,   
  dbo.wsm_Contact.Name "ContactName",   
  ts.Created,   
  wb.Name AS BuildingName,   
  ts.Contact,   
 ts.CreatedBy,   
  ts.ContactEmail,   
  dbo.wsm_Contact.TradingAs,  
  --wsm_Contact_User.UserId "RequestedByUserId",  
  c2.Name "RequestedByUser",  
  dbo.wsm_Contact.ContactID  
  FROM   
   dbo.TSP_TSR_Job ts  
  LEFT OUTER JOIN   
   dbo.wsm_Ref_Buildings wb ON ts.BuildingID = wb.BuildingId   
  LEFT OUTER JOIN   
   dbo.wsm_Contact ON ts.TenancyID = dbo.wsm_Contact.ContactID   
  LEFT OUTER JOIN   
   dbo.TSP_TSR_JobStatus ON ts.JobStatusID = dbo.TSP_TSR_JobStatus.JobStatusID   
  LEFT OUTER JOIN   
   dbo.TSP_CAT_Category ON ts.CategoryID = dbo.TSP_CAT_Category.CategoryID  
  LEFT OUTER JOIN   
   dbo.wsm_Contact_User ON UserID = ts.ContactEmail COLLATE SQL_Latin1_General_CP1_CI_AS  
  LEFT OUTER JOIN   
   wsm_Contact c2 ON c2.ContactID = wsm_Contact_User.ContactID  

  WHERE     
   -- JobId criteria  
    (@JobID = 0 OR JobID = @JobId)  

   AND (@TenancyId = '0' OR TenancyId in (select Item from Split_fn(@TenancyID,',')))  

   AND (@TradingAs = '0' OR wsm_Contact.ContactID in (select Item from Split_fn(@TradingAs,',') ))  
   --RequestedBy    
   AND (@vRequestedBy IS NULL OR @vRequestedBy = '0' OR ts.ContactEmail in (Select distinct Email from dbo.wsm_Contact WHere Email in (select Item from Split_fn(@vRequestedBy,',')) ))  
   -- Job Category  
   AND (@CategoryId = '0' OR ts.CategoryID in (select Item from Split_fn(@CategoryId,',') ))  
   -- Contact Id (always filter on this, enough security?!)    
   AND ts.BuildingID IN (SELECT distinct b.BuildingId    
            FROM   
             wsm_ContactSite s   
            INNER JOIN   
             wsm_Contact c ON c.ContactID = s.ContactID  
            INNER JOIN   
             wsm_Ref_Buildings b ON b.SiteId = s.SiteID  
            WHERE   
             c.ContactID = @ContactUserId)  

   AND wsm_Contact.FloorID  IN     (SELECT t.FloorID   
              FROM wsm_Contact_Tenancy t  
             WHERE t.ContactID = @ContactUserId)  

   AND wsm_Contact.OCCPSTAT NOT IN ('I', 'P')  

   AND (@vJobType IS NULL OR ts.JobType in (select Item from Split_fn(@vJobType,',')))  
   AND (ts.Created between @CreatedFrom and DATEADD(DD,1,@CreatedTo))  
   ORDER BY   
    JobID

统计数据：

 SQL Server parse and compile time: 
       CPU time = 0 ms, elapsed time = 0 ms.
    SQL Server parse and compile time: 
       CPU time = 0 ms, elapsed time = 0 ms.

     SQL Server Execution Times:
       CPU time = 0 ms,  elapsed time = 0 ms.

     SQL Server Execution Times:
       CPU time = 0 ms,  elapsed time = 0 ms.
    Table 'TSP_CAT_Category'. Scan count 1, logical reads 3, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
    Table 'wsm_Contact_Tenancy'. Scan count 1, logical reads 3, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
    Table 'wsm_Contact'. Scan count 2, logical reads 3822, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
    Table 'wsm_ContactSite'. Scan count 1, logical reads 5, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
    Table 'wsm_Ref_Buildings'. Scan count 3, logical reads 2811, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
    Table 'Workfile'. Scan count 0, logical reads 0, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
    Table 'Worktable'. Scan count 2, logical reads 341364, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
    Table '#AFEC4F2F'. Scan count 2, logical reads 524444, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
    Table 'TSP_TSR_Job'. Scan count 3, logical reads 58210, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
    Table 'wsm_Contact_User'. Scan count 2, logical reads 2300, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
    Table 'TSP_TSR_JobStatus'. Scan count 2, logical reads 650, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
    Table '1159564537'. Scan count 0, logical reads 0, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
    Table 'Workfile'. Scan count 0, logical reads 0, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
    Table 'Worktable'. Scan count 0, logical reads 0, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
    Table '#BB5E01DB'. Scan count 0, logical reads 0, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
    Table '#BA69DDA2'. Scan count 0, logical reads 0, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
    Table '#B1D497A1'. Scan count 0, logical reads 0, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.
    Table '#B0E07368'. Scan count 0, logical reads 0, physical reads 0, read-ahead reads 0, lob logical reads 0, lob physical reads 0, lob read-ahead reads 0.

     SQL Server Execution Times:
       CPU time = 8391 ms,  elapsed time = 5792 ms.

     SQL Server Execution Times:
       CPU time = 8391 ms,  elapsed time = 5793 ms.

【问题讨论】：

您的查询非常大，我不确定是否有人能够从 Stack Overflow 页面调整整个内容。您可能想开始阅读诸如索引和调优之类的内容。
第一：能否提供执行计划？第二：桌子“工作台”在哪里发挥作用？是否涉及访问“工作表”的视图？您的查询至少不包含任何工作表。第三：ts.ContactEmail COLLATE SQL_Latin1_General_CP1_CI_AS 可能执行得非常糟糕，因为这可能涉及转换过程。请确保所有数据具有相同的排序规则。
@JosefBiehler brentozar.com/pastetheplan/?id=ry0ay7Gtm
也可能是参数嗅探问题。你能展示你的存储过程的声明吗？
你读过这个吗：sqlperformance.com/2012/10/t-sql-queries/sp_prefix

标签： sql-server query-performance

【解决方案1】：

为wsm_Ref_Buildings 添加聚集索引
将每个IN 变成EXISTS
添加OPTION RECOMPILE，因为您真的需要所有这些ORs
将Split_fn 正文替换为来自互联网的内容（有更干净 此类代码的示例以及SQL SERVER SPLIT_STRING 函数附带），确保Split_fn 是DETERMINISTIC（需要WITH SCHEMABINDING 选项）
尽量摆脱所有DISTINCTs
修复所有关于索引和缺失统计信息的警告（至少其中任何警告）（如TSP_CAT_Category table）
粘贴新的实际执行计划（不是估计一个）
也许有一天会将所有 PK 从字符串转换为整数

DISTINCTs 在您的IN 子查询中只会做无意义的额外排序，这对查询逻辑或输出没有影响。最上面的DISTINCT 修复（或仍然不做）设计糟糕的查询：重复行（如果有）是由必须修复的错误定义的连接产生的（例如OUTER APPLY (SELECT TOP 1...)）。

更新

Exists 示例：

WHERE ts.CategoryID in (select Item from Split_fn(@CategoryId,',') )

-->>

WHERE EXISTS (select 1 from Split_fn(@CategoryId,',') s WHERE s.Item = ts.CategoryID)

【讨论】：

我应该用什么代替 Distinct
TenancyId EXISTS (select Item from Split_fn(@TenancyID,','))) 在需要表达式的地方指定非布尔类型的表达式时出错
嗯，这不是EXISTS 的有效语法和用法。 docs.microsoft.com/en-us/sql/t-sql/language-elements/…
在答案中添加了示例@nikhiljain

【解决方案2】：

Ivan Starostin 有一组很好的建议，我不会在此重复，但我想请您考虑一下为什么您觉得首先需要使用“distinct”。

select distinct 增加查询的时间和精力，这不是“宽查询”（多列查询）的好方法，尤其是在查询涉及多个连接表的情况下。虽然将表连接在一起是常见且必要的，但不要忘记连接通常具有乘以行数的效果。因此，如果行数过多，请在完全依赖 select distinct 作为灵丹妙药之前重新考虑连接。

例如，这需要“distinct”吗？

SELECT
    ts.JobID
  , ts.Contact
  , ts.ContactEmail
  , ts.Created
  , ts.CreatedBy
  , ts.JobType
FROM dbo.TSP_TSR_Job ts
WHERE (ts.Created BETWEEN @CreatedFrom AND DATEADD(DD, 1, @CreatedTo))
AND (@JobID = 0 OR ts.JobID = @JobId)
AND (@TenancyId = '0' OR ts.TenancyId IN (
    SELECT
        Item
    FROM Split_fn(@TenancyID, ',')
    )
)

如果（我怀疑）不需要“distinct”，则将其用作子查询，然后添加剩余的表。您还可以在不添加更多行的情况下合并类别和状态的查找表，例如

SELECT
    ts.JobID
  , ts.Contact
  , ts.ContactEmail
  , ts.Created
  , ts.CreatedBy
  , ts.JobType
  , dbo.TSP_CAT_Category.Category
  , dbo.TSP_TSR_JobStatus.JobStatus
FROM dbo.TSP_TSR_Job ts
LEFT OUTER JOIN dbo.TSP_TSR_JobStatus    ON ts.JobStatusID = dbo.TSP_TSR_JobStatus.JobStatusID
LEFT OUTER JOIN dbo.TSP_CAT_Category     ON ts.CategoryID = dbo.TSP_CAT_Category.CategoryID
WHERE (ts.Created BETWEEN @CreatedFrom AND DATEADD(DD, 1, @CreatedTo))
AND (@JobID = 0 OR ts.JobID = @JobId)
AND (@TenancyId = '0' OR ts.TenancyId IN (
    SELECT
        Item
    FROM Split_fn(@TenancyID, ',')
    )
)
-- Job Category  
AND (@CategoryId = '0'
OR ts.CategoryID IN (
    SELECT
        Item
    FROM Split_fn(@CategoryId, ',')
)
)
AND (@vJobType IS NULL
OR ts.JobType IN (
    SELECT
        Item
    FROM Split_fn(@vJobType, ',')
)
)

如果不需要“distinct”，则将其作为子查询（“派生表”或“公用表表达式”），然后尝试一一添加每个附加连接（即添加连接，以及相关的 where子句过滤并为此表添加选定的列）。然后，如果您在该额外连接之后开始在结果中看到不需要的重复，您就会知道重复来自何处。您可能需要对该表使用完全不同的方法来解决此问题（例如，加入使用 row_number() 的子查询以仅获取“最新”联系人）。

【讨论】：

【解决方案3】：

您可以为 wsm_Contact 、 TSP_TSR_JobStatus 、TSP_CAT_Category 、 wsm_Contact_User 创建别名，并在查询中使用他们的别名

【讨论】：