如何优化查询，使手动工作可以动态和优化的方式进行

Question

我有以下 table 喜欢：SQL fiddle

我可以通过 XML 获得此输出，但我不确定如何为更多用户（大约 200 万用户）正确获得低于输出的值。

稍后我想通过每个 id 的计数获得前 3 个名称，因此 RANK 或 OrderBy 子句将进入 SQL 并且不确定当数据量很大时需要多少次迭代用户。

我试过的工作代码：

-----------SQL Raw Table Creation------------------------
CREATE TABLE tb
(
    Id INT,
    Name VARCHAR(50) NOT NULL
);

INSERT INTO tb (Id, Name) VALUES (1,  'aa');
INSERT INTO tb (Id, Name) VALUES (1,  'aa');
INSERT INTO tb (Id, Name) VALUES (1,  'aa');
INSERT INTO tb (Id, Name) VALUES (1,  'aa');
INSERT INTO tb (Id, Name) VALUES (1,  'aa');
INSERT INTO tb (Id, Name) VALUES (1,  'bb');
INSERT INTO tb (Id, Name) VALUES (1,  'cc');
INSERT INTO tb (Id, Name) VALUES (1,  'cc');
INSERT INTO tb (Id, Name) VALUES (1,  'dd');
INSERT INTO tb (Id, Name) VALUES (1,  'dd');
INSERT INTO tb (Id, Name) VALUES (1,  'dd');

INSERT INTO tb (Id, Name) VALUES (2,  'aa');
INSERT INTO tb (Id, Name) VALUES (2,  'bb');
INSERT INTO tb (Id, Name) VALUES (2,  'bb');
INSERT INTO tb (Id, Name) VALUES (2,  'ee');
INSERT INTO tb (Id, Name) VALUES (2,  'ee');
INSERT INTO tb (Id, Name) VALUES (2,  'ee');
INSERT INTO tb (Id, Name) VALUES (2,  'ee');

INSERT INTO tb (Id, Name) VALUES (3,  'aa');
INSERT INTO tb (Id, Name) VALUES (3,  'bb');
INSERT INTO tb (Id, Name) VALUES (3,  'cc');
INSERT INTO tb (Id, Name) VALUES (3,  'dd');
INSERT INTO tb (Id, Name) VALUES (3,  'dd');
INSERT INTO tb (Id, Name) VALUES (3,  'dd');




-----------------Want to RANK or get only top 3 rows for each Id when group by Name--------------
select f.* into #t1
from(
    select f.* 
    from(
        select f.* 
        from (
        
        select top 3 id,name,count(name) as total 
        from tb 
        where id = 1
        group by id,name 
        order by id,total desc
        )f
        Union 
    
        select top 3 id,name,count(name) as total 
        from tb 
        where id = 2
        group by id,name 
        order by id,total desc
    )f
    Union 

    select top 3 id,name,count(name) as total 
    from tb 
    where id = 3
    group by id,name 
    order by id,total desc

    ) f

/* Output is moved in temp table #t1 which looks like 
id  name    total
1   aa  5
1   cc  2
1   dd  3
2   aa  1
2   bb  2
2   ee  4
3   bb  1
3   cc  1
3   dd  3 
 */

---------Final Joining for each Top3Names and RespectiveTotal -----

select a.id as ID, a.listStr as Top3Names , b.Total as RespectiveTotal
from
    (
    SELECT  id,STUFF((SELECT  ',' + name
                FROM #t1 EE
                WHERE  EE.id=E.id
           FOR XML PATH('')), 1, 1, '') AS listStr
    FROM  #t1 E
    GROUP BY E.id
    )a
    left Join
    (
    SELECT  id,STUFF((SELECT  ',' + cast(total as Varchar)
                FROM #t1 EE
                WHERE  EE.id=E.id
           FOR XML PATH('')), 1, 1, '') AS Total
    FROM  #t1 E
    GROUP BY E.id
    )b 
    on a.id=b.id

输出：

ID  Top3Names   RespectiveTotal
1   aa,cc,dd    5,2,3
2   aa,bb,ee    1,2,4
3   bb,cc,dd    1,1,3

这里我为每个 ID 使用 UNION，这是不正确的做法。我想要一个优化的方式。我还使用临时 table 来存储我的结果。这是一个好方法吗？让我知道任何正确的解决方案或替代方案，以便我可以在更大的设备上进行测试。

Answer 1

在我的 SQL 服务器机器上，对于给定的示例数据，您的查询统计信息如下所示：

逻辑读取总数：13
总 CPU 时间：00:00:00.007

如果您使用的是 SQL SERVER 2017+，您可以使用 STRING_AGG 功能：

SELECT
    id
    , STRING_AGG(name,',') WITHIN GROUP (order by name asc) Top3Names  
    , STRING_AGG(countx,',')  WITHIN GROUP (order by name asc) RespectiveTotal
FROM (
    SELECT 
      id 
      ,  name 
      , count(*) countx 
      , ROW_NUMBER() over (partition by id order by count(*)  desc) rownumber
    FROM tb
    GROUP BY name, id
    ) result1
WHERE 
    result1.rownumber < 4
GROUP BY id

统计数据如下：

逻辑读取总数：1
总 CPU 时间：00:00:00.000

对于 SQL 服务器 2016- :

select id
    ,  STUFF((
       SELECT ',' + t1.Name
         FROM cte t1
        WHERE t1.id = t2.id 
        and t1.rownumber <  4
        ORDER BY t1.name
          FOR XML PATH('')), 1, LEN(','), '') AS Top3Names

    , STUFF((
       SELECT ',' + cast(t1.countx as varchar(50))
         FROM cte t1
        WHERE t1.id = t2.id --and t1.name = t2.name 
        and t1.rownumber <  4
        ORDER BY t1.name
          FOR XML PATH('')), 1, LEN(','), '') AS RespectiveTotal
from cte t2
group by id

统计数据如下：

逻辑读取总数：7
总CPU时间：00:00:00.006

因此无论 sql 服务器版本如何，它都会提高性能，如果您使用 sql 服务器 2017 或更高版本使用上面的查询，您将获得最佳性能。

如何优化查询，使手动工作可以动态和优化的方式进行

How to optimize the query so the manual work can be dyanmic and in optimized way

sql-server

sql-order-by

query-optimization

union

for-xml