Steven Hibble's questions

我正在做一些字符串聚合，并注意到对我来说看起来像一个错误。STRING_AGG在具有不同分隔符参数的同一查询中使用了两次。但是，两者都产生了相同的结果（在两种情况下都使用了“第一个”分隔符）。它似乎只在某些情况下发生。这是一个错误还是这是记录在案的行为？

首先，设置：

CREATE TABLE #Data
  ([Group] INT
 , Member  CHAR(1));

INSERT INTO #Data
VALUES (1, 'a'), (1, 'b')
     , (2, 'c'), (2, 'd');

从表与VALUES构造函数。没有WITHIN GROUP ORDER BY，一切都很好。有了它，bug就出现了

SELECT Commas = STRING_AGG(Member, ', ') --WITHIN GROUP(ORDER BY Member)
     , Colons = STRING_AGG(Member, '::') --WITHIN GROUP(ORDER BY Member)
FROM #Data;

SELECT Commas = STRING_AGG(Member, ', ') --WITHIN GROUP(ORDER BY Member)
     , Colons = STRING_AGG(Member, '::') --WITHIN GROUP(ORDER BY Member)
FROM (VALUES (1, 'a'), (1, 'b')
           , (2, 'c'), (2, 'd')) [v] ([Group], Member);

添加 aGROUP BY添加排序，现在两种情况都存在错误：

SELECT [Group]
     , Commas = STRING_AGG(Member, ', ')
     , Colons = STRING_AGG(Member, '::')
FROM #Data
GROUP BY [Group];

SELECT [Group]
     , Commas = STRING_AGG(Member, ', ')
     , Colons = STRING_AGG(Member, '::')
FROM (VALUES (1, 'a'), (1, 'b')
           , (2, 'c'), (2, 'd')) [v] ([Group], Member)
GROUP BY [Group];

加上它肯定坏了……就像内心深处一样。将两个表达式组合在 a 中HAVING，您将一无所获。单独使用其中一个，您就可以获得数据。

SELECT [Group]
     , Colons = STRING_AGG(Member, '::')
     , Commas = STRING_AGG(Member, ', ')
FROM #Data
GROUP BY [Group]
HAVING STRING_AGG(Member, ', ') LIKE '%, %'
   AND STRING_AGG(Member, '::') LIKE '%::%';

切换列的顺序会切换使用的分隔符。

我使用这些跟踪标志进行调查，但我找不到任何关于“优化”第二个聚合的信息。

OPTION (QUERYTRACEON 3604 -- Output info to client
      , QUERYTRACEON 8619 -- Show applied optimization rules
      , QUERYTRACEON 8606 -- Show logical query trees
      , QUERYTRACEON 8607 -- Show physical query tree
      , RECOMPILE);

我有 3 个“大”表，它们连接在一对列（两者int）上。

• 表 1 有约 2 亿行
• 表 2 有约 150 万行
• Table3 有大约 600 万行

每个表在 , 上都有一个聚集索引Key1，Key2然后是一列。Key1具有低基数并且非常偏斜。它总是在WHERE子句中引用。Key2条款中从未提及WHERE。每个连接都是多对多的。

问题在于基数估计。每个连接的输出估计变得更小而不是更大。当实际结果达到数百万时，这导致最终估计为低数百。

我有什么办法可以让行政长官做出更好的估计吗？

SELECT 1
FROM Table1 t1
     JOIN Table2 t2
       ON t1.Key1 = t2.Key1
          AND t1.Key2 = t2.Key2
     JOIN Table3 t3
       ON t1.Key1 = t3.Key1
          AND t1.Key2 = t3.Key2
WHERE t1.Key1 = 1;

我尝试过的解决方案：

• 在 上创建多列统计信息Key1，Key2
• 创建大量过滤后的统计信息Key1（这很有帮助，但我最终会在数据库中获得数千个用户创建的统计信息。）

屏蔽的执行计划（抱歉屏蔽不好）

在我正在查看的情况下，结果有 900 万行。新的 CE 估计有 180 行；旧版 CE 估计有 6100 行。

这是一个可重现的示例：

DROP TABLE IF EXISTS #Table1, #Table2, #Table3;
CREATE TABLE #Table1 (Key1 INT NOT NULL, Key2 INT NOT NULL, T1Key3 INT NOT NULL, CONSTRAINT pk_t1 PRIMARY KEY CLUSTERED (Key1, Key2, T1Key3));
CREATE TABLE #Table2 (Key1 INT NOT NULL, Key2 INT NOT NULL, T2Key3 INT NOT NULL, CONSTRAINT pk_t2 PRIMARY KEY CLUSTERED (Key1, Key2, T2Key3));
CREATE TABLE #Table3 (Key1 INT NOT NULL, Key2 INT NOT NULL, T3Key3 INT NOT NULL, CONSTRAINT pk_t3 PRIMARY KEY CLUSTERED (Key1, Key2, T3Key3));

-- Table1 
WITH Numbers
     AS (SELECT TOP (1000000) Number = ROW_NUMBER() OVER(ORDER BY t1.number)
         FROM master..spt_values t1
              CROSS JOIN master..spt_values t2),
     DataSize (Key1, NumberOfRows)
     AS (SELECT 1, 2000 UNION
         SELECT 2, 10000 UNION
         SELECT 3, 25000 UNION
         SELECT 4, 50000 UNION
         SELECT 5, 200000)
INSERT INTO #Table1
SELECT Key1
     , Key2 = ROW_NUMBER() OVER (PARTITION BY Key1, T1Key3 ORDER BY Number)
     , T1Key3
FROM DataSize
     CROSS APPLY (SELECT TOP(NumberOfRows) 
                         Number
                       , T1Key3 = Number%(Key1*Key1) + 1 
                  FROM Numbers
                  ORDER BY Number) size;

-- Table2 (same Key1, Key2 values; smaller number of distinct third Key)
WITH Numbers
     AS (SELECT TOP (1000000) Number = ROW_NUMBER() OVER(ORDER BY t1.number)
         FROM master..spt_values t1
              CROSS JOIN master..spt_values t2)
INSERT INTO #Table2
SELECT DISTINCT 
       Key1
     , Key2
     , T2Key3
FROM #Table1
     CROSS APPLY (SELECT TOP (Key1*10) 
                         T2Key3 = Number
                  FROM Numbers
                  ORDER BY Number) size;

-- Table2 (same Key1, Key2 values; smallest number of distinct third Key)
WITH Numbers
     AS (SELECT TOP (1000000) Number = ROW_NUMBER() OVER(ORDER BY t1.number)
         FROM master..spt_values t1
              CROSS JOIN master..spt_values t2)
INSERT INTO #Table3
SELECT DISTINCT 
       Key1
     , Key2
     , T3Key3
FROM #Table1
     CROSS APPLY (SELECT TOP (Key1) 
                         T3Key3 = Number
                  FROM Numbers
                  ORDER BY Number) size;


DROP TABLE IF EXISTS #a;
SELECT col = 1 
INTO #a
FROM #Table1 t1
     JOIN #Table2 t2
       ON t1.Key1 = t2.Key1
          AND t1.Key2 = t2.Key2
WHERE t1.Key1 = 1;

DROP TABLE IF EXISTS #b;
SELECT col = 1 
INTO #b
FROM #Table1 t1
     JOIN #Table2 t2
       ON t1.Key1 = t2.Key1
          AND t1.Key2 = t2.Key2
     JOIN #Table3 t3
       ON t1.Key1 = t3.Key1
          AND t1.Key2 = t3.Key2
WHERE t1.Key1 = 1;