执行计划中的估计行数有偏差

Question

我遇到的情况是执行计划中的估计行数有偏差

我在联接中的列是varchar(50)。我尝试了不同的索引，但它并没有减少这个问题。我什至尝试过使用临时表上的索引。我还能做什么？

PS 这是估计数字开始漂移的第一个地方......而且表格并不大（48000 行）。

代码是：

SELECT DISTINCT householdnumber, householdid, primaryCustomerID
INTO #Households
FROM TableA

    
SELECT 
A.*,
MIN(B.[ProfileCreatedDate]) PROFILECREATEDDATE 
INTO #Profile
from #Households AS a 
LEFT JOIN TableA AS B
    
ON A.[HouseholdNumber]= B.[HouseholdNumber] and A.[HouseholdId]=B.[HouseholdId]
GROUP BY a.householdnumber, a.householdid, a.primaryCustomerID;

我知道这似乎可以重写为：

SELECT householdnumber, householdid, primaryCustomerID, MIN([ProfileCreatedDate]) AS PROFILECREATEDDATE 
INTO #Profile2
from TableA
GROUP BY householdnumber, householdid, primaryCustomerID;

但结果并不相同，我不想更改结果，因为我不确定这段代码的创建者是否知道他们在做什么。

关于列的一些统计信息： householdnumber 始终等于 householdid。 householdid 是 nvarchar(50) 但 householdnumber 是 varchar(40)。该表有 48877 行。 householdnumber, householdid, primaryCustomerID 的不同组合有 48029 行。 primaryCustomerID 的不同数量是 47152。

Answer 1

关于代码 - 似乎较大（原始）版本和您更简单的 GROUP BY 版本之间的区别在于，原始版本为该家庭中的任何人找到了最小值 profilecreateddate，而您的更简单的版本为特定的primarycustomerid 找到profilecreateddate。

例如（使用更简单的数据）

CREATE TABLE #TableA (householdnumber int, householdid int, primaryCustomerID int, ProfileCreatedDate datetime);
INSERT INTO #TableA (householdnumber, householdid, primaryCustomerID, ProfileCreatedDate) VALUES
(1, 1, 1, '20201001'),
(1, 1, 1, '20201002'),
(1, 1, 2, '20201003');

SELECT DISTINCT householdnumber, householdid, primaryCustomerID
INTO #Households
FROM #TableA;

SELECT 
    A.*,
    MIN(B.[ProfileCreatedDate]) PROFILECREATEDDATE 
INTO #Profile
from #Households AS a 
    LEFT JOIN #TableA AS B  
      ON A.[HouseholdNumber]= B.[HouseholdNumber] and A.[HouseholdId]=B.[HouseholdId]
GROUP BY a.householdnumber, a.householdid, a.primaryCustomerID;

SELECT * FROM #Profile;
/* -- Results
householdnumber  householdid  primaryCustomerID  PROFILECREATEDDATE
1                1            1                  2020-10-01 00:00:00.000
1                1            2                  2020-10-01 00:00:00.000
*/

SELECT householdnumber, householdid, primaryCustomerID, MIN([ProfileCreatedDate]) AS PROFILECREATEDDATE 
INTO #Profile2
from #TableA
GROUP BY householdnumber, householdid, primaryCustomerID;

SELECT * FROM #Profile2;
/* -- Results
householdnumber  householdid  primaryCustomerID  PROFILECREATEDDATE
1                1            1                  2020-10-01 00:00:00.000
1                1            2                  2020-10-03 00:00:00.000
*/

如果您在上面注意到，第 2 行的 PROFILECREATEDATE 是不同的。

因此，您可以尝试以下代码，它应该会给出与原始集相同的结果 - 看看它在一段时间内的表现如何（并确认它与原始结果匹配）。

SELECT DISTINCT t1.householdnumber, t1.householdid, primaryCustomerID, 
        MIN([ProfileCreatedDate]) OVER (PARTITION BY t1.householdnumber, t1.householdid) AS PROFILECREATEDDATE 
INTO #Profile3
FROM #TableA t1;

SELECT * FROM #Profile3;
/* -- Results
householdnumber  householdid  primaryCustomerID  PROFILECREATEDDATE
1                1            1                  2020-10-01 00:00:00.000
1                1            2                  2020-10-01 00:00:00.000
*/