聚合日志 table 记录以避免冗余

Question

我有一个用于产品变更跟踪的 table，如下所示：

CREATE TABLE ProductHistory (
  ProductId INT NOT NULL,
  Name NVARCHAR(50) NOT NULL,
  Price MONEY NOT NULL,
  StartDate DATETIME NOT NULL,
  EndDate DATETIME NOT NULL
)

INSERT INTO ProductHistory VALUES
(1, 'Phone',       100, '2020-11-20 00:00', '2020-11-20 01:00'), /* initial */
(1, 'Phone',       100, '2020-11-20 01:01', '2020-11-20 02:00'), /* no change */
(1, 'Phone',       200, '2020-11-20 02:01', '2020-11-20 03:00'), /* no change, current */

(2, 'Apple',          5, '2020-11-20 00:00', '2020-11-20 01:00'), /* initial */
(2, 'Apple',         10, '2020-11-20 01:01', '2020-11-20 02:00'), /* changed */
(2, 'Pineapple',     10, '2020-11-20 02:01', '2020-11-20 03:00'), /* no change, current */

(3, 'Orange juice', 100, '2020-11-21 00:00', '2020-11-21 01:00'), /* initial */
(3, 'Orange juice', 100, '2020-11-21 01:01', '2020-11-21 02:00'), /* no change */
(3, 'Orange juice', 100, '2020-11-21 02:01', '2020-11-21 03:00')  /* no change, current */

我希望提出一个查询来获得以下结果。请注意，没有实际更改的记录应该合并在一起，这样就没有冗余。

  ProductId       Name       Price                StartDate                               EndDate                 
 ----------- -------------- ------- -------------------------------------- -------------------------------------- 
          1   Phone            100   2020-11-20 00:00:00.000 (first row)    2020-11-20 02:00:00.000 (second row)  
          1   Phone            200   2020-11-20 02:01:00.000 (third row)    2020-11-20 03:00:00.000 (third row)   

          2   Apple              5   2020-11-20 00:00:00.000 (first row)    2020-11-20 01:00:00.000 (first row)   
          2   Apple             10   2020-11-20 01:01:00.000 (second row)   2020-11-20 02:00:00.000 (second row)  
          2   Pineapple         10   2020-11-20 02:01:00.000 (third row)    2020-11-20 03:00:00.000 (third row)   

          3   Orange juice     100   2020-11-21 00:00:00.000 (first row)    2020-11-20 03:00:00.000 (third row)

我最接近的是：

; WITH history AS (
    SELECT
        ProductId,
        Name,
        Price,
        StartDate,
        EndDate
    FROM (
        SELECT
            ROW_NUMBER() OVER (PARTITION BY ProductId ORDER BY StartDate DESC) 'RowNumber',
            *
        FROM ProductHistory
    ) history
    WHERE history.RowNumber = 1 -- select newest row per ProductId

    UNION ALL

    SELECT
        previous.ProductId,
        previous.Name,
        previous.Price,
        previous.StartDate,
        EndDate
    FROM (
        SELECT
            ROW_NUMBER() OVER (PARTITION BY previous.ProductId ORDER BY previous.StartDate DESC) 'RowNumber',
            previous.*
        FROM history [current]
        INNER JOIN ProductHistory previous
            ON previous.ProductId = [current].ProductId
            AND previous.StartDate < [current].StartDate
            AND (
                previous.Name <> [current].Name
                OR previous.Price <> [current].Price
            )   
    ) previous
    WHERE previous.RowNumber = 1 -- select previous row of each ProductId, recursively
)

SELECT *
FROM history
ORDER BY
    ProductId,
    StartDate

  ProductId       Name       Price           StartDate                  EndDate          
 ----------- -------------- -------- ------------------------- ------------------------- 
          1   Phone          100,00   2020-11-20 01:01:00.000   2020-11-20 02:00:00.000  
          1   Phone          200,00   2020-11-20 02:01:00.000   2020-11-20 03:00:00.000  

          2   Apple            5,00   2020-11-20 00:00:00.000   2020-11-20 01:00:00.000  
          2   Apple           10,00   2020-11-20 01:01:00.000   2020-11-20 02:00:00.000  
          2   Pineapple       10,00   2020-11-20 02:01:00.000   2020-11-20 03:00:00.000  

          3   Orange juice   100,00   2020-11-21 02:01:00.000   2020-11-21 03:00:00.000

虽然 Name 和 Price 列值是正确的，但我不确定如何聚合 StartDate 和 EndDate 列以获得我需要的值。 fiddle 中提供了所有代码，如果有帮助的话。

Answer 1

这是一种间隙和孤岛问题。可能最简单的方法就是行数的不同：

select productid, name, price, min(startdate), max(enddate)
from (select ph.*,
             row_number() over (partition by productid order by startdate) as seqnum,
             row_number() over (partition by productid, name, price order by startdate) as seqnum_2
      from producthistory
     ) ph
group by productid, name, price, (seqnum - seqnum_2);

这假设时间范围内没有间隙——这对于此数据模型来说似乎是合理的。

为什么这行得通？这有点难以解释。但是，如果您查看子查询的结果，您将看到相邻行的两个行号之间的差异是如何保持不变的，其中 name 和 price 相同。

聚合日志 table 记录以避免冗余

Aggregate log table records to avoid redundancy

sql

sql-server

common-table-expression