T-SQL：與TIES進行分頁

Question

我正在嘗試實現一個有點不同的分頁例程。

對於一個簡單的例子起見，讓我們假設我有一個表中定義和填充如下：

DECLARE @Temp TABLE 
(
    ParentId INT, 
    [TimeStamp] DATETIME, 
    Value  INT 
); 

INSERT INTO @Temp VALUES (1, '1/1/2013 00:00', 6); 
INSERT INTO @Temp VALUES (1, '1/1/2013 01:00', 7); 
INSERT INTO @Temp VALUES (1, '1/1/2013 02:00', 8); 
INSERT INTO @Temp VALUES (2, '1/1/2013 00:00', 6); 
INSERT INTO @Temp VALUES (2, '1/1/2013 01:00', 7); 
INSERT INTO @Temp VALUES (2, '1/1/2013 02:00', 8); 
INSERT INTO @Temp VALUES (3, '1/1/2013 00:00', 6); 
INSERT INTO @Temp VALUES (3, '1/1/2013 01:00', 7); 
INSERT INTO @Temp VALUES (3, '1/1/2013 02:00', 8); 

TimeStamp將始終是相同的時間間隔，例如每日數據，1小時數據，1分鐘數據等，不會混雜在一起。

報告和演示的目的，我想實現分頁是：

1. 訂單由TimeStamp
2. 開始時使用建議pageSize（比如4），但會自動調整以包括其他記錄匹配的上TimeStamp。換句話說，如果一個ParentId包含1/1/2013 01:00，建議的pageSize將被覆蓋，所有ParentId's的時間01:00的記錄將包含在內。這幾乎就像TOP WITH TIES選項。

因此，使用4的pageSize運行此查詢將返回6條記錄。有3小時00:00和1小時01:00默認情況下，但因爲有更多小時01:00's，pageSize將被覆蓋返回所有小時00:00和01:00。

這是我到目前爲止，我認爲我已經接近，因爲它適用於第一次迭代，但後續pageSize+行的後續查詢不起作用。

WITH CTE AS 
(
    SELECT ParentId, [TimeStamp], Value, 
    RANK() OVER(ORDER BY [TimeStamp]) AS rnk, 
    ROW_NUMBER() OVER(ORDER BY [TimeStamp]) AS rownum 
    FROM @Temp 
) 

SELECT * 
FROM CTE 
WHERE (rownum BETWEEN 1 AND 4) OR (rnk BETWEEN 1 AND 4) 
ORDER BY TimeStamp, ParentId 

ROW_NUMBER確保滿足最小pageSize，但RANK將包含其他關係。

Answer 1

我認爲你使用row_number()和rank()的策略是過於複雜的事情。

只需從數據中選擇前4個時間戳即可。然後選擇與這些匹配的時間戳：

select * 
from @temp 
where [timestamp] in (select top 4 [timestamp] from @temp order by [TimeStamp]) 

Answer 2

declare @Temp as Table (ParentId Int, [TimeStamp] DateTime, [Value] Int); 
insert into @Temp (ParentId, [TimeStamp], [Value]) values 
(1, '1/1/2013 00:00', 6), 
(1, '1/1/2013 01:00', 7), 
(1, '1/1/2013 02:00', 8), 
(2, '1/1/2013 00:00', 6), 
(2, '1/1/2013 01:00', 7), 
(2, '1/1/2013 02:00', 8), 
(3, '1/1/2013 00:00', 6), 
(3, '1/1/2013 01:00', 7), 
(3, '1/1/2013 02:00', 8); 

declare @PageSize as Int = 4; 
declare @Page as Int = 1; 

with Alpha as (
    select ParentId, [TimeStamp], Value, 
     Rank() over (order by [TimeStamp]) as Rnk, 
     Row_Number() over (order by [TimeStamp]) as RowNum 
    from @Temp), 
    Beta as (
    select Min(Rnk) as MinRnk, Max(Rnk) as MaxRnk 
     from Alpha 
     where (@Page - 1) * @PageSize < RowNum and RowNum <= @Page * @PageSize) 
    select A.* 
     from Alpha as A inner join 
      Beta as B on B.MinRnk <= A.Rnk and A.Rnk <= B.MaxRnk 
     order by [TimeStamp], ParentId; 

編輯： 的是，指定頁碼，因爲它去，這樣一個/上一個頁面可以不重疊的行來實現另一種查詢：

with Alpha as (
    select ParentId, [TimeStamp], Value, 
     Rank() over (order by [TimeStamp]) as Rnk, 
     Row_Number() over (order by [TimeStamp]) as RowNum 
    from @Temp), 
    Beta as (
    select ParentId, [TimeStamp], Value, Rnk, RowNum, 1 as Page, 1 as PageRow 
     from Alpha 
     where RowNum = 1 
    union all 
    select A.ParentId, A.[TimeStamp], A.Value, A.Rnk, A.RowNum, 
     case when B.PageRow >= @PageSize and A.TimeStamp <> B.TimeStamp then B.Page + 1 else B.Page end, 
     case when B.PageRow >= @PageSize and A.TimeStamp <> B.TimeStamp then 1 else B.PageRow + 1 end 
     from Alpha as A inner join 
      Beta as B on B.RowNum + 1 = A.RowNum 
    ) 
    select * from Beta 
     option (MaxRecursion 0) 

需要注意的是遞歸的CTE往往規模不佳。