在Microsoft SQL Server中，有没有比我目前所做的更好的方法来在两个方向上填充空值？

我正在尝试找出如何在Microsoft SQL Server中向上填充空值。这在甲骨文中非常简单，但老板并不同意。经过六个小时的谷歌搜索，我终于决定发帖

在结果集中

值是我们的时间戳数据，包含所有可能的日期时间

VALUE2是我们的填充空向下方法

VALUE3是我们的向上填充空值方法

考虑到这张假烂桌子

-- Test Data & Table 
DECLARE @TEST001 TABLE
(TIME_STAMP datetime,
    TagA integer,
    TagB integer,
    TagC integer) 
-- Insert test Values       
 INSERT INTO @TEST001
 VALUES
    ('2017-01-21 00:01:00.042', NULL, NULL, 87),
    ('2017-01-21 00:04:10.155', NULL, 1239, NULL),
    ('2017-01-21 00:04:10.959', NULL, NULL, 86),
    ('2017-01-21 00:06:49.401', NULL, 1240, NULL),
    ('2017-01-21 00:06:59.301', NULL, 1239, NULL),
    ('2017-01-21 00:07:10.124', 108, NULL, NULL),
    ('2017-01-21 00:12:11.789', 109, NULL, NULL),
    ('2017-01-21 00:16:12.190', 108, NULL, NULL),
    ('2017-01-21 00:16:13.987', 107, NULL, NULL),
    ('2017-01-21 00:17:31.410', NULL, 1260, NULL),
    ('2017-01-21 00:17:32.511', NULL, 1261, 87),
    ('2017-01-21 00:17:32.966', NULL, 1262, NULL)

给定这个查询，它在伪垃圾表中查询TagA

-- Start of Query used in VBS ADODB CONN.
declare @s datetime
declare @e datetime
set @s = '2017-01-21 00:00:00'
set @e = '2017-01-21 23:59:59'
;
-- We need to get all intervals between our two dates.
WITH ALL_INTERVALS AS (
    SELECT TOP (datediff(mi,@s,@e))
    TIMES = dateadd(mi,CONVERT(INT, ROW_NUMBER() OVER (ORDER BY s1.[object_id])),@s)
        FROM sys.all_objects AS s1
        CROSS JOIN
        sys.all_objects AS s2
),
-- We need to include both our real data and all possible intervals.
ALL_TIMES AS (
SELECT
    Time_Stamp as TIMES,
    TagA AS VALUE
FROM @TEST001 H
WHERE Time_Stamp BETWEEN @s and @e
UNION ALL
SELECT
    TIMES AS TIMES,
    NULL AS VALUE
    FROM ALL_INTERVALS
),
-- We need to find the real first value and fill all nulls with this value until we hit a new value.
FILL_DOWN AS ( SELECT 
            TIMES, 
            VALUE,
            ISNULL(VALUE, (SELECT TOP 1 VALUE FROM ALL_TIMES WHERE TIMES < AT.TIMES AND VALUE IS NOT NULL ORDER BY TIMES DESC)) AS VALUE2
            FROM ALL_TIMES AT
),
-- Our fill up method does not work if our first set of values is null. UGH, this does not work either....crap. It fills our first set
-- With our last real value....
FILL_UP AS  ( 
        SELECT 
            TIMES, 
            VALUE, 
            VALUE2,
            ISNULL(VALUE2, (SELECT TOP 1 VALUE2 FROM FILL_DOWN WHERE TIMES > FD.TIMES AND VALUE2 IS NOT NULL ORDER BY TIMES DESC)) AS VALUE3
            FROM FILL_DOWN FD
)
SELECT * 
FROM FILL_UP 
ORDER BY TIMES ASC

结果

TIMES                   VALUE   VALUE2  VALUE3
2017-01-2100:01:00.000  NULL    NULL    107 <---------- This should be 108 from our fake crappy table.
2017-01-2100:01:00.043  NULL    NULL    107 
2017-01-2100:02:00.000  NULL    NULL    107 
2017-01-2100:03:00.000  NULL    NULL    107 
2017-01-2100:04:00.000  NULL    NULL    107 
2017-01-2100:04:10.157  NULL    NULL    107 
2017-01-2100:04:10.960  NULL    NULL    107 
2017-01-2100:05:00.000  NULL    NULL    107 
2017-01-2100:06:00.000  NULL    NULL    107 
2017-01-2100:06:49.400  NULL    NULL    107 
2017-01-2100:06:59.300  NULL    NULL    107 
2017-01-2100:07:00.000  NULL    NULL    107 
2017-01-2100:07:10.123  108     108     108 
2017-01-2100:08:00.000  NULL    108     108 
2017-01-2100:09:00.000  NULL    108     108 
2017-01-2100:10:00.000  NULL    108     108 
2017-01-2100:11:00.000  NULL    108     108 
2017-01-2100:12:00.000  NULL    108     108 
2017-01-2100:12:11.790  109     109     109 
2017-01-2100:13:00.000  NULL    109     109 
2017-01-2100:14:00.000  NULL    109     109 
2017-01-2100:15:00.000  NULL    109     109 
2017-01-2100:16:00.000  NULL    109     109 
2017-01-2100:16:12.190  108     108     108 
2017-01-2100:16:13.987  107     107     107 
2017-01-2100:17:00.000  NULL    107     107 
2017-01-2100:17:31.410  NULL    107     107 
2017-01-2100:17:32.510  NULL    107     107 
2017-01-2100:17:32.967  NULL    107     107 
2017-01-2100:18:00.000  NULL    107     107 
2017-01-2100:19:00.000  NULL    107     107 
2017-01-2100:20:00.000  NULL    107     107 
2017-01-2100:21:00.000  NULL    107     107 
2017-01-2100:22:00.000  NULL    107     107 
2017-01-2100:23:00.000  NULL    107     107 
2017-01-2100:24:00.000  NULL    107     107 
2017-01-2100:25:00.000  NULL    107     107 
2017-01-2100:26:00.000  NULL    107     107

编辑：

我今天在玩非常大的数据集，发现这是有史以来最快的事情。它在一秒钟内完全符合我的要求。你看不到这个小表格中的速度差异，但是对于大数据集来说，1秒与24分钟是非常大的！谢谢你的帮助，祝你一切顺利

-- Test Data & Table 
DECLARE @TEST001 TABLE
(TIME_STAMP datetime,
    TagA integer,
    TagB integer,
    TagC integer) 
-- Insert test Values       
 INSERT INTO @TEST001
 VALUES
    ('2017-01-21 00:01:00.042', NULL, NULL, 87),
    ('2017-01-21 00:04:10.155', NULL, 1239, NULL),
    ('2017-01-21 00:04:10.959', NULL, NULL, 86),
    ('2017-01-21 00:06:49.401', NULL, 1240, NULL),
    ('2017-01-21 00:06:59.301', NULL, 1239, NULL),
    ('2017-01-21 00:07:10.124', 108, NULL, NULL),
    ('2017-01-21 00:12:11.789', 109, NULL, NULL),
    ('2017-01-21 00:16:12.190', 108, NULL, NULL),
    ('2017-01-21 00:16:13.987', 107, NULL, NULL),
    ('2017-01-21 00:17:31.410', NULL, 1260, NULL),
    ('2017-01-21 00:17:32.511', NULL, 1261, 87),
    ('2017-01-21 00:17:32.966', NULL, 1262, NULL)
-- Start of Query used in VBS ADODB CONNN
declare @s datetime
declare  @e datetime
set @s = '2017-01-01 00:00:00'
set @e = '2017-01-31 23:59:59'
;
WITH ALL_INTERVALS 
AS ( SELECT TOP (datediff(mi,@s,@e))
    TIMES = dateadd(mi,CONVERT(INT, ROW_NUMBER() OVER (ORDER BY s1.[object_id])),@s),
    NULL AS VALUE
        FROM sys.all_objects AS s1
        CROSS JOIN
        sys.all_objects AS s2
),
ALL_TIMES 
AS ( SELECT
    H.TIME_STAMP as TIMES,
    TagA AS VALUE
    FROM @TEST001 H
    WHERE TIME_STAMP BETWEEN @s and @e
    UNION ALL
    SELECT 
        AI.TIMES AS TIMES, 
        AI.VALUE AS VALUE
        FROM ALL_INTERVALS AI
),
-- JUST INCASE OUR REAL TIME == INTERVAL TIME exactly.
FILL_ACCROSS AS( SELECT TIMES AS TIMES,
                ISNULL(AT.VALUE, (SELECT TOP 1 VALUE FROM ALL_TIMES WHERE TIMES = AT.TIMES AND VALUE IS NOT NULL ORDER BY TIMES ASC)) AS VALUE
                FROM ALL_TIMES AT

),
-- FILL UP AND FILL DOWN.
FILL_UP_DOWN AS ( SELECT 
                TIMES, 
                VALUE,
                ISNULL(FA.VALUE, (SELECT TOP 1 VALUE FROM ALL_TIMES WHERE TIMES > FA.TIMES AND VALUE IS NOT NULL ORDER BY TIMES ASC)) AS VALUE2,
                ISNULL(FA.VALUE, (SELECT TOP 1 VALUE FROM ALL_TIMES WHERE TIMES < FA.TIMES AND VALUE IS NOT NULL ORDER BY TIMES DESC)) AS VALUE3
                FROM FILL_ACCROSS FA
)
--Just a nice display of what is going on.
select FD.TIMES,
       FD.VALUE,
       FD.VALUE2,
       FD.VALUE3,
       CASE
          WHEN FD.VALUE3 IS NULL THEN (FD.VALUE2)
          ELSE (FD.VALUE3)
          END AS VALUE4
  FROM FILL_UP_DOWN FD order by TIMES ASC

---

如果您正在寻找替代方案，请尝试此分组

-- Start of Query used in VBS ADODB CONN.
declare @s datetime
declare @e datetime
set @s = '2017-01-21 00:00:00.001'
set @e = '2017-01-21 23:59:59'
;
--select datepart(millisecond,@s);
-- We need to get all intervals between our two dates.
WITH ALL_INTERVALS AS (
    SELECT TOP (datediff(mi,@s,@e))
    TIMES = dateadd(mi,CONVERT(INT, ROW_NUMBER() OVER (ORDER BY s1.[object_id])),@s)
        FROM sys.all_objects AS s1
        CROSS JOIN
        sys.all_objects AS s2
),
-- We need to include both our real data and all possible intervals.
ALL_TIMES AS (
SELECT
    Time_Stamp as TIMES,
    TagA AS VALUE
FROM @TEST001 H
WHERE Time_Stamp BETWEEN @s and @e
UNION ALL
SELECT
    TIMES AS TIMES,
    NULL AS VALUE
    FROM ALL_INTERVALS
),
g1 AS ( 
    SELECT 
        TIMES, 
        VALUE,
        diff = case 
               when value=lag(value,1) over(order by TIMES) 
                  or coalesce (value, lag(value,1) over(order by TIMES)) is null
               then 0 else 1 end
     FROM ALL_TIMES AT
),
g2 as ( 
    select TIMES, VALUE
         , grp = sum(diff) over(order by TIMES)
         , direction = case sum(diff) over(order by TIMES) when 0 then 1 else -1 end 
    from g1
)
select 
    TIMES, 
    VALUE, 
    grp,
    ISNULL(VALUE,
           (SELECT TOP 1 VALUE FROM g2 repl WHERE grp = g2.grp + g2.direction AND VALUE IS NOT NULL ORDER BY TIMES ) 
           )  AS VALUE3
from g2;

---

我真的不明白你想做什么。您可以缩短示例数据并添加所需结果吗？如果您将相关子查询中的顺序更改为从DESC获取VALUE3到ASC，您应该会得到所需的结果。目前，它正在寻找给定日期后的最后一个非空值，而从外观上看，你想要的是第一个非空值。在填充顺序中删除DESC byI冷静下来，吃了一顿冷笑，谢谢@GarethD和Serg，执行仍然需要5秒，但至少它能工作！我发誓我试过删除DESC，但是太晚了，我看不清楚。太棒了，我喜欢不同的方法，谢谢你的发帖！我发现在使用大型数据集时，速度与我最初的查询相同。但是在玩了一整天之后，我想到了上面的编辑。在我的大数据集上完成几乎需要一秒钟。干杯再次感谢您指出需要通过以下方式取消或更改订单。那天我做得不太好，弄不明白。