重新排序记录的巧妙方法？答案

【问题标题】：Clever way to reorder records?重新排序记录的巧妙方法？
【发布时间】：2021-02-02 10:38:04
【问题描述】：

我有一个包含以下数据的表格：

VehicleID       Time                    ContextID                   Value
--------------- ----------------------- --------------------------  -----
359586015047188 2021-02-01 07:27:14.777 SafeProtectDeviceConnected  False
359586015047188 2021-02-01 07:53:38.000 SafeProtectProtectionLevel  5
359586015047188 2021-02-01 07:53:47.777 SafeProtectDeviceConnected  True
359586015047188 2021-02-01 10:24:20.777 SafeProtectDeviceConnected  False
359586015047188 2021-02-01 10:26:46.000 SafeProtectProtectionLevel  5
359586015047188 2021-02-01 10:26:55.777 SafeProtectDeviceConnected  True
359586015047188 2021-02-01 10:43:53.777 SafeProtectDeviceConnected  False
359586015047188 2021-02-01 10:46:01.000 SafeProtectProtectionLevel  5
359586015047188 2021-02-01 10:46:09.777 SafeProtectDeviceConnected  True
359586015047188 2021-02-01 11:02:16.777 SafeProtectDeviceConnected  False
359586015047188 2021-02-01 14:39:41.777 SafeProtectProtectionLevel  5
359586015047188 2021-02-01 14:39:42.777 SafeProtectDeviceConnected  True
359586015047188 2021-02-01 14:55:48.777 SafeProtectDeviceConnected  False
359586015047188 2021-02-02 07:52:12.777 SafeProtectDeviceConnected  True
359586015047188 2021-02-02 07:52:12.777 SafeProtectProtectionLevel  5
359586015047188 2021-02-02 07:52:32.777 SafeProtectDeviceConnected  False
359586015047188 2021-02-02 07:53:57.000 SafeProtectProtectionLevel  5
359586015047188 2021-02-02 07:54:10.777 SafeProtectDeviceConnected  True

如您所见，数据是及时插入的。但有时，时间是不正确的。我需要一个查询来修复它。

如果数据是一致的，我总是会有一行 ContextID 的值是 SafeProtectDeviceConnected 并且 Value 的值是 True 后跟一行 ContextID 的值 SafeProtectProtectionLevel 无论 Value 的值是什么列。

我发现我可以使用 LAG 分析函数来访问前一行的值，并且我还可以将一个或多个 CASE 放入 order by clause

那么在前一个结果集上应用修复查询后的正确结果将是：

VehicleID       Time                    ContextID                   Value
--------------- ----------------------- --------------------------  -----
359586015047188 2021-02-01 07:27:14.777 SafeProtectDeviceConnected  False
359586015047188 2021-02-01 07:53:47.777 SafeProtectDeviceConnected  True
359586015047188 2021-02-01 07:53:38.000 SafeProtectProtectionLevel  5
359586015047188 2021-02-01 10:24:20.777 SafeProtectDeviceConnected  False
359586015047188 2021-02-01 10:26:55.777 SafeProtectDeviceConnected  True
359586015047188 2021-02-01 10:26:46.000 SafeProtectProtectionLevel  5
359586015047188 2021-02-01 10:43:53.777 SafeProtectDeviceConnected  False
359586015047188 2021-02-01 10:46:09.777 SafeProtectDeviceConnected  True
359586015047188 2021-02-01 10:46:01.000 SafeProtectProtectionLevel  5
359586015047188 2021-02-01 11:02:16.777 SafeProtectDeviceConnected  False
359586015047188 2021-02-01 14:39:42.777 SafeProtectDeviceConnected  True
359586015047188 2021-02-01 14:39:41.777 SafeProtectProtectionLevel  5
359586015047188 2021-02-01 14:55:48.777 SafeProtectDeviceConnected  False
359586015047188 2021-02-02 07:52:12.777 SafeProtectDeviceConnected  True
359586015047188 2021-02-02 07:52:12.777 SafeProtectProtectionLevel  5
359586015047188 2021-02-02 07:52:32.777 SafeProtectDeviceConnected  False
359586015047188 2021-02-02 07:54:10.777 SafeProtectDeviceConnected  True
359586015047188 2021-02-02 07:53:57.000 SafeProtectProtectionLevel  5

基本上，如果我们查看上面重新排序的结果集，特别是从上到下的 Value 列，我们应该有一个 True 值，后跟一个或多个带有数值的行，然后是一个 False 值

到目前为止我尝试过的是使用ORDER BY t.VehiculeID, /*dbo.ContextDetail.Time*/ CASE WHEN t.ContextID='SafeProtectDeviceConnected' AND t.Value='True' THEN 1 END, CASE WHEN t.ContextID='SafeProtectProtectionLevel' THEN 2 END, CASE WHEN t.ContextID='SafeProtectDeviceConnected' AND t.Value='False' THEN 3 END

但它给了我（显然）所有False 行，然后是所有numeric rows，然后是剩余的True 值。

这个问题是缝隙和孤岛问题吗？
解决这个问题的正确方法是什么？

【问题讨论】：

试试这个：ORDER BY VehicleID, CAST(CONVERT(NVARCHAR(8), [Time], 112) AS BIGINT)*10000 + DATEPART(HOUR, [Time])* 100 + DATEPART(MINUTE , [Time]), CASE WHEN ContextID = 'SafeProtectDeviceConnected' AND [Value] = 'False' THEN 1 WHEN ContextID = 'SafeProtectDeviceConnected' AND [Value] = 'True' THEN 2 ELSE 3 END
@Tyron78 这似乎可行，但我想了解原因。
我添加了我的查询作为答案和一个小解释 - 请检查。如有其他问题，请随时提问。

标签： sql-server gaps-and-islands

【解决方案1】：

试试这个：

ORDER BY VehicleID, CAST(CONVERT(NVARCHAR(8), [Time], 112) AS BIGINT)*10000 + DATEPART(HOUR, [Time])* 100 + DATEPART(MINUTE, [Time]), CASE WHEN ContextID = 'SafeProtectDeviceConnected' AND [Value] = 'False' THEN 1 WHEN ContextID = 'SafeProtectDeviceConnected' AND [Value] = 'True' THEN 2 ELSE 3 END

与您的方法的主要区别：首先，我详细说明了订单中小于分钟的所有内容，因为您提到了，那个时间可能是错误的。但是，据我（之后）注意到，时间似乎还可以，但不是唯一的-因此按时间排序也应该可以解决问题。其次，我创建了一个 case 语句来考虑上下文和值，而不是每列一个 case - 您希望按两列的组合排序，而不是单独按每一列排序。因此，Connected + False 的组合得到 1，Connected + True 得到 2，其他一切都得到 3。在您的查询中，您创建了三个返回值或 NULL 的案例（CASE... END 没有其他），所以毕竟您添加了三个值来排序。

【讨论】：

【解决方案2】：

只要必须分组的行之间的时间不重叠，那么您可以使用Time 列对行进行分组，并使用case 表达式来确定单个组内的排序。下面的解决方案在一个公用表表达式中计算这些新列，以便于选择和排序。

样本数据

create table log
(
  VehicleID bigint,
  Time datetime,
  ContextID nvarchar(50),
  Value nvarchar(10)
);

insert into log (VehicleID, Time, ContextID, Value) values
(359586015047188, '2021-02-01 07:27:14.777', 'SafeProtectDeviceConnected', 'False'),
(359586015047188, '2021-02-01 07:53:38.000', 'SafeProtectProtectionLevel', '5'),
(359586015047188, '2021-02-01 07:53:47.777', 'SafeProtectDeviceConnected', 'True'),
(359586015047188, '2021-02-01 10:24:20.777', 'SafeProtectDeviceConnected', 'False'),
(359586015047188, '2021-02-01 10:26:46.000', 'SafeProtectProtectionLevel', '5'),
(359586015047188, '2021-02-01 10:26:55.777', 'SafeProtectDeviceConnected', 'True'),
(359586015047188, '2021-02-01 10:43:53.777', 'SafeProtectDeviceConnected', 'False'),
(359586015047188, '2021-02-01 10:46:01.000', 'SafeProtectProtectionLevel', '5'),
(359586015047188, '2021-02-01 10:46:09.777', 'SafeProtectDeviceConnected', 'True'),
(359586015047188, '2021-02-01 11:02:16.777', 'SafeProtectDeviceConnected', 'False'),
(359586015047188, '2021-02-01 14:39:41.777', 'SafeProtectProtectionLevel', '5'),
(359586015047188, '2021-02-01 14:39:42.777', 'SafeProtectDeviceConnected', 'True'),
(359586015047188, '2021-02-01 14:55:48.777', 'SafeProtectDeviceConnected', 'False'),
(359586015047188, '2021-02-02 07:52:12.777', 'SafeProtectDeviceConnected', 'True'),
(359586015047188, '2021-02-02 07:52:12.777', 'SafeProtectProtectionLevel', '5'),
(359586015047188, '2021-02-02 07:52:32.777', 'SafeProtectDeviceConnected', 'False'),
(359586015047188, '2021-02-02 07:53:57.000', 'SafeProtectProtectionLevel', '5'),
(359586015047188, '2021-02-02 07:54:10.777', 'SafeProtectDeviceConnected', 'True');

解决方案

备注：你可以将 case 表达式缩小一点。为了清楚起见，这里完整地写出来。

with cte as
(
  select l.VehicleID,
         l.Time,
         l.ContextID,
         l.Value,
         (row_number() over(order by l.Time)-1)/3 as GroupNum,
         case
           when l.ContextID = 'SafeProtectDeviceConnected' and l.Value = 'False' then 1
           when l.ContextID = 'SafeProtectDeviceConnected' and l.Value = 'True'  then 2
           when l.ContextID = 'SafeProtectProtectionLevel'                       then 3
         end as GroupSort
  from log l
)
select cte.VehicleID,
       cte.Time,
       cte.ContextID,
       cte.Value
from cte
order by cte.GroupNum,
         cte.GroupSort;

结果

VehicleID        Time                     ContextID                   Value
---------------  -----------------------  --------------------------  -----
359586015047188  2021-02-01 07:27:14.777  SafeProtectDeviceConnected  False
359586015047188  2021-02-01 07:53:47.777  SafeProtectDeviceConnected  True
359586015047188  2021-02-01 07:53:38.000  SafeProtectProtectionLevel  5
359586015047188  2021-02-01 10:24:20.777  SafeProtectDeviceConnected  False
359586015047188  2021-02-01 10:26:55.777  SafeProtectDeviceConnected  True
359586015047188  2021-02-01 10:26:46.000  SafeProtectProtectionLevel  5
359586015047188  2021-02-01 10:43:53.777  SafeProtectDeviceConnected  False
359586015047188  2021-02-01 10:46:09.777  SafeProtectDeviceConnected  True
359586015047188  2021-02-01 10:46:01.000  SafeProtectProtectionLevel  5
359586015047188  2021-02-01 11:02:16.777  SafeProtectDeviceConnected  False
359586015047188  2021-02-01 14:39:42.777  SafeProtectDeviceConnected  True
359586015047188  2021-02-01 14:39:41.777  SafeProtectProtectionLevel  5
359586015047188  2021-02-01 14:55:48.777  SafeProtectDeviceConnected  False
359586015047188  2021-02-02 07:52:12.777  SafeProtectDeviceConnected  True
359586015047188  2021-02-02 07:52:12.777  SafeProtectProtectionLevel  5
359586015047188  2021-02-02 07:52:32.777  SafeProtectDeviceConnected  False
359586015047188  2021-02-02 07:54:10.777  SafeProtectDeviceConnected  True
359586015047188  2021-02-02 07:53:57.000  SafeProtectProtectionLevel  5

Fiddle 了解实际情况。

【讨论】：

能否请您澄清在 GroupNum 计算中使用的 -1 和 3 值的选择？
row_number() 函数产生1, 2, 3, 4, 5, 6, ... 减一使其从零开始0, 1, 2, 3, 4, 5, ... 整数除以三（一组中的行数）将其变为0, 0, 0, 1, 1, 1, ... 没有初始减法只有 2 个零，分组不正确。
谢谢，我必须确认我在真实数据中的一组只有3行。
我发现我的“SafeProtectProtectionLevel”行数可能未知。所以@Tyron78 的答案最适合我的情况。