以下是一种在一个查询中格式化数据的解决方案。对于 4 行的小型数据集（我在您的示例中添加了一行），性能不是问题，但我猜这对于数百万条记录来说不会很快。
此处提供的解决方案以common table expressions (CTE) 的形式生成不同的数据集，并使用其他 StackOverflow 对remove the seconds 和concatenate the row values 的回答中的一些技术。最后加上cross apply。
可以按照与连续 CTE 的/连接对应的步骤来描述该方法：
- 为每个产品创建一组属性。
- 为每个产品创建一组周期开始时刻（省略秒数）。
- 将每个产品的属性与每个时期相结合，并寻找合适的值。
- 使用一些 XML 函数来格式化单行中的属性值。
- 使用
cross apply 获取期末。
完整解决方案：
-- sample data
declare @data table
(
ID int,
ProductId int,
Attribute nvarchar(10),
Value nvarchar(10),
SysStartTime datetime2(0),
SysEndTime datetime2(0)
);
insert into @data (ID, ProductId, Attribute, Value, SysStartTime, SysEndTime) values
(1, 8, 'Size', 'S', '2020-07-06 05:00:00', '9999-12-31 23:59:59'),
(2, 8, 'Color', 'Blue', '2020-07-06 05:00:01', '2020-07-09 11:11:11'),
(2, 8, 'Color', 'Green', '2020-07-09 11:11:11', '9999-12-31 23:59:59'),
(2, 8, 'Weight', 'Light', '2020-07-10 10:11:12', '9999-12-31 23:59:59'); -- additional data to have extra attribute not available from start
-- solution
with prodAttrib as -- attributes per product
(
select d.ProductId, d.Attribute
from @data d
group by d.ProductId, d.Attribute
),
prodPeriod as -- periods per product
(
select d.ProductId,
dateadd(minute, datediff(minute, 0, d.SysStartTime), 0) as 'SysStartTimeNS' -- start time No Seconds
from @data d
group by ProductId, dateadd(minute, datediff(minute, 0, d.SysStartTime), 0)
),
prodResult as -- attribute value per period per product
(
select pp.ProductId,
convert(nvarchar(16), pp.SysStartTimeNS, 120) as 'FromDateTime',
convert(nvarchar(16), coalesce(pe.SysEndTime, '9999-12-31 23:59:59'), 120) as 'ToDateTime',
pa.Attribute,
av.Value
from prodPeriod pp
join prodAttrib pa
on pa.ProductId = pp.ProductId
outer apply ( select top 1 d.Value
from @data d
where d.ProductId = pp.ProductId
and d.Attribute = pa.Attribute
and dateadd(minute, datediff(minute, 0, d.SysStartTime), 0) <= pp.SysStartTimeNS
order by d.SysStartTime desc ) av -- attribute values per product
outer apply ( select top 1 dateadd(second, -1, d.SysStartTime) as 'SysEndTime'
from @data d
where d.ProductId = pp.ProductId
and dateadd(minute, datediff(minute, 0, d.SysStartTime), 0) > pp.SysStartTimeNS
order by d.SysStartTime ) pe -- period end
),
prodResultFormat as -- concatenate attribute values per period
(
select pp.ProductId,
convert(nvarchar(16), pp.SysStartTimeNS, 120) as 'FromDateTime',
(
select pr.Attribute + ' = ' + coalesce(pr.Value,'') + ', ' as [text()]
from prodResult pr
where pr.ProductId = pp.ProductId
and pr.FromDateTime = convert(nvarchar(16), pp.SysStartTimeNS, 120)
order by pr.Attribute
for xml path('')
) as 'Attributes'
from prodPeriod pp
)
select prf.ProductId,
prf.FromDateTime,
x.ToDateTime,
left(prf.Attributes, len(prf.Attributes)-1) as 'Attributes'
from prodResultFormat prf
cross apply ( select top 1 pr.ToDateTime
from prodResult pr
where pr.ProductId = prf.ProductId
and pr.FromDateTime = prf.FromDateTime ) x
order by prf.ProductId, prf.FromDateTime;
扩展示例数据的结果：
ProductId FromDateTime ToDateTime Attributes
----------- ---------------- ---------------- ----------------------------------------
8 2020-07-06 05:00 2020-07-09 11:11 Color = Blue, Size = S, Weight =
8 2020-07-09 11:11 2020-07-10 10:11 Color = Green, Size = S, Weight =
8 2020-07-10 10:11 9999-12-31 23:59 Color = Green, Size = S, Weight = Light
附：如果您真的需要NULL 值，请将x.EndDateTime 替换为case when x.ToDateTime = '9999-12-31 23:59' then NULL else x.ToDateTime end as 'ToDateTime'。