Create a unique identifier for groups from union - sql-server-2008-r2

Question

Is there a way to get a unique ID for each [Group] to be generated dynamically? I have tried newid() but the result is a different value for each row. The intent is to not have to worry about adding a unique value for each set of T (as indicated by the set which is commented out). (The initial character for each value of [Value] (ex., 1) corresponds to the hard-coded value of [Group] only for readability purposes when viewing the results.)
select
[Name],
[Value],
[Group]
from (
select
[Name],
[Value],
1 as [Group] --newid() as [Group] --
from (
values
('abc', '1qwert'),
('def', '1yuiop')
) as T ([Name], [Value])
union all
select
[Name],
[Value],
2 as [Group] --newid() as [Group] --
from (
values
('abc', '2asdfg'),
('def', '2hjkl')
) as T ([Name], [Value])
union all
select
[Name],
[Value],
3 as [Group] --newid() as [Group] --
from (
values
('abc', '3zxcv'),
('def', '3bnm')
) as T ([Name], [Value])
--union all
--select
-- [Name],
-- [Value],
-- ? as [Group] --newid() as [Group] --
--from (
-- values
-- ('abc', '?poiuy'),
-- ('def', '?trewq')
-- ) as T ([Name], [Value])
) as TT
order by
[Group],
[Name];

Answer 1

Generally speaking, the function NEWID() generate the unique value for every row, when we put it as a column in the select list:
select
NEWID() as column_name
from table
There is a common, well known issue with the strange behavior of the function NEWID().
So we actually can't force query optimizer to materialize NEWID() value inside from clause:
select
t.*,
t2.col
from table t
cross join (select NEWID() as col) t2
In this case we get again the same unique values for every row.
Still, we have to find a way to force optimizer to calculate the NEWID() value inside the from clause.
As we know from logical query processing order the "ORDER" operator executed almost at the end of the query, after SELECT clause. We have to know values before we would like to sort them.
So first, we should somehow sort "select" statement inside query by NEWID() value.
The only legal possibility to do this is to add TOP 1 operator.
Second, to be sure that the optimizer won't skip ordering due to only 1 value, we add another NEWID() value.
So, to every union select we should add:
cross join (select top 1 t.group_id from (values (NEWID()), (NEWID()))as t(group_id) ORDER BY group_id) as tt
In this case the optimizer should, no matter what, sort values, which it should know beforehand.
And the query looks like this:
select
[Name],
[Value],
[Group]
from (
select
[Name],
[Value],
tt.group_id as [Group]
from (
values
('abc', '1qwert'),
('def', '1yuiop')
) as T ([Name], [Value])
cross join (select top 1 t.group_id from (values (NEWID()), (NEWID()))as t(group_id) ORDER BY group_id) as tt
union all
select
[Name],
[Value],
tt.group_id as [Group]
from (
values
('abc', '2asdfg'),
('def', '2hjkl')
) as T ([Name], [Value])
cross join (select top 1 t.group_id from (values (NEWID()), (NEWID()))as t(group_id) ORDER BY group_id) as tt
union all
select
[Name],
[Value],
tt.group_id as [Group]
from (
values
('abc', '3zxcv'),
('def', '3bnm')
) as T ([Name], [Value])
cross join (select top 1 t.group_id from (values (NEWID()), (NEWID()))as t(group_id) ORDER BY group_id) as tt
) as t
In this case, we shouldn't materialize random unique values in the table.
And the final correction is to add
select top 1 t.group_id from (values (NEWID()), (NEWID()))as t(group_id) ORDER BY group_id
into CTE.
The final query will look like this:
;with unique_group as (
select top 1
t.group_id
from (values
(NEWID()),
(NEWID())
)as t(group_id)
order by group_id
)
select
[Name],
[Value],
[Group]
from (
select
[Name],
[Value],
tt.group_id as [Group]
from (
values
('abc', '1qwert'),
('def', '1yuiop')
) as T ([Name], [Value])
cross join unique_group as tt
union all
select
[Name],
[Value],
tt.group_id as [Group]
from (
values
('abc', '2asdfg'),
('def', '2hjkl')
) as T ([Name], [Value])
cross join unique_group as tt
union all
select
[Name],
[Value],
tt.group_id as [Group]
from (
values
('abc', '3zxcv'),
('def', '3bnm')
) as T ([Name], [Value])
cross join unique_group as tt
) as t

Answer 2

Give each subquery within your union a unique reference number, relate that number to a small set of newid() values that also share the same set of integers.
;with cte as (
select 1 as gid, newid() as [group]
union all
select gid + 1, newid()
from cte
where gid < 3
)
select
[Name],[Value],[Group]
from (
select
[Name],[Value], 1 as gid
from (
values
('abc', '1qwert'),
('def', '1yuiop')
) as T ([Name], [Value])
union all
select
[Name],[Value], 2 as gid
from (
values
('abc', '2asdfg'),
('def', '2hjkl')
) as T ([Name], [Value])
union all
select
[Name], [Value], 3 as gid
from (
values
('abc', '3zxcv'),
('def', '3bnm')
) as T ([Name], [Value])
) d
inner join cte on d.gid = cte.gid
see: https://rextester.com/IEPJC57700
+----+------+--------+--------------------------------------+
| | Name | Value | Group |
+----+------+--------+--------------------------------------+
| 1 | abc | 1qwert | 68f1884f-c3af-44b6-9856-2ca71615e4aa |
| 2 | def | 1yuiop | 68f1884f-c3af-44b6-9856-2ca71615e4aa |
| 3 | abc | 2asdfg | bf36a235-ef68-4c02-a502-8ba0009aa302 |
| 4 | def | 2hjkl | bf36a235-ef68-4c02-a502-8ba0009aa302 |
| 5 | abc | 3zxcv | 221ca4d2-ed7f-4654-9234-7081360f693c |
| 6 | def | 3bnm | 221ca4d2-ed7f-4654-9234-7081360f693c |
+----+------+--------+--------------------------------------+