使用 SQL 根据各个组的传递/合并分配组号

Question

Consider the following table

create table #table 
(
     Name varchar(100), 
     Group1 bigint, 
     Group2 bigint, 
     Group3 bigint
)

insert into #table values('Adam', 276328, 00001, 0)
insert into #table values('Bob', 276328, 00002, 0)
insert into #table values('Catherine', 927356, 00002, 0)
insert into #table values('Dave', 927356, 00003, 0)
insert into #table values('Eleanor', 927379, 00003, 0)
insert into #table values('Krampus', 927390, 00004, 0)

I'm trying to update Group3 column, with a newly autogenerated group number.

For instance, initially the names are pre-grouped as "Adam, Bob", "Catherine, Dave", "Eleanor" and then "Krampus" which is evident from Group 1 .

Group 2 then groups the Names as "Adam", "Bob & Catherine", "Dave & Eleanor" and then "Krampus"

What I'm trying to do is, use Group 1 and Group 2 to assign a new "Group 3" that transitively groups all of them.

Something like "Adam & Bob & Catherine & Dave & Eleanor" because Adam is grouped with Bob on Group 1 and Bob is grouped with Catherine on Group 2 which means, they're all transitively grouped together.

An output like:

'Adam', 276328, 00001, 00001
'Bob', 276328, 00002, 00001
'Catherine', 927356, 00002, 00001
'Dave', 927356, 00003, 00001
'Eleanor', 927379, 00003, 00001
'Krampus', 927390, 00004, 00002

I've played around with Rank() , Dense_Rank() - but cannot find a way to transitively group all of them.

Answer 1

This is a graph walking problem, because A and B could be in group 1. Then A and C could be in group 2. Then C and D could be in group 1.

That suggests a recursive CTE:

with cte as (
      select name, group1, group2, convert(varchar(max), ',' + name + ',') as visited, name as grouping
      from t
      union all
      select cte.name, t.group1, t.group2, concat(visited, t.name, ','),
             (case when cte.name < t.name then cte.name else t.name end)
      from cte join
           t
           on cte.group1 = t.group1 or
              cte.group2 = t.group2
      where visited not like '%,' + t.name + ',%'
     )
select name, dense_rank() over (order by min(grouping)) as grp
from cte
group by name;

Here is a db<>fiddle.