SQL server: Question about query group and auto id - sql

Question

i have a table which has 2 columns like this, picture is input and output:
Explain input: 2 column is 2 person who relation ship together. Exam: A relation with B, C,D,H
Output: i want to merge 2 column , with Column group ID auto and column RelationShip
ID group auto: i tried query: row_number() OVER (ORDER BY [columnA]) n
RelationShip: is group all person relatioship together and group them to 1 ID group. Exam person A,B,C,D,H realation ship together and group id auto is 1
I have a demo in https://dbfiddle.uk/?rdbms=sqlserver_2019&fiddle=f81ff386b07589654ad133a8e4b30472
In my query, first i turn the data into a single column, second i count letter, next step i don't know query.
I tried recursive CTE but error (perhaps I don't understand clearly about recursive)
I want to solve this problem in sql query.

Answer 1

You can use a recursive CTE to assign each "column" value a number. This starts by created edges in both directions and then following them:
with pairs as (
select columnA, columnB from myTable
union -- on purpose to remove duplicates
select columnB, columnA from myTable
),
cte as (
select distinct columnA, columnA as columnB,
convert(varchar(max), ',' + columnA + ',') as visited,
1 as lev
from pairs
union all
select cte.columnA, p.columnB, concat(visited, p.columnA, ','), lev + 1
from cte join
pairs p
on cte.columnB = p.columnA
where cte.visited not like concat('%,', p.columnB, ',%') and
lev < 5
)
select cte.columnA, min(cte.columnB),
dense_rank() over (order by min(cte.columnB))
from cte
group by cte.columnA;
You can then join this in to assign a group id:
with pairs as (
select columnA, columnB from myTable
union -- on purpose to remove duplicates
select columnB, columnA from myTable
),
cte as (
select distinct columnA, columnA as columnB,
convert(varchar(max), ',' + columnA + ',') as visited,
1 as lev
from pairs
union all
select cte.columnA, p.columnB, concat(visited, p.columnA, ','), lev + 1
from cte join
pairs p
on cte.columnB = p.columnA
where cte.visited not like concat('%,', p.columnB, ',%') and
lev < 5
),
groups as (
select cte.columnA, min(cte.columnB) as min_columnB,
dense_rank() over (order by min(cte.columnB)) as group_id
from cte
group by cte.columnA
)
select t.*, g.group_id
from mytable t join
groups g
on g.columnA = t.columnA;
Here is a db<>fiddle.