How to find missing pairs in many-to-many table? - sql-server

Question

Given: Users and Groups, as well as many-to-many GroupUsers table.
It is necessary that each pair of users has its own paired group, provided that there can be groups with more or less users.
How to check and create the missing paired groups?
Link to SQL Fiddle
create table dbo.Users (
Id int not null,
Name nvarchar(50) not null,
constraint PK_Users primary key clustered (Id)
);
create table dbo.Groups (
Id int not null,
Name nvarchar(50) not null,
constraint PK_Groups primary key clustered (Id)
);
create table dbo.GroupUsers (
GroupId int not null ,
UserId int not null ,
constraint PK_GroupUsers primary key clustered (GroupId, UserId),
constraint FK_GroupUsers_GroupId foreign key (GroupId) references dbo.Groups (Id),
constraint FK_GroupUsers_UserId foreign key (UserId) references dbo.Users (Id)
);
insert into dbo.Users values (1, 'Anna'), (2, 'Berta'), (3, 'Carlie'), (4, 'Dana'), (5, 'Emil');
insert into dbo.Groups values (1, 'Anna-Berta'), (2, 'Anna-Carlie'), (3, 'Anna-Berta-Carlie');
insert into dbo.GroupUsers values
(1,1), (1, 2), -- 'Anna-Berta' group
(2,1), (2, 3), -- 'Anna-Carlie' group
(3,1), (3, 2), (3, 3); -- 'Anna-Berta-Carlie' group
How to find and create missing "Pair Group" for user Anna?
So Anna has a Pair Group with all other users
and all the Anna's Pair Groups must have exactly two users,
although any user including Anna may have a group which includes more or less than two users.
UPDATE 2019-12-15:
So here (Link to SQL Fiddle) is my favorite solution (so far) of how to find missing paired groups and their users (Thanks to answer from #Kari F.)
declare #UserId int = 1;
with cte as
(
select
VirtualGroupId = row_number() over(order by p.Id desc) * -1,
GroupName = concat( u.Name, '_', p.Name),
CurrentUserId = u.Id,
OtherUserId = p.Id
from
dbo.Users u inner join dbo.Users p on u.Id = #UserId and p.Id <> #UserId
where
concat('_', u.Id, '_', p.Id) not in
(
select
(
select concat('_', gu.UserId)
from dbo.GroupUsers gu
where gu.GroupId = g.Id
order by case when gu.UserId = #UserId then 0 else 1 end
for xml path ('')
)
from
dbo.Groups g
)
)
select VirtualGroupId, GroupName, UserId = CurrentUserId from cte
union all
select VirtualGroupId, GroupName, UserId = OtherUserId from cte
order by VirtualGroupId, UserId

Answer 1

What about this simple solution:
select u.Id, p.Id
from dbo.Users u
cross join dbo.Users p
where u.Id = 1 and p.Id <> 1
and u.Name + '-' + p.Name not in (select g.Name from dbo.Groups g);
http://sqlfiddle.com/#!18/3c30f/16
Second solution taking into account you comment
with required_groups as (
select u.Id as userId, p.Id pairId
from dbo.Users u
cross join dbo.Users p
where u.Id = 1 and p.Id <> 1
),
existing_groups as (
select annas_groups.UserId as userId, pairs_groups.UserId as pairId
from (
select *
from GroupUsers gu
where gu.UserId = 1
and gu.GroupId in (
select x.GroupId from GroupUsers x
group by x.GroupId
having count(*) = 2
)
) annas_groups
inner join (
select *
from GroupUsers gu
where
gu.UserId <> 1
and gu.GroupId in (
select x.GroupId from GroupUsers x
group by x.GroupId
having count(*) = 2
)
) pairs_groups on annas_groups.GroupId = pairs_groups.GroupId
)
select *
from required_groups rg
where not exists (
select 1
from existing_groups eg
where eg.UserId = rg.UserId
and eg.PairId = rg.PairId
)
http://sqlfiddle.com/#!18/3c30f/34
A bit more complex but still readable, I think.

Answer 2

I have tried to solve the problem using a loop approach, this is because the id from the groups table must be supplied and is not an identity column otherwise the implementation becomes simpler.
I've also used a keyuser variable to handle the fact that it is "Anna" who must be paired and not some other user. This avoids simply hard coding the userid value.
create table #Users (
Id int not null,
Name nvarchar(50) not null,
constraint PK_Users primary key clustered (Id)
);
create table #Groups (
Id int not null,
Name nvarchar(50) not null,
constraint PK_Groups primary key clustered (Id)
);
create table #GroupUsers (
GroupId int not null ,
UserId int not null ,
constraint PK_GroupUsers primary key clustered (GroupId, UserId),
constraint FK_GroupUsers_GroupId foreign key (GroupId) references #Groups (Id),
constraint FK_GroupUsers_UserId foreign key (UserId) references #Users (Id)
);
insert into #Users values (1, 'Anna'), (2, 'Berta'), (3, 'Carlie'), (4, 'Dana'), (5, 'Emil');
insert into #Groups values (1, 'Anna-Berta'), (2, 'Anna-Carlie'), (3, 'Anna-Berta-Carlie');
insert into #GroupUsers values
(1,1), (1, 2), -- 'Anna-Berta' group
(2,1), (2, 3), -- 'Anna-Carlie' group
(3,1), (3, 2), (3, 3); -- 'Anna-Berta-Carlie' group
declare #sql nvarchar(max) = '';
declare #nextgroupid int = 0;
declare #keyuser int = 0;
select #keyuser = ID from #Users where [Name]='Anna';
while exists (select 1 from #Users u inner join #Users u2 on u.Id=1 where u2.Id not in (select userid from #GroupUsers))
begin
select #nextgroupid = MAX(id)+1 from #Groups;
set #sql = 'insert #groups (id, [name]) select top 1 ' + CAST(#nextgroupid as nvarchar(3)) + ', u.[name] + ''-'' + u2.[name] as missingusergroup from #Users u inner join #Users u2 on u.Id= ' + CAST(#keyuser as nvarchar(3)) + ' where u2.Id not in (select userid from #GroupUsers) order by u2.id;';
exec(#sql);
set #sql = 'insert #groupusers (groupid, userid) select top 1 ' + CAST(#nextgroupid as nvarchar(3)) + ', u2.id from #Users u inner join #Users u2 on u.Id= ' + CAST(#keyuser as nvarchar(3)) + ' where u2.Id not in (select userid from #GroupUsers) union select top 1 '+ CAST(#nextgroupid as nvarchar(3)) +', u.id from #Users u where u.id= ' + CAST(#keyuser as nvarchar(3)) + ' order by u2.id;';
exec(#sql);
end
select * from #GroupUsers;
select * from #Groups;
drop table #Groups, #GroupUsers, #Users;