Find rows with a group values from other table - sql

Question

id
role
Group_ID
1
A
1
2
B
1
3
A
2
4
D
2
5
A
3
6
B
3
7
C
3
8
C
4
...
User_id
role
user1
A
user1
B
user2
C
user2
D
user3
A
user3
D
user4
C
user5
A
user5
B
user5
C
user5
D
...
I have 2 tables Table1 and Table2 as shown above.
My requirement is to get the User_ID from the table2 which has all the roles from a group. Additionally, only those groups need to be checked which has at least 2 roles. If a group_ID has only 1 role then it should not be considered
For example, this is how the result will look like from above 2 tables
user1 has both the roles from group 1 (A,B) -> therefore it is in the results.
user3 has both the roles from group 2 (A,D) -> therefore it is in the results.
user5 has all the roles from group 1(A,B), 2(A,D) and 3(A,B,C) -> therefore it is in the results.
User2 has role C and D which is not a group, hence not shown in the result
User4 has role C which is a group (Group_ID = 4), but the group should have at least 2 roles, hence not shown in the result
User_id
Group_ID
user1
1
user3
2
user5
1
user5
2
user5
3
....
Select Table2.USER_ID,Table1.GROUP_ID
from Table2,
Table1
Where Table2.ROLE = Table1.ROLE
group by Table1.GROUP_ID,Table2.USER_ID
With the above query, I am able to get the records with user_id assigned any of the role. However, I want to get the User_ID from the table2 which has all the roles from a group.
Any help is much appreciated. I will also make sure to accept the answer leading me to the solution

Answer 1

A direct translation of the requirement into SQL (Oracle style)
with g as (
select group_id, count(*) role_cnt from table1 group by group_id having count(*) > 1),
u as (
select u.user_id, g.group_id, count(*) usergroup_role_cnt from table1 g, table2 u
where u.role=g.role
group by u.user_id, g.group_id)
select u.user_id, g.group_id from g, u
where u.group_id=g.group_id
and u.usergroup_role_cnt=g.role_cnt;
Edit:
There has been concern about query performance using "group by" when tables are "very large". I don't know how "very large" is defined. Anyway, I did a test on Oracle Clould Free Tier ATP database 19c, using tables with tens of thousand rows. Here is the result.
select count(*) from table1;
COUNT(*)
--------
289327
Elapsed: 00:00:00.007
1 rows selected.
select count(*) from table2;
COUNT(*)
--------
1445328
Elapsed: 00:00:00.024
1 rows selected.
with g as (
select group_id, count(*) role_cnt from table1 group by group_id having count(*) > 1),
u as (
select u.user_id, g.group_id, count(*) usergroup_role_cnt from table1 g, table2 u
where u.role=g.role
group by u.user_id, g.group_id)
select u.user_id, g.group_id from g, u
where u.group_id=g.group_id
and u.usergroup_role_cnt=g.role_cnt;
USER_ID GROUP_ID
------- --------
user99 994
user97 462
user97 199
user87 35
user99 462
user87 179
user99 199
user98 199
user96 199
user87 813
user87 941
user96 994
user97 994
user96 462
user98 462
user98 994
Elapsed: 00:00:04.770
16 rows selected.
select a2.user_id, a1.group_id
from (
select user_id, cast(collect(role) as role_list) as user_roles
from table2
group by user_id
) a2
inner join
(
select group_id, cast(collect(role) as role_list) as group_roles
from table1
group by group_id
) a1
on a1.group_roles submultiset of a2.user_roles
order by user_id, group_id
;
USER_ID GROUP_ID
------- --------
user87 35
user87 179
user87 813
user87 941
user96 199
user96 462
user96 994
user97 199
user97 462
user97 994
user98 199
user98 462
user98 994
user99 199
user99 462
user99 994
Elapsed: 00:01:35.395
16 rows selected.
select q2.user_id, q1.group_id
from (select distinct user_id from table2) q2
cross join
(select distinct group_id from table1) q1
where not exists
(
select role
from table1
where group_id = q1.group_id
and role not in
(
select role
from table2
where user_id = q2.user_id
)
)
order by user_id, group_id
;
(Manually cancelled after 10 min)
Data is generated quite randomly and is somewhat skewed. But in reality, skewed data is not avoidable. Statistics should minimize the impact. (Both tables were analyzed and have no indexes)

Answer 2

Here is a solution that will work in Oracle Database; adapt it for SQL Server (if that is possible; I don't know that dialect).
Test data (others may use this too):
create table table1(id number, role varchar2(10), group_id number);
insert into table1 (id, role, group_id)
select 1, 'A', 1 from dual union all
select 2, 'B', 1 from dual union all
select 3, 'A', 2 from dual union all
select 4, 'D', 2 from dual union all
select 5, 'A', 3 from dual union all
select 6, 'B', 3 from dual union all
select 7, 'C', 3 from dual
;
create table table2 (user_id varchar2(20), role varchar2(10));
insert into table2 (user_id, role)
select 'user1', 'A' from dual union all
select 'user1', 'B' from dual union all
select 'user2', 'C' from dual union all
select 'user2', 'D' from dual union all
select 'user3', 'A' from dual union all
select 'user3', 'D' from dual union all
select 'user4', 'C' from dual union all
select 'user5', 'A' from dual union all
select 'user5', 'B' from dual union all
select 'user5', 'C' from dual union all
select 'user5', 'D' from dual
;
commit;
Create user-defined data type (collection of strings representing roles):
create or replace type role_list as table of varchar2(10);
/
Query and output:
select a2.user_id, a1.group_id
from (
select user_id, cast(collect(role) as role_list) as user_roles
from table2
group by user_id
) a2
inner join
(
select group_id, cast(collect(role) as role_list) as group_roles
from table1
group by group_id
) a1
on a1.group_roles submultiset of a2.user_roles
order by user_id, group_id
;
USER_ID GROUP_ID
------------ ----------
user1 1
user3 2
user5 1
user5 2
user5 3
The strategy is pretty obvious, and should be easy to read directly from the code. Group the roles by group_id in the first table and by user_id in the second. Identify all the pairs (user, group) where all roles for the group are found in the role list of the user - that is exactly what the submultiset comparison operator does.
A more rudimentary query (harder to follow and maintain, and likely slower), but perhaps helpful as it is likely to work with very few changes - if any - in pretty much all SQL dialects, might look like this. Assuming role can't be null in table2 (to make the query slightly simpler):
select q2.user_id, q1.group_id
from (select distinct user_id from table2) q2
cross join
(select distinct group_id from table1) q1
where not exists
(
select role
from table1
where group_id = q1.group_id
and role not in
(
select role
from table2
where user_id = q2.user_id
)
)
order by user_id, group_id
;