Zip/repeat join? - sql

Question

Let's say I have a simple table of documents with a type column:
Documents
Id Type
1 A
2 A
3 B
4 C
5 C
6 A
7 A
8 A
9 B
10 C
Users have permissions to access different types of documents:
Permissions
Type User
A John
A Jane
B Sarah
C Peter
C John
C Mark
And I need to distribute those documents among the users as tasks:
Tasks
Id T DocId UserId
1 A 1 John
2 A 2 Jane
3 B 3 Sarah
4 C 4 Peter
5 C 5 John
6 A 6 John
7 A 7 Jane
8 A 8 John
9 B 9 Sarah
10 C 10 Mark
How do I do that? How do I get the Tasks?

Answer 1

You can enumerate the rows and then use modulo arithmetic for the matching:
with d as (
select d.*,
row_number() over (partition by type order by newid()) as seqnum,
count(*) over (partition by type) as cnt
from documents d
),
u as (
select u.*,
row_number() over (partition by type order by newid()) as seqnum,
count(*) over (partition by type) as cnt
from users u
)
select d.*
from d join
u
on d.type = u.type and
u.seqnum = (d.seqnum % u.cnt) + 1

Answer 2

Great question.
This solution returns all possible distributions, ordered by priority which is determined by information such as number of user involved, minimum documents per user, standard deviation of tasks per user etc.
I'm not counting on document.id to be a sequence of numbers starting with 1, therfore the use of dense_rank.
The core of the solutions is the iterative CTE which generates the record sets of all possible distributions.
Execution time on my laptop is around 20 seconds, (the iterative part takes 5 seconds)
with doc_user as
(
select d."id" as docid
,p."user" as userid
,dense_rank () over (order by d."id") as doc_seq
from documents d
left join permissions p
on p.type = d.type
)
,it_cte as
(
select docid
,userid
,doc_seq
,cast (coalesce(userid,'') as varchar(max)) as path
,'A' as cte_part
from doc_user
where doc_seq = 1
union all
select r.docid
,r.userid
,du.doc_seq
,r.path + ',' + coalesce (du.userid,'')
,'B'
from it_cte as r
cross join doc_user as du
where du.doc_seq = r.doc_seq + 1
union all
select du.docid
,du.userid
,du.doc_seq
,r.path + ',' + coalesce (du.userid,'')
,'C'
from it_cte as r
cross join doc_user as du
where du.doc_seq = r.doc_seq + 1
and r.cte_part in ('A','C')
)
,result_sets as
(
select dense_rank () over (order by path) as set_id
,docid
,userid
from it_cte
where doc_seq = (select count(*) from documents)
)
,result_sets_stat as
(
select set_id
,count (distinct userid) as users_involved
from result_sets
group by set_id
)
,result_sets_users_stat as
(
select set_id
,min (doc) min_doc_per_user
,stdevp (doc) stdevp_doc_per_user
from (select set_id
,userid
,count (*) as doc
from result_sets
group by set_id
,userid
) t
group by set_id
)
select s.set_priority
,r.docid
,r.userid
,s.users_involved
,s.min_doc_per_user
,s.stdevp_doc_per_user
from (select s.set_id
,s.users_involved
,u.min_doc_per_user
,u.stdevp_doc_per_user
,row_number () over
(
order by s.users_involved desc
,u.min_doc_per_user desc
,u.stdevp_doc_per_user
,s.set_id
) as set_priority
from result_sets_stat as s
join result_sets_users_stat as u
on u.set_id =
s.set_id
) s
join result_sets as r
on r.set_id =
s.set_id
order by s.set_priority
,r.docid
option (merge join)
;