Iterating over groups in table - sql

Question

I have the following data:
cte1
===========================
m_ids |p_id |level
---------|-----------|-----
{123} |98 |1
{123} |111 |2
{432,222}|215 |1
{432,222}|215 |1
{432,222}|240 |2
{432,222}|240 |2
{432,222}|437 |3
{432,222}|275 |3
I have to perform the following operation:
Extract p_id by the following algorithm
For every row with same m_ids
In each group:
2.I. Group records by p_id
2.II. Order desc records by level
2.III. Select p_id with exact count as the m_ids length and with the biggest level
So far I fail to write this algorithm completely, but I wrote (probably wrong where I'm getting array_length) this for the last part of it:
SELECT id
FROM grouped_cte1
GROUP BY id,
level
HAVING Count(*) = array_length(grouped_cte1.m_ids, 1)
ORDER BY level DESC
LIMIT 1
where grouped_cte1 for m_ids={123} is
m_ids |p_id |level
---------|-----------|-----
{123} |98 |1
{123} |111 |2
and for m_ids={432,222} is
m_ids |p_id |level
---------|-----------|-----
{432,222}|215 |1
{432,222}|215 |1
{432,222}|240 |2
{432,222}|240 |2
{432,222}|437 |3
{432,222}|275 |3
etc.
2) Combine query from p.1 with the following. The following extracts p_id with level=1 for each m_ids:
select m_ids, p_id from cte1 where level=1 --also selecting m_ids for joining later`
which results in the following:
m_ids |p_id
---------|----
{123} |98
{432,222}|215
Desirable result:
m_ids |result_1 |result_2
---------|-----------|--------
{123} |111 |98
{432,222}|240 |215
So could anyone please help me solve the first part of algorithm and (optionally) combine it in a single query with the second part?
EDIT: So far I fail at:
1. Breaking the presented table into subtables by m_ids while iterating over it.
2. Performing computation of array_length(grouped_cte1.m_ids, 1) for corresponding rows in query.

Answer 1

For the first part of the query you're on the right track, but you need to change the grouping logic and then join again to the table to filter it out by highest level per m_ids for which you could use DISTINCT ON clause combined with proper sorting:
select
distinct on (t.m_ids)
t.m_ids, t.p_id, t.level
from cte1 t
join (
select
m_ids,
p_id
from cte1
group by m_ids, p_id
having count(*) = array_length(m_ids, 1)
) as g using (m_ids, p_id)
order by t.m_ids, t.level DESC;
This would give you:
m_ids | p_id | level
-----------+------+-------
{123} | 111 | 2
{432,222} | 240 | 2
And then when combined with second query (using FULL JOIN for displaying purposes, when the first query is missing such conditions) which I modified by adding distinct since there can be (and in fact is) more than one record for m_ids, p_id pair with first level it would look like:
select
coalesce(r1.m_ids, r2.m_ids) as m_ids,
r1.p_id AS result_1,
r2.p_id AS result_2
from (
select
distinct on (t.m_ids)
t.m_ids, t.p_id, t.level
from cte1 t
join (
select
m_ids,
p_id
from cte1
group by m_ids, p_id
having count(*) = array_length(m_ids, 1)
) as g using (m_ids, p_id)
order by t.m_ids, t.level DESC
) r1
full join (
select distinct m_ids, p_id
from cte1
where level = 1
) r2 on r1.m_ids = r2.m_ids
giving you result:
m_ids | result_1 | result_2
-----------+----------+----------
{123} | 111 | 98
{432,222} | 240 | 215
that looks different from what you've expected but from my understanding of the logic it is the correct one. If I misunderstood anything, please let me know.
Just for the sake of logic explanation, one point:
Why m_ids with {123} returns 111 for result_1?
for group of m_ids = {123} we have two distinct p_id values
both 98 and 111 account for the condition of equality count with the m_ids length
p_id = 111 has a higher level, thus is chosen for the result_1