Join a dynamic number of rows in postgres - sql

Question

Let's say I have the following tables:
Batch Items
---+----- ---+----------+--------
id | size id | batch_id | quality
---+----- ---+----------+--------
1 | 10 1 | 1 | 9
2 | 2 2 | 1 | 10
3 | 2 | 1
4 | 2 | 2
5 | 2 | 1
6 | 2 | 9
I have batches of items. They are sent by batches of size batch.size. An item is broken if it's quality is <= 3.
I want to know the number of broken items in the last batches sent:
batch_id | broken_item_count
---------+---------------------
1 | 0
2 | 2 (and not 3)
My idea is the following:
SELECT batch.id as batch_id, COUNT(broken_items.*) as broken_item_count
FROM batch
INNER JOIN (
SELECT id
FROM items
WHERE items.quality <= 3
ORDER BY items.id asc
LIMIT batch.size -- invalid reference to FROM-clause entry for table "batch"
) broken_items ON broken_items.batch_id = batch.id
(I would ORDER BY items.shipped_at. But for simplicity, I order by items.id)
But this query shows me the error I put as the comment.
How can I limit the number of joined items based on the batch.size that is different for each row ?
Is there any other way to achieve what I want ?

Answer 1

SELECT b.id AS batch_id
, count(i.quality < 4 OR NULL) AS broken_item_count
FROM batch b
LEFT JOIN (
SELECT batch_id, quality
, row_number() OVER (PARTITION BY batch_id ORDER BY id DESC) AS rn
FROM items
) i ON i.batch_id = b.id
AND i.rn <= b.size
GROUP BY 1
ORDER BY 1;
SQL Fiddle with added examples.
This is much like #Clodoaldos's answer, but with a couple of differences. Most importantly:
You want to count the broken items in the last batches sent, so we have to ORDER BY id DESC
If there can be batches without items at all you need to use LEFT JOIN instead of a plain JOIN or those batches are excluded.
Consequently, the check i.rn <= b.size needs to move from the WHERE clause to the JOIN clause.

Answer 2

SQL Fiddle
select
b.id as batch_id,
count(quality <= 3 or null) as broken_item_count
from
batch b
inner join (
select
id, quality, batch_id,
row_number() over (partition by batch_id order by id) as rn
from items
) i on i.batch_id = b.id
where rn <= b.size
group by b.id
order by b.id

Answer 3

From what I understand the count of defective items cannot be greater than the batch size.
EDIT: After reading your comments, I think using the RANK() function, and then join by rank and size should work for you. The following query attempts that.
SELECT b.id,
SUM(CASE WHEN i1.quality <= 3 THEN 1 ELSE 0END) as broken_item_count
FROM BATCH as b
LEFT JOIN (SELECT i.id, i.batch_id, i.quality,
RANK() OVER(PARTITION BY i.batch_id ORDER BY i.id) as RANK
FROM ITEMS as i) as i1 ON b.id = i1.batch_id AND i1.RANK <= b.size
GROUP BY b.id
EDIT2: Updated the query with a LEFT JOIN to cover the case where there are no samples in some batch.