How can I merge these two left joins into a single one? - mysql

Question

How can I merge these two left joins: http://sqlfiddle.com/#!9/1d2954/69/0
SELECT d.`id`, (adcount + bdcount)
FROM `docs` d
LEFT JOIN
(
SELECT da.`doc_id`, COUNT(da.`doc_id`) AS adcount FROM `docs_scod_a` da
INNER JOIN `scod_a` a ON a.`id` = da.`scod_a_id`
WHERE a.`ver_a` IN ('AA', 'AB')
GROUP BY da.`doc_id`
) ad ON ad.`doc_id` = d.`id`
LEFT JOIN
(
SELECT db.`doc_id`, COUNT(db.`doc_id`) AS bdcount FROM `docs_scod_b` db
INNER JOIN `scod_b` b ON b.`id` = db.`scod_b_id`
WHERE b.`ver_b` IN ('BA', 'BB')
GROUP BY db.`doc_id`
) bd ON bd.`doc_id` = d.`id`
to be a Single left join just to ease its use in my code, while making it no less slower?

Answer 1

Let me first emphasize that your method of doing the calculation is the better method. You have two separate dimensions and aggregating them separately is often the most efficient method for doing the calculation. It is also the most scalable method.
That said, your query should be equivalent to this version:
SELECT d.id,
count(distinct a.id),
count(distinct b.id)
FROM docs d left join
docs_scod_a da
ON da.doc_id = d.id LEFT JOIN
scod_a a
ON a.id = da.scod_a_id AND a.ver_a IN ('AA', 'AB') LEFT JOIN
docs_scod_b db
ON db.doc_id = d.id LEFT JOIN
scod_b b
ON b.id = db.scod_b_id AND b.ver_b IN ('BA', 'BB')
GROUP BY d.id
ORDER BY d.id;
This query is more expensive than it looks, because the COUNT(DISTINCT) incurs additional overhead compared to COUNT().
And here is the SQL Fiddle.
And, because LEFT JOIN can return NULL values, your query is more correctly written as:
SELECT d.`id`, COALESCE(adcount, 0) + COALESCE(bdcount, 0)
If you were having problems with the results, this small change might fix those problems.

Answer 2

Performance may be a big problem, depending on sizes of each table. It appears to be an "inflate-deflate" situation since it first "inflates" the number of rows via JOIN, then "deflates" via GROUP BY. The formulation below avoids inflation-deflation.
But first, if I understand this subquery correctly, this
SELECT da.`doc_id`, COUNT(da.`doc_id`) AS adcount
FROM `docs_scod_a` da
INNER JOIN `scod_a` a ON a.`id` = da.`scod_a_id`
WHERE a.`ver_a` IN ('AA', 'AB')
GROUP BY da.`doc_id`
can be rewritten as
SELECT `doc_id`,
( SELECT COUNT(*)
FROM `scod_a`
WHERE `id` = da.`scod_a_id`
AND `ver_a` IN ('AA', 'AB')
) AS adcount
FROM `docs_scod_a` AS da
If that is correct, then the entire query becomes
SELECT d.id,
( SELECT COUNT(*)
FROM docs_scod_a ds
JOIN scod_a s ON s.id = ds.scod_a_id
WHERE ds.doc_id = d.id
AND s.ver_a IN ('AA', 'AB')
) +
( SELECT COUNT(*)
FROM docs_scod_b ds
JOIN scod_b s ON s.id = ds.scod_b_id
WHERE ds.doc_id = d.id
AND s.ver_b IN ('BA', 'BB')
)
FROM docs AS d
Which needs these indexes:
docs_scod_a: (doc_id, scod_a_id), (scod_a_id, doc_id)
docs_scod_b: (doc_id, scod_b_id), (scod_b_id, doc_id)
scod_a: (ver_a, id)
scod_b: (ver_b, id)
docs: -- presumably has PRIMARY KEY(id)
Note the lack of GROUP BY.
docs_scod_a smells like a many-to-many mapping table. I recommend you follow the tips here.
(No COALESCE is needed since COUNT will simply return zero.)
(I don't know whether my version is better (faster or whatever) than Gordon's, nor whether my indexes will help his formulation.)