仅在GROUP BY上不存在行A的情况下如何选择行B

Question

I'm passing through the following situation and have not found a good solution to this problem. I am going through a optimization of a API so am looking for fastest possible solution.

The following description is not exactly what I am doing, but I think it represents the problem well.

Let's say I have a table of products:

+----+----------+
| id |   name   |
+----+----------+
|  1 | product1 |
|  2 | product2 |
+----+----------+

And I have a table of attachments to each product, separate by language:

+----+----------+------------+-----------------------+
| id | language | product_id |     attachment_url    |
+----+----------+------------+-----------------------+
|  1 |    bb    |      1     |     image1_bb.jpg     |
|  1 |    en    |      1     |     image1_en.jpg     |
|  1 |    pt    |      1     |     image1_pt.jpg     |
|  2 |    bb    |      1     |     image2_bb.jpg     |
|  2 |    pt    |      1     |     image2_pt.jpg     |
+----+----------+------------+-----------------------+

What I intend to do is to get the correct attachment according to the language selected on the request. As you can see above, I can have several attachments to each product. We use Babel ( bb ) as a generic language, so every time I don't have a attachment to the right language, I should get the babel version. Is also important to consider that the Primary Key of the attachments table is a composite of id + language .

So, supposing I try to get all the data in pt , my first option to create a SQL query was:

SELECT p.id, p.name, 
    GROUP_CONCAT( '{',a.id,',',a.attachment_url, '}' ) as attachments_list 
FROM products p 
LEFT JOIN attachments a 
    ON (a.product_id=p.id AND (a.language='pt' OR a.language='bb')) 

The problem is that, with this query I always get the bb data and I only want to get it when there is no attachment on the right language.

I already tried to do a subquery changing attachments for:

(SELECT * FROM attachments GROUP BY id ORDER BY id ASC, language DESC)

but it doubles the time of the request.

I also tried using DISTINCT inside the GROUP_CONCAT , but it only works if the whole result of each row is equal, so it does not work for me.

Does anyone knows any other solution that I can apply directly into the query?

EDIT:

Combining the answers of @Vulcronos and @Barmar made the final solution at least 2x faster than the one I first suggested.

Just to add some context, for anybody else who is looking for it. I am using Phalcon. Because of it, I had a lot of trouble putting the pieces together, as Phalcon PHQL does not support subqueries, nor a lot of the other stuff I had to use.

For my scenario , where I had to deliver approximatelly 1.2MB of JSON content, with more than 2100 objects, using custom queries made the total request time up to 3x faster than Phalcon native relations management methods ( hasMany() , hasManyToMany() , etc.) and 10x faster than my original solution (which used a lot the find() method).

Answer 1

Try doing two joins instead of one:

SELECT p.id, p.name, 
    GROUP_CONCAT( '{',COALESCE(a.id, b.id),',',COALESCE(a.attachment_url, b.attachment_url), '}' ) as attachments_list 
FROM products p 
LEFT JOIN attachments a 
    ON (a.product_id=p.id AND a.language='pt') 
LEFT JOIN attachments b
    ON (a.product_id=p.id AND a.language='bb') 

and then using COALESCE to return b instead of a if a doesn't exist. You can also do it with a subselect if the above doesn't work.

Answer 2

OR conditions tend to make queries slow, because it's hard to optimize them with indexes. Try joining separately using the two different languages.

SELECT p.id, p.name, 
    IFNULL(apt.attachment_url, abb.attachment_url) AS attachment_url
FROM products AS p
JOIN attachments AS abb ON abb.product_id = p.id
LEFT JOIN attachments AS apt ON alang.product_id = p.id AND apt.language = 'pt'
WHERE abb.language = 'bb'

This assumes that all products have a bb attachment, while pt is optional.

Answer 3

I left out the join of Product, because it's not relevant for this problem. It's only needed to include the product name in the resultset.

SELECT a.product_id, a.id, a.attachment_url FROM attachments a
WHERE a.language = ?
OR (a.language = 'bb' 
   AND NOT EXISTS
       (SELECT * FROM attachments
        WHERE language = ?
        AND id = a.id
        AND product_id = a.product_id));

Notes: problems like this usually have many possible solutions. This is not necessarily the most efficient one.