提高LEFT JOIN / GROUP BY的性能

Question

I have two tables snippets and platforms . Each snippet belongs to a platform (fork_id is nullable and links to another record on the same table). Structure:

PLATFORMS (id, name, slug, syntax)
SNIPPETS (id, platform_id, fork_id, private etc.) 

I'm now trying to run a query to get the total number of snippets for each platform. The query is slow (between 10 - 20 seconds) when the snippets table has a million records.

SELECT platforms.id, name, slug, syntax, COUNT(*) AS total FROM platforms 
LEFT JOIN snippets on platforms.id = snippets.platform_id
WHERE fork_id IS NULL
AND private = 0
GROUP BY platforms.id, name
ORDER BY total DESC, name asc;

Some additional information:

• snippets.id and platform.id have primary key indexes.
• fork_id and platform_id have foreign key indexes.
• private and platforms.name have indexes.

Running an EXPLAIN query offers the following:

How can I get the performance to an acceptable level? Thanks!

Answer 1

In MySQL, this type of query can be faster with a correlated subquery:

SELECT p.id, p.name, p.slug, p.syntax,
       (SELECT COUNT(*)
        FROM snippets s
        WHERE p.id = s.platform_id AND
              s.fork_id IS NULL AND
              s.private = 0
       ) AS total
FROM platforms  p
ORDER BY total DESC, name asc;

Then, you want an index on snippets(platform_id, fork_id, private) .

I should note that your original query is equivalent to:

SELECT p.id, p.name, p.slug, p.syntax, COUNT(*) AS total
FROM platforms p JOIN
     snippets s
     on p.id = s.platform_id
WHERE s.fork_id IS NULL AND s.private = 0
GROUP BY p.id, p.name
ORDER BY total DESC, name asc;

Because the WHERE clause turns the LEFT JOIN into an INNER JOIN . For this query, you can try an index on snippets(private, fork_id, platform_id) .

Answer 2

Two things going on here I can see. One is the counting, and the other is the display of the details from the platforms table.

Let's count first.

                     SELECT platform_id, COUNT(*) snips
                       FROM snippets
                      WHERE fork_id IS NULL
                        AND private = 0
                      GROUP BY platform_id

To make this as fast as possible, create a compound index on the ( private, fork_id, platform_id) columns of the snippets table. That way the inner query can do a so-called index scan, about which you can read.

Let's report the details now that we have the counts.

       SELECT a.id, a.name, a.slug, a.syntax, b.snips
         FROM platforms a
    LEFT JOIN (
                     SELECT platform_id, COUNT(*) snips
                       FROM snippets
                      WHERE fork_id IS NULL
                        AND private = 0
                      GROUP BY platform_id
              ) b ON a.platform_id = b.platform_id
       ORDER BY b.snips DESC, a.name ASC;

The trick is: simplify simplify simplify when aggregating (grouping) large tables.