MySQL: How do I Optimize this JOIN Query?-CodePudding

I have 3 tables; artist, album, song_cover, and song. I would like to select an album and the total number of songs in that album. Am currently using this query, but it is logged in the mysql-slow.log file. In the PHPMyAdmin, the query speed is inconsistent. Sometimes it will execute for 0.0005 seconds and other times, 2 seconds or more.

SELECT /*  MAX_EXECUTION_TIME(1000) */ album.*, 
artist_id, artist_aka, artist_slug, artist_profile_image, cover_filename,
(
   SELECT COUNT(*)
   FROM song
   WHERE song.song_album_id = album.album_id
) AS TotalSongs
FROM album 
LEFT JOIN artist ON album.album_artist = artist.artist_id 
LEFT JOIN song_cover ON album.album_cover_id = song_cover.cover_id 
ORDER BY album_id DESC LIMIT 0, 11

ROWS artist: 15,978, album: 14,167, song: 67,559, song_cover: 12,668

EXPLAIN

Thank you in advance.

CodePudding user response：

I would write it this way:

EXPLAIN SELECT b.*,
  a.artist_id, a.artist_aka, a.artist_slug, a.artist_profile_image, 
  c.cover_filename,
  COUNT(*) AS TotalSongs
FROM album AS b
INNER JOIN artist AS a ON b.album_artist = a.artist_id
LEFT OUTER JOIN song AS s ON s.song_album_id = b.album_id
LEFT OUTER JOIN song_cover AS c ON b.album_cover_id = c.cover_id
GROUP BY b.album_id
ORDER BY b.album_id DESC LIMIT 0, 11;

This eliminates the dependent subquery, in favor of another join and GROUP BY.

Here's the EXPLAIN report as near as I can guess at it:

 ---- ------------- ------- ------------ -------- ------------------------------------- --------------- --------- ----------------------- ------ ---------- --------------------- 
| id | select_type | table | partitions | type   | possible_keys                       | key           | key_len | ref                   | rows | filtered | Extra               |
 ---- ------------- ------- ------------ -------- ------------------------------------- --------------- --------- ----------------------- ------ ---------- --------------------- 
|  1 | SIMPLE      | b     | NULL       | index  | PRIMARY,album_cover_id,album_artist | PRIMARY       | 4       | NULL                  |    1 |   100.00 | Backward index scan |
|  1 | SIMPLE      | a     | NULL       | eq_ref | PRIMARY                             | PRIMARY       | 4       | test.b.album_artist   |    1 |   100.00 | NULL                |
|  1 | SIMPLE      | s     | NULL       | ref    | song_album_id                       | song_album_id | 4       | test.b.album_id       |    1 |   100.00 | Using index         |
|  1 | SIMPLE      | c     | NULL       | eq_ref | PRIMARY                             | PRIMARY       | 4       | test.b.album_cover_id |    1 |   100.00 | NULL                |
 ---- ------------- ------- ------------ -------- ------------------------------------- --------------- --------- ----------------------- ------ ---------- ---------------------

I have no data in my tables, so the row counts are trivial.

There's still a problem that it's doing an index-scan of album, which in your case is 14,167 rows. That could be costly.

But the other joins are all using indexes. Two of them are type: eq_ref, indicating that it's joining to the primary key of those tables.

I changed the join to artist to an inner join. I can't see how an album could not reference an artist. But I suppose it's possible for an album to have no songs, hence the outer join.

I find it strange that you join album directly to song_cover. Wouldn't song_cover also need to reference the original song it's a cover of?