How to speed up left join queries by indexing? - mysql

Question

At the moment I am experiencing some slower MySQL queries in my application which I want to speed up. Unfortunately I’m not quite sure which is the correct way to do it.
I have the following (fictitious) tables: Book, Page and Word.
Word is child of Page by word_page_id
Page is child of Book by page_book id
I already have individual indexes on page_book_id, word_page_id, book_user_id and book_flag_delete.
SELECT `book`.*, COUNT(word_id) AS `word_amount` FROM `book`
LEFT JOIN `page` ON page_book_id = book_id
LEFT JOIN `word` ON word_page_id = paragraph_id
WHERE (book_user_id = 1) AND (book_flag_delete IS NULL)
GROUP BY `book_id`
ORDER BY `book_id` ASC LIMIT 100
SELECT COUNT(DISTINCT `book_id`) AS `book_row_count` FROM `book`
LEFT JOIN `page` ON page_book_id = book_id
LEFT JOIN `word` ON word_page_id = page_id
WHERE (book_user_id = 59) AND (book_flag_delete IS NULL)
Any ideas how to speed up such queries?
Is there extra indexing involved?

Answer 1

Set indexes on the fields you use for joining.
Further make sure that these have both the same datatype, encoding, and collation, else the index will also not be used.
mysql> EXPLAIN <query> will show you the actually used fields (key column in output) and the available indexes (possible_keys output field).

Answer 2

For this query:
SELECT b.*, COUNT(w.word_id) AS `word_amount`
FROM `book` b LEFT JOIN
`page` p
ON p.page_book_id = b.book_id LEFT JOIN
`word` w
ON w.word_page_id = p.paragraph_id
WHERE (b.book_user_id = 1) AND (b.book_flag_delete IS NULL)
GROUP BY b.`book_id`
ORDER BY b.`book_id` ASC
LIMIT 100;
The best indexes are: book(user_id, book_flag_delete, book_id), page(page_book_id, paragraph_id), and word(word_page_id, word_id).
However, the overall group by might be expensive. You might try writing the query as:
SELECT b.*,
(SELECT COUNT(w.word_id)
FROM `page` p JOIN
`word` w
ON w.word_page_id = p.paragraph_id
WHERE p.page_book_id = b.book_id
) AS `word_amount`
FROM `book` b LEFT JOIN
WHERE (b.book_user_id = 1) AND (b.book_flag_delete IS NULL)
ORDER BY b.`book_id` ASC
LIMIT 100;
The same indexes indexes work here. But, this query should avoid a group by on all the data at once (instead, it uses the indexes for the aggregation).

Answer 3

The optimal schema for a many-to-many mapping table is
CREATE TABLE XtoY (
# No surrogate id for this table
x_id MEDIUMINT UNSIGNED NOT NULL, -- For JOINing to one table
y_id MEDIUMINT UNSIGNED NOT NULL, -- For JOINing to the other table
# Include other fields specific to the 'relation'
PRIMARY KEY(x_id, y_id), -- When starting with X
INDEX (y_id, x_id) -- When starting with Y
) ENGINE=InnoDB;
The details on 'why' are in my index cookbook

Answer 4

In your select you're gonna want to refrain from using the wildcard "*" to grab columns. Plus utilize aliases ALWAYS!! This will keep your db from having to create a "virtual" alias.
select book1.column1, book1.column2, page1.column1
from book book1
left join page page1
on page1.page_book_id = book1.book_id
..... blah