Reputation: 20100
This query is very slow. It is pretty simple and the 3 tables used are indexed on all columns in JOIN and WHERE clauses. How can I optimize my query, or my tables for this query?
This is the slow query. It takes 15-20 seconds to run.
SELECT
user.id,
user.name,
user.key,
user.secret,
account.id,
account.name,
account.admin,
setting.attribute,
setting.value
FROM user
INNER JOIN account ON account.id = user.account_id
INNER JOIN setting ON setting.user_id = user.id
AND setting.deleted = 0
WHERE user.deleted = 0
It is likely issue is caused by join on the setting table, because the below two queries take about 5 seconds total. Although, 5 seconds still seems a little long?
SELECT
user.id,
user.name,
user.user_key,
user.secret,
account.id,
account.name,
account.admin
FROM user
INNER JOIN account ON account.user_id = user.id
WHERE user.deleted = 0
SELECT
setting.user_id,
setting.attribute,
setting.value
FROM setting
WHERE setting.deleted = 0
The explain for the slow query:
id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra
1, 'SIMPLE', 'user', 'ALL', 'PRIMARY,idx_id,idx_deleted', null, null, null, 600, 'Using where'
1, 'SIMPLE', 'account', 'eq_ref', 'PRIMARY', 'PRIMARY', '8', 'user.account_id', 1, null
1, 'SIMPLE', 'setting', 'ref', 'attribute_version_unique,idx_user_id,indx_deleted', 'attribute_version_unique', '8', 'user.id', 35, 'Using where'
The schema:
CREATE TABLE user
(
id BIGINT(20) unsigned PRIMARY KEY NOT NULL AUTO_INCREMENT,
name VARCHAR(45) NOT NULL,
user_key VARCHAR(45) NOT NULL,
secret VARCHAR(16),
account_id BIGINT(20) unsigned NOT NULL,
name VARCHAR(40) NOT NULL,
demo TINYINT(1) DEFAULT '0' NOT NULL,
details VARCHAR(4000),
date_created DATETIME NOT NULL,
date_modified DATETIME NOT NULL,
deleted TINYINT(1) DEFAULT '0' NOT NULL
);
CREATE INDEX idx_date_modified ON user (date_modified);
CREATE INDEX idx_deleted ON user (deleted);
CREATE INDEX idx_id ON pub_application (id);
CREATE UNIQUE INDEX idx_name_unique ON user (user_key);
CREATE TABLE account
(
id BIGINT(20) unsigned PRIMARY KEY NOT NULL AUTO_INCREMENT,
name VARCHAR(100) NOT NULL,
display_name VARCHAR(100),
admin TINYINT(1) DEFAULT '0' NOT NULL,
visibility VARCHAR(15) DEFAULT 'public',
cost DOUBLE,
monthly_fee VARCHAR(300),
date_created DATETIME NOT NULL,
date_modified DATETIME NOT NULL,
deleted TINYINT(1) DEFAULT '0'
);
CREATE INDEX idx_date_modified ON account (date_modified);
CREATE TABLE setting
(
id BIGINT(20) unsigned PRIMARY KEY NOT NULL AUTO_INCREMENT,
user_id BIGINT(20) unsigned NOT NULL,
attribute VARCHAR(45) NOT NULL,
value VARCHAR(4000),
date_created DATETIME NOT NULL,
date_modified DATETIME NOT NULL,
deleted TINYINT(1) DEFAULT '0' NOT NULL
);
CREATE UNIQUE INDEX attribute_version_unique ON setting (user_id, attribute);
CREATE INDEX idx_user_id ON setting (user_id);
CREATE INDEX idx_date_modified ON setting (date_modified);
CREATE INDEX indx_deleted ON setting (deleted);
Upvotes: 3
Views: 169
Reputation: 108839
With respect, you've stumbled across a common antipattern. Indexing "all columns" ordinarily is a useless move. MySQL (as of late 2016) can exploit at most one index per table when satisfying a query. So the extra indexes are likely to help no queries, and definitely add overhead on INSERT
and UPDATE
operations.
This query might be improved by some purpose-designed compound covering indexes.
Try this index on your user
table. It's a covering index: intended to contain all the columns necessary to satisfy the query. It's organized in an order that matches your WHERE
clause.
CREATE INDEX idx_user_account_setting
ON user (deleted , account_id, id, name, key, secret);
This covering index might help on your setting
table
CREATE INDEX idx_setting_user
ON setting (user_id, deleted , attribute, value);
Also try this one, switching the order of the first two columns, if the first one doesn't help.
CREATE INDEX idx_setting_user_alt
ON setting (deleted, user_id, attribute, value);
Finally try this one on account
.
CREATE INDEX idx_account_user
ON account (id, name, admin);
Please, if these suggestions help leave a brief comment telling how much they helped.
Read this. http://use-the-index-luke.com/
Upvotes: 3