MySQL Improve performance execution time - mysql

Question

I am trying to improve performance of following query which took 93.2 sec to execute query below:
SELECT year(date), month(date), `country_name_name`,
CEIL(count(res.`user_xmpp_login`) /DAY(LAST_DAY(date))) as avgUser,
CEIL(count(res.user)/DAY(LAST_DAY(date))) as avgPurchase
FROM
( SELECT DATE(`user_registration_timestamp`) as date,
user_country,
NULL as user, `user_xmpp_login`
FROM users
WHERE `user_registration_timestamp` >= "2015-01-01 00:00:00"
AND `user_registration_timestamp` < "2016-01-01 00:00:00"
UNION ALL
SELECT DATE(`ts`) as date, user_country, user, NULL as `user_xmpp_login`
FROM purchase_log p
INNER JOIN users u ON u.`user_xmpp_login` = p.`user`
WHERE `ts` >= "2015-01-01 00:00:00"
AND `ts` < "2016-01-01 00:00:00"
AND result in ('ok', 'cancelled', 'pending')
) AS res
INNER JOIN countries c ON c.`country_id` = res.`user_country`
INNER JOIN country_names cn
ON (cn.`country_name_country` = c.`country_id`
AND cn.`country_name_language` = 'en')
GROUP BY 1,2,3
ORDER BY 4 DESC,5 DESC, 3 ASC;
Explain command shows:
And structure of each table is:
purchase table:
CREATE TABLE `purchase` (
`id` int(11) NOT NULL AUTO_INCREMENT,
`user` varchar(255) COLLATE utf8_unicode_ci NOT NULL,
`ts` timestamp NOT NULL DEFAULT CURRENT_TIMESTAMP,
`result` varchar(32) COLLATE utf8_unicode_ci NOT NULL,
PRIMARY KEY (`id`),
KEY `iuser` (`user`),
) ENGINE=InnoDB AUTO_INCREMENT=12710221 DEFAULT CHARSET=utf8 COLLATE=utf8_unicode_ci;
users table:
CREATE TABLE `users` (
`user_id` int(11) NOT NULL AUTO_INCREMENT,
`user_country` int(11) DEFAULT NULL,
`user_xmpp_login` varchar(255) COLLATE utf8_unicode_ci DEFAULT NULL,
`user_registration_timestamp` timestamp NOT NULL DEFAULT CURRENT_TIMESTAMP,
PRIMARY KEY (`user_id`),
UNIQUE KEY `user_xmpp_login_UNIQUE` (`user_xmpp_login`),
KEY `user_country_FK` (`user_country`),
KEY `user_registration_timestamp` (`user_registration_timestamp`),
CONSTRAINT `users_country_FK` FOREIGN KEY (`user_country`)
REFERENCES `countries` (`country_id`) ON UPDATE CASCADE
) ENGINE=InnoDB AUTO_INCREMENT=33504745 DEFAULT CHARSET=utf8 COLLATE=utf8_unicode_ci;
countries table
CREATE TABLE `countries` (
`country_id` int(11) NOT NULL AUTO_INCREMENT,
`country_code` varchar(2) COLLATE utf8_unicode_ci NOT NULL,
PRIMARY KEY (`country_id`),
) ENGINE=InnoDB AUTO_INCREMENT=508 DEFAULT CHARSET=utf8 COLLATE=utf8_unicode_ci;
country names
CREATE TABLE `country_names` (
`country_name_id` int(11) NOT NULL AUTO_INCREMENT,
`country_name_country` int(11) NOT NULL,
`country_name_language` char(2) COLLATE utf8_unicode_ci NOT NULL,
`country_name_name` varchar(255) COLLATE utf8_unicode_ci NOT NULL,
PRIMARY KEY (`country_name_id`),
UNIQUE KEY `country_name_country_language_UNIQUE`
(`country_name_country`,`country_name_language`),
KEY `country_name_language` (`country_name_language`),
CONSTRAINT `country_name_country` FOREIGN KEY (`country_name_country`)
REFERENCES `countries` (`country_id`) ON DELETE CASCADE ON UPDATE CASCADE
) ENGINE=InnoDB AUTO_INCREMENT=45793 DEFAULT CHARSET=utf8 COLLATE=utf8_unicode_ci;
Is there any recommendations?

Answer 1

If you time each subquery, I think you will find users is the slowest component.
The purchase_log subquery can probably be improved with this "covering" INDEX(result, ts, user).
Combine the two "country" tables!. Use CHAR(2) CHARACTER SET ascii for the PRIMARY KEY and the JOINs to other tables. It is only 2 bytes, unlike INT, which is 4 bytes and VARCHAR..., which is 3 bytes (in this case).
You mention ts, but I don't see where it is coming from. If it is in purchase_log, then that table needs INDEX(user, ts).
What percentage of the users involved 2015? If it is more than about 20%, the INDEX(user_registration_timestamp) won't help.
Consider: Get rid of PRIMARY KEY (country_name_id), and promote the UNIQUE key to PRIMARY.

Answer 2

The biggest problem seems to be in your users table. Remember, mysql can only use one index per table for most situations. On your users table, the user_xmpp_login_UNIQUE column has been used to join it to the purchase_log table. There fore, the user_registration_timestamp index is not being used on the comparison involving the timestamp column.
One suggestion is to create a composite index on the user_xmpp_login and user_registration_timestamp columns.