Mysql过滤器行为没有意义

2jcobegt  于 2023-06-21  发布在  Mysql
关注(0)|答案(1)|浏览(136)

我有以下疑问

SELECT `users`.`id`, `users`.`vPlatform`, `users`.`vLanguage` 
FROM `users` FORCE INDEX (test1)
WHERE
    EXISTS (
        SELECT `orders`.* FROM `orders` WHERE `orders`.`is_deleted` = FALSE
        AND `orders`.`iCourierId` = `users`.`id`
        AND `orders`.`vStatus` = 'delivered'
        AND `orders`.`updated_at` >= '2022-12-06 09:05:27'
    )
    AND `users`.`login_status` != FALSE
    AND `users`.`vPlatform` IN ('300', '301')
    AND NOT ((`users`.`vDeviceToken` = '' OR `users`.`vDeviceToken` IS NULL))
    AND `users`.`vCurrency` = 'EGP'
    AND `users`.`is_deleted` = FALSE
    AND `users`.`id` BETWEEN 17395001 AND 17400000 ORDER BY `users`.`id` ASC;

它的explain analyze显示以下内容

  • 用于订单表的索引为CourierId_idx (iCourierId,vStatus,updated_at,created_at)
-> Nested loop semijoin  (cost=11633.55 rows=254) (actual time=0.743..8.440 rows=8 loops=1)
    -> Filter: ((users.vPlatform in ('300','301')) and (users.vDeviceToken <> '') and (users.vDeviceToken is not null))  (cost=3636.72 rows=73) (actual time=0.059..4.504 rows=904 loops=1)
        -> Index range scan on users using test1 over (17395001 <= id <= 17400000 AND NULL < login_status), with index condition: ((users.is_deleted = false) and (users.login_status <> false) and (users.vCurrency = 'EGP') and (users.id between 17395001 and 17400000))  (cost=3636.72 rows=5000) (actual time=0.057..4.257 rows=914 loops=1)
    -> Filter: (orders.is_deleted = false)  (cost=346.09 rows=3) (actual time=0.004..0.004 rows=0 loops=904)
        -> Index lookup on orders using CourierId_idx (iCourierId=users.id, vStatus='delivered'), with index condition: (orders.updated_at >= TIMESTAMP'2022-12-06 09:05:27')  (cost=346.09 rows=105) (actual time=0.004..0.004 rows=0 loops=904)

这表明orders.is_deleted被过滤了,因为它不是使用的索引的一部分,将其添加到索引中会使下面的explain analyze变慢

  • 用于订单表的索引为test3(iCourierId,vStatus,created_at,updated_at,is_deleted)
-> Nested loop semijoin  (cost=4184.56 rows=92) (actual time=0.854..9.845 rows=8 loops=1)
    -> Filter: ((users.vPlatform in ('300','301')) and (users.vDeviceToken <> '') and (users.vDeviceToken is not null))  (cost=3636.72 rows=73) (actual time=0.045..6.321 rows=904 loops=1)
        -> Index range scan on users using test1 over (17395001 <= id <= 17400000 AND NULL < login_status), with index condition: ((users.is_deleted = false) and (users.login_status <> false) and (users.vCurrency = 'EGP') and (users.id between 17395001 and 17400000))  (cost=3636.72 rows=5000) (actual time=0.043..6.071 rows=914 loops=1)
    -> Filter: ((orders.is_deleted = false) and (orders.updated_at >= TIMESTAMP'2022-12-06 09:05:27'))  (cost=1.46 rows=1) (actual time=0.004..0.004 rows=0 loops=904)
        -> Covering index lookup on orders using test3 (iCourierId=users.id, vStatus='delivered')  (cost=1.46 rows=64) (actual time=0.003..0.004 rows=0 loops=904)

使用的索引包括orders.is_deletedorders.updated_at,但它被过滤了,而orders.updated_atwith index condition移动到filter为什么?

guykilcj

guykilcj1#

除非您需要vDeviceToken = ''来表示不同的含义,否则请选择其中一个。使用OR会降低性能。
这些措施应有助于:

users:  INDEX(login_status, vCurrency, is_deleted, vDeviceToken, vPlatform, id)
orders:  INDEX(is_deleted, iCourierId, vStatus, updated_at)

去掉FORCE INDEX和这些,如果你碰巧有它们:INDEX(login_status)INDEX(is_deleted)
为了提高效率,索引中的列应该从用=测试的列开始。更多:Index Cookbook

相关问题