大表百万条以上记录非主键索引删除大量数据的操作方法
众所周知,上百万表数据类似上述写法,哪怕where条件命中索引,执行时间也不会低于10分钟;考虑对数据的操作会引起两表的排他锁,造成业务被暂停!从服务器资源的影响:IOPS升高,IO争用造成CPU等待(高CPU);删除表存在的大量空间碎片,造成的空间浪费及性能低下
这种需求可以采用中间表rename方法,将要保留的数据写入中间表,写入完成后将原表删除,并将中间表改名为原始表;
具体操作步骤如下:
1.SHOW CREATE TABLE ooo_scrm_customer; 2.创建同ooo_scrm_customer表结构一样的新表:_ooo_scrm_customer_new REATE TABLE _ooo_scrm_customer_new ( `id` int (11) NOT NULL AUTO_INCREMENT, ..... ..... ..... PRIMARY KEY (`id`) ) ENGINE=InnoDB DEFAULT CHARSET=utf8; 3.创建增删改触发器(主要用于主表向中间表写数据时,业务方请求修改的数据要一起同步到中间表) CREATE TRIGGER `ooo_scrm_customer_del` AFTER DELETE ON ooo_scrm_customer FOR EACH ROW DELETE IGNORE FROM _ooo_scrm_customer_new WHERE _ooo_scrm_customer_new.`id` <=> OLD.`id` CREATE TRIGGER `ooo_scrm_customer_upd` AFTER UPDATE ON ooo_scrm_customer FOR EACH ROW REPLACE INTO _ooo_scrm_customer_new(`id`, `其他列`) VALUES (NEW.`id`, NEW.`其他列`) CREATE TRIGGER `ooo_scrm_customer_ins` AFTER INSERT ON ooo_scrm_customer FOR EACH ROW REPLACE INTO _ooo_scrm_customer_new(`id`, `其他列`) VALUES (NEW.`id`, NEW.`其他列`) 4.将要保留的数据写入中间表 INSERT LOW_PRIORITY IGNORE INTO _ooo_scrm_customer_new (`id`, `其他列`) SELECT `id`, `其他列` FROM ooo_scrm_customer LOCK IN SHARE MODE # where 条件优化,按分批执行思想,稳步执行! 5.优化中间表 ANALYZE TABLE_ooo_scrm_customer_new 6.改名并删除操作记录 RENAME TABLE ooo_scrm_customer TO _ooo_scrm_customer_old, _ooo_scrm_customer_new TO ooo_scrm_customer DROP TRIGGER IF EXISTS `mysqlcheck`.`ooo_scrm_customer_del` DROP TRIGGER IF EXISTS `mysqlcheck`.`ooo_scrm_customer_upd` DROP TRIGGER IF EXISTS `mysqlcheck`.`ooo_scrm_customer_ins` |
其他应用案例
该思想起源facebook的mysql path,也是大名鼎鼎pt-online-schema-change工具的实现原理。
可应用于在线改表结构,在线修改表数据,原则上来讲可以实现秒级别的表锁定,对业务的0伤害
同学们在程序实现时还可灵活的使用其他方法:
SET SESSION lock_wait_timeout=60;SET SESSION innodb_lock_wait_timeout=1;SET SESSION wait_timeout=10000
(innodb_lock_wait_timeout指的是事务等待获取资源等待的最长时间,超过这个时间还未分配到资源则会返回应用失败;)
如有类似需求,请将该思想运用到你们的项目中吧。
OVER!!