背景
XX实例(一主一从)xxx告警中每天凌晨在报SLA报警,该报警的意思是存在一定的主从延迟(若在此时发生主从切换,需要长时间才可以完成切换,要追延迟来保证主从数据的一致性)
XX实例的慢查询数量最多(执行时间超过1s的sql会被记录),XX应用那方每天晚上在做删除一个月前数据的任务
基于 Spring Boot + MyBatis Plus + Vue & Element 实现的后台管理系统 + 用户小程序,支持 RBAC 动态权限、多租户、数据权限、工作流、三方登录、支付、短信、商城等功能
项目地址:https://github.com/YunaiV/ruoyi-vue-pro
视频教程:https://doc.iocoder.cn/video/
分析
使用pt-query-digest工具分析最近一周的mysql-slow.log
pt-query-digest--since=148hmysql-slow.log|less
结果第一部分
最近一个星期内,总共记录的慢查询执行花费时间为25403s,最大的慢sql执行时间为266s,平均每个慢sql执行时间5s,平均扫描的行数为1766万
结果第二部分
select arrival_record操作记录的慢查询数量最多有4万多次,平均响应时间为4s,delete arrival_record记录了6次,平均响应时间258s。
select xxx_record语句
select arrival_record 慢查询语句都类似于如下所示,where语句中的参数字段是一样的,传入的参数值不一样select count(*) from arrival_record where product_id=26 and receive_time between '2019-03-25 1400' and '2019-03-25 1500' and receive_spend_ms>=0G
select arrival_record 语句在mysql中最多扫描的行数为5600万、平均扫描的行数为172万,推断由于扫描的行数多导致的执行时间长
查看执行计划
explainselectcount(*)fromarrival_recordwhereproduct_id=26andreceive_timebetween'2019-03-251400'and'2019-03-251500'andreceive_spend_ms>=0G; ***************************1.row*************************** id:1 select_type:SIMPLE table:arrival_record partitions:NULL type:ref possible_keys:IXFK_arrival_record key:IXFK_arrival_record key_len:8 ref:const rows:32261320 filtered:3.70 Extra:Usingindexcondition;Usingwhere 1rowinset,1warning(0.00sec)
用到了索引IXFK_arrival_record,但预计扫描的行数很多有3000多w行
showindexfromarrival_record; +----------------+------------+---------------------+--------------+--------------+-----------+-------------+----------+--------+------+------------+---------+---------------+ |Table|Non_unique|Key_name|Seq_in_index|Column_name|Collation|Cardinality|Sub_part|Packed|Null|Index_type|Comment|Index_comment| +----------------+------------+---------------------+--------------+--------------+-----------+-------------+----------+--------+------+------------+---------+---------------+ |arrival_record|0|PRIMARY|1|id|A|107990720|NULL|NULL||BTREE||| |arrival_record|1|IXFK_arrival_record|1|product_id|A|1344|NULL|NULL||BTREE||| |arrival_record|1|IXFK_arrival_record|2|station_no|A|22161|NULL|NULL|YES|BTREE||| |arrival_record|1|IXFK_arrival_record|3|sequence|A|77233384|NULL|NULL||BTREE||| |arrival_record|1|IXFK_arrival_record|4|receive_time|A|65854652|NULL|NULL|YES|BTREE||| |arrival_record|1|IXFK_arrival_record|5|arrival_time|A|73861904|NULL|NULL|YES|BTREE||| +----------------+------------+---------------------+--------------+--------------+-----------+-------------+----------+--------+------+------------+---------+---------------+ showcreatetablearrival_record; .......... arrival_spend_msbigint(20)DEFAULTNULL, total_spend_msbigint(20)DEFAULTNULL, PRIMARYKEY(id), KEYIXFK_arrival_record(product_id,station_no,sequence,receive_time,arrival_time)USINGBTREE, CONSTRAINTFK_arrival_record_productFOREIGNKEY(product_id)REFERENCESproduct(id)ONDELETENOACTIONONUPDATENOACTION )ENGINE=InnoDBAUTO_INCREMENT=614538979DEFAULTCHARSET=utf8COLLATE=utf8_bin|
该表总记录数约1亿多条,表上只有一个复合索引,product_id字段基数很小,选择性不好
传入的过滤条件 where product_id=26 and receive_time between '2019-03-25 1400' and '2019-03-25 1500' and receive_spend_ms>=0 没有station_nu字段,使用不到复合索引 IXFK_arrival_record的 product_id,station_no,sequence,receive_time 这几个字段
根据最左前缀原则,select arrival_record只用到了复合索引IXFK_arrival_record的第一个字段product_id,而该字段选择性很差,导致扫描的行数很多,执行时间长
receive_time字段的基数大,选择性好,可对该字段单独建立索引,select arrival_record sql就会使用到该索引
现在已经知道了在慢查询中记录的select arrival_record where语句传入的参数字段有 product_id,receive_time,receive_spend_ms,还想知道对该表的访问有没有通过其它字段来过滤了?
神器tcpdump出场的时候到了
使用tcpdump抓包一段时间对该表的select语句
tcpdump-ibond0-s0-l-w-dstport3316|strings|grepselect|egrep-i'arrival_record'>/tmp/select_arri.log
获取select 语句中from 后面的where条件语句
IFS_OLD=$IFS IFS=$' ' foriin`cat/tmp/select_arri.log`;doecho${i#*'from'};done|less IFS=$IFS_OLD
arrival_recordarrivalrec0_wherearrivalrec0_.sequence='2019-03-2708:40'andarrivalrec0_.product_id=17andarrivalrec0_.station_no='56742' arrival_recordarrivalrec0_wherearrivalrec0_.sequence='2019-03-2708:40'andarrivalrec0_.product_id=22andarrivalrec0_.station_no='S7100' arrival_recordarrivalrec0_wherearrivalrec0_.sequence='2019-03-2708:40'andarrivalrec0_.product_id=24andarrivalrec0_.station_no='V4631' arrival_recordarrivalrec0_wherearrivalrec0_.sequence='2019-03-2708:40'andarrivalrec0_.product_id=22andarrivalrec0_.station_no='S9466' arrival_recordarrivalrec0_wherearrivalrec0_.sequence='2019-03-2708:40'andarrivalrec0_.product_id=24andarrivalrec0_.station_no='V4205' arrival_recordarrivalrec0_wherearrivalrec0_.sequence='2019-03-2708:40'andarrivalrec0_.product_id=24andarrivalrec0_.station_no='V4105' arrival_recordarrivalrec0_wherearrivalrec0_.sequence='2019-03-2708:40'andarrivalrec0_.product_id=24andarrivalrec0_.station_no='V4506' arrival_recordarrivalrec0_wherearrivalrec0_.sequence='2019-03-2708:40'andarrivalrec0_.product_id=24andarrivalrec0_.station_no='V4617' arrival_recordarrivalrec0_wherearrivalrec0_.sequence='2019-03-2708:40'andarrivalrec0_.product_id=22andarrivalrec0_.station_no='S8356' arrival_recordarrivalrec0_wherearrivalrec0_.sequence='2019-03-2708:40'andarrivalrec0_.product_id=22andarrivalrec0_.station_no='S8356' select该表where条件中有product_id,station_no,sequence字段,可以使用到复合索引IXFK_arrival_record的前三个字段
综上所示,优化方法为,删除复合索引IXFK_arrival_record,建立复合索引idx_sequence_station_no_product_id,并建立单独索引indx_receive_time
delete xxx_record语句
该delete操作平均扫描行数为1.1亿行,平均执行时间是262s
delete语句如下所示,每次记录的慢查询传入的参数值不一样
deletefromarrival_recordwherereceive_time< STR_TO_DATE('2019-02-23', '%Y-%m-%d')G
执行计划
explainselect*fromarrival_recordwherereceive_time< STR_TO_DATE('2019-02-23', '%Y-%m-%d')G *************************** 1. row *************************** id: 1 select_type: SIMPLE table: arrival_record partitions: NULL type: ALL possible_keys: NULL key: NULL key_len: NULL ref: NULL rows: 109501508 filtered: 33.33 Extra: Using where 1 row in set, 1 warning (0.00 sec)
该delete语句没有使用索引(没有合适的索引可用),走的全表扫描,导致执行时间长
优化方法也是 建立单独索引indx_receive_time(receive_time)
基于 Spring Cloud Alibaba + Gateway + Nacos + RocketMQ + Vue & Element 实现的后台管理系统 + 用户小程序,支持 RBAC 动态权限、多租户、数据权限、工作流、三方登录、支付、短信、商城等功能
项目地址:https://github.com/YunaiV/yudao-cloud
视频教程:https://doc.iocoder.cn/video/
测试
拷贝arrival_record表到测试实例上进行删除重新索引操作XX实例arrival_record表信息
du-sh/datas/mysql/data/3316/cq_new_cimiss/arrival_record* 12K/datas/mysql/data/3316/cq_new_cimiss/arrival_record.frm 48G/datas/mysql/data/3316/cq_new_cimiss/arrival_record.ibd selectcount()fromcq_new_cimiss.arrival_record; +-----------+ |count()| +-----------+ |112294946| +-----------+ 1亿多记录数 SELECT table_name, CONCAT(FORMAT(SUM(data_length)/1024/1024,2),'M')ASdbdata_size, CONCAT(FORMAT(SUM(index_length)/1024/1024,2),'M')ASdbindex_size, CONCAT(FORMAT(SUM(data_length+index_length)/1024/1024/1024,2),'G')AStable_size(G), AVG_ROW_LENGTH,table_rows,update_time FROM information_schema.tables WHEREtable_schema='cq_new_cimiss'andtable_name='arrival_record'; +----------------+-------------+--------------+------------+----------------+------------+---------------------+ |table_name|dbdata_size|dbindex_size|table_size(G)|AVG_ROW_LENGTH|table_rows|update_time| +----------------+-------------+--------------+------------+----------------+------------+---------------------+ |arrival_record|18,268.02M|13,868.05M|31.38G|175|109155053|2019-03-2612:40:17| +----------------+-------------+--------------+------------+----------------+------------+---------------------+
磁盘占用空间48G,mysql中该表大小为31G,存在17G左右的碎片,大多由于删除操作造成的(记录被删除了,空间没有回收)
备份还原该表到新的实例中,删除原来的复合索引,重新添加索引进行测试
mydumper并行压缩备份
user=root passwd=xxxx socket=/datas/mysql/data/3316/mysqld.sock db=cq_new_cimiss table_name=arrival_record backupdir=/datas/dump_$table_name mkdir-p$backupdir nohupecho`date+%T`&&mydumper-u$user-p$passwd-S$socket-B$db-c-T$table_name-o$backupdir-t32-r2000000&&echo`date+%T`&
并行压缩备份所花时间(52s)和占用空间(1.2G,实际该表占用磁盘空间为48G,mydumper并行压缩备份压缩比相当高!)
Starteddumpat:2019-03-2612:46:04 ........ Finisheddumpat:2019-03-2612:46:56 du-sh/datas/dump_arrival_record/ 1.2G/datas/dump_arrival_record/
拷贝dump数据到测试节点
scp-rp/datas/dump_arrival_recordroot@10.230.124.19:/datas
多线程导入数据
timemyloader-uroot-S/datas/mysql/data/3308/mysqld.sock-P3308-proot-Btest-d/datas/dump_arrival_record-t32
real 126m42.885suser 1m4.543ssys 0m4.267s
逻辑导入该表后磁盘占用空间
du-h-d1/datas/mysql/data/3308/test/arrival_record.* 12K/datas/mysql/data/3308/test/arrival_record.frm 30G/datas/mysql/data/3308/test/arrival_record.ibd 没有碎片,和mysql的该表的大小一致 cp-rp/datas/mysql/data/3308/datas
分别使用online DDL和 pt-osc工具来做删除重建索引操作先删除外键,不删除外键,无法删除复合索引,外键列属于复合索引中第一列
nohupbash/tmp/ddl_index.sh& 2019-04-04-10:41:39beginstopmysqld_3308 2019-04-04-10:41:41beginrm-rfdatadirandcp-rpdatadir_bak 2019-04-04-10:46:53startmysqld_3308 2019-04-04-10:46:59onlineddlbegin 2019-04-04-11:20:34onlieddlstop 2019-04-04-11:20:34beginstopmysqld_3308 2019-04-04-11:20:36beginrm-rfdatadirandcp-rpdatadir_bak 2019-04-04-11:22:48startmysqld_3308 2019-04-04-11:22:53pt-oscbegin 2019-04-04-12:19:15pt-oscstop onlineddl花费时间为34分钟,pt-osc花费时间为57分钟,使用onlneddl时间约为pt-osc工具时间的一半
*做DDL 参考 *
实施
由于是一主一从实例,应用是连接的vip,删除重建索引采用online ddl来做。停止主从复制后,先在从实例上做(不记录binlog),主从切换,再在新切换的从实例上做(不记录binlog)
functionred_echo(){ localwhat="$*" echo-e"$(date+%F-%T)${what}" } functioncheck_las_comm(){ if["$1"!="0"];then red_echo"$2" echo"exit1" exit1 fi } red_echo"stopslave" mysql-uroot-p$passwd--socket=/datas/mysql/data/${port}/mysqld.sock-e"stopslave" check_las_comm"$?""stopslavefailed" red_echo"onlineddlbegin" mysql-uroot-p$passwd--socket=/datas/mysql/data/${port}/mysqld.sock-e"setsql_log_bin=0;selectnow()asddl_start;ALTERTABLE$db_.`${table_name}`DROPFOREIGNKEYFK_arrival_record_product,dropindexIXFK_arrival_record,addindexidx_product_id_sequence_station_no(product_id,sequence,station_no),addindexidx_receive_time(receive_time);selectnow()asddl_stop">>${log_file}2>&1 red_echo"onlieddlstop" red_echo"addforeignkey" mysql-uroot-p$passwd--socket=/datas/mysql/data/${port}/mysqld.sock-e"setsql_log_bin=0;ALTERTABLE$db_.${table_name}ADDCONSTRAINT_FK_${table_name}_productFOREIGNKEY(product_id)REFERENCEScq_new_cimiss.product(id)ONDELETENOACTIONONUPDATENOACTION;">>${log_file}2>&1 check_las_comm"$?""addforeignkeyerror" red_echo"addforeignkeystop" red_echo"startslave" mysql-uroot-p$passwd--socket=/datas/mysql/data/${port}/mysqld.sock-e"startslave" check_las_comm"$?""startslavefailed"
*执行时间 *
2019-04-08-1136 stop slavemysql: [Warning] Using a password on the command line interface can be insecure.ddl_start2019-04-08 1136ddl_stop2019-04-08 11132019-04-08-1113 onlie ddl stop2019-04-08-1113 add foreign keymysql: [Warning] Using a password on the command line interface can be insecure.2019-04-08-1248 add foreign key stop2019-04-08-1248 start slave
*再次查看delete 和select语句的执行计划 *
explainselectcount(*)fromarrival_recordwherereceive_time< STR_TO_DATE('2019-03-10', '%Y-%m-%d')G *************************** 1. row *************************** id: 1 select_type: SIMPLE table: arrival_record partitions: NULL type: range possible_keys: idx_receive_time key: idx_receive_time key_len: 6 ref: NULL rows: 7540948 filtered: 100.00 Extra: Using where; Using index explain select count(*) from arrival_record where product_id=26 and receive_time between '2019-03-25 1400' and '2019-03-25 1500' and receive_spend_ms>=0G; ***************************1.row*************************** id:1 select_type:SIMPLE table:arrival_record partitions:NULL type:range possible_keys:idx_product_id_sequence_station_no,idx_receive_time key:idx_receive_time key_len:6 ref:NULL rows:291448 filtered:16.66 Extra:Usingindexcondition;Usingwhere 都使用到了idx_receive_time索引,扫描的行数大大降低
索引优化后
delete 还是花费了77s时间
deletefromarrival_recordwherereceive_time< STR_TO_DATE('2019-03-10', '%Y-%m-%d')G
delete 语句通过receive_time的索引删除300多万的记录花费77s时间*
delete大表优化为小批量删除
*应用端已优化成每次删除10分钟的数据(每次执行时间1s左右),xxx中没在出现SLA(主从延迟告警) *
*另一个方法是通过主键的顺序每次删除20000条记录 *
#得到满足时间条件的最大主键ID #通过按照主键的顺序去顺序扫描小批量删除数据 #先执行一次以下语句 SELECTMAX(id)INTO@need_delete_max_idFROM`arrival_record`WHEREreceive_time<'2019-03-01' ; DELETE FROM arrival_record WHERE id<@need_delete_max_id LIMIT 20000; select ROW_COUNT(); #返回20000 #执行小批量delete后会返回row_count(), 删除的行数 #程序判断返回的row_count()是否为0,不为0执行以下循环,为0退出循环,删除操作完成 DELETE FROM arrival_record WHERE id<@need_delete_max_id LIMIT 20000; select ROW_COUNT(); #程序睡眠0.5s
总结
表数据量太大时,除了关注访问该表的响应时间外,还要关注对该表的维护成本(如做DDL表更时间太长,delete历史数据)。
对大表进行DDL操作时,要考虑表的实际情况(如对该表的并发表,是否有外键)来选择合适的DDL变更方式。
对大数据量表进行delete,用小批量删除的方式,减少对主实例的压力和主从延迟。
审核编辑:刘清
-
SQL
+关注
关注
1文章
760浏览量
44080 -
DDL
+关注
关注
0文章
12浏览量
6320 -
MYSQL数据库
+关注
关注
0文章
95浏览量
9382
原文标题:面试官:MySQL 上亿大表,如何深度优化?
文章出处:【微信号:芋道源码,微信公众号:芋道源码】欢迎添加关注!文章转载请注明出处。
发布评论请先 登录
相关推荐
评论