故障案例 | 慢SQL引发MySQL高可用切换排查全过程

关注

原创最新推荐文章于 2026-08-11 14:20:02 发布 615 阅读 AI 写同款· GEO 优化 ›

MySQL之高可用性终极实践：故障转移与架构冗余设计 高可用性架构的构建是一个系统性工程，需综合考虑业务需求、技术成本与风险承受能力。通过主备复制结合自动化工具（如MHA）可快速实现基础高可用，而多主集群方案则适用于对一致性与扩展性要求极高的场景。在实践中，需定期进行故障演练，完善监控体系，并建立应急响应流程。未来，随着云原生技术的发展，容器化部署与自动化运维将进一步提升MySQL高可用性的实现效率，降低人工干预成本。写作不易，每一个架构设计和代码示例都凝聚了对高可用性的深入理解。阅读详情

作者：梁行万里数据库DBA，擅长数据库性能问题诊断、事务与锁问题的分析等，负责处理客户MySQL日常运维中的问题，对开源数据库相关技术非常感兴趣。

GreatSQL社区原创内容未经授权不得随意使用，转载请联系小编并注明来源。

一、现象说明

在排查问题时发现MySQL主备做了切换，而查看MySQL服务是正常的，DBA也没有做切换操作，服务器也无维护操作，万幸的是业务还没有受到大的波及。这到底是是为什么呢？

假设原主服务器地址为：172.16.87.72，原备主服务器地址为：172.16.87.123。

二、排查思路

通过监控查看MySQL的各个指标查看双主（keepalived）服务切换日志 MySQL错误日志信息

三、问题排查

3.1 通过监控系统查看MySQL监控指标，判断故障发生的具体时间（通过流量判断大致切换时间点）

通过监控查看现在主库MySQL（支撑业务）的监控指标，看一天的、十五天的 (以每秒查询数量速率为例)

file

发现从22号开始陡增，初步怀疑，可能22号发生了主备切换，再看一下备库的折线图，进一步确认查看双主（keepalived）服务切换日志，确定主从切换时间

3.2 查看keepalived判断为什么会做主从切换

先登录87.72查看keepalived的切换日志，日志信息如下：注：系统是ubuntu而且keepalived没有指定输出日志输出，所以keepalived会将日志输出到系统默认日志文件syslog.log中

shell> cat syslog.6|grep "vrrp"
Oct 23 15:50:34 bjuc-mysql1 Keepalived_vrrp: VRRP_Script(check_mysql) failed
Oct 23 15:50:34 bjuc-mysql1 syslog-ng[31634]: Error opening file for writing; filename='/var/log/tang/Keepalived_vrrp.log', error='No such file or directory (2)'
Oct 23 15:50:35 bjuc-mysql1 Keepalived_vrrp: VRRP_Instance(VI_74) Received higher prio advert
Oct 23 15:50:35 bjuc-mysql1 Keepalived_vrrp: VRRP_Instance(VI_74) Entering BACKUP STATE
Oct 23 15:50:35 bjuc-mysql1 syslog-ng[31634]: Error opening file for writing; filename='/var/log/tang/Keepalived_vrrp.log', error='No such file or directory (2)'
Oct 23 15:50:35 bjuc-mysql1 syslog-ng[31634]: Error opening file for writing; filename='/var/log/tang/Keepalived_vrrp.log', error='No such file or directory (2)'
Oct 23 15:50:35 bjuc-mysql1 syslog-ng[31634]: Error opening file for writing; filename='/var/log/tang/Keepalived_vrrp.log', error='No such file or directory (2)'
Oct 23 15:50:35 bjuc-mysql1 syslog-ng[31634]: Error opening file for writing; filename='/var/log/tang/Keepalived_vrrp.log', error='No such file or directory (2)'
Oct 23 15:50:56 bjuc-mysql1 Keepalived_vrrp: VRRP_Script(check_mysql) succeeded

日志分析结果

从日志中可以发现，在Oct 23 15:50:34 因为检测脚本检测MySQL服务失败，导致了vip切换再去87.123上查看keepalvied的切换，确定一下切换时间

Oct 23 15:50:35 prod-UC-yewu-db-slave2 Keepalived_vrrp: VRRP_Instance(VI_74) forcing a new MASTER election 
Oct 23 15:50:36 prod-UC-yewu-db-slave2 Keepalived_vrrp: VRRP_Instance(VI_74) Transition to MASTER STATE 
Oct 23 15:50:37 prod-UC-yewu-db-slave2 Keepalived_vrrp: VRRP_Instance(VI_74) Entering MASTER STATE

通过以上日志信息，我们发现双主是在Oct 23 15:50:35发生了切换

3.3 疑问？那么检测脚本为什么会检测失败呢？

查看检测脚本内容

shell> cat check_mysql.sh
#!/bin/bash
export PATH=$PATH:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
USER=xxx
PASSWORD=xxx
PORT=3306
IP=127.0.0.1

mysql  --connect-timeout=1 -u$USER -p$PASSWORD -h$IP -P$PORT -e "select 1;"
exit $?

我们发现，检测脚本很简单，执行“select 1”命令，超时时间为1s，这个sql不影响MySQL任何操作，耗费性能很小，那为什么还会失败呢？我们去MySQL查看一下日志，看看有没有发现

3.4 查看MySQL日志

191023 15:50:54 [ERROR] /usr/sbin/mysqld: Sort aborted: Server shutdown in progress
191023 15:50:55 [ERROR] /usr/sbin/mysqld: Sort aborted: Server shutdown in progress
191023 15:50:55 [ERROR] /usr/sbin/mysqld: Sort aborted: Server shutdown in progress
191023 15:50:56 [ERROR] /usr/sbin/mysqld: Sort aborted: Server shutdown in progress
191023 15:50:56 [ERROR] /usr/sbin/mysqld: Sort aborted: Server shutdown in progress
191023 15:50:57 [ERROR] /usr/sbin/mysqld: Sort aborted: Server shutdown in progress
191023 15:50:58 [ERROR] /usr/sbin/mysqld: Sort aborted: Server shutdown in progress
191023 15:50:58 [ERROR] /usr/sbin/mysqld: Sort aborted: Server shutdown in progress
191023 15:50:59 [ERROR] /usr/sbin/mysqld: Sort aborted: Server shutdown in progress
191023 15:51:28 [ERROR] Slave SQL: Error 'Lock wait timeout exceeded; try restarting transaction' on query. Default database: 'statusnet'. Query: 'UPDATE ticket SET max_id=max_id+step WHERE biz_tag='p2p_file'', Error_code: 1205
191023 15:54:43 [ERROR] Slave SQL thread retried transaction 10 time(s) in vain, giving up. Consider raising the value of the slave_transaction_retries variable.
191023 15:54:43 [Warning] Slave: Lock wait timeout exceeded; try restarting transaction Error_code: 1205
191023 15:54:43 [ERROR] Error running query, slave SQL thread aborted. Fix the problem, and restart the slave SQL thread with "SLAVE START". We stopped at log 'mysql-bin.001432' position 973809430
191023 16:04:13 [ERROR] Error reading packet from server: Lost connection to MySQL server during query ( server_errno=2013)
191023 16:04:13 [Note] Slave I/O thread killed while reading event
191023 16:04:13 [Note] Slave I/O thread exiting, read up to log 'mysql-bin.001432', position 998926454
191023 16:04:20 [Note] Slave SQL thread initialized, starting replication in log 'mysql-bin.001432' at position 973809430, relay log './mysql-relay-bin.000188' position: 54422
191023 16:04:20 [Note] 'SQL_SLAVE_SKIP_COUNTER=1' executed at relay_log_file='./mysql-relay-bin.000188', relay_log_pos='54422', master_log_name='mysql-bin.001432', master_log_pos='973809430' and new position at relay_log_file='./mysql-relay-bin.000188', relay_log_pos='54661', master_log_name='mysql-bin.001432', master_log_pos='973809669'
191023 16:04:20 [Note] Slave I/O thread: connected to master 'zhu@x:3306',replication started in log 'mysql-bin.001432' at position 998926454

从日志中可以发现，MySQL在15：50：54报 Server shutdown in progress错误，在15:54:43 主从关系挂了，在16:04:20的时候主从关系自动恢复正常。那么这个报错是什么呢，Server shutdown in progress是什么意思呢？这个报错信息是因为在MySQL 5.5时，如果主动kill一个长的慢查询，而且这个sql还用到了sort buffer，就会出现类似MySQL重启的情况，连接也会关闭。

官方bug地址：https://bugs.mysql.com/bug.php?id=18256

赶紧去线上环境看看自己的环境是不是MySQL 5.5

mysql> \s
--------------
mysql  Ver 14.14 Distrib 5.5.22, for debian-linux-gnu (x86_64) using readline 6.2

Connection id:  76593111
Current database:
Current user:  xxx@127.0.0.1
SSL:   Not in use
Current pager:  stdout
Using outfile:  ''
Using delimiter: ;
Server version:  5.5.22-0ubuntu1-log (Ubuntu)
Protocol version: 10
Connection:  127.0.0.1 via TCP/IP
Server characterset: utf8
Db     characterset: utf8
Client characterset: utf8
Conn.  characterset: utf8
TCP port:  3306
Uptime:   462 days 11 hours 50 min 11 sec

Threads: 9  Questions: 14266409536  Slow queries: 216936  Opens: 581411  Flush tables: 605  Open tables: 825  Queries per second avg: 357.022

我们看到线上环境是MySQL 5.5版本，运行时间为462 days 11 hours 50 min 11 sec，证明MySQL进程没有重启，那么通过刚才error日志（主从关系重启）验证了MySQL的这个bug so，keepalived在这一个发现MySQL不可用，就产生了failed。

3.5 那么我们再深挖一下，是什么慢sql导致被kill呢？

我们继续查找相应时间点的slow日志

# Time: 191023 15:50:39
# User@Host: dstatusnet[dstatusnet] @  [x]
# Query_time: 35.571088  Lock_time: 0.742899 Rows_sent: 1  Rows_examined: 72
SET timestamp=1571817039;
SELECT g.id,g.group_pinyin, g.external_type,g.external_group_id,g.mount_id,g.is_display as group_type,g.nickname,g.site_id,g.group_logo,g.created,g.modified,g.misc,g.name_update_flag as name_flag, g.logo_level,g.group_status,g.conversation, g.only_admin_invite,g.reach_limit_count,g.water_mark  FROM user_group AS g LEFT JOIN group_member as gm ON g.id = gm.group_id WHERE g.is_display in (1,2,3) AND gm.profile_id = 7382782 AND gm.join_state != 2 AND UNIX_TIMESTAMP(g.modified) > 1571722305 ORDER BY g.modified ASC LIMIT 0, 1000;
# User@Host: dstatusnet[dstatusnet] @  [x]
# Query_time: 34.276504  Lock_time: 8.440094 Rows_sent: 1  Rows_examined: 1
SET timestamp=1571817039;
SELECT conversation, is_display as group_type, group_status FROM user_group WHERE id=295414;
# Time: 191023 15:50:40
# User@Host: dstatusnet[dstatusnet] @  [x]
# Query_time: 31.410245  Lock_time: 0.561083 Rows_sent: 0  Rows_examined: 7
SET timestamp=1571817040;
SELECT id, clientScope, msgTop FROM uc_application where site_id=106762 AND msgTop=1;
# User@Host: dstatusnet[dstatusnet] @  [x]
# Query_time: 38.442446  Lock_time: 2.762945 Rows_sent: 2  Rows_examined: 38
SET timestamp=1571817040;
SELECT g.id,g.group_pinyin, g.external_type,g.external_group_id,g.mount_id,g.is_display as group_type,g.nickname,g.site_id,g.group_logo,g.created,g.modified,g.misc,g.name_update_flag as name_flag, g.logo_level,g.group_status,g.conversation, g.only_admin_invite,g.reach_limit_count,g.water_mark  FROM user_group AS g LEFT JOIN group_member as gm ON g.id = gm.group_id WHERE g.is_display in (1,2,3) AND gm.profile_id = 9867566 AND gm.join_state != 2 AND UNIX_TIMESTAMP(g.modified) > 1571796860 ORDER BY g.modified ASC LIMIT 0, 1000;
# Time: 191023 15:50:46
# User@Host: dstatusnet[dstatusnet] @  [x]
# Query_time: 40.481602  Lock_time: 8.530303 Rows_sent: 0  Rows_examined: 1
SET timestamp=1571817046;
update heartbeat set modify=1571817004 where session_id='89b480936c1384305e35b531221b705332a4aebf';
# Time: 191023 15:50:41
# User@Host: dstatusnet[dstatusnet] @  [x]
# Query_time: 32.748755  Lock_time: 0.000140 Rows_sent: 0  Rows_examined: 3
SET timestamp=1571817041;
SELECT id,nickname,mount_id,is_display,external_type,misc,site_id,conversation as conv_id,only_admin_invite,reach_limit_count,water_mark, group_logo,logo_level,created,modified,name_update_flag as name_flag, group_pinyin, group_status,external_group_id,event_id FROM user_group WHERE id in(1316262,1352179,1338753) AND UNIX_TIMESTAMP(modified) > 1571023021 ORDER BY created ASC;
# Time: 191023 15:50:46
# User@Host: dstatusnet[dstatusnet] @  [x]
# Query_time: 40.764872  Lock_time: 4.829707 Rows_sent: 0  Rows_examined: 7
SET timestamp=1571817046;
SELECT id,nickname,mount_id,is_display,external_type,misc,site_id,conversation as conv_id,only_admin_invite,reach_limit_count,water_mark, group_logo,logo_level,created,modified,name_update_flag as name_flag, group_pinyin, group_status,external_group_id,event_id FROM user_group WHERE id in(673950,1261923,1261748,1262047,1038545,1184675,1261825) AND UNIX_TIMESTAMP(modified) > 1571744114 ORDER BY created ASC;

如上所示: 发现在这个时间段这条sql被执行了多次，时间都超过了30~60s，这个属于严重慢sql了

3.6 那么是什么操作执行了kill呢？

经过询问及排查，发现我们通过pt-kill用到了自动查杀功能，配置的查询超时时间为30s

shell> ps -aux|grep 87.72
Warning: bad syntax, perhaps a bogus '-'? See /usr/share/doc/procps-3.2.8/FAQ
root      3813  0.0  0.0 218988  8072 ?        Ss   Oct19   9:54 perl /usr/bin/pt-kill --busy-time 30 --interval 10 --match-command Query --ignore-user rep|repl|dba|ptkill --ignore-info 40001 SQL_NO_CACHE --victims all --print --kill --daemonize --charset utf8 --log=/tmp/ptkill.log.kill_____xxx_____172.16.87.72 -h172.16.87.72 -uxxxx -pxxxx

shell> cat /tmp/ptkill.log.kill_____xxx_____172.16.87.72
2019-10-23T15:45:37 KILL 75278028 (Query 39 sec) statusnet SELECT t.conv_id, t.notice_id, t.thread_type, t.parent_conv_id, t.seq, t.reply_count, t.last_reply_seq1, t.last_reply_seq2, t.revocation_seq, t.creator_id, t.status, t.operator_id, t.extra1 FROM notice_thread t WHERE parent_conv_id = 3075773 and notice_id in (1571816367355,1571814620251,1571814611880,1571814601809,1571814589604,1571814578823,1571814568507,1571814559399,1571812785487,1571812769810)

2019-10-23T15:45:37 KILL 75277932 (Query 39 sec) statusnet SELECT t.conv_id, t.notice_id, t.thread_type, t.parent_conv_id, t.seq, t.reply_count, t.last_reply_seq1, t.last_reply_seq2, t.revocation_seq, t.creator_id, t.status, t.operator_id, t.extra1 FROM notice_thread t WHERE parent_conv_id = 6396698 and notice_id in (18606128)

2019-10-23T15:45:37 KILL 75191686 (Query 39 sec) statusnet SELECT id,nickname,mount_id,is_display,external_type,misc,site_id,conversation as conv_id,only_admin_invite,reach_limit_count,water_mark, group_logo,logo_level,created,modified,name_update_flag as name_flag, group_pinyin, group_status,external_group_id,event_id FROM user_group WHERE id in(988091,736931,843788,739447,737242,1252702,736180,1150776) AND UNIX_TIMESTAMP(modified) > 1571740050 ORDER BY created ASC

发现果然是pt-kill主动去干掉了，结合之前的发现，真相已经在我们眼前了

3.7 那为什么这个sql跑了这么长时间，就单单这个时间报了呢？

我们结合其他时间的slow日志和当前的时间做个对比

# User@Host: dstatusnet[dstatusnet] @  [x]
# Query_time: 2.046068  Lock_time: 0.000079 Rows_sent: 1  Rows_examined: 1
SET timestamp=1572306496;
SELECT ... FROM user_group WHERE id in(693790) AND UNIX_TIMESTAMP(modified) > 0 ORDER BY created ASC;

我们发现一个规律，当where id in（单个值时）时，基本上是两秒，多个值时，时间就不固定了，这个结论有待其他证据进行进一步确认