参考:
https://www.jianshu.com/p/ed4bb55f44da
前因背景
当系统服务的MySQL错误日志中,发现大量以下类似信息:经常收到客户关于通信故障错误的问题—客户面临间歇性的”Got an error reading communication packet”错误,这里分析这个错误出现的原因,以及如何解决这个问题。
Aborted connection 1055898 to db: 'xxx' user: 'yyy' host: 'xxx.xxx.xxx.xxx' (Got timeout reading communication packets)
官方解释
下面看看官网怎么说:
Aborted_connects:
If a client is unable even to connect, the server increments the Aborted_connects status variable.
A client attempts to access a database but has no privileges for it.
#客户端没有权限但是尝试访问MySQL数据库
A client uses an incorrect password.
#客户端输入的密码有误。
A connection packet does not contain the right information.
#连接包不包含正确信息
takes more than connect_timeout seconds to obtain a connect packet.
#超过连接时间限制,主要是这个系统变量connect_timeout控制(mysql默认是10s,基本上,除非网络环境极端不好,一般不会超时。)
Aborted_clients:
lIf a client successfully connects but later disconnects improperly or is terminated, the server increments the Aborted_clients status variable
The client program did not call mysql_close() before exiting…
#客户端没有进行关闭
The client had been sleeping more than wait_timeout or interactive_timeout seconds without issuing any requests to the server.
#客户端睡眠时间超过了wait_timeout或interactive_timeout秒,而没有向服务器发出任何请求。
The client program ended abruptly in the middle of a data transfer.
#客户端程序在数据传输过程中突然终止。
Aborted_connects OR Aborted_clients:
Other reasons for problems with aborted connections or aborted clients:
the max_allowed_packet variable value is too small or queries require more memory than you have allocated for mysqld
#max_allow_packet设置过小
Use of Ethernet protocol with Linux, both half and full duplex. Some Linux Ethernet drivers have this bug
#Linux以太网驱动程序有这个bug
A problem with the thread library that causes interrupts on reads.#线程库中导致读取中断的问题。
Badly configured TCP/IP. #tcp/iip 配置信息混乱
Faulty Ethernets, hubs, switches, cables, and so forth. This can be diagnosed properly only by replacing hardware.
#故障的以太网、集线器、交换机、电缆等等
https://dev.mysql.com/doc/refman/5.7/en/communication-errors.html
Aborted connection情况下,这也意味着以下几个问题:
- 客户端正常连接,但是被异常结束(可能是程序没有正常关闭连接)
- 客户端sleep的时间超过了wait_timeout、或interactive_timeout的值(这会导致连接被mysql强制关闭)
- 客户端异常终端,或者查询超出max_allowed_packet的值
临时配置解决办法
设置最大包大小
set global max_allowed_packet = 1024*1024*1024;
查看包大小
mysql> show variables like '%max_allowed_packet%';
+--------------------------+------------+
| Variable_name | Value |
+--------------------------+------------+
| max_allowed_packet | 16777216 |
| slave_max_allowed_packet | 1073741824 |
+--------------------------+------------+
永久配置解决办法
[mysqld]
max_allowed_packet=256M
当然,也可能是其它原因导致的。坦白讲,异常中断是很难诊断的,也有可能是和网络、防火墙有关。
从以下几个方面考虑:
- 如果有大量的连接进程处于sleep状态时间较长,也就意味着应用没有正确、及时关闭数据库连接。强烈建议在应用中能恰当地关闭数据库连接,否则就需要依赖mysql的wait_timeout的设置来关闭连接了。
- 建议检查max_allowed_packet的值,确保该值设置的合理,这样客户端就不会接收到"packet too large"消息提示。如果设置不合理,会异常中断连接。
- 建议关注线程的time_wait数量。如果netstat发现有大量的连接处于time_wait状态,表示该建议应用端调整连接关闭问题了。
TIME_WAIT的处理方案
netstat -ano|grep TIME_WAIT
tcp 0 0 xxx.xxx.xxx.xxx:10054 xxx.xxx.xxx.xxx:55586 TIME_WAIT timewait (32.97/0/0)
tcp 0 0 xxx.xxx.xxx.xxx:10054 xxx.xxx.xxx.xxx:55367 TIME_WAIT timewait (27.82/0/0)
tcp 0 0 xxx.xxx.xxx.xxx:10054 xxx.xxx.xxx.xxx:55776 TIME_WAIT timewait (37.09/0/0)
tcp 0 0 xxx.xxx.xxx.xxx:10054 xxx.xxx.xxx.xxx:56505 TIME_WAIT timewait (54.61/0/0)
tcp 0 0 xxx.xxx.xxx.xxx:10054 xxx.xxx.xxx.xxx:55553 TIME_WAIT timewait (31.94/0/0)
tcp 0 0 xxx.xxx.xxx.xxx:10054 xxx.xxx.xxx.xxx:56643 TIME_WAIT timewait (57.73/0/0)
tcp 0 0 xxx.xxx.xxx.xxx:10054 xxx.xxx.xxx.xxx:55221 TIME_WAIT timewait (23.70/0/0)
tcp 0 0 xxx.xxx.xxx.xxx:10054 xxx.xxx.xxx.xxx:55920 TIME_WAIT timewait (41.18/0/0)
网络问题分析
- 检查DNS配置是否有延迟问题。
- 检查是否同时配置了skip_name_resolve,且使用IP验证主机而不是使用主机名。设置该参数后,使用ip验证主机,而不是使用主机名。使用该参数后,mysql授权表中的host列必须是IP地址或者localhost。
- 增加net_read_timeout、net_write_timeout的值,并观察是否还有该错误发生;
- net_read_timeout很少会导致出错,除非网络环境非常差。
mysql的参数设置:
mysql> show variables like '%timeout%';
+-----------------------------+----------+
| Variable_name | Value |
+-----------------------------+----------+
| connect_timeout | 10 |
| interactive_timeout | 1800 |
| lock_wait_timeout | 31536000 |
| net_read_timeout | 30 |
| net_write_timeout | 60 |
| wait_timeout | 1800 |
+-----------------------------+----------+
- 连接异常中断是因为连接没有被正常关闭。
- server端不会导致连接abort,除非客户端/服务器端发生了网络问题。而不是server端的问题。
tcpdump,netstat -s
log_warnings掩耳盗铃
image
image
mysql> show global variables like '%log_warning%';
+---------------+-------+
| Variable_name | Value |
+---------------+-------+
| log_warnings | 2 |
+---------------+-------+
1 row in set (0.00 sec)
如果log_warnings的值大于1,mysql会将类似信息写入错误日志:
[Warning] Aborted connection 305628 to db: 'db' user: 'dbuser' host: 'hostname' (Got an error reading communication packets)
[Warning] Aborted connection 305627 to db: 'db' user: 'dbuser' host: 'hostname' (Got an error reading communication packets)
可以修改一下log_waring的值:
set global log_warnings=1;
但这样直接修改,重启后会失效,修改配置文件mysql.cnf log_warnings = 1