学步园

这两天线上服务报了非常多的Error，具体内容如下：

SQLSTATE[HY000] [2013] Lost connection to MySQL server at 'reading initial communication packet', system error: 104

DBA说没有找到MySQL错误日志，网上搜索了一圈也没有找到满意答案。只好自己动手查MySQL源码，查到如下代码：

./libmysql/client.c

/*<br /> Part 1: Connection established, read and parse first packet<br /> */</p> <p> if ((pkt_length=cli_safe_read(mysql)) == packet_error)<br /> {<br /> //调用cli_safe_read，若返回的last_errno是CR_SERVER_LOST，则抛出我遇到的那个错误<br /> if (mysql->net.last_errno == CR_SERVER_LOST)<br /> set_mysql_extended_error(mysql, CR_SERVER_LOST, unknown_sqlstate,<br /> ER(CR_SERVER_LOST_EXTENDED),<br /> "reading initial communication packet",<br /> errno);<br /> goto error;<br /> }

继续跟cli_safe_read函数，是这么定义的：

/*****************************************************************************<br /> Read a packet from server. Give error message if socket was down<br /> or packet is an error message<br /> *****************************************************************************/</p> <p>ulong<br /> cli_safe_read(MYSQL *mysql)<br /> {<br /> NET *net= &mysql->net;<br /> ulong len=0;<br /> init_sigpipe_variables</p> <p> /* Don't give sigpipe errors if the client doesn't want them */<br /> set_sigpipe(mysql);<br /> //开始读数据<br /> if (net->vio != 0)<br /> len=my_net_read(net);<br /> reset_sigpipe(mysql);</p> <p> if (len == packet_error || len == 0)<br /> {<br /> DBUG_PRINT("error",("Wrong connection or packet. fd: %s len: %lu",<br /> vio_description(net->vio),len));<br /> #ifdef MYSQL_SERVER<br /> if (net->vio && vio_was_interrupted(net->vio))<br /> return (packet_error);<br /> #endif /*MYSQL_SERVER*/<br /> end_server(mysql);</p> <p> //当读出的last_errno不是ER_NET_PACKET_TOO_LARGE的，时候，都会置为CR_SERVER_LOST<br /> set_mysql_error(mysql, net->last_errno == ER_NET_PACKET_TOO_LARGE ?<br /> CR_NET_PACKET_TOO_LARGE: CR_SERVER_LOST, unknown_sqlstate);<br /> return (packet_error);<br /> }<br /> if (net->read_pos[0] == 255)<br /> {<br /> if (len > 3)<br /> {<br /> char *pos=(char*) net->read_pos+1;<br /> net->last_errno=uint2korr(pos);<br /> pos+=2;<br /> len-=2;<br /> if (protocol_41(mysql) && pos[0] == '#')<br /> {<br /> strmake(net->sqlstate, pos+1, SQLSTATE_LENGTH);<br /> pos+= SQLSTATE_LENGTH+1;<br /> }<br /> (void) strmake(net->last_error,(char*) pos,<br /> min((uint) len,(uint) sizeof(net->last_error)-1));<br /> }<br /> else<br /> set_mysql_error(mysql, CR_UNKNOWN_ERROR, unknown_sqlstate);<br /> /*<br /> Cover a protocol design error: error packet does not<br /> contain the server status. Therefore, the client has no way<br /> to find out whether there are more result sets of<br /> a multiple-result-set statement pending. Luckily, in 5.0 an<br /> error always aborts execution of a statement, wherever it is<br /> a multi-statement or a stored procedure, so it should be<br /> safe to unconditionally turn off the flag here.<br /> */<br /> mysql->server_status&= ~SERVER_MORE_RESULTS_EXISTS;</p> <p> DBUG_PRINT("error",("Got error: %d/%s (%s)",<br /> net->last_errno, net->sqlstate, net->last_error));<br /> return(packet_error);<br /> }<br /> return len;<br /> }

注释是我添加的，继续往下跟my_net_read函数，函数在libmysql/net.c 中，有点长，不全贴出来了。

在my_net_read函数中，有几种情况会报error：

• 啥也没读到，会重试，重试也不行，会报：ER_NET_FCNTL_ERROR、ER_NET_READ_INTERRUPTED或者ER_NET_READ_ERROR
• 读到了，但是package有问题，会报：ER_NET_PACKETS_OUT_OF_ORDER
• 之后在net_realloc的时候，还会报两种错误。

 没有环境权限、没法复现，dba说明天抓包分析一下，只能先这样了。