reading initial communication packet

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

/* Part 1: Connection established, read and parse first packet */ if ((pkt_length=cli_safe_read(mysql)) == packet_error) { //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 if (mysql->net.last_errno == CR_SERVER_LOST) set_mysql_extended_error(mysql, CR_SERVER_LOST, unknown_sqlstate, ER(CR_SERVER_LOST_EXTENDED), "reading initial communication packet", errno); goto error; }

继续跟cli_safe_read函数，是这么定义的：

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

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。