#!/bin/sh
#
# created by yejr, 2007/06/03
#
# 本脚本用于监控MySQL 复制是否运行，并且根据具体的错误代码自动判断是否忽略
#
now=`date +"%Y%m%d%H%M%S"`
StatFile="./slave_status.$now"
echo "show slave status\G" | mysql -uroot -pmypasswd > $StatFile
#取得 io_thread, sql_thread, last_errno 的状态
IoStat=`cat $StatFile | grep Slave_IO_Running | awk '{print $2}'`
SqlStat=`cat $StatFile | grep Slave_SQL_Running | awk '{ print $2}'`
Errno=`cat $StatFile | grep Last_Errno | awk '{print $2}'`
Behind=`cat $StatFile | grep Seconds_Behind_Master | awk '{print $2}'`
#IoStat=`cat $StatFile | head -n 12 | tail -n 1 | awk '{print $2}'`
#SqlStat=`cat $StatFile | head -n 13 | tail -n 1 | awk '{print $2}'`
#Errno=`cat $StatFile | head -n 20 | tail -n 1 | awk '{print $2}'`
if [ $IoStat = 'No' ] || [ $SqlStat = 'No' ] ; then
echo "[chkslave]"
date
 #如果错误代码为 0，则可能是因为网络等原因导致复制中断，直接重新启动复制即可
 if [ "$Errno" -eq 0 ] ; then
  echo "start slave io_thread; start slave sql_thread;" | mysql -uroot -pmypasswd
  echo "start slave io_thread; start slave sql_thread;"
 #如果是一些不是很要紧的错误代码，也可以直接略过
 elif [ "$Errno" -eq 1007 ] || [ "$Errno" -eq 1053 ] || [ "$Errno" -eq 1062 ] || [ "$Errno" -eq 1213 ] \
|| [ "$Errno" -eq 1158 ] || [ "$Errno" -eq 1159 ] || [ "$Errno" -eq 1008 ] ; then
  echo "stop slave; set global sql_slave_skip_counter=1; slave start;" | mysql -uroot -pmypasswd
  echo "stop slave; set global sql_slave_skip_counter=1; slave start;"
 else
  echo `date` "slave is down!!!"
 fi
 # 远远落后于 master
 if [ $Behind -gt 200 ] ; then
   echo `date` "slave is behind master $Behind seconds!!!"
 fi
#删除临时状态文件
rm -f $StatFile
echo "[/chkslave]"
fi