监控系统脚本

#!/bin/sh
#FileName: Monitor.sh
#make by GuoYanLiang
#source ~/.bash_profile
export LANG="zh_CN.GB2312"
export LC_ALL="zh_CN.GB2312"
export LC_CTYPE="zh_CN.GB2312"
#locale
CheckDir="/root/gyl/Monitor"
[ ! -d "$CheckDir" ] && mkdir -p $CheckDir
cd $CheckDir
MailFile=/root/gyl/Monitor/MailSendFile.log
MonitorTMP=/root/gyl/Monitor/MonitorTMP.log
MonitorLog=/root/gyl/Monitor/logs/Monitor_$(date +"%Y%m%d").log
cp -f /dev/null $MailFile
cp -f /dev/null $MonitorTMP
#cp -f /dev/null $MonitorLog
#===================================================================init===================================================================
init()
{
IPAddress=`/sbin/ifconfig | grep 'inet addr:' | head -1 | awk '{print $2}' | sed 's/addr://g'`;
NOWTIME=$(date "+%F %T");
MAILTO="280637803@qq.com,13866112949@139.com";
MAIL_SUB="服务器性能监控";
MAIL_FROM="运维组";
MAIL_SERVER="192.168.18.8"
MAIL_SERVER_PORT="22"
ROOT_ID_RSA=~/.ssh/18.8_id_rsa
unset StateTMP
unset State
MailSendFlag=0
}
#===================================================================mhead===================================================================
mhead()
{
echo "" >>$MailFile
echo "服务器 : $IPAddress" >>$MailFile
echo "时  间 : $NOWTIME" >>$MailFile
echo -e "状  态 : $State" >>$MailFile
echo "" >>$MailFile
}
#===================================================================smail===================================================================
smail()
{
scp -i $ROOT_ID_RSA -P $MAIL_SERVER_PORT $MailFile root@$MAIL_SERVER:$MailFile
ssh -i $ROOT_ID_RSA -p $MAIL_SERVER_PORT root@$MAIL_SERVER  "export LANG=zh_CN.GB2312;export LC_ALL=zh_CN.GB2312;export LC_CTYPE=zh_CN.GB2312;mail -s $MAIL_SUB $MAILTO -F $MAIL_FROM < $MailFile;cp /dev/null $MailFile;"
}
#===================================================================Monitor_System===================================================================
Monitor_System()
{
distrib="unknown"
if [ -f /etc/SuSE-release ] ; then
distrib="`head -1 /etc/SuSE-release`"
fi
if [ -f /etc/redhat-release ] ; then
distrib="`head -1 /etc/redhat-release`"
fi
HostName=`hostname`
KernelVer=`uname -r`
CPUNum=`cat /proc/cpuinfo | grep "processor" | wc -l `
CPUMHz=`cat /proc/cpuinfo | grep MHz | uniq | awk -F: '{print $2}'`
MemTotal=`free -m | head -2 | tail -1 | awk '{print $2}'`
MemFree=`free -m | head -3 | tail -1 | awk '{print $3}'`
echo "系统检测时间 : " `date +%Y"."%m"."%d" "%k":"%M":"%S` >> $MonitorTMP
echo "主机 : " $HostName >> $MonitorTMP
echo "IP地址 : " $IPAddress >> $MonitorTMP
echo "系统版本 : " $distrib >> $MonitorTMP
echo "系统内核 : " $KernelVer >> $MonitorTMP
echo "CPU数目 : " $CPUNum 个 >> $MonitorTMP
echo "CPU频率 : " $CPUMHz MHz >> $MonitorTMP
echo "总内存 : " $MemTotal M >> $MonitorTMP
echo "空闲内存 : " $MemFree M >> $MonitorTMP
echo "系统进程数 : " `ps -efl | wc -l` 个 >> $MonitorTMP
echo "系统启动时间 : " `uptime | awk -F, '{print $1,$2}'` >> $MonitorTMP
echo "系统检测时间 : " `date +%Y"."%m"."%d" "%k":"%M":"%S` >> $MailFile
echo "主机 : " $HostName >> $MailFile
echo "IP地址 : " $IPAddress >> $MailFile
echo "系统版本 : " $distrib >> $MailFile
echo "系统内核 : " $KernelVer >> $MailFile
echo "CPU数目 : " $CPUNum 个 >> $MailFile
echo "CPU频率 : " $CPUMHz MHz >> $MailFile
echo "总内存 : " $MemTotal M >> $MailFile
echo "空闲内存 : " $MemFree M >> $MailFile
echo "系统进程数 : " `ps -efl | wc -l` 个 >> $MailFile
echo "系统启动时间 : " `uptime | awk -F, '{print $1,$2}'` >> $MailFile
Upday=`uptime | awk '{print $3}'`
if [ $Upday -lt 1 ]
then
MailSendFlag=1
echo "服务器最近有重启,请检查" >> $MonitorTMP
echo "服务器最近有重启,请检查" >> $MailFile
StateTMP=" 系统启动时间 : "`uptime | awk -F, '{print $1,$2}'`",服务器最近有重启,请检查"
fi
}
#===================================================================Monitor_memory===================================================================
# Monitor the MemFree utilization if exceed the threshold
Monitor_memory()
{
MemThreshold=45
MemTotal=`free -m | head -2 | tail -1 | awk '{print $2}'`
MemFree=`free -m | head -3 | tail -1 | awk '{print $3}'`
MemPercentage=`expr $MemFree * 100 / $MemTotal`
echo "" >>$MonitorTMP;
echo "内存使用情况:" >>$MonitorTMP;
free -m >>$MonitorTMP;
echo "" >>$MonitorTMP;
echo "内存占用率: $MemPercentage % = ($MemFree/$MemTotal)*100%" >>$MonitorTMP
if [ $MemPercentage -gt $MemThreshold ]; then
MailSendFlag=1
echo "内存占用率 $MemPercentage 高于阀值 $MemThreshold %" >>$MonitorTMP
echo "" >>$MailFile
echo "内存使用情况:" >>$MailFile
free -m >>$MailFile
echo "" >>$MailFile
echo "内存占用率 $MemPercentage 高于阀值 $MemThreshold %" >>$MailFile
StateTMP="$StateTMP 内存占用为 $MemPercentage 高于阀值 $MemThreshold %"
fi
}
#===================================================================Monitor_cpu===================================================================
# Monitor the CPU Max,Avg utilization if exceed the threshold
Monitor_cpu()
{
CPUAvgThreshold=8
CPUMaxThreshold=15
if [ -x /usr/bin/vmstat ] ; then
CPUMax=0
CPUTotal=0
for x in `vmstat 2 6 | tail -5 | awk '{print 100-$15}'`
do
if [ $x -gt $CPUMax ]; then
CPUMax="$x"
fi
CPUTotal=`expr $CPUTotal + $x`
done
CPUAvg=`expr $CPUTotal / 5`
echo "" >>$MonitorTMP
echo "CPU最大占用率 : " $CPUMax >>$MonitorTMP
if [ $CPUMax -gt $CPUMaxThreshold ]; then
MailSendFlag=1
echo "CPU最大占用率 $CPUMax 高于阀值 $CPUMaxThreshold %" >>$MonitorTMP
echo "" >>$MailFile
echo "CPU最大占用率 $CPUMax 高于阀值 $CPUMaxThreshold %" >>$MailFile
StateTMP="$StateTMP CPU最大占用率 $CPUMax 高于阀值 $CPUMaxThreshold %"
fi
echo "" >>$MonitorTMP
echo "CPU平均占用率 : " $CPUAvg >>$MonitorTMP
if [ $CPUAvg -gt $CPUAvgThreshold ]; then
MailSendFlag=1
echo "CPU平均占用率 $CPUAvg 高于阀值 $CPUAvgThreshold %" >>$MonitorTMP
echo "" >>$MailFile
echo "CPU平均占用率 $CPUAvg 高于阀值 $CPUAvgThreshold %" >>$MailFile
StateTMP="$StateTMP CPU平均占用率 $CPUAvg 高于阀值 $CPUAvgThreshold %"
fi
fi
}
#===================================================================Monitor_Process_CPU===================================================================
# Check the ProcessCPU Maximum utilization if exceed the threshold
Monitor_Process_CPU()
{
ProcessCPUThreshold=30
echo "" > Process_CPU.log
echo " Top5: 单个进程CPU占用率 : " >> Process_CPU.log
ps aux | head -1 >> Process_CPU.log
ps aux | grep -v "ps aux" | grep -v "Monitor.sh" | grep -v USER | grep -v wireless-bak | sort -nr -k 3 | head -5 >> Process_CPU.log
cat Process_CPU.log >>$MonitorTMP
ProcessCPU=`cat Process_CPU.log | grep -v Top5 | grep -v USER | head -2 | tail -1 | awk '{print $3}' | cut -f 1 -d "."`
if [ $ProcessCPU -gt $ProcessCPUThreshold ]; then
MailSendFlag=1
echo "" >>$MonitorTMP
echo "单个进程CPU占用率最大为 $ProcessCPU 高于阀值 $ProcessCPUThreshold %" >> $MonitorTMP
echo "" >>$MailFile
cat Process_CPU.log >>$MailFile
echo "" >>$MailFile
echo "单个进程CPU占用率最大为 $ProcessCPU 高于阀值 $ProcessCPUThreshold %" >> $MailFile
StateTMP="$StateTMP 单个进程CPU占用率最大为 $ProcessCPU 高于阀值 $ProcessCPUThreshold %"
fi
}
#===================================================================Monitor_Process_MEM===================================================================
# Check the ProcessMEM Maximum utilization if exceed the threshold
Monitor_Process_MEM()
{
ProcessMEMThreshold=20
echo "" > Process_MEM.log
echo " Top5: 单个进程内存占用率 : " >> Process_MEM.log
ps aux | head -1 >> Process_MEM.log
ps aux | grep -v "ps aux" | grep -v "Monitor.sh" | grep -v USER | grep -v wireless-bak | sort -nr -k 4 | head -5 >> Process_MEM.log
cat Process_MEM.log >>$MonitorTMP
ProcessMEM=`cat Process_MEM.log | grep -v Top5 | grep -v USER | head -2 | tail -1 | awk '{print $4}' | cut -f 1 -d "."`
if [ $ProcessMEM -gt $ProcessMEMThreshold ]; then
MailSendFlag=1
echo "" >>$MonitorTMP
echo "单个进程内存占用率最大为 $ProcessMEM 高于阀值 $ProcessMEMThreshold %" >> $MonitorTMP
echo "" >>$MailFile
cat Process_MEM.log >>$MailFile
echo "" >>$MailFile
echo "单个进程内存占用率最大为 $ProcessMEM 高于阀值 $ProcessMEMThreshold %" >> $MailFile
StateTMP="$StateTMP 单个进程内存占用率最大为 $ProcessMEM 高于阀值 $ProcessMEMThreshold %"
fi
}
#===================================================================Monitor_Disk===================================================================
Monitor_Disk()
{
DiskThreshold=88
DiskNum=0
echo "" >> $MonitorTMP
echo "磁盘分区占用率:" >> $MonitorTMP
df -hP >> $MonitorTMP
for x in `df -hP | grep / | awk '{print $5}' | sed 's/%//g'`
do
if [ $x -gt $DiskThreshold ]; then
DiskNum=`expr $DiskNum + 1`
fi
done
if [ $DiskNum -gt 0 ] ; then
MailSendFlag=1
echo "" >> $MonitorTMP
echo "磁盘空间占用高于阀值 $DiskThreshold %" >> $MonitorTMP
echo "" >> $MailFile
echo "磁盘分区占用率:" >> $MailFile
df -hP >> $MailFile
echo "" >> $MailFile
echo "$DiskNum 个磁盘分区占用率高于阀值 $DiskThreshold %" >> $MailFile
StateTMP="$StateTMP $DiskNum 个磁盘分区占用率高于阀值 $DiskThreshold %"
fi
}
#===================================================================Monitor_process===================================================================
Monitor_process()
{
OS_Process=(java httpd)
OS_Process_Num=${#OS_Process[@]}
for (( i=0; i<$OS_Process_Num; i++ ))
do
Process_Num=0
Process_Num=`ps -efl | grep ${OS_Process[i]} | grep -v grep | wc -l`
echo "" >>$MonitorTMP
echo "${OS_Process[i]} 有 $Process_Num 个进程运行." >>$MonitorTMP
if [ $Process_Num -eq 0 ] ; then
MailSendFlag=1
echo "${OS_Process[i]} 进程不存在" >>$MonitorTMP
echo "" >>$MailFile
echo "业务进程信息:" >>$MailFile
echo "${OS_Process[i]} 有 $Process_Num 个进程运行." >>$MailFile
echo "${OS_Process[i]} 进程不存在" >>$MailFile
StateTMP="$StateTMP ${OS_Process[i]} 进程不存在"
fi
done
}
#===================================================================Monitor_Ping===================================================================
Monitor_Ping()
{
Ping_Num=10
Ping_Time=0.02
DesIP=(192.168.28.22 192.168.28.26 192.168.18.49)
DesIP_Num=${#DesIP[@]}
echo "" >>$MonitorTMP
echo "网络Ping检测:" >>$MonitorTMP
for (( j=0; j<$DesIP_Num; j++ ))
do
ping -i $Ping_Time -c $Ping_Num ${DesIP[j]} > Ping_Out.log
echo "" >>$MonitorTMP
echo `expr $j + 1`" ,Ping测试: ping -i $Ping_Time -c $Ping_Num ${DesIP[j]}" >>$MonitorTMP
cat Ping_Out.log >>$MonitorTMP
Ping_Loss=`cat Ping_Out.log | grep % | awk '{print $6}' | sed 's/%//g'`
if [ $Ping_Loss -gt 0 ] ; then
MailSendFlag=1
echo "" >>$MailFile
echo "Ping测试: ping -i $Ping_Time -c $Ping_Num ${DesIP[j]} 丢包率: $Ping_Loss % ,请检查网络" >>$MailFile
cat Ping_Out.log >>$MailFile
StateTMP="$StateTMP Ping ${DesIP[j]} 丢包率: $Ping_Loss % "
fi
done
}
#===================================================================Monitor_Ifconfig===================================================================
Monitor_Ifconfig()
{
ErrorsNum=0
echo "" >>$MonitorTMP
echo "网卡信息(ifconfig):" >> $MonitorTMP
/sbin/ifconfig >> $MonitorTMP
for x in `ifconfig | grep errors | awk '{print $3}' | sed 's/errors://g'`
do
if [ $x -gt 0 ]; then
ErrorsNum=`expr $ErrorsNum + 1`
fi
done
if [ $ErrorsNum -gt 0 ] ; then
MailSendFlag=1
echo "网卡数据包存在错误" >> $MonitorTMP
echo "" >> $MailFile
echo "网卡信息(ifconfig):" >> $MailFile
ifconfig >> $MailFile
echo "网卡数据包存在错误" >> $MailFile
StateTMP="$StateTMP 网卡数据包存在错误"
fi
}
#===================================================================Monitor_Config===================================================================
Monitor_Config()
{
ConfigName=(ifcfg-bond0 ifcfg-eth0 ifcfg-eth1 network resolv.conf hosts rc.local mount fstab crontab iptables route)
ConfigName_Num=${#ConfigName[@]}
ConfigCheck=("cat /etc/sysconfig/network-scripts/ifcfg-bond0" "cat /etc/sysconfig/network-scripts/ifcfg-eth0" "cat /etc/sysconfig/network-scripts/ifcfg-eth1" "cat /etc/sysconfig/network" "cat /etc/resolv.conf" "cat /etc/hosts" "cat /etc/rc.local" "mount" "cat /etc/fstab" "crontab -l" "/sbin/iptables -L" "/sbin/route")
echo "" >>$MonitorTMP
echo "监控系统配置信息:" >>$MonitorTMP
for (( k=0; k<$ConfigName_Num; k++ ))
do
echo "" > ${ConfigName[k]}.log
${ConfigCheck[k]} >> ${ConfigName[k]}.log
echo "" >>$MonitorTMP
echo `expr $k + 1`" , ${ConfigName[k]} 配置:" >> $MonitorTMP
cat ${ConfigName[k]}.log >> $MonitorTMP
if [ -f ${ConfigName[k]}.ref ]
then
diff ${ConfigName[k]}.log ${ConfigName[k]}.ref > ${ConfigName[k]}.diff
if [ `cat ${ConfigName[k]}.diff | wc -l` -gt 0 ]
then
MailSendFlag=1
 
`expr $SectionN + 1`
 
echo "" >>$MonitorTMP
echo `expr $k + 1`" , ${ConfigName[k]} 原配置:" >>$MonitorTMP
cat ${ConfigName[k]}.ref >>$MonitorTMP
echo "" >>$MonitorTMP
echo `expr $k + 1`" , ${ConfigName[k]} 配置更改内容:" >>$MonitorTMP
cat ${ConfigName[k]}.diff >>$MonitorTMP
echo "" >> $MailFile
echo "${ConfigName[k]} 现配置:" >> $MailFile
cat ${ConfigName[k]}.log >> $MailFile
echo "" >> $MailFile
echo "${ConfigName[k]} 原配置:" >> $MailFile
cat ${ConfigName[k]}.ref >> $MailFile
echo "" >> $MailFile
echo "${ConfigName[k]} 配置更改内容:" >> $MailFile
cat ${ConfigName[k]}.diff >> $MailFile
StateTMP="$StateTMP ${ConfigName[k]} 配置已经更改"
fi
cp -f ${ConfigName[k]}.log ${ConfigName[k]}.ref
else
cp -f ${ConfigName[k]}.log ${ConfigName[k]}.ref
fi
done
}
#===================================================================Monitor_Main===================================================================
init;
Monitor_System;
Monitor_memory;
Monitor_cpu;
Monitor_Process_CPU;
Monitor_Process_MEM;
Monitor_Disk;
Monitor_process;
Monitor_Ping;
Monitor_Ifconfig;
Monitor_Config;
echo "" >> $MonitorLog
echo "" >> $MonitorLog
echo ===================================================== 开始: $NOWTIME ===================================================== >> $MonitorLog
echo "" >> $MonitorLog
cat $MonitorTMP >> $MonitorLog
echo "" >> $MonitorLog
echo ===================================================== 结束: $(date "+%F %T") ===================================================== >> $MonitorLog
echo "" >> $MonitorLog
echo "" >> $MonitorLog
if [ $MailSendFlag != 0 ] ; then
State=$StateTMP
mhead;
smail;
echo OK
fi