Shell 系统信息脚本，资源瓶颈告警脚本
linux分析系统资源性能瓶颈
1、查看CPU利用率与负载(top、vmstat、sar)
2、查看磁盘、Inode利用率与I/O负载（df、iostat、iotop、sar、dstat）
3、查看内存利用率（free、vmstat）
4、查看TCP连接状态（netstat、 ss）
5、查看CPU与内存占用最高的10个进程（top、ps）
6、查看网络流量（ifconfig、iftop、iptraf）
实例1：系统信息脚本
#!/bin/bash
#show system information
#设置一个函数：检测当前系统版本，并根据系统版本选择yum安装时的指定命令
os_check() {
if [ -e /etc/redhat-release ];then
REDHAT=`cat /etc/redhat-release | cut -d' ' -f1`
else
DEBIAN=`cat /etc/issue | cut -d' ' -f1`
fi
if [ "$REDHAT" == "CentOS" -o "$REDHAT" == "Red" ];then
P_M=yum
elif [ "$DEBIAN"== "Ubuntu" -o "$DEBIAN" == "ubutnu" ];then
P_M=apt-get
else
Operating system does not support.
exit 1
fi
}
case_check() {
case $input in
cpu_load)
#监控CPU的负载情况
echo "-----------------------------------"
for i in `seq -w 1 3`;do
echo -e "\033[32m ${i} \033[0m"
UTIL=`vmstat | awk '{if(NR==3)print 100-$15"%"}'`
USER=`vmstat | awk '{if(NR==3)print $13"%"}'`
SYS=`vmstat |awk '{if(NR==3)print $14"%"}'`
IOWAIT=`vmstat |awk '{if(NR==3)print $16"%"}'`
echo "util:$UTIL"
echo "User use:$USER"
echo "System use:$SYS"
echo "I/O wait:$IOWAIT"
let ++i
sleep 1
done
echo "-----------------------------------"
break
;;
disk_load)
#监控磁盘负载情况
echo "------------------------------------"
for i in `seq -w 1 3`;do
echo -e "\033[32m ${i} \033[0m"
UTIL=`iostat -x -k |awk '/^[v|s]/{OFS=": ";print $1,$NF"%"}'`
READ=`iostat -x -k |awk '/^[v|s]/{OFS=": ";print $1,$6"KB"}'`
WRITE=`iostat -x -k |awk '/^[v|s]/{OFS=": ";print $1,$7"KB"}'`
IOWAIT=`vmstat |awk '{if(NR==3)print $16"%"}'`
echo -e "util:\n${UTIL}"
echo -e "Read:\n$READ"
echo -e "Write:\n$WRITE"
echo -e "I/O wait:$IOWAIT"
let ++i
sleep 1
done
echo "------------------------------------"
break
;;
disk_use)
#监控磁盘利用率的使用情况
DISK_LOG=/tmp/disk_use.tmp
DISK_TOTAL=`fdisk -l |awk '/^Disk.*bytes/&&/\/dev/{printf $2" ";print $3$4}'`
USE_RATE=`df -h | grep -v /dev/sr0 | awk '/^\/dev/{print int($5)}'`
for i in $USE_RATE;do
if [ $i -gt 10 ];then
PART=`df -h | awk '{if(int($5)=='''$i''')print $6}'`
echo "$(date +%F-%H:%S):$PART = $i%" >> $DISK_LOG
fi
done
echo "-----------------------------------"
echo -e "Disk total:\n$DISK_TOTAL"
if [ -f $DISK_LOG ];then
echo "------------------------------"
cat $DISK_LOG
echo "------------------------------"
rm -rf $DISK_LOG
else
echo "---------------------------------"
fi
break
;;
disk_inode)
#监控磁盘Inode利用率使用情况
INODE_LOG=/tmp/inode_use.tmp
INODE_TOTAL=`df -i |awk '/^\/dev/{print $1": "int($5) "%"}'`
INODE_USE=`df -i |awk '/^\/dev/{print int($5)}'`
for i in $INODE_USE;do
if [ $i -gt 90 ];then
PART=`df -h |awk '{if(int($5)=='''$i''')print $6}'`
echo "$(date +%F-%H%S)$PART = ${i}%" >> $INODE_LOG
fi
done
echo -e "Disk inode: \n$INODE_TOTAL"
if [ -f $INODE_LOG ];then
echo "------------------------------"
rm -rf $INODE_LOG
else
echo "------------------------------"
echo "Inode use rate no than 90% of the partition."
echo "------------------------------"
fi
break
;;
mem_use)
#监控内存使用情况
MEM_TOTAL=`free -m | awk '{if(NR==2)printf "%.1f",$2} END{print "M"}'`
USE=`free -m | awk '{if(NR==2)printf "%.1f",$3} END{print "M"}'`
FREE=`free -m | awk '{if(NR==2)printf "%.1f",$4} END{print "M"}'`
CACHE=`free -m | awk '{if(NR==2)printf "%.1f",$6} END{print "M"}'`
echo -e "Total: $MEM_TOTAL"
echo -e "Use: $USE"
echo -e "Free: $FREE"
echo -e "Cache: $CACHE"
echo "----------------------------"
break
;;
tcp_status)
#监控当前系统TCP连接数量
echo "---------------------------"
COUNT=`ss -ant |awk '!/State/{state[$1]++} END {for(i in state)print i,state[i]}'`
echo -e "TCP connection status:\n$COUNT"
echo "---------------------------"
;;
cpu_top10)
#监控CPU占用率最高的前10个进程以及文件
echo "----------------------------"
CPU_LOG=/tmp/cpu_top.tmp
for i in `seq -w 1 3`;do
#ps aux | awk '{if($3>0.1)printf "PID:" $2 "CPU:" $3 "-->",$11"\n"}' | sort -k2 -nr | head -n 10 > $CPU_LOG'
ps aux | awk '{if($3>0.1) {{ printf "PID:" $2 "CPU:" $3 "-->" }for(i=11;i<=NF;i++)if(i==NF)printf $i"\n";else printf $11}}' | sort -k4 -nr | head -10 > $CPU_LOG
#
if [[ -n `cat $CPU_LOG` ]];then
echo -e "\033[32m ${i} \033[0m"
cat $CPU_LOG
rm -rf $CPU_LOG
else
echo "No process using the CPU."
break
fi
let i++
sleep 1
echo "----------------------------"
done
;;
mem_top10)
#监控内存占用率最高的前10个进程以及文件
echo "----------------------------"
MEM_LOG=/tmp/mem_top.tmp
for i in `seq -w 1 3`;do
ps aux | awk '{if($4>0.1) {{ printf "PID:" $2 "Memory:" $4 "-->" }for(i=11;i<=NF;i++)if(i==NF)printf $i"\n";else printf $11}}' | sort -k4 -nr | head -10 > $MEM_LOG
if [[ -n `cat $MEM_LOG` ]];then
echo -e "\033[32m ${i} \033[0m"
cat $MEM_LOG
rm -rf $MEM_LOG
else
echo "No process using the Memory."
break
fi
let i++
sleep 1
echo "-----------------------------"
done
;;
traffic)
#监控网络流量
echo "----------------------------"
while true;do
read -p "Please enter the network card name(ens[0-9] or em[0-9]):" ens
if [ `ifconfig | grep -c "\<$ens\>"` -eq 1 ];then
break
else
echo "Input format error or Don't have the card name,please input agin"
fi
done
echo "----------------------------"
echo -e "In -------- Out"
for i in `seq -w 1 3`;do
OLD_IN=`ifconfig $ens |awk '/bytes/{if(NR==8)print $4;else if(NR==5)print $5}'`
OLD_OUT=`ifconfig $ens |awk '/bytes/{if(NR==8)print $4;else if(NR==7)print $5}'`
sleep 1
NEW_IN=`ifconfig $ens |awk '/bytes/{if(NR==8)print $4;else if(NR==5)print $5}'`
NEW_OUT=`ifconfig $ens |awk '/bytes/{if(NR==8)print $4;else if(NR==7)print $5}'`
IN=`awk 'BEGIN{printf "%.1f\n",'$((${NEW_IN}-${OLD_IN}))'/1024/128}'`
OUT=`awk 'BEGIN{printf "%.1f\n",'$((${NEW_OUT}-${OLD_OUT}))'/1024/128}'`
echo "${IN}MB/s ${OUT}MB/s"
sleep 1
done
echo "--------------------------"
break
;;
quit)
exit 0
;;
*)
echo "error"
esac
}
#如果没有vmstat命令则进行安装
which vmstat &>/dev/null
if [ $? -ne 0 ];then
echo "vmstat command not found,now the install."
sleep 1
os_check
$P_M install procps -y
echo "-----------------------------------------"
fi
#如果没有iostat命令则进行安装
which iostat &>/dev/null
if [ $? -ne 0 ];then
echo "iostat command not found,now the install."
sleep 1
os_check
$P_M install sysstat -y
echo "-----------------------------------------"
fi
#循环调用函数进行选择
while true
do
select input in cpu_load disk_load disk_use disk_inode mem_use tcp_status cpu_top10 mem_top10 traffic quit
do
case_check
done
done
实例2：如果磁盘根分区使用量超过90%就报警
磁盘根分区使用量： df -Th | grep “/$” | awk ‘{print $6}’ —13%
#!/bin/bash
disk=`df -hT | grep "/$" | awk '{print$6}' | awk -F"%" '{print$1}'`
if [ $disk -ge 13 ];then
echo "`date +%F.%H:%M` 当前根分区使用量已到达13%" | mail -s "disl war ..." root
fi实例3：如果内存使用率超过60%就报警
#!/bin/bash
mem_user=`free -m | grep ^Mem | awk '{print $3}'`
mem_total=`free -m | grep ^Mem | awk '{print $2}'`
mem_percent=$[ mem_user*10 / mem_total ]
war_file=/tmp/mem_war.txt
rm -rf $war_file
if [ $mem_percent -ge 1 ];then
echo "`date +%F.%H:%M` memory的百分比为${mem_percent}%" > $war_file
fi
if [ -f $war_file ];then
mail -s "mem is war..." root < $war_file
rm -rf $war_file
fi