Linux 系統監控shell腳本

比較粗略的一個腳本：主要監控系統磁盤、CPU、內存、網絡流量、tcp連接數等

代碼如下：

[root@test system_monitor_shell_script]# cat system_monitor.sh
#!/bin/bash
#. /home/hduser/modify_system_variable.sh
. /data/script_ssh/system_monitor_shell_script/modify_system_variable.sh
#1、系統基本信息
################################################################
echo -e "\033[32m############# 系統時間 #############\033[0m"
echo System_Now_Time: `date "+%Y-%m-%d %H:%M:%S"`
#獲取系統時間也可以使用who -b
echo System_Boot_Time: `date -d "$(awk -F. '{print $1}' /proc/uptime) second ago" +"%Y-%m-%d %H:%M:%S"`
echo -e "\033[32m############ 主機名 ###############\033[0m"
echo  Hostname: `hostname`





echo -e "\033[32m########## Ip地址和網卡流量 ##########\033[0m"
#獲取服務器的網卡和IP地址，以便在網卡流量監控時使用。



for Network in `ifconfig |awk -F : '/^[[:alpha:]]/{print $1}'`
do
  FLOWA=/tmp/.$Network
  grep $Network: /proc/net/dev|awk -F: '{print $2}'|awk '{print "INPUT "  $1 " "  "OUTPUT " $9}' > $FLOWA
  #第一次獲取網卡流量，單位字節
  INPUTA=`awk '{print $2}' $FLOWA`
  OUTPUTA=`awk '{print $4}' $FLOWA`
  #獲取網卡流量間隔時間
  sleep $NETTIME
  #第二次獲取網卡流量，單位字節
  grep $Network: /proc/net/dev|awk -F: '{print $2}'|awk '{print "INPUT "  $1 " "  "OUTPUT " $9}' > $FLOWA
  INPUTB=`awk '{print $2}' $FLOWA`
  OUTPUTB=`awk '{print $4}' $FLOWA`
  #計算
  #轉換網卡流量單位為KB
  INPUTC=`echo -e "$INPUTB-$INPUTA" | bc`
  OUTPUTC=`echo -e "$OUTPUTB-$OUTPUTA"| bc`
  INPUTMBA=`echo -e $INPUTC |awk '{printf "%0.3f\n",$1/1024}'`
  OUTPUTMBA=`echo -e $OUTPUTC |awk '{printf "%0.3f\n",$1/1024}'`
  INPUT=`echo -e $INPUTC |awk '{printf "%0.3f\n",$1/1024/1024}'`
  OUTPUT=`echo -e $OUTPUTB |awk '{printf "%0.3f\n",$1/1024/1204}'`
  #判斷網卡流量使用量
  if [ $INPUTC -le 1048576 ];then
        if [ $OUTPUTC -le 1048576 ];then
            echo  Network_Card: $Network    Ip: `/sbin/ifconfig $Network|grep "inet "|awk '{print $2}'`  "輸入流量: " ${INPUTMBA}K/S  "輸出流量: " "${OUTPUTMBA}K/S "
        else
            echo  Network_Card: $Network    Ip: `/sbin/ifconfig $Network|grep "inet "|awk '{print $2}'`  "輸入流量: " ${INPUTMBA}K/S  "輸出流量: " "${OUTPUT}M/S "
        fi
    elif [ $INPUTC -gt 1048576 ];then
        if [ $OUTPUTC -gt 1048576 ];then
            echo  Network_Card: $Network    Ip: `/sbin/ifconfig $Network|grep "inet "|awk '{print $2}'`  "輸入流量: " ${INPUT}M/S "輸出流量: " "${OUTPUTMBA}K/S "
        else
            echo  Network_Card: $Network    Ip: `/sbin/ifconfig $Network|grep "inet "|awk '{print $2}'`  "輸入流量: " ${INPUT}M/S "輸出流量:" "${OUTPUT}M/S "
        fi
    fi
done

#2、CPU監控
##############################################################
echo -e "\033[32m############### CPU平均負載和系統進程數 ##############\033[0m"
#us(user time)：用戶進程執行消耗cpu時間；sy(system time)：系統進程執行消耗cpu時間；id：空閑時間（包括IO等待時間）；wa：等待IO時間。
#CPULIMIT=90
CPU_us=$(vmstat | awk '{print $13}' | sed -n '$p')
CPU_sy=$(vmstat | awk '{print $14}' | sed -n '$p')
CPU_id=$(vmstat | awk '{print $15}' | sed -n '$p')
CPU_wa=$(vmstat | awk '{print $16}' | sed -n '$p')
CPU_st=$(vmstat | awk '{print $17}' | sed -n '$p')
#計算服務器CPU使用率
CPU1=`cat /proc/stat | grep 'cpu ' | awk '{print $2" "$3" "$4" "$5" "$6" "$7" "$8}'`
sleep $CPUTIME
CPU2=`cat /proc/stat | grep 'cpu ' | awk '{print $2" "$3" "$4" "$5" "$6" "$7" "$8}'`
IDLE1=`echo -e $CPU1 | awk '{print $4}'`
IDLE2=`echo -e $CPU2 | awk '{print $4}'`
CPU1_TOTAL=`echo -e $CPU1 | awk '{print $1+$2+$3+$4+$5+$6+$7}'`
CPU2_TOTAL=`echo -e $CPU2 | awk '{print $1+$2+$3+$4+$5+$6+$7}'`
IDLE=`echo -e "$IDLE2-$IDLE1" | bc`
CPU_TOTAL=`echo -e "$CPU2_TOTAL-$CPU1_TOTAL" | bc`
RATE=`echo -e "scale=4;($CPU_TOTAL-$IDLE)/$CPU_TOTAL*100" | bc | awk '{printf "%.2f",$1}'`

echo "用戶進程占用CPU時間: us=$CPU_us" ; echo "系統進程消耗CPU時間: sy=$CPU_sy";
echo "CPU空閑時間:         id=$CPU_id" ; echo "等待I/O時間:         wa=$CPU_wa";
echo "CPU使用率:           ${RATE}%"
CPU_RATE=`echo -e $RATE | cut -d. -f1`
load_average=$(uptime |gawk -F':' '{print $NF}')
running_process=$(top -b -d 1 -n 1 |sed -n '2p' |awk -F',' '{print $2}' |awk '{print $1}')
#running_process=`ps aux |wc -l`
total_process=$(ps -ef |wc -l)
echo  
echo  "CPU_load_average: ${load_average}"
echo  "running_process:""    " ${running_process}
echo  "total_process:""      " ${total_process}

Host_running_time=$(uptime |sed 's/^.*up//' | awk -F "," '{print $1}')
User_connection_number=$(uptime |cut -d , -f 3|awk '{print $1}')
#cat /proc/uptime| awk -F. '{run_days=$1 / 86400;run_hour=($1 % 86400)/3600;run_minute=($1 % 3600)/60;run_second=$1 % 60;printf("系統已運行：%d天%d時%d分%d秒\n",run_days,run_hour,run_minute,run_second)}'
echo -e   "Host_running_time:    `cat /proc/uptime| awk -F. '{run_days=$1 / 86400;run_hour=($1 % 86400)/3600;run_minute=($1 % 3600)/60;run_second=$1 % 60;printf("%d天%d時%d分%d秒\n",run_days,run_hour,run_minute,run_second)}'`"
echo -e   "User_connection_number:    ${User_connection_number}\n" 
#CPU占用最多的十個進程
CPUA=`expr $RATE \> $CPULIMIT`
if [ $CPUA -eq 1  ];then
   echo "$time1 CPU警告:當前CPU使用率$RATE%,大於$CPULIMIT%"; echo \ ;
   echo "process使用CPU情況:"
   ps aux |sort -k3nr |head |awk 'BEGIN{printf "%-10s%-10s%-10s%-10s\n","USER","PID","%CPU","COMMAND"}{printf "%-10s%-10d%-10.1f%-10s\n",$1,$2,$3,$11}'
fi
echo

#3、內存監控
####################################################################
echo -e "\033[32m########### 系統內存使用情況 ##############\033[0m" 
#內存使用情況
#MEMLIMIT=80
Total_Mem=$(free -m |sed -n '2p' |awk '{print $2}')
Usage_Mem=$(free -m |sed -n '2p' |awk '{print $3}')
Free_Mem=$(free -m |sed -n '2p' |awk '{print $4}')
Mem_Usage_Percent=`free -m |sed -n '2p'|awk '{printf "%-1d",$3/$2*100}'`
#交換分區使用情況
Swap_Total_Mem=$(free -m |grep Swap |sed -n 'p' |awk '{print $2}')
Swap_Usage_Mem=$(free -m |grep Swap |sed -n 'p' |awk '{print $3}')
Swap_Free_Mem=$(free -m |grep Swap |sed -n 'p' |awk '{print $4}')
Swap_Mem_Usage_Percent=`free -m |grep Swap|sed -n 'p'|awk '{printf "%-1d",$3/$2*100}'`
echo "總內存:"${Total_Mem}M
echo "使用內存: ${Usage_Mem}M"; echo "剩余內存: ${Free_Mem}M"
echo -e "內存使用率: ${Mem_Usage_Percent}%\n" 
echo "交換分區使用情況:"
echo "總內存: ${Swap_Total_Mem}M" ;echo "使用內存: ${Swap_Usage_Mem}M"
echo "剩余內存: ${Swap_Free_Mem}M";echo -e "使用率: ${Swap_Mem_Usage_Percent}%\n"
#判斷內存使用是否超過80%
MA=`expr $Mem_Usage_Percent \> $MEMLIMIT`
if [ $MA -eq 1 ];then
    echo "$time1 內存警告,當前內存使用$Mem_Usage_Percent% 大於$MEMLIMIT%"
fi

MB=`expr $Swap_Mem_Usage_Percent \> $MEMLIMIT`
if [ $MB -eq 1 ];then
   echo "$time1 內存警告,當前內存$Swap_Mem_Usage_Percent% 大於$MEMLIMIT%"
fi
echo "process使用內存情況:"
ps aux |sort -k4nr |head |awk 'BEGIN{printf "%-10s%-10s%-10s%-10s\n","USER","PID","%MEM","COMMAND"}{printf "%-10s%-10d%-10.1f%-10s\n",$1,$2,$4,$11}'
#PCPID=`ps aux |sort -k4nr |head |awk 'BEGIN{printf "%-10s%-10s%-10s%-10s\n","USER","PID","%MEM","COMMAND"}{printf "%-10s%-10d%-10.1f%-10s\n",$1,$2,$4,$11}' |grep -v -i user |awk '{print $2}'`
#PDMEM=`ps aux |sort -k4nr |head |awk 'BEGIN{printf "%-10s%-10s%-10s%-10s\n","USER","PID","%MEM","COMMAND"}{printf "%-10s%-10d%-10.1f%-10s\n",$1,$2,$4,$11}' |grep -v -i user |awk '{print $3}'`
#for i in $PCPID
#do
#    echo $i;ps -ef |grep $i |awk '{print $NF}' |head -1
#done
echo 
#4、磁盤監控
#################################################################
#ROOT=80
#VAR=90
#HOME=90
#BOOT=80
#DATA=90
NULL=/dev/null
echo -e "\033[32m############### 磁盤使用情況 ################\033[0m"
disk=/tmp/.disk
#parted適用於Ubuntu
#parted -l |grep -i 'disk' |grep '/dev/[a-z]d[a-z]' |awk -F: '{print $1}' |awk '{print $2}' > /tmp/.disknumber
#/sbin/fdisk -l |grep -i 'disk' |grep '/dev/[a-z]d[a-z]' |awk -F: '{print $1}' |awk '{print $2}' > /tmp/.disknumber
   disk_data_usage=`df -h  |grep  "data$"  |awk '{print $5}'|awk -F% '{print $1}'`
   disk_root_usage=`df -h  |grep  "/$"    |awk '{print $5}'|awk -F% '{print $1}'`
   disk_home_usage=`df -h  |grep  "home$"    |awk '{print $5}'|awk -F% '{print $1}'`
   disk_var_usage=`df -h   |grep  "var"   |awk '{print $5}'|awk -F% '{print $1}'`
   disk_boot_usage=`df -h  |grep  "boot"  |awk '{print $5}'|awk -F% '{print $1}'`
   disk_data2_usage=`df -h |grep  "data2" |awk '{print $5}'|awk -F% '{print $1}'`
   disk_data3_usage=`df -h |grep  "data3" |awk '{print $5}'|awk -F% '{print $1}'`
   disk_data4_usage=`df -h |grep  "data4" |awk '{print $5}'|awk -F% '{print $1}'`
   disk_data5_usage=`df -h |grep  "data5" |awk '{print $5}'|awk -F% '{print $1}'`
#獲取磁盤分區使用情況
df -h |awk '{print $NF}' > /tmp/.disk
grep home $disk > $NULL
if [ $? -eq 0 ];then
    if [ $disk_home_usage -ge $HOME ];then
       echo "    "/home分區警告:當前使用$disk_home_usage%，大於$HOME
    else
       echo "    "當前/home分區使用:"         "$disk_home_usage%
    fi
fi
grep  "/$" $disk > $NULL
if [ $? -eq 0 ];then
    if [ $disk_root_usage -ge $ROOT ];then
       echo "    "/分區警告:當前使用$disk_root_usage%，大於$ROOT
    else
       echo "    "當前/分區使用:"         "$disk_root_usage%
    fi

fi
grep  "var" $disk > $NULL
if [ $? -eq 0 ];then
    if [ $disk_var_usage -ge $VAR ];then
       echo "    "/var分區警告:當前使用$disk_var_usage%，大於$VAR
    else
       echo "    "當前/var分區使用:"         "$disk_var_usage%
    fi

fi
grep  "boot" $disk > $NULL
if [ $? -eq 0 ];then
    if [ $disk_boot_usage -ge $BOOT ];then
       echo "    "/boot分區警告:當前使用$disk_boot_usage%，大於$BOOT
    else
       echo "    "當前/boot分區使用:"         "$disk_boot_usage%
    fi
fi
grep  "data" $disk > $NULL
if [ $? -eq 0 ];then
    if [ $disk_data_usage -ge $DATA ];then
       echo "    "/data分區警告:當前使用$disk_data_usage%，大於$DATA
    else
       echo "    "當前/data分區使用:"         "$disk_data_usage%
    fi
fi
grep  "data2" $disk > $NULL
if [ $? -eq 0 ];then
    if [ $disk_data2_usage -ge $DATA ];then
       echo "    "/data2分區警告:當前使用$disk_data2_usage%，大於$DATA
    else
       echo "    "當前/data2分區使用:"         "$disk_data2_usage%
    fi
fi
grep  "data3" $disk > $NULL
if [ $? -eq 0 ];then
    if [ $disk_data3_usage -ge $DATA ];then
       echo "    "/data3分區警告:當前使用$disk_data3_usage%，大於$DATA
    else
       echo "    "當前/data3分區使用:"         "$disk_data3_usage%
    fi
fi
grep  "data4" $disk > $NULL
if [ $? -eq 0 ];then
    if [ $disk_data4_usage -ge $DATA ];then
       echo "    "/data4分區警告:當前使用$disk_data4_usage%，大於$DATA
    else
       echo "    "當前/data4分區使用:"         "$disk_data4_usage%
    fi
fi
grep  "data5" $disk > $NULL
if [ $? -eq 0 ];then
    if [ $disk_data5_usage -ge $DATA ];then
       echo "    "/data5分區警告:當前使用$disk_data5_usage%，大於$DATA
    else
       echo "    "當前/data5分區使用:"         "$disk_data5_usage%
    fi
fi
echo



#5、連接數
######################################################################
tcpfile=/tmp/.tcp
#查看並發連接數
#描述
#CLOSED：無連接是活動的或正在進行 
#LISTEN：服務器在等待進入呼叫 
#SYN_RECV：一個連接請求已經到達，等待確認 
#SYN_SENT：應用已經開始，打開一個連接 
#ESTABLISHED：正常數據傳輸狀態 
#FIN_WAIT1：應用說它已經完成 
#FIN_WAIT2：另一邊已同意釋放 
#ITMED_WAIT：等待所有分組死掉 
#CLOSING：兩邊同時嘗試關閉 
#TIME_WAIT：另一邊已初始化一個釋放 
#LAST_ACK：等待所有分組死掉
netstat -n | awk '/^tcp/ {++S[$NF]} END {for(a in S) print a, S[a]}' > $tcpfile
grep TIME_WAIT $tcpfile > $NULL
if [ $? -eq 1 ] ; then
echo "TIME_WAIT 0 " >> $tcpfile
fi
grep FIN_WAIT1 $tcpfile > $NULL
if [ $? -eq 1 ] ; then
echo "FIN_WAIT1 0 " >> $tcpfile
fi
grep FIN_WAIT2 $tcpfile  > $NULL
if [ $? -eq 1 ] ; then
echo "FIN_WAIT2 0 " >> $tcpfile
fi
grep CLOSE_WAIT $tcpfile  > $NULL
if [ $? -eq 1 ] ; then
echo "CLOSE_WAIT 0 " >> $tcpfile
fi
grep LAST_ACK $tcpfile  > $NULL
if [ $? -eq 1 ] ; then
echo "LAST_ACK 0 " >> $tcpfile
fi
grep SYN_RECV $tcpfile  > $NULL
if [ $? -eq 1 ] ; then
echo "SYN_RECV 0 " >> $tcpfile
fi
grep CLOSING $tcpfile  > $NULL
if [ $? -eq 1 ] ; then
echo "CLOSING 0 " >> $tcpfile
fi
grep ESTABLISHED $tcpfile  > $NULL
if [ $? -eq 1 ] ; then
echo "ESTABLISHED 0 " >> $tcpfile
fi

TIME_WAITV=`grep TIME_WAIT $tcpfile | awk '{print $2}'`
FIN_WAIT1V=`grep FIN_WAIT1 $tcpfile | awk '{print $2}'`
FIN_WAIT2V=`grep FIN_WAIT2 $tcpfile | awk '{print $2}'`
ESTABLISHEDV=`grep ESTABLISHED $tcpfile | awk '{print $2}'`
SYN_RECVV=`grep SYN_RECV $tcpfile | awk '{print $2}'`
CLOSINGV=`grep CLOSING $tcpfile | awk '{print $2}'`
CLOSE_WAITV=`grep CLOSE_WAIT $tcpfile | awk '{print $2}'`
LAST_ACKV=`grep LAST_ACK $tcpfile | awk '{print $2}'`
echo -e "\033[32m###########Tcp連接數##################\033[0m"
echo "    "當前TIME_WAIT"  " 連接數為 $TIME_WAITV 個。
echo "    "當前FIN_WAIT1"  " 連接數為 $FIN_WAIT1V 個。
echo "    "當前FIN_WAIT2"  " 連接數為 $FIN_WAIT2V 個。
echo "    "當前CLOSE_WAIT" " 連接數為 $CLOSE_WAITV 個。
echo "    "當前ESTABLISHED"" 連接數為 $ESTABLISHEDV 個。
echo "    "當前SYN_RECV"   " 連接數為 $SYN_RECVV 個。
echo "    "當前LAST_ACKV"  " 連接數為 $LAST_ACKV 個。
echo "    "當前CLOSING"    " 連接數為 $CLOSINGV 個。; echo \ ;
echo -e "\033[31m ----------------------------END--------------------------\033[0m"

modify_system_variable.sh文件如下

#!/bin/bash
#計算CPU利用率的時間間隔。
CPUTIME=2     # 單位 s
#CPU限制，當系統使用CPU達到下面值時報警。
CPULIMIT=90 #單位 %
#內存限制，當系統使用內存達到下面值時報警。
MEMLIMIT=80   #單位 %
#磁盤監控。

ROOT=80       # / 分區使用限制。單位 %
VAR=90        # var 分區使用限制。 單位 %
HOME=90        # home 分區使用限制。單位 %
BOOT=80       # boot 分區使用限制。 單位 %
DATA=90       # data1 分區使用限制。單位 %
#計算網卡流量時間間隔
NETTIME=3     #單位 s