Linux Health Check Scripts

针对Linux的检查，编写了shell脚本，如下：

#!/bin/sh
###############################################################################################################
#Script name: linuxHealthCheck.sh
#Script description: for Linux Health Check
#Current Release Version: 1.0.0
#Script Owner: hbhe0316
#Latest editor: hbhe0316
#Support platform:  Linux OS for redhat and centos.
#Date: 2021/10/15---first Version for linuxHealthCheck.sh.
#Change log:
#Descript:sh linuxHealthCheck.sh
#
###############################################################################################################

export LANG=en_US
osbox=`uname`
RHversion=$(cat /proc/version | sed 's/[^0-9]//g' | cut -b -3)

###打印日志函数
log_info(){
    DATE_N=`date "+%Y-%m-%d %H:%M:%S"`
    USER_N=`whoami`
    echo "${DATE_N} ${USER_N} execute $0 [INFO] $@"  
}

log_error(){
    DATE_N=`date "+%Y-%m-%d %H:%M:%S"`
    USER_N=`whoami`
    echo -e "/033[41;37m ${DATE_N} ${USER_N} execute $0 [ERROR] $@ /033[0m" 
}

fn_log(){
    if [[ $? -eq 0 ]];then
        log_info "$@ sucessed."
        echo -e "/033[32m $@ sucessed. /033[0m"
    else
        log_error "$@ failed."
        echo -e "/033[41;37m $@ failed. /033[0m"
    fi
}

checkExecuteUser(){
    uid=`id -u`
    if [[ $uid == 0 ]];then
        continue
    else
        fn_log "Current excute user is not root ,shell will exist."
        exit 1
    fi
}

###Create /tmp/log if not exist.
mkdirLogPath(){
    if [[ ! -d /tmp/log ]];then
        mkdir -p /tmp/log
        chmod 767 /tmp/log
    fi
    CheckLinuxLogPath="/tmp/log"
    CheckLinuxLog=`hostname`_$(date +%y%m%d).txt    
}

###Get OS Arch Linux or not
getOsArch(){
    if [[ "$osbox" == "Linux" ]];then
        continue
    else
        fn_log "Current OS is $osbox,shell is exit now."
        echo 0
        exit 0
    fi
}

###Get redhat or centos
getOsCentosOrRedhat(){
    cat /proc/version | grep -iE "redhat|centos" > /dev/null
    if [[ $? == 0 ]];then
        continue
    else
        echo "Current OS is not centos or redhat."
        echo 1
        exit 1
    fi
}



networkinfo(){
#先过滤网卡名称,存到数组DEVNAME中
    DEVNAME=(`ifconfig | grep ^[a-z] | awk -F: '{print $1}'`)
#在拿到IP地址，存到数组IPADDR中
    IPADDR=(`ifconfig | grep 'inet' | sed 's/^.*inet //g' | sed 's/ *netmask.*$//g' | grep -v "inet6"`)
##get RX
    RX=(`ifconfig | grep 'RX' | awk '{print $3/1024/1024}'`)
    TX=(`ifconfig | grep 'TX' | awk '{print $3/1024/1024}'`)
    RX_ERROR=(`ifconfig | grep "RX errors" | awk -F ' ' {'print $3'}`)
    TX_ERROR=(`ifconfig | grep "TX errors" | awk -F ' ' {'print $3'}`)

    for ((i=0;i<${#DEVNAME[@]};i++))
    do
        echo ${DEVNAME[$i]} ${IPADDR[$i]} ${RX[$i]} ${TX[$i]} ${RX_ERROR[$i]} ${TX_ERROR[$i]}
    done
}

osSystem(){
    os_type=`uname`
    echo "Current System is:$os_type"
    os_banben=`cat /etc/redhat-release`
    echo "Current System Version is:$os_banben"
    os_neihe=`uname -r`
    echo "Current System Kernel is:$os_neihe"
    os_time=`date +%F_%T`
    echo "Current System Time is:$os_time"
    os_last=`uptime |awk '{print $3}'`
    echo "Current System Last Reboot Time is:$os_last Days."
    os_hostname=`hostname`
    echo "Current System Hostname is:$os_hostname"
}


#硬件信息
cpuInfo(){
    #cpu的物理个数
    #cpu_phy=`cat /proc/cpuinfo | grep "physical id" | wc -l`
    #echo "当前系统物理cpu为:$cpu_phy"
    #cpu的核心数
    #cpu_core=`cat /proc/cpuinfo | grep "core" | wc -l`
    #echo "当前系统cpu核心数为:$cpu_core"
    #cpu的型号
    #cpu_model=`cat /proc/cpuinfo | grep "model" | sed -n '2p' | awk -F : '{print $2}'`
    #echo "当前系统cpu型号是:$cpu_model"
    cpu_idle=$(top -b -n 1 | grep Cpu | awk '{print $2}' | cut -f 1 -d ".")
    if [[ "$cpu_idle" -gt 80 ]]; then
        echo "Check System CPU used is geater than 80%,warning."
    else
        echo "Check System CPU used is $cpu_idle%,success."
    fi
}
##内存信息
memInfo(){
    mem_total=`free -m | grep Mem | awk '{print $2}'`
#    echo "内存总大小为:$mem_total"
    mem_used=`free -m | grep Mem | awk '{print $3}'`
#    echo "已使用内存为:$mem_used"
    mem_free=`free -m | grep Mem | awk '{print $4}'`
#    echo "剩余内存为:$mem_free"

    #百分比
    mem_used_percent=`echo "scale=2;$mem_used/$mem_total*100"|bc`
    mem_used_percent=`echo $mem_used_percent | awk -F '.' {'print $1'}`
#    echo "已使用内存百分比为:$mem_used_percent"
    if [[ $mem_used_percent -gt 80 ]];then
        echo "Check System have memory used percent geater than 80%,warning."
    else
        echo "Check System have memory used percent is $mem_used_percent%,success."
    fi
}


swapInfo(){
    swap_total=`free -m | grep Swap | awk '{print $2}'`
    swap_used=`free -m | grep Swap | awk '{print $3}'`
    swap_used_percent=`echo "scale=2;$swap_used/$swap_total*100"|bc`
    swap_used_percent=`echo $swap_used_percent | awk -F '.' {'print $1'}`
    if [[ $swap_used_percent -gt 80 ]];then
        echo "Check System have swap used percent geater than 80%,warning."
    else
        echo "Check Swap Used is $swap_used_percent%,success."
    fi
}

##get df -h infomation
dfInfo(){
    percent=`df -h | awk '{print int($5)}'`
    for each_one in $percent
    do
        if [[ $each_one -gt 80 ]];then
            echo "Check System have filedisk used geater than 80%,warning."
        else
            echo "Check File System success."
        fi
    done
}

userLogin(){
    userNum=`uptime |awk '{print $6}'`
    if [[ $userNum -gt 3 ]];then
        echo "Check System User Login Number is greater than 3,warning."
    else
        echo "Check Current User Login is $userNum,success."
    fi
}

checkDefunct(){
    DefunctNum=`ps -A -o stat,ppid,pid,cmd | grep -e '^[Zz]' | wc -l`
    if [[ $DefunctNum == 0 ]];then
        echo "Check Defunct process,success."
    else
        echo "Check System  have Defunct process,warning."
    fi
}

checkIOInfo(){
    IOPercent=`iostat | head -4  | awk 'END {print}' | awk -F ' ' {'print $6'} | sed 's/[ /t]//g'`
    IOPercent=`echo $IOPercent | awk -F '.' {'print $1'}`
    if [[ $IOPercent -lt 70 ]];then
        echo "Check System is letter than 70%,warning."
    else
        echo "Check iostat is $IOPercent%,success."
    fi
}

checkNtp(){
    ntpstat > /dev/null 2>&1
    if [[ $? == 0 ]];then
        echo "Check ntp status success."
    else
        echo "Check ntp status warning."
    fi
}


main(){
    checkExecuteUser
    getOsArch
    getOsCentosOrRedhat
    mkdirLogPath
    osSystem
    cpuInfo
    memInfo
    swapInfo
    dfInfo
    checkIOInfo
#    networkinfo
    userLogin
    checkDefunct
    checkNtp
}

main