#!/bin/bash
# 通用单位转换函数
convert_to_bytes() {
  local value="$1"
  local number=$(echo "$value" | grep -o -E '[0-9]+(\.[0-9]+)?')
  local unit=$(echo "$value" | grep -o -E '[a-zA-Z]+' | tr '[:lower:]' '[:upper:]')
  case "$unit" in
    GB)
      awk "BEGIN {printf \"%.0f\", $number * 1024 * 1024 * 1024}"
      ;;
    MB)
      awk "BEGIN {printf \"%.0f\", $number * 1024 * 1024}"
      ;;
    KB)
      awk "BEGIN {printf \"%.0f\", $number * 1024}"
      ;;
    B)
      echo $number
      ;;
    TB)
      awk "BEGIN {printf \"%.0f\", $number * 1024 * 1024 * 1024 * 1024}"
      ;;
    *)
      echo "Invalid unit: $unit"
      exit 1
      ;;
  esac
}

# 监控机器型号
model=($( dmidecode -t 1 | grep -E "Manufacturer|Product Name" | awk -F"[: ]+" 'NR==1 {first=$2"_"$3; next}{print first"_"$3"_"$4}' ))
sn=($( dmidecode -t 1 | grep -E "Serial Number" | awk -F"[: ]+" '{print $3}' ))
echo "node_machine_info{model=\"$model\",sn=\"$sn\"}" 1

# 监控出口公网IP
echo $(timeout 2 curl -s ifconfig.me | awk '{ip=$1; gsub(/\./, "", ip); print "node_snat_ip{ip=\"" $1 "\"} " ip}')

# 监控物理内存大小及槽位
locators=($( dmidecode -t 17|grep "Locator:"|grep -v Bank|awk '{print $2$3}' ))
sizes=($( dmidecode -t 17|grep "Size:"|grep -Ev "Non-Volatile|Volatile|Cache|Logical"|awk '{print $2$3}' ))
for i in "${!locators[@]}"; do
    locator=${locators[$i]}; size=${sizes[$i]};
    if [[ $size == *"B"* ]];then
        bytes=($( convert_to_bytes $size ));
        echo "node_mem_info{locator=\"$locator\"} $bytes";
    fi;
done

# 监控挂载点读写正常与否
awk '$3 ~ /^(ext2|ext3|ext4|xfs|nfs|cifs|ossfs|glusterfs)$/' /proc/mounts | \
egrep -v "kubelet|run|tmp|nfsd|rpc_pipefs|overlay2|devicemapper" | \
while read -r device mountpoint fs _; do
    status=$(timeout 5 sh -c "touch \"$mountpoint/.disk.tmp\" && rm -f \"$mountpoint/.disk.tmp\"" &>/dev/null && echo "1" || echo "0")
  echo "node_disk_mount{device=\"$device\",fstype=\"$fs\",mountpoint=\"$mountpoint\"} $status"
done

# 监控挂载点是否丢失
awk '$2 != "/" && $2 != "swap" && $2 != "" && !/^#/ {print $2}' /etc/fstab | while read -r mountpoint; do
  if ! grep -q "$mountpoint" /proc/mounts; then
    awk -v mp="$mountpoint" '$2 == mp {print "node_disk_volume_loss{device=\"" $1 "\",mountpoint=\"" $2 "\"} 0"}' /etc/fstab
  fi
done

# 监控物理磁盘
command -v lspci >/dev/null 2>&1 || yum -y install pciutils
if [ -f "/opt/MegaRAID/MegaCli/MegaCli64" ]; then
    slot_numbers=($( /opt/MegaRAID/MegaCli/MegaCli64 -PDList -aALL -NoLog | grep "Slot Number" | awk '{print $3}' ))
    raw_sizes=($( /opt/MegaRAID/MegaCli/MegaCli64 -PDList -aALL -NoLog | grep "Raw" | awk '{print $3 $4}' ))
    firmware_states=($( /opt/MegaRAID/MegaCli/MegaCli64 -PDList -aALL -NoLog | grep "Firmware state" | awk -F"[ ,]+" '{print $3}' ))
    media_errors=($( /opt/MegaRAID/MegaCli/MegaCli64 -PDList -aALL -NoLog | grep "Media Error Count" | awk '{print $4}' ))
    other_errors=($( /opt/MegaRAID/MegaCli/MegaCli64 -PDList -aALL -NoLog | grep "Other Error Count" | awk '{print $4}' ))
    predictive_failures=($( /opt/MegaRAID/MegaCli/MegaCli64 -PDList -aALL -NoLog | grep "Predictive Failure Count" | awk '{print $4}' ))
    for i in "${!slot_numbers[@]}"; do
        slot=${slot_numbers[$i]}
        size=${raw_sizes[$i]}
        state=${firmware_states[$i]}
        media_error=${media_errors[$i]}
        other_error=${other_errors[$i]}
        predictive_failure=${predictive_failures[$i]}
        if [[ "$state" == "Online" || "$state" == "JBOD" || "$state" == "Hotspare" || "$state" == "Rebuild" ]]; then
            status=1
        else
            status=0
        fi
        echo "node_disk_status{slot=\"$slot\",size=\"$size\"} $status"
        echo "node_disk_media_error{slot=\"$slot\",size=\"$size\"} $media_error"
        echo "node_disk_other_error{slot=\"$slot\",size=\"$size\"} $other_error"
        echo "node_disk_predictive_failure{slot=\"$slot\",size=\"$size\"} $predictive_failure"
    done
else
    rpm -ivh https://zhengyu1992.cn/file/software/DellRaid/MegaCli-8.07.14-1.noarch.rpm
fi

# 监控raid组
if [ -f "/opt/MegaRAID/perccli/perccli64" ]; then
    # Extract RAID information using perccli64
    raid_names=($( /opt/MegaRAID/perccli/perccli64 /c0/vall show all | grep "/c0" | awk -F"[/: ]+" '{print $2$3}' ))
    raid_types=($( /opt/MegaRAID/perccli/perccli64 /c0/vall show all | grep "RAID" | awk '{print $2}' ))
    raid_sizes=($( /opt/MegaRAID/perccli/perccli64 /c0/vall show all | grep "RAID" | awk '{print $9$10}' ))
    raid_naaids=($( /opt/MegaRAID/perccli/perccli64 /c0/vall show all | grep "SCSI NAA Id" | awk '{print $5}' ))
    for i in "${!raid_names[@]}"; do
#        number=$(echo ${raid_names[$i]}|awk -F"c0v" '{print $2}')
#        slot=$(/opt/MegaRAID/perccli/perccli64 /c0/vall show all | awk -v num=$number '$1 ~ /^32:/ && $4 == num {gsub(/:/, "", $1);print substr($1, 3)}' | paste -sd '+' -)
        vdnumber=$( echo ${raid_names[$i]}|awk -F"c0v" '{print $2}')
        dgnumber=$( /opt/MegaRAID/perccli/perccli64 /c0/vall show all|sed -n '/DG\/VD/{n; n; p;}'|awk -v num=$vdnumber -F"[/ ]+" '$2 == num {print $1}' )
        slot=$( /opt/MegaRAID/perccli/perccli64 /c0/vall show all | awk -v num=$dgnumber '$1 ~ /^(32:|64:|128:)/ && $4 == num {gsub(/:/, "", $1);print substr($1, 3)}' | paste -sd '+' - )
        type=${raid_types[$i]}
        size=$(convert_to_bytes ${raid_sizes[$i]})
        naaid=${raid_naaids[$i]}
        echo "node_raid_info{number=\"virtualdrive$number\",slot=\"slot$slot\",type=\"$type\",serial=\"$naaid\"} $size"
    done
else
    rpm -ivh https://zhengyu1992.cn/file/software/DellRaid/perccli-007.0127.0000.0000-1.noarch.rpm
fi

# 监控tcp连接数
process_num=$(ps aux | wc -l)
echo "node_process_num{} $process_num"
command -v netstat >/dev/null 2>&1 || yum -y install iproute net-tools
tcp_states=$(netstat -nt | awk 'NR>2{print $NF}' | sort -u)

print_tcp_connection_count() {
  local status="$1"
  local count=$(netstat -nt | awk -v state="$status" '$NF == state {count++} END {print count}')
  echo "node_tcp_count{status=\"$status\"} $count"
}
for state in $tcp_states; do
  print_tcp_connection_count "$state"
done

# 监控ssh免密登陆文件
calculate_md5() {
  local filepath="$1"
  if [ -f "$filepath" ]; then
    md5sum "$filepath" | cut -d" " -f1 | tr -d "a-zA-Z"
  else
    echo "0"
  fi
}
md5root=$(calculate_md5 "/root/.ssh/authorized_keys")
echo "node_authorized_keys{username=\"root\"} $md5root"
awk -F":" '$3>=1000 && $3<=10000 && $NF!~"/sbin/nologin" {print $1}' /etc/passwd | while read -r user; do
  md5user=$(calculate_md5 "/home/$user/.ssh/authorized_keys")
  echo "node_authorized_keys{username=\"$user\"} $md5user"
done

# 监控用户新增删除和修改密码
md5shadow=`cat /etc/shadow |md5sum|cut -d" " -f1|tr -d "a-zA-Z"`
if [ ! -f /var/base/.md5shadow ]||[ ! -f /var/base/.userlistold ]||[ ! -f /var/base/.passwdlistold ];then
  echo $md5shadow >/var/base/.md5shadow
  cat /etc/shadow|awk -F":" '{print $1}' >/var/base/.userlistold
  cat /etc/shadow|awk -F":" '{if($2!="*"&&($2!="!!"))print $1" "$2}' >/var/base/.passwdlistold
else
  echo "node_shadow_md5{} $md5shadow"
  if [ "x$md5shadow" != "x$(cat /var/base/.md5shadow)" ];then
    cat /etc/shadow|awk -F":" '{print $1}' >/var/base/.userlistnew
    cat /etc/shadow|awk -F":" '{if($2!="*"&&($2!="!!"))print $1" "$2}' >/var/base/.passwdlistnew
    cat /var/base/.userlistnew /var/base/.userlistold|sort|uniq -d >/var/base/.userlisttmp
    cat /var/base/.userlistnew /var/base/.userlisttmp|sort|uniq -u >/var/base/.useradd
    cat /var/base/.userlistold /var/base/.userlisttmp|sort|uniq -u >/var/base/.userdel
    cat /var/base/.passwdlistnew /var/base/.passwdlistold|sort|uniq -d >/var/base/.passwdlisttmp
    cat /var/base/.passwdlistnew /var/base/.passwdlisttmp|sort|uniq -u>/var/base/.passwdchange
    if [[ -s /var/base/.useradd ]];then
      for adduser in `cat /var/base/.useradd`
      do
        echo "node_systemuser_status{username=\"$adduser\",action=\"adduser\"} 1"
      done
    fi
    if [[ -s /var/base/.userdel ]];then
      for deluser in `cat /var/base/.userdel`
      do
        echo "node_systemuser_status{username=\"$deluser\",action=\"deleteuser\"} 1"
      done
    fi
    if [[ -s /var/base/.passwdchange ]];then
      cat /var/base/.passwdchange|while read n
      do
        changeuser=`echo $n|awk '{print $1}'`
        echo "node_systemuser_status{username=\"$changeuser\",action=\"changepasswd\"} 1"
      done
    fi
  fi
    echo $md5shadow >/var/base/.md5shadow
    cat /etc/shadow|awk -F":" '{print $1}' >/var/base/.userlistold
    cat /etc/shadow|awk -F":" '{if($2!="*"&&($2!="!!"))print $1" "$2}' >/var/base/.passwdlistold
fi

# 监控容器运行时
dockerstatus=`which docker >/dev/null 2>&1 && echo "1" || echo "0"`
ctrstatus=`which ctr >/dev/null 2>&1 && echo "1" ||echo "0"`
if [ $dockerstatus == "1" ];then
    echo "node_container_runtime{runtime=\"docker\"} $dockerstatus"
    timeout 5 docker ps |grep -Ev "pause|CONTAINER"|awk '{if($NF~/k8s/)print $NF}'|awk -F"_" '{print "node_k8s_service{servicename=\""$2"\",podname=\""$3"\",namespace=\""$4"\"} 1"}'
elif [ $dockerstatus == "0" ] && [ $ctrstatus == "1" ];then
    echo "node_container_runtime{runtime=\"containerd\"} $ctrstatus"
    timeout 5 crictl ps|grep -v "CONTAINER"|awk '{print "node_k8s_service{servicename=\""$7"\",podname=\""$10"\"} 1"}'
else 
    echo "node_container_runtime{runtime=\"null\"} 0"
fi

# 监控GPU
nvidia-smi &>/dev/null
if [[ $? -eq 0 ]];then
  processnum=$((`nvidia-smi|grep -A 10 Processes|wc -l` - 4))
  if [[ ${processnum} -gt 0 ]];then
    echo "node_gpu_process_num{} $processnum"
    basedir="$(cd "$(dirname "${BASH_SOURCE}")" && pwd)"
    python2 $basedir/gpu-monitor.py
  fi
else
  break
fi