docker热重启，临时解决docker内存泄漏问题

#!/bin/bash

#####################################自定义变量区
#内存阈值
host_mem_limit=60 #主机内存占用阈值，单位%(百分比)

max_mem_gb=5 ##dockerd内存占用阈值，单位GB

###########################业务方法区####################################
hotRestartDockerd(){

    echo "hotRestartDockerd"

    DOCKERD_CONF='/etc/docker/daemon.json'

    [ ! -f "$DOCKERD_CONF" ] && echo "error, failed to find $DOCKERD_CONF " && exit 1
    [ ! -s "$DOCKERD_CONF" ] && echo "error, $DOCKERD_CONF is empty" &&  exit 2

    DOCKERD_PID=$(systemctl status docker | grep "Main PID" | awk '{print $3}'）
    [ -z "$DOCKERD_PID" ] && echo "error, dockerd is not running" && exit 3

    # enable Live Restore
    CHANGED=""
    if !( docker info | grep -i "Live Restore Enabled" | grep -i "true" > /dev/null ) ; then
        echo "enable Live Restore "
        cp -f $DOCKERD_CONF $DOCKERD_CONF-backup
        sed -i '/live-restore/ d'  $DOCKERD_CONF
        sed -i -E 's/^\{$/\{\n    "live-restore": true , /' $DOCKERD_CONF
        CHANGED="1"

        kill -SIGHUP $DOCKERD_PID
        sleep 5
        if !( docker info | grep -i "Live Restore Enabled" | grep -i "true" > /dev/null ) ; then
            echo "error, failed to enable Live Restore "
            exit 4
        fi
    fi

    systemctl restart docker
    echo "restarting docker "
    
    OK=""
    for((i=0;i<10;i++)) ; do
        sleep 4
        ( systemctl status docker | grep "Active: active (running)" &> /dev/null  ) \
            && OK="1" && break
    done
    [ -z "$OK" ] && echo "error, failed to restart docker" && exit 5

    if [ -n "$CHANGED" ] ; then
        sed -i '/live-restore/ d'  $DOCKERD_CONF
        sed -i -E 's/^\{$/\{\n    "live-restore": false , /' $DOCKERD_CONF
        kill -SIGHUP $(systemctl status docker | grep "Main PID" | awk '{print $3}')

        rm -f $DOCKERD_CONF
        mv  $DOCKERD_CONF-backup $DOCKERD_CONF
        sleep 5
    fi
    echo "finished restarting dockerd"
}


restartSpecialCon(){
    echo "restart related contariners which wants to communicate with "
    LIST=` docker inspect $(docker ps -q)|grep 'Source.*docker.sock\|Name": "/'|grep sock -B1|grep Name  | awk -F'"' '{print $4}' | sed 's/^\///' `
    [ -z "$LIST" ] && echo "no container needed to restart , finish" && return 0
    for NAME in $LIST ; do
        docker restart $NAME
    done

}

#判断docker的文件系统是否为xfs,非xfs如果daemon.json设置了overlay2.size会导致docker重启异常
checkFileSystem(){
	local daemonPath="/etc/docker/daemon.json"
	backing_fs=$(docker info | grep "Backing Filesystem" | awk '{print $3}')
	if [ "${backing_fs}" = "extfs" ]; then
		# 检查/etc/docker/daemon.json文件中是否配置了overlay2.size
		if grep -q "overlay2.size" ${daemonPath}; then
			sed -i '/"storage-opts": \[/,/]/d' ${daemonPath}
		fi
	fi

    # 检查倒数第二行是否以逗号结尾
    second_last_line=$(tail -n 2 "${daemonPath}" | head -n 1)
    if [[ ${second_last_line} =~ ,$ ]]; then
    # 删除倒数第二行的逗号
        sed -i '$!N;$s/,\n/\n/;P;D' "${daemonPath}"
        echo "已删除倒数第二行的逗号"
    else
        echo "倒数第二行不以逗号结尾，无需删除"
    fi
}

##############################执行逻辑区#########################
  mem_used=$(free | awk '/^Mem/ {print $3/$2 * 100.0}')
  mem_used=${mem_used%.*}

  if [ ${mem_used} -gt ${host_mem_limit} ]; then
	# 获取进程内存占用值（单位为 GB）
	dockerd_mem_gb=$(ps -p $(systemctl status docker | grep "Main PID" | awk '{print $3}') -o rss= | awk '{ printf "%.2f", $1/1024/1024 }')
	echo -e "${dockerd_mem_gb%.*} ${max_mem_gb}"
	if [ ${dockerd_mem_gb%.*} -ge ${max_mem_gb} ]; then
  		echo -e "$(date +"%Y-%m-%d %H:%M:%S") $(hostname) 主机内存占用率超过:${limit}%  dockerd 进程内存占用超过 ${max_mem_gb} GB，执行热重启"
  		#执行热重启前先校验docker文件系统，若非xfs需要修改daemon.json文件避免热重启失败
		checkFileSystem
		# 执行dockerd热重启
		hotRestartDockerd
		restartSpecialCon
		echo -e  "$(date +"%Y-%m-%d %H:%M:%S") $(hostname) finished"
	else
		echo -e "$(date +"%Y-%m-%d %H:%M:%S") $(hostname) 主机内存使用量超出${host_mem_limit}%的限制范围，dockerd服务内存占用在${max_mem_gb} GB范围内，当前占用为${dockerd_mem_gb} GB."
	fi
  else
	  echo -e  "$(date +"%Y-%m-%d %H:%M:%S") $(hostname) 内存使用量在${host_mem_limit}%的限制范围内."
  fi