背景:線上zk都部分機器老是磁碟慢,故寫了個監控磁碟閾值進行自動清理的指令碼,本指令碼可以調節磁碟上限和保留檔案數量.

#!/bin/bash
#monitor available disk space

dir='/data/zookeeper/data/version-2/'
least_retain_date=20
threshold=85
retain_num=500
least_retain_num=400

function cur_date(){
    cur_date=$(date "+%Y-%m-%d %H:%M:%S")
}

function DiskUseRate(){
    SPACE=$(df -P| sed -n '/\/data$/p' | gawk '{print $5}' | sed  's/%//')
}

function CalDiffDay(){
    cd $dir
    oldest_file=$(ls -alt snapshot.*| tail -1| gawk '{print $9}')
    oldest_date=$(stat $oldest_file| grep ^Modify| gawk '{print$2}'| sed 's/-//g')
    cur_time=$(date +%s)
    oldest_time=$(date -d ${oldest_date} +%s)
    diff_time=$(($cur_time - $oldest_time))
    diff_date=$(($diff_time / 86400))
}

function CalSnapshotNum(){
    cd $dir
    total_num=$(ls -lt|egrep -i "snapshot"|wc -l)
}

function DeleteNum(){
    echo -e "\033[33mzookeeper snapshot is bigger!Begin clean it which delete with number!\033[0m"
    cd $dir
    CalSnapshotNum
    DiskUseRate
    if [ $SPACE -ge $threshold ];then
        delete_num=$(expr $total_num - $retain_num)
        echo -e "Delete number:\033[31m\033[01m\033[05m$delete_num\033[0m.Retain number is \033[31m\033[01m\033[05m$retain_num\033[0m."
        for each_snapshot in `ls -lt|egrep -i "snapshot"|tail -n $delete_num|awk '{print $NF}'`;do
            sudo rm -rf $each_snapshot
        done
    fi
    DiskUseRate
    if [ $SPACE -ge $threshold ];then
        CalSnapshotNum
        delete_num=$(expr $total_num - $least_retain_num)
        echo -e "Delete number:\033[31m\033[01m\033[05m$delete_num\033[0m.Retain number is \033[31m\033[01m\033[05m$least_retain_num\033[0m."
        for each_snapshot in `ls -lt|egrep -i "snapshot"|tail -n $delete_num|awk '{print $NF}'`;do
            sudo rm -rf $each_snapshot
        done
    fi
}

cur_date_d=`cur_date`
echo "==========$cur_date_d start=========="

DiskUseRate
if [ $SPACE -ge $threshold ];then
    CalDiffDay
    if [ $diff_date -ge $least_retain_date ];then
        remain_date=$least_retain_date
    fi
    cmd_num=`sudo find $dirsnapshot* -mtime +$remain_date|wc -l`

    CalSnapshotNum
    delete_least_num=$(expr $total_num - $least_retain_num)
    if [ $cmd_num -ge $delete_least_num ];then
        cmd="sudo find $dirsnapshot* -mtime +$remain_date -exec rm -rvf {} \;"
        echo "Space is use more than $threshold%! Rmove some files!"
        echo $cmd
        echo "---detail---"
        eval $cmd
    fi

    DiskUseRate
    if [ $SPACE -ge $threshold ];then
        DeleteNum
    fi
else
    echo "$SPACE% used!"
fi
echo "=======end======="

#disk clean
*/30 * * * * /opt/ops/disk_clean.sh >> disk_auto_clean.log