oom_of_squid
用于监控服务器上的squid进程，保证对用户服务的稳定。
此进程意欲以守护进程模式运行，可以这样使用：./oom_of_squid &
主要功能：
监控配置过的每个squid进程，保证它的内存占用看似正常；
确保进程PID和PID文件一致；
squid进程挂掉后启动之；
squid进程占用的内存超过设置的阀值则重启之；
系统可用内存低于阀值，则选择一个squid进程重启之；
默认每3秒检查一次；

注意事项：
`echo squid7[0-9].conf`在gentoo和CentOS上的行为不一样，到CentOS下应用可能需要改为`ls squid7[0-9].conf`；
/proc/$PID/comm 在较新的内核上才有这个接口，使用旧内核需要另想办法；
/etc/init.d/squid_multi_instance是我重写过的进程管理脚本（还有bug），使用“标准”脚本的人需要自己改一下；
系统可用内存阀值应该适当高于系统自身的oom阀值；
用于全内存式缓存更好，因为不用顾忌状态文件的同步。
我设置的配置文件中，squid7[0-9].conf是carp-child，用于缓存文件；squid8[0-9].conf是carp-parent，用于决定URL在集群中的分布。

附件为脚本。望用的上的同仁多提宝贵意见！！！

#!/bin/bash
# 功能:
#    在系统可用内存很小时选取一个squid进程重启
#    检查各进程健康状况

MAIL="[email protected]"        # 设置自己的邮箱
MAIL_FROM="[email protected]"
CHECK_INTERVAL="3s"        # 检查时间间隔
THRESHOLD_LOW_MEM="90"            # MB,低内存阀值; 必须大于vm.min_free_kbytes
THRESHOLD_CHILD_MAX_MEM="200"    # MB,child持有内存大于此值就重启
THRESHOLD_EMERG="100"            # MB,LEVEL="emerg"时重启进程的阀值
LEVEL="normal"    # 内存级别,可用内存较低时设置为emerg; 正常情况下应该是空变量
LOG="$0.log"

# 探测配置文件,70-79之间的端口为parent,80-87为child
CONFIG_LOCATION="/etc/squid/"
PARENT_CONF=$(cd $CONFIG_LOCATION; echo squid7[0-9].conf)
CHILD_CONF=$(cd $CONFIG_LOCATION; echo squid8[0-9].conf)
ALL_CONF="$PARENT_CONF $CHILD_CONF"

# disable killing by oom
echo "-17" > /proc/self/oom_adj

all_free_mem() {
    free -m | awk '/^Mem:/ {print ($4 + $6 + $7)}'    # 单位是MB
}
squid_instance_mem() {
    INSTANCE=${CONF%.conf}
    PID_FILE="/var/run/$INSTANCE.pid"
    [[ -e $PID_FILE ]] && PID_FROM_FILE=`cat $PID_FILE` || PID_FROM_FILE=""
    # 检查名称和进程号是否对应
    if [ -e $PID_FILE ]; then
        [[ x$PID_FROM_FILE != "x" ]] && grep -q squid /proc/$PID_FROM_FILE/comm || false
    else
        echo "$INSTANCE pid file not match to squid" >> $LOG
    fi
    # 占用的内存数量
    RSS=`awk '/^VmRSS:/ {printf "%d\n", $2/1024}' /proc/$PID_FROM_FILE/status`
}
keep_processes_health() {
    # 检查所有进程,如果挂了,修正一下
    for CONF in $ALL_CONF; do
        INSTANCE=${CONF%.conf}
        PID_FILE="/var/run/$INSTANCE.pid"
        STATE_FILE="/var/lib/init.d/started/$INSTANCE"
        [[ -e $PID_FILE ]] && PID_FROM_FILE=`cat $PID_FILE` || PID_FROM_FILE=""
        PID_RUNNING=`ps axo user,pid,cmd | awk '/^squid/ && /'$CONF'/ {print $2}'`

        if [ x$PID_RUNNING = "x" ]; then
            # 进程不存在: 启动进程
            process_state="not_running"
            echo "`date +%F\ %T` $INSTANCE state is $process_state, restarted" >> $LOG
            [[ -e $PID_FILE ]] && /bin/rm -f $PID_FILE
            [[ -e $STATE_FILE ]] && /bin/rm -f $STATE_FILE
            /etc/init.d/squid_multi_instance start ${INSTANCE#squid} >/dev/null 2>&1
            notify_admin &
        elif [ x$PID_FROM_FILE = "x" ]; then
            # pid文件有问题: 修正
            process_state="bad_pid_file"
            if echo $PID_RUNNING > $PID_FILE; then
                process_state="good"
                echo "fixed pid file of $INSTANCE at `date +%F\ %T`" >> $LOG
            else
                echo "can not write $PID_FILE" >> $LOG
                process_state="pid_file_not_writeable"
                notify_admin &
            fi
        elif [ x$PID_FROM_FILE != x$PID_RUNNING ]; then
            # 什么情况下会这样呢...
            process_state="pid_not_equal"
            if echo $PID_RUNNING > $PID_FILE; then
                echo "fixed $process_state of $INSTANCE at `date +%F\ %T`" >> $LOG
                notify_admin &
            else
                echo "can not write $PID_FILE" >> $LOG
                process_state="$process_state pid_file_not_writeable"
                notify_admin &
            fi
        elif [ x$PID_FROM_FILE = x$PID_RUNNING ]; then
            # 运行良好
            process_state="good"
            continue
        else
            process_state="unknow"
            notify_admin &
        fi
    done
    unset process_state
}
restart_process() {
    INSTANCE=${CONF%.conf}
    /etc/init.d/squid_multi_instance restart ${INSTANCE#squid} >/dev/null 2>&1
    notify_admin &
}
pick_and_restart_parent() {
    # 重启占用内存最大的进程
    for CONF in $PARENT_CONF; do
        squid_instance_mem
        PARENT_MEM="$PARENT_MEM\n$RSS $CONF"
    done
    PARENT_MEM_MAX=`echo -e $PARENT_MEM | sort -n | tail -1`
    if [ -n $PARENT_MEM_MAX ]; then
        CONF=`echo $PARENT_MEM_MAX | awk '{print $2}'`
        restart_process
        echo "restarted parent ${CONF%.conf} at `date +%F\ %T`" >> $LOG
    else
        echo "unknow error in pick_and_restart_parent" >> $LOG
    fi
}
pick_and_restart_child() {
    # 正常情况下重启内存大于300M的进程; 紧急情况下重启大于100M的进程
    for CONF in $CHILD_CONF; do
        squid_instance_mem
        if [ $RSS -gt $THRESHOLD_CHILD_MAX_MEM ]; then
            process_state="over_THRESHOLD_CHILD_MAX_MEM"
            echo "$process_state restarted child ${CONF%.conf} at `date +%F\ %T`" >> $LOG
            restart_process
        else
            true
        fi
    done
    if [ x$LEVEL = "xemerg" ]; then
        for CONF in $CHILD_CONF; do
            squid_instance_mem
            [[ $RSS -gt $THRESHOLD_EMERG ]] && (process_state="over_THRESHOLD_EMERG"; restart_process)
        done
    else
        true
    fi
}
pick_and_restart_one() {
    # restart a child first
    pick_and_restart_child
    
    # memory still low, restart a parent
    [[ `all_free_mem` -lt $THRESHOLD_LOW_MEM ]] && pick_and_restart_parent
    
    # memory still low, restart all child
    [[ `all_free_mem` -lt $THRESHOLD_LOW_MEM ]] && \
    (CONF="child"; restart_process)
}
notify_admin() {
    sendmail -t -f $MAIL_FROM <<EOF
To: $MAIL
From: $MAIL_FROM
Subject: $INSTANCE on `hostname` restarted

重启过了 `hostname` 上的 $INSTANCE
进程 $INSTANCE 的最后状态为: $process_state

当前系统总剩余内存为: `all_free_mem`
.
EOF
}
while true; do
    keep_processes_health

    # child占用内存大于THRESHOLD_CHILD_MAX_MEM就重启,不等总内存过低
    pick_and_restart_child

    if [ `all_free_mem` -lt $THRESHOLD_LOW_MEM ]; then
        LEVEL="emerg"
        echo "low memory at `date +%F\ %T`" >> $LOG
        pick_and_restart_one
    else
        LEVEL="normal"
        sleep $CHECK_INTERVAL
    fi
done
# vim: set sw=4 ts=4: