场景：

   CentOS6.3的二次封装版本，安装hortonworks的hadoop发行版，并按照其官方文档实现高可用，但无fence设备支持，因此导致断网和断电测试时，备用HA节点无法得到资源释放的通知。

   因此，笔者简单写了一个脚本，让HA的备节点在主节点断网和断电时能够通过简单测试得知并获取资源，以实现真正意义的HA。


思考：

   如何判断网络的连通性？

   如何判断VIP的可用性？

   如何判断自身的故障？

   如何尽可能防止脑裂？


实现：

   1、当自身节点IP均不正常时，使本节点重启

   2、当检测自身网络正常，且VIP不在线，且日志指出对方HA节点故障时获取资源

   3、ganglia的VIP连接修复（不在本文主要思考范围）


代码如下：


#!/bin/bash
STD=$$
Cron_if=`ps aux|grep $0|grep -vE "grep|$STD"|/usr/bin/wc -l`
[[ "$Cron_if" -ge 2 ]] && exit 2
sleep 30
VIP=${VIP:-192.168.1.198}
RE_PRD=${RE_PRD:-10}
SAFE_TIME=${SAFE_TIME:-60}
NMK=${NMK:-8}
NUL=/dev/null
Date="/bin/date +%k:%M:%S/%Y-%m-%d"
[ ! -f /etc/sysconfig/hdp.conf ] && echo "`$Date` Error: No such config fil." >> $Mlog && exit 1
[ ! -f /etc/init.d/cman ] && echo "`$Date` War: `uname -n` Invalid HA node." >> $Mlog && exit 2
[ ! -f /etc/init.d/rgmanager ] && echo "`$Date` War: `uname -n` Invalid HA node." >> $Mlog && exit 2
[ -f /etc/sysconfig/hdp.conf ] && . /etc/sysconfig/hdp.conf
while :;do
    RQE1=`/etc/init.d/rgmanager status 2> $Mlog|grep "is running."`
    RQE2=`/etc/init.d/cman status 2> $Mlog|grep "is running."` 
    RQE3=`/etc/init.d/ricci status 2> $Mlog|grep "is running."`
    RQE4=`/etc/init.d/modclusterd status 2> $Mlog|grep "is running."`
    [ -z "$RQE2" ] && /etc/init.d/cman start &> $NUL
    [ -z "$RQE1" ] && /etc/init.d/rgmanager start &> $NUL
    [ -z "$RQE3" ] && /etc/init.d/ricci start &> $NUL
    [ -z "$RQE4" ] && /etc/init.d/modclusterd start &> $NUL
    if [[ -n "$RQE1" && -n "$RQE2" ]];then
        break
    else
        sleep $SAFE_TIME
        continue
    fi
done
NODE=(`grep clusternode /etc/cluster/cluster.conf|grep nodeid|awk -F\" '{print $2}'`)
for i in ${NODE[@]};do
    NODE_IP=`grep $i /etc/hosts|awk '{print $1}'`
    JUDG_VAR=`/sbin/ip a|grep $NODE_IP`
    [ -n "$JUDG_VAR" ] && N_NAME=$i
    [ -z "$JUDG_VAR" ] && R_NAME=$i
done
Node_Reboot(){
    W_VIP=`/sbin/ip a show $Ne|grep $VIP`
    Nic_File=/etc/sysconfig/network-scripts/ifcfg-$Ne
    PHY_IP_FILE=`grep IPADDR $Nic_File|awk -F\= '{print $2}'`
    IP_FILE_IF=`/sbin/ifconfig $Ne|grep $PHY_IP_FILE`
    if [[ -z "$W_VIP" && -z "$IP_FILE_IF" ]];then
        KILL_PID=(`ps aux|grep -E "rgmanager|fenced|dlm_controld|gfs_controld|corosync"|grep -v grep|awk '{print $2}'`)
        for i in ${KILL_PID[@]};do
            kill -9 $i  &> $NUL
        done
        /etc/init.d/rgmanager stop &> $NUL
        /etc/init.d/cman stop &> $NUL
        /sbin/reboot &> $NUL
    fi
}
FREE_RESOURCE(){
    DFGW=`/sbin/route -n|grep $Ne|awk '{print $2}'|grep -v "0.0.0.0"`
    NFS_IP=`grep netfs /etc/cluster/cluster.conf|awk -F\" '{print $8}'|grep -v "^$"`
    P_CMD="/bin/ping -c 3 -W 1"
    if ! $P_CMD $VIP &>$NUL;then
        if $P_CMD $DFGW &>$NUL || $P_CMD $R_NAME &>$NUL || $P_CMD $NFS_IP &>$NUL ;then
            if ! $P_CMD $VIP &>$NUL;then
                DOWN_LOG=`/usr/bin/tail -1 /var/log/cluster/rgmanager.log|grep "$R_NAME DOWN"`
                FENCE_LOG=`tail -1 /var/log/cluster/fenced.log|grep "$R_NAME failed"`
                if [[ -n "$DOWN_LOG" && -n "$FENCE_LOG" ]];then
                    echo absolutely|/usr/sbin/fence_ack_manual $R_NAME
                fi
            fi
        else
            echo "`$Date` Gateway:$DFGW and HA_node:$R_NAME and Nfs:$NFS_IP Offline!!" >> $Mlog
        fi
    fi
}
                                                         
GGA(){
    RE=`netstat -anup|grep gmond|grep $VIP|/usr/bin/wc -l`
    [ "$RE" -eq 4 ] && return 0
    MGF=/etc/ganglia/gmond.conf
    RE=`grep -A 2 -E "udp_send_channel|tcp_accept_channel" $MGF|grep $VIP|/usr/bin/wc -l`
    if [ "$RE" -ne 2 ];then
        sed -i "/^udp_accept_channel/a\ \ bind = $VIP" $MGF
        sed -i "/^tcp_accept_channel/a\ \ bind = $VIP" $MGF
    fi
    GFL=(`find /etc/ganglia/hdp -name "gmond.slave.conf"`)
    for g in ${GFL[@]};do
        if grep "bind = $VIP" $g &> $NUL;then
            continue
        fi
        sed -i "/\<host\>/i\ \ bind = $VIP" $g
    done
    /etc/init.d/gmond restart &> $NUL
    /etc/init.d/hdp-gmond restart &> $NUL
}
while :;do
    GGA
    FREE_RESOURCE
    Node_Reboot
    sleep $RE_PRD
done