場景：
CentOS6.3的二次封裝版本，安裝hortonworks的hadoop發行版，並按照其官方文檔實現高可用，但無fence設備支持，因此導致斷網和斷電測試時，備用HA節點無法得到資源釋放的通知。
因此，筆者簡單寫了一個腳本，讓HA的備節點在主節點斷網和斷電時能夠通過簡單測試得知並獲取資源，以實現真正意義的HA。
思考：
如何判斷網絡的連通性？
如何判斷VIP的可用性？
如何判斷自身的故障？
如何盡可能防止腦裂？
實現：
1、當自身節點IP均不正常時，使本節點重啟
2、當檢測自身網絡正常，且VIP不在線，且日志指出對方HA節點故障時獲取資源
3、ganglia的VIP連接修復（不在本文主要思考范圍）
代碼如下：
#!/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
