HA 模式下的 Hadoop+ZooKeeper+HBase 启动关闭管理脚本

HA 集群启动与关闭的顺序请参考：http://blog.csdn.net/u011414200/article/details/50437356

一. 集群介绍

1.1 集群规划

1.2 正确的启动顺序

1. ZooKeeper -> Hadoop -> HBase

2. ZooKeeper -> JournalNode (Hadoop) -> NameNode (Hadoop) -> DataNode (Hadoop) -> 主 ResourceManager/NodeManager (Hadoop) -> 备份 ResourceManager (Hadoop) -> ZKFC (Hadoop) -> MapReduce JobHistory (Hadoop) -> 主 Hmaster/HRegionServer (HBase) ->备份 Hmaster (HBase)

关闭集群的顺序则相反

二. 集群管理脚本

笔者的脚本部署在下面的路径下

/usr/local/cluster/cluster-manager

2.1 Slaves 配置文件

1. 文件名：slaves

2. 功能：记录运行 datanode 、nodemanager、HRegionServer、QuorumPeerMain、JournalNode 的节点 IP 地址或主机名

3. 内容

slave51
slave52
slave53

2.2 ZooKeeper 管理脚本

1. 文件名：zk-manager.sh

2. 功能：启动、关闭与重启 ZooKeeper 集群，并可查看运行 ZK 服务的的模式（leader or follower？）

3. 内容

#!/bin/bash 
SLAVES=$(cat slaves)

start_time=`date +%s`
for slave in $SLAVES
do
        case $1 in
                start)  ssh -t $slave "source ~/.bash_profile;zkServer.sh start" 1>/dev/null;;
                stop)   ssh -t $slave "source ~/.bash_profile;zkServer.sh stop" 1>/dev/null ;;
                status) ssh -t $slave "source ~/.bash_profile;zkServer.sh status" ;;
                restart)ssh -t $slave "source ~/.bash_profile;zkServer.sh restart" 1>/dev/null;;
                *)      echo -e "Usage: sh zk-manager.sh {start|stop|restart} ^_^\n" && exit ;;
        esac
done
end_time=`date +%s`
elapse_time=$((${end_time}-${start_time}))
echo -e "\n$1 ZooKeeper Server takes ${elapse_time} seconds\n"

2.3 JournalNode 管理脚本

1. 文件名：journal-manager.sh

2. 功能：启动、关闭运行在各个 slaves 上的 JournalNode 进程

3. 内容

#!/bin/bash 
start_time=`date +%s`

SLAVES=$(cat slaves)

for slave in $SLAVES
do
        case $1 in
                start)  ssh -t $slave "/usr/local/cluster/hadoop/sbin/hadoop-daemon.sh start journalnode" ;;
                stop)   ssh -t $slave "/usr/local/cluster/hadoop/sbin/hadoop-daemon.sh stop journalnode" ;;
                *)      echo -e "Usage: sh journal-manager.sh {start|stop} ^_^\n" && exit ;;
        esac
done

end_time=`date +%s`
elapse_time=$((${end_time}-${start_time}))
echo -e "\n$1 JournalNode Server takes ${elapse_time} seconds\n"

2.4 Hadoop 管理脚本

1. 文件名：hadoop-manager.sh

2. 功能：管理 hadoop 的启动与关闭

3. 内容

# which machine to be active NameNode
NameNode_1=master5
# which machine to be standy NameNode
NameNode_2=master52
# which machine to be active ResourceManager
ResourceManager_1=master5
# which machine to be standby ResourceManager
ResourceManager_2=master52
# which machine to be JobHistoryServer
HistoryServer=master5


# make sure which namenode is active and which resourcemanager is active
function getServiceState () {

        hdfs haadmin -getServiceState ${NameNode_1} | grep 'active' >> /dev/null && NameNode_Active=${NameNode_1} && NameNode_Standby=${NameNode_2}
        hdfs haadmin -getServiceState ${NameNode_2} | grep 'active' >> /dev/null && NameNode_Active=${NameNode_2} && NameNode_Standby=${NameNode_1}
        yarn rmadmin -getServiceState rm1 | grep 'active' >> /dev/null && ResourceManager_Active=${ResourceManager_1} && ResourceManager_Standby=${ResourceManager_2}
        yarn rmadmin -getServiceState rm2 | grep 'active' >> /dev/null && ResourceManager_Active=${ResourceManager_2} && ResourceManager_Standby=${ResourceManager_1}

}


case $1 in

        start)  ssh -t ${NameNode_1} "/usr/local/cluster/hadoop/sbin/hadoop-daemon.sh start namenode" ;
                ssh -t ${NameNode_1} "/usr/local/cluster/hadoop/sbin/hadoop-daemon.sh start zkfc" ;
                ssh -t ${NameNode_2} "/usr/local/cluster/hadoop/sbin/hadoop-daemon.sh start namenode" ;
                ssh -t ${NameNode_2} "/usr/local/cluster/hadoop/sbin/hadoop-daemon.sh start zkfc" ;
                ssh -t ${NameNode_1} "/usr/local/cluster/hadoop/sbin/hadoop-daemons.sh start datanode" ;
                ssh -t ${ResourceManager_1}     "/usr/local/cluster/hadoop/sbin/start-yarn.sh" ;
                ssh -t ${ResourceManager_2}     "/usr/local/cluster/hadoop/sbin/yarn-daemon.sh start resourcemanager" ;
                ssh -t ${HistoryServer}         "/usr/local/cluster/hadoop/sbin/mr-jobhistory-daemon.sh start historyserver" ;
                ;;

        stop)   getServiceState
                ssh -t ${HistoryServer} "/usr/local/cluster/hadoop/sbin/mr-jobhistory-daemon.sh stop historyserver" ;
                ssh -t ${ResourceManager_Standby} "/usr/local/cluster/hadoop/sbin/yarn-daemon.sh stop resourcemanager" ;
                ssh -t ${ResourceManager_Active} "/usr/local/cluster/hadoop/sbin/stop-yarn.sh" ;
                ssh -t ${NameNode_Active} "/usr/local/cluster/hadoop/sbin/hadoop-daemons.sh stop datanode" ;
                ssh -t ${NameNode_Standby} "/usr/local/cluster/hadoop/sbin/hadoop-daemon.sh stop namenode" ;
                ssh -t ${NameNode_Active} "/usr/local/cluster/hadoop/sbin/hadoop-daemon.sh stop namenode" ;
                ssh -t ${NameNode_Standby} "/usr/local/cluster/hadoop/sbin/hadoop-daemon.sh stop zkfc" ;
                ssh -t ${NameNode_Active} "/usr/local/cluster/hadoop/sbin/hadoop-daemon.sh stop zkfc" ;
                ;;

        * )     echo -e "Usage: hadoop-manager.sh {start|stop} ^_^\n" && exit ;
                ;;
esac

end_time=`date +%s`
elapse_time=$((${end_time}-${start_time}))
echo -e "$1 Hadoop Server takes ${elapse_time} seconds\n"

4. 注意事项

• NameNode_1、NameNode_2、ResourceManager_1、ResourceManager_2、HistoryServer 按实际填写，比如想让 master5 称为 active 则将其赋值给 NameNode_1
• 上述的启动顺序也许和 HA 集群启动与关闭的顺序 略微有点不同，这是因为笔者要在故障自动切换下实现特意选择其中一台作为 active

2.5 HBase 管理脚本

1. 文件名：hbase-manager.sh

2. 功能：启动、关闭 HBase 集群

3. 内容

#!/bin/bash 
HMaster_1=master5
HMaster_2=master52


start_time=`date +%s`

SLAVES=$(cat slaves)

function HRegionServer (){

        for slave in $SLAVES
        do
                ssh -t $slave "/usr/local/cluster/hbase/bin/hbase-daemon.sh $1 regionserver"
        done
}

case $1 in

        start)  ssh -t ${HMaster_1} "/usr/local/cluster/hbase/bin/hbase-daemon.sh start master" ;
                ssh -t ${HMaster_2} "/usr/local/cluster/hbase/bin/hbase-daemon.sh start master" ;
                HRegionServer   start
                ;;

        stop)   HRegionServer   stop
                ssh -t ${HMaster_2} "/usr/local/cluster/hbase/bin/hbase-daemon.sh stop master" ;
                ssh -t ${HMaster_1} "/usr/local/cluster/hbase/bin/hbase-daemon.sh stop master" ;
                ;;

        * )     echo -e "Usage: sh hbase-manager.sh {start|stop} ^_^\n" && exit;;
esac

end_time=`date +%s`
elapse_time=$((${end_time}-${start_time}))
echo -e "\n$1 Hbase Server takes ${elapse_time} seconds\n"

2.7 整个集群管理脚本

1. 文件名：Bigdata-Cluster.sh

2. 功能：统一启动、关闭及查看 ZooKeeper+Hadoop+HBase 大集群

3. 内容

#!/bin/bash 
CLUSTER_CONF_PATH=$(cd "$(dirname "$0")"; pwd)
NameNode_1=master5
NameNode_2=master52
SLAVE_1=slave51
start_time=`date +%s`

function showJps() {

        echo -e "\n**********************************************************************************"
        echo -e "当前 ${NameNode_1} 上的进程为:" && ssh -t ${NameNode_1} "source ~/.bash_profile; jps"
        echo -e "\n**********************************************************************************"
        echo -e "当前 ${NameNode_2} 上的进程为:" && ssh -t ${NameNode_2} "source ~/.bash_profile; jps"
        echo -e "\n**********************************************************************************"
        echo -e "当前 ${SLAVE_1} 上的进程为:" && ssh -t ${SLAVE_1} "source ~/.bash_profile; jps"
        echo -e "************************************************************************************"
}


case $1 in

        start)  sh $CLUSTER_CONF_PATH/zk-manager.sh  start ;
                sh $CLUSTER_CONF_PATH/journal-manager.sh  start ;
                sh $CLUSTER_CONF_PATH/hadoop-manager.sh  start ;
                sh $CLUSTER_CONF_PATH/hbase-manager.sh  start ;
                showJps
                ;;

        stop)   sh $CLUSTER_CONF_PATH/hbase-manager.sh  stop ;
                sh $CLUSTER_CONF_PATH/hadoop-manager.sh  stop ;
                sh $CLUSTER_CONF_PATH/journal-manager.sh  stop ;
                sh $CLUSTER_CONF_PATH/zk-manager.sh  stop ;
                showJps
                ;;

        status) showJps
                ;;

        *)      echo -e "Usage: sh Bigdata-Cluster.sh {start|stop|status} ^_^\n"  ;;

esac

end_time=`date +%s`
elapse_time=$((${end_time}-${start_time}))
echo -e "\n$1 Bigdata Cluster takes ${elapse_time} seconds\n"

2.8 配置文件统一分发管理脚本

1. 文件名：cluster-rsync.sh

2. 功能：在已经安装 rsync 服务器的情况下，在配合该脚本，即可实现在主节点上修改配置文件，其他节点迅速更新 (因为采用的 增量备份)

3. 内容

!/bin/bash

RSYN_SECRET_PATH=/home/hadoop5/softwares/rsyncd/rsyncd.secrets

NN_STANDBY=master52
SLAVES=$(cat slaves)

RSYNC_PORT=1873
RSYNC_SERVER=master5

function updateHadoopConf () {

        echo -e "\n****************************** Hadoop **********************************************\n"
        for slave in $SLAVES
        do
                ssh -t ${slave} "rsync -avz --port=${RSYNC_PORT} --password-file=${RSYN_SECRET_PATH} ${RSYNC_SERVER}::hadoop $HADOOP_HOME/etc/hadoop" #1>/dev/null
                echo -e "\n"
        done

        echo -e "\n"
        ssh -t ${NN_STANDBY} "rsync -avz --port=${RSYNC_PORT} --password-file=${RSYN_SECRET_PATH} ${RSYNC_SERVER}::hadoop $HADOOP_HOME/etc/hadoop" #1>/dev/null
}

function updateHbaseConf () {

        echo -e "\n***************************** Hbase **********************************************\n"
        for slave in $SLAVES
        do
                ssh -t ${slave} "rsync -avz --port=${RSYNC_PORT} --password-file=${RSYN_SECRET_PATH} ${RSYNC_SERVER}::hbase $HBASE_HOME/conf" #1>/dev/null
                echo -e "\n"
        done

        echo -e "\n"
        ssh -t ${NN_STANDBY} "rsync -avz --port=${RSYNC_PORT} --password-file=${RSYN_SECRET_PATH} ${RSYNC_SERVER}::hbase $HBASE_HOME/conf" #1>/dev/null
}


case $1 in
        hadoop) updateHadoopConf ;;
        hbase)  updateHbaseConf ;;
        all )   updateHadoopConf;updateHbaseConf ;;
        *)      echo -e "Usage: sh cluster-rsync.sh {hadoop|hbase|all} ^_^\n" && exit ;;
esac

三. 验证

3.1 同步集群配置文件

sh cluster-rsync.sh  all

3.2 启动整个大集群

1. 启动前查看大集群的状态

sh Bigdata-Cluster.sh status

2. 启动大集群

sh Bigdata-Cluster.sh start

3. 启动后的进程

sh Bigdata-Cluster.sh status

4. 关闭大集群

sh Bigdata-Cluster.sh stop