Spark集群之yarn提交作业优化案例

　　　　　　　　　　Spark集群之yarn提交作业优化案例

　　　　　　　　　　　　　　　　　　　　　　　　　　　　作者：尹正杰

版权声明：原创作品，谢绝转载！否则将追究法律责任。

 

 

一.启动Hadoop集群

1>.自定义批量管理脚本

[yinzhengjie@s101 ~]$ more `which xzk.sh`
#!/bin/bash
#@author :yinzhengjie
#blog:http://www.cnblogs.com/yinzhengjie
#EMAIL:[email protected]

#判断用户是否传参
if [ $# -ne 1 ];then
    echo "无效参数，用法为: $0  {start|stop|restart|status}"
    exit
fi

#获取用户输入的命令
cmd=$1

#定义函数功能
function zookeeperManger(){
    case $cmd in
    start)
        echo "启动服务"        
        remoteExecution start
        ;;
    stop)
        echo "停止服务"
        remoteExecution stop
        ;;
    restart)
        echo "重启服务"
        remoteExecution restart
        ;;
    status)
        echo "查看状态"
        remoteExecution status
        ;;
    *)
        echo "无效参数，用法为: $0  {start|stop|restart|status}"
        ;;
    esac
}


#定义执行的命令
function remoteExecution(){
    for (( i=102 ; i<=104 ; i++ )) ; do
            tput setaf 2
            echo ========== s$i zkServer.sh  $1 ================
            tput setaf 9
            ssh s$i  "source /etc/profile ; zkServer.sh $1"
    done
}

#调用函数
zookeeperManger
[yinzhengjie@s101 ~]$ 

[yinzhengjie@s101 ~]$ more `which xzk.sh` （zookeeper集群管理脚本）

[yinzhengjie@s101 ~]$ more `which xcall.sh`
#!/bin/bash
#@author :yinzhengjie
#blog:http://www.cnblogs.com/yinzhengjie
#EMAIL:[email protected]


#判断用户是否传参
if [ $# -lt 1 ];then
        echo "请输入参数"
        exit
fi

#获取用户输入的命令
cmd=$@

for (( i=101;i<=105;i++ ))
do
        #使终端变绿色 
        tput setaf 2
        echo ============= s$i $cmd ============
        #使终端变回原来的颜色，即白灰色
        tput setaf 7
        #远程执行命令
        ssh s$i $cmd
        #判断命令是否执行成功
        if [ $? == 0 ];then
                echo "命令执行成功"
        fi
done
[yinzhengjie@s101 ~]$ 

[yinzhengjie@s101 ~]$ more `which xcall.sh` （批量执行命令的脚本）

2>.启动zookeeper集群

[yinzhengjie@s101 ~]$ xzk.sh start
启动服务
========== s102 zkServer.sh start ================
ZooKeeper JMX enabled by default
Using config: /soft/zk/bin/../conf/zoo.cfg
Starting zookeeper ... STARTED
========== s103 zkServer.sh start ================
ZooKeeper JMX enabled by default
Using config: /soft/zk/bin/../conf/zoo.cfg
Starting zookeeper ... STARTED
========== s104 zkServer.sh start ================
ZooKeeper JMX enabled by default
Using config: /soft/zk/bin/../conf/zoo.cfg
Starting zookeeper ... STARTED
[yinzhengjie@s101 ~]$ 

3>.启动hdfs分布式文件系统

[yinzhengjie@s101 ~]$ start-dfs.sh 
SLF4J: Class path contains multiple SLF4J bindings.
SLF4J: Found binding in [jar:file:/soft/hadoop-2.7.3/share/hadoop/common/lib/slf4j-log4j12-1.7.10.jar!/org/slf4j/impl/StaticLoggerBinder.class]
SLF4J: Found binding in [jar:file:/soft/apache-hive-2.1.1-bin/lib/log4j-slf4j-impl-2.4.1.jar!/org/slf4j/impl/StaticLoggerBinder.class]
SLF4J: See http://www.slf4j.org/codes.html#multiple_bindings for an explanation.
SLF4J: Actual binding is of type [org.slf4j.impl.Log4jLoggerFactory]
Starting namenodes on [s101 s105]
s101: starting namenode, logging to /soft/hadoop-2.7.3/logs/hadoop-yinzhengjie-namenode-s101.out
s105: starting namenode, logging to /soft/hadoop-2.7.3/logs/hadoop-yinzhengjie-namenode-s105.out
s102: starting datanode, logging to /soft/hadoop-2.7.3/logs/hadoop-yinzhengjie-datanode-s102.out
s103: starting datanode, logging to /soft/hadoop-2.7.3/logs/hadoop-yinzhengjie-datanode-s103.out
s104: starting datanode, logging to /soft/hadoop-2.7.3/logs/hadoop-yinzhengjie-datanode-s104.out
Starting journal nodes [s102 s103 s104]
s102: starting journalnode, logging to /soft/hadoop-2.7.3/logs/hadoop-yinzhengjie-journalnode-s102.out
s104: starting journalnode, logging to /soft/hadoop-2.7.3/logs/hadoop-yinzhengjie-journalnode-s104.out
s103: starting journalnode, logging to /soft/hadoop-2.7.3/logs/hadoop-yinzhengjie-journalnode-s103.out
SLF4J: Class path contains multiple SLF4J bindings.
SLF4J: Found binding in [jar:file:/soft/hadoop-2.7.3/share/hadoop/common/lib/slf4j-log4j12-1.7.10.jar!/org/slf4j/impl/StaticLoggerBinder.class]
SLF4J: Found binding in [jar:file:/soft/apache-hive-2.1.1-bin/lib/log4j-slf4j-impl-2.4.1.jar!/org/slf4j/impl/StaticLoggerBinder.class]
SLF4J: See http://www.slf4j.org/codes.html#multiple_bindings for an explanation.
SLF4J: Actual binding is of type [org.slf4j.impl.Log4jLoggerFactory]
Starting ZK Failover Controllers on NN hosts [s101 s105]
s101: starting zkfc, logging to /soft/hadoop-2.7.3/logs/hadoop-yinzhengjie-zkfc-s101.out
s105: starting zkfc, logging to /soft/hadoop-2.7.3/logs/hadoop-yinzhengjie-zkfc-s105.out
[yinzhengjie@s101 ~]$ 

4>.启动yarn集群

[yinzhengjie@s101 ~]$ start-yarn.sh 
starting yarn daemons
s101: starting resourcemanager, logging to /soft/hadoop-2.7.3/logs/yarn-yinzhengjie-resourcemanager-s101.out
s105: starting resourcemanager, logging to /soft/hadoop-2.7.3/logs/yarn-yinzhengjie-resourcemanager-s105.out
s102: starting nodemanager, logging to /soft/hadoop-2.7.3/logs/yarn-yinzhengjie-nodemanager-s102.out
s104: starting nodemanager, logging to /soft/hadoop-2.7.3/logs/yarn-yinzhengjie-nodemanager-s104.out
s103: starting nodemanager, logging to /soft/hadoop-2.7.3/logs/yarn-yinzhengjie-nodemanager-s103.out
[yinzhengjie@s101 ~]$ 

5>.查看集群是否启动成功

[yinzhengjie@s101 ~]$ xcall.sh jps
============= s101 jps ============
3065 ResourceManager
2602 NameNode
2907 DFSZKFailoverController
3374 Jps
命令执行成功
============= s102 jps ============
2356 JournalNode
2277 DataNode
2202 QuorumPeerMain
2476 NodeManager
2589 Jps
命令执行成功
============= s103 jps ============
2595 Jps
2197 QuorumPeerMain
2358 JournalNode
2279 DataNode
2478 NodeManager
命令执行成功
============= s104 jps ============
2272 DataNode
2197 QuorumPeerMain
2469 NodeManager
2583 Jps
2346 JournalNode
命令执行成功
============= s105 jps ============
2640 Jps
2258 NameNode
2358 DFSZKFailoverController
命令执行成功
[yinzhengjie@s101 ~]$ 

　　检查WebUI是否正常打开：

 

 

二.Spark集群的运行模式

1>.local

　　本地模式,不需要启动任何进程.使用jvm多个线程模拟worker。

2>.standalone

　　独立模式，master + worker，启动方式：spark-submit --master spark://s101:7077

 

3>.yarn

 　　不需要启动任务spark进程，不需要安装spark集群，启动方式如：spark-submit --master yarn | yarn-client | yarn-cluster

1.yarn-client 
　　driver运行在client，appmaster只负责请求资源列表。

2.yarn-cluster
      appmaster除了请求资源列表之外，还要运行driver程序。

 

三.使用yarn操作步骤

　　我们需要停止spark集群，只需要安装Spark软件并且启动hadoop集群即可。

四.优化yarn集群配置案例