HDFS-HA原理及配置
1.HDFS-HA架构原理介绍
hadoop2.x之后，Clouera提出了QJM/Qurom Journal Manager，这是一个基于Paxos算法实现的HDFS HA方案，它给出了一种较好的解决思路和方案,示意图如下：
- 基本原理就是用2N+1台 JN 存储EditLog，每次写数据操作有大多数（>=N+1）返回成功时即认为该次写成功，数据不会丢失了。当然这个算法所能容忍的是最多有N台机器挂掉，如果多于N台挂掉，这个算法就失效了。这个原理是基于Paxos算法
- 在HA架构里面SecondaryNameNode这个冷备角色已经不存在了，为了保持standby NN时时的与主Active NN的元数据保持一致，他们之间交互通过一系列守护的轻量级进程JournalNode
- 任何修改操作在 Active NN上执行时，JN进程同时也会记录修改log到至少半数以上的JN中，这时 Standby NN 监测到JN 里面的同步log发生变化了会读取 JN 里面的修改log，然后同步到自己的的目录镜像树里面，如下图：
当发生故障时，Active的 NN 挂掉后，Standby NN 会在它成为Active NN 前，读取所有的JN里面的修改日志，这样就能高可靠的保证与挂掉的NN的目录镜像树一致，然后无缝的接替它的职责，维护来自客户端请求，从而达到一个高可用的目的。
2.HDFS-HA 详细配置
1）环境准备
根据以上介绍，要完成HA的配置则必须要添加一个NameNode（2号节点）和三个JournalNode。为了和我们之前配置的非HA避免冲突，我们选择对原来的环境进行备份，然后在备份的基础上重新配置HA环境，即两个环境隔离开互不影响。
[kfk@bigdata-pro01 etc]$ lshadoop[kfk@bigdata-pro01 etc]$ cp -r hadoop/ dist-hadoop[kfk@bigdata-pro01 etc]$ lsdist-hadoop hadoop[kfk@bigdata-pro01 etc]$ cd ..[kfk@bigdata-pro01 hadoop-2.6.]$ lsbin data etc include lib libexec LICENSE.txt logs NOTICE.txt README.txt sbin share[kfk@bigdata-pro01 hadoop-2.6.]$ cd data/[kfk@bigdata-pro01 data]$ lstmp[kfk@bigdata-pro01 data]$ mv tmp/ dist-tmp[kfk@bigdata-pro01 data]$ mkdir tmp[kfk@bigdata-pro01 data]$ lsdist-tmp tmp
2）修改hdfs-site.xml配置文件
vi hdfs-site.xml<configuration> <property> <name>dfs.replication</name> <value></value> </property> <property> <name>dfs.permissions</name> <value>false</value> </property> <property> <name>dfs.permissions.enabled</name> <value>false</value> </property> <property> <name>dfs.nameservices</name> <value>ns</value> </property> <property> <name>dfs.ha.namenodes.ns</name> <value>nn1,nn2</value> </property> <property> <name>dfs.namenode.rpc-address.ns.nn1</name> <value>bigdata-pro01.kfk.com:</value> </property> <property> <name>dfs.namenode.rpc-address.ns.nn2</name> <value>bigdata-pro02.kfk.com:</value> </property> <property> <name>dfs.namenode.http-address.ns.nn1</name> <value>bigdata-pro01.kfk.com:</value> </property> <property> <name>dfs.namenode.http-address.ns.nn2</name> <value>bigdata-pro02.kfk.com:</value> </property> <property> <name>dfs.namenode.shared.edits.dir</name> <value>qjournal://bigdata-pro01.kfk.com:8485;bigdata-pro02.kfk.com:8485;bigdata-pro03.kfk.com:8485/ns</value> </property> <property> <name>dfs.journalnode.edits.dir</name> <value>/opt/modules/hadoop-2.6./data/jn</value> </property> <property> <name>dfs.client.failover.proxy.provider.ns</name> <value>org.apache.hadoop.hdfs.server.namenode.ha.ConfiguredFailoverProxyProvider</value> </property> <property> <name>dfs.ha.automatic-failover.enabled.ns</name> <value>true</value> </property> <property> <name>ha.zookeeper.quorum</name> <value>bigdata-pro01.kfk.com:,bigdata-pro02.kfk.com:,bigdata-pro03.kfk.com:</value> </property> <property> <name>dfs.ha.fencing.methods</name> <value>sshfence</value> </property> <property> <name>dfs.ha.fencing.ssh.private-key-files</name> <value>/home/kfk/.ssh/id_rsa</value> </property></configuration>
然后创建JournalNode日志目录：
[kfk@bigdata-pro01 data]$ mkdir jn[kfk@bigdata-pro01 data]$ lsdist-tmp jn tmp[kfk@bigdata-pro01 data]$ cd jn[kfk@bigdata-pro01 jn]$ pwd/opt/momdules/hadoop-2.6./data/jn
3）修改core-site.xml配置文件
<configuration> <property> <name>fs.defaultFS</name> <value>hdfs://ns</value> </property> <property> <name>hadoop.http.staticuser.user</name> <value>kfk</value> </property> <property> <name>hadoop.tmp.dir</name> <value>/opt/modules/hadoop-2.6./data/tmp</value> </property> <property> <name>dfs.namenode.name.dir</name> <value>file://${hadoop.tmp.dir}/dfs/name</value> </property></configuration>
4）将修改的配置分发到其他节点
先同样对非HA环境进行备份：
然后再将HA环境分发给其他节点：
scp -r hadoop/ bigdata-pro02.kfk.com:/opt/modules/hadoop-2.6./etcscp -r hadoop/ bigdata-pro03.kfk.com:/opt/modules/hadoop-2.6./etc
3.HDFS-HA 服务启动及自动故障转移测试
1）启动所有节点上面的Zookeeper进程
zkServer.sh start（本次在前面的过程中已经启动了，以后注意启动顺序）
2）启动所有节点上面的journalnode进程
sbin/hadoop-daemon.sh start journalnode
3）在[nn1]上，对namenode进行格式化，并启动
#namenode 格式化bin/hdfs namenode -format#格式化高可用并启动1和2节点的zkfcbin/hdfs zkfc -formatZKsbin/hadoop-daemon.sh start zkfc#启动节点一的namenodesbin/hadoop-daemon.sh start namenode
4）在[nn2]上，同步nn1元数据信息
bin/hdfs namenode -bootstrapStandby然后启动节点二的namenodesbin/hadoop-daemon.sh start namenode
5）启动所有节点的DataNode
sbin/hadoop-daemon.sh start datanode
然后通过命令上传文件至hdfs，检查hdfs是否可用。
[kfk@bigdata-pro01 hadoop-2.6.]$ bin/hdfs dfs -mkdir -p /user/kfk/data[kfk@bigdata-pro01 hadoop-2.6.]$ bin/hdfs dfs -put /opt/momdules/hadoop-2.6./etc/hadoop/core-site.xml /user/kfk/data
hdfs启动之后，kill其中active状态的namenode，观察另外一个NameNode是否会自动切换为active状态。然后在节点1（停掉的NameNode）上查看我们刚才上传的文件，如果成功表示HA配置是成功的！
[kfk@bigdata-pro01 hadoop-2.6.]$ sbin/hadoop-daemon.sh stop namenodestopping namenode[kfk@bigdata-pro01 hadoop-2.6.]$ bin/hdfs dfs -text /user/kfk/data/core-site.xml
成功读取！并且两个节点的状态也发生了改变。
YARN-HA原理及配置
1.YARN-HA架构原理及介绍
ResourceManager HA 由一对Active，Standby结点构成，通过RMStateStore存储内部数据和主要应用的数据及标记。目前支持的可替代的RMStateStore实现有：基于内存的MemoryRMStateStore，基于文件系统的FileSystemRMStateStore，及基于zookeeper的ZKRMStateStore。 ResourceManager HA的架构模式同NameNode HA的架构模式基本一致，数据共享由RMStateStore，而ZKFC成为 ResourceManager进程的一个服务，非独立存在。
2.YARN-HA详细配置
修改yarn-site.xml配置文件
<configuration> <property> <name>yarn.nodemanager.aux-services</name> <value>mapreduce_shuffle</value> </property> <property> <name>yarn.resourcemanager.ha.enabled</name> <value>true</value> </property> <property> <name>yarn.resourcemanager.cluster-id</name> <value>rs</value> </property> <property> <name>yarn.resourcemanager.ha.rm-ids</name> <value>rm1,rm2</value> </property> <property> <name>yarn.resourcemanager.hostname.rm1</name> <value>bigdata-pro01.kfk.com</value> </property> <property> <name>yarn.resourcemanager.hostname.rm2</name> <value>bigdata-pro02.kfk.com</value> </property> <property> <name>yarn.resourcemanager.zk-address</name> <value>bigdata-pro01.kfk.com:,bigdata-pro02.kfk.com:,bigdata-pro03.kfk.com:</value> </property> <property> <name>yarn.resourcemanager.recovery.enabled</name> <value>true</value> </property> <property> <name>yarn.resourcemanager.store.class</name> <value>org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore</value> </property> <property> <name>yarn.nodemanager.aux-services.mapreduce_shuffle.class</name> <value>org.apache.hadoop.mapred.ShuffleHandler</value> </property> <property> <name>yarn.log-aggregation-enable</name> <value>true</value> </property> <property> <name>yarn.log-aggregation.retain-seconds</name> <value></value> </property></configuration>
3）将修改的配置分发到其他节点
scp yarn-site.xml bigdata-pro02.kfk.com:/opt/modules/hadoop-2.6./etc/hadoop/scp yarn-site.xml bigdata-pro03.kfk.com:/opt/modules/hadoop-2.6./etc/hadoop/
3.YARN-HA服务启动及自动故障转移测试
1）在rm1节点上启动yarn服务
sbin/start-yarn.sh
2）在rm2节点上启动ResourceManager服务
sbin/yarn-daemon.sh start resourcemanager
3）查看yarn的web界面
http://bigdata-pro01.kfk.com:8088
http://bigdata-pro02.kfk.com:8088
4）查看ResourceManager主备节点状态
#bigdata-pro01.kfk.com节点上执行bin/yarn rmadmin -getServiceState rm1
#bigdata-pro02.kfk.com节点上执行bin/yarn rmadmin -getServiceState rm2
5）hadoop集群测试WordCount运行
[kfk@bigdata-pro01 hadoop-2.6.]$ bin/yarn jar share/hadoop/mapreduce/hadoop-mapreduce-examples-2.6..jar wordcount /user/kfk/data/wc.input /user/kfk/data/output// :: WARN util.NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable// :: INFO input.FileInputFormat: Total input paths to process : // :: INFO mapreduce.JobSubmitter: number of splits:// :: INFO mapreduce.JobSubmitter: Submitting tokens for job: job_1540197665543_0001// :: INFO impl.YarnClientImpl: Submitted application application_1540197665543_0001// :: INFO mapreduce.Job: The url to track the job: http://bigdata-pro01.kfk.com:8088/proxy/application_1540197665543_0001/// :: INFO mapreduce.Job: Running job: job_1540197665543_0001// :: INFO mapreduce.Job: Job job_1540197665543_0001 running in uber mode : false// :: INFO mapreduce.Job: map % reduce %// :: INFO mapreduce.Job: map % reduce %// :: INFO mapreduce.Job: map % reduce %// :: INFO mapreduce.Job: Job job_1540197665543_0001 completed successfully// :: INFO mapreduce.Job: Counters: File System Counters FILE: Number of bytes read= FILE: Number of bytes written= FILE: Number of read operations= FILE: Number of large read operations= FILE: Number of write operations= HDFS: Number of bytes read= HDFS: Number of bytes written= HDFS: Number of read operations= HDFS: Number of large read operations= HDFS: Number of write operations= Job Counters Launched map tasks= Launched reduce tasks= Data-local map tasks= Total time spent by all maps in occupied slots (ms)= Total time spent by all reduces in occupied slots (ms)= Total time spent by all map tasks (ms)= Total time spent by all reduce tasks (ms)= Total vcore-seconds taken by all map tasks= Total vcore-seconds taken by all reduce tasks= Total megabyte-seconds taken by all map tasks= Total megabyte-seconds taken by all reduce tasks= Map-Reduce Framework Map input records= Map output records= Map output bytes= Map output materialized bytes= Input split bytes= Combine input records= Combine output records= Reduce input groups= Reduce shuffle bytes= Reduce input records= Reduce output records= Spilled Records= Shuffled Maps = Failed Shuffles= Merged Map outputs= GC time elapsed (ms)= CPU time spent (ms)= Physical memory (bytes) snapshot= Virtual memory (bytes) snapshot= Total committed heap usage (bytes)= Shuffle Errors BAD_ID= CONNECTION= IO_ERROR= WRONG_LENGTH= WRONG_MAP= WRONG_REDUCE= File Input Format Counters Bytes Read= File Output Format Counters Bytes Written=[kfk@bigdata-pro01 hadoop-2.6.]$ bin/hdfs dfs -text /user/kfk/data/output/par*// :: WARN util.NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicablehadoop hbase hive java spark
以上就是博主为大家介绍的这一板块的主要内容，这都是博主自己的学习过程，希望能给大家带来一定的指导作用，有用的还望大家点个支持，如果对你没用也望包涵，有错误烦请指出。如有期待可关注博主以第一时间获取更新哦，谢谢！同时也欢迎转载，但必须在博文明显位置标注原文地址，解释权归博主所有！
