Step-by-Step Guide to Building a High-Availability Hadoop HDFS and YARN Cluster

The article explains how to build a High‑Availability (HA) Hadoop cluster (HDFS + YARN) after Hadoop 2.x, detailing each required step.

Download links

http://www.trieuvan.com/apache/zookeeper/stable/zookeeper-3.4.6.tar.gz

http://www.trieuvan.com/apache/hadoop/common/stable/hadoop-2.6.0.tar.gz

http://download.oracle.com/otn-pub/java/jdk/7u75-b13/jdk-7u75-linux-x64.tar.gz

2.1 Create Hadoop user

useradd hadoop

passwd hadoop

chmod +w /etc/sudoers

hadoop ALL=(root)NOPASSWD:ALL

chmod -w /etc/sudoers

2.2 Install JDK

sudo vi /etc/profile

export JAVA_HOME=/usr/java/jdk1.7

export PATH=$PATH:$JAVA_HOME/bin

source /etc/profile

java -version

2.3 Configure hosts

10.211.55.12    nna   # NameNode Active

10.211.55.13    nns   # NameNode Standby

10.211.55.14    dn1   # DataNode1

10.211.55.15    dn2   # DataNode2

10.211.55.16    dn3   # DataNode3

scp /etc/hosts hadoop@nns:/etc/

2.4 Install SSH and configure password‑less login

ssh-keygen -t rsa

cat ~/.ssh/id_rsa.pub >> ~/.ssh/authorized_keys

chmod 600 ~/.ssh/authorized_keys

scp ~/.ssh/authorized_keys hadoop@nns:~/.ssh/

ssh nns

2.5 Disable firewall (for testing)

chkconfig iptables off

2.6 Set timezone to Shanghai

# cp /usr/share/zoneinfo/Asia/Shanghai /etc/localtime

# vi /etc/sysconfig/clock

ZONE="Asia/Shanghai"

UTC=false

2.7 Install, start and verify Zookeeper

tar -zxvf zk-{version}.tar.gz

# edit conf/zoo.cfg (tickTime, initLimit, syncLimit, dataDir, clientPort, server.X entries)

bin/zkServer.sh start

jps   # should show QuorumPeerMain

bin/zkServer.sh status

2.8 HDFS HA architecture diagram (image omitted)

2.9 Role assignment (image omitted)

2.10 Environment variable configuration

export JAVA_HOME=/usr/java/jdk1.7

export HADOOP_HOME=/home/hadoop/hadoop-2.6.0

export ZK_HOME=/home/hadoop/zookeeper-3.4.6

export PATH=$PATH:$JAVA_HOME/bin:$HADOOP_HOME/bin:$HADOOP_HOM

2.11 Core configuration files

mkdir -p /home/hadoop/tmp

mkdir -p /home/hadoop/data/tmp/journal

mkdir -p /home/hadoop/data/dfs/name

mkdir -p /home/hadoop/data/dfs/data

mkdir -p /home/hadoop/data/yarn/local

mkdir -p /home/hadoop/log/yarn

core-site.xml

<configuration>
  <property>
    <name>fs.defaultFS</name>
    <value>hdfs://cluster1</value>
  </property>
  ...
  <property>
    <name>ha.zookeeper.quorum</name>
    <value>dn1:2181,dn2:2181,dn3:2181</value>
  </property>
</configuration>

hdfs-site.xml

<configuration>
  <property>
    <name>dfs.nameservices</name>
    <value>cluster1</value>
  </property>
  <property>
    <name>dfs.ha.namenodes.cluster1</name>
    <value>nna,nns</value>
  </property>
  ...
  <property>
    <name>dfs.namenode.shared.edits.dir</name>
    <value>qjournal://dn1:8485;dn2:8485;dn3:8485/cluster1</value>
  </property>
</configuration>

map-site.xml

<configuration>
  <property>
    <name>mapreduce.framework.name</name>
    <value>yarn</value>
  </property>
  <property>
    <name>mapreduce.jobhistory.address</name>
    <value>nna:10020</value>
  </property>
</configuration>

yarn-site.xml

<configuration>
  <property>
    <name>yarn.resourcemanager.ha.enabled</name>
    <value>true</value>
  </property>
  <property>
    <name>yarn.resourcemanager.ha.rm-ids</name>
    <value>rm1,rm2</value>
  </property>
  ...
  <property>
    <name>yarn.resourcemanager.address.rm1</name>
    <value>nna:8132</value>
  </property>
  <property>
    <name>yarn.resourcemanager.address.rm2</name>
    <value>nns:8132</value>
  </property>
</configuration>

hadoop-env.sh / yarn-env.sh

# The java implementation to use.
export JAVA_HOME=/usr/java/jdk1.7

2.12 Edit slave file

dn1
dn2
dn3

2.13 Start services (order matters)

On each DataNode: bin/zkServer.sh start and verify with jps (QuorumPeerMain).

On NameNode(s): hadoop-daemons.sh start journalnode (or hadoop-daemon.sh start journalnode ).

Format HDFS (first time): hadoop namenode -format .

Format Zookeeper for HA: hdfs zkfc -formatZK .

Start HDFS and YARN: start-dfs.sh and start-yarn.sh . Verify on active NN with jps (DFSZKFailoverController, NameNode, ResourceManager).

On standby NN (NNS) manually start hadoop-daemon.sh start namenode and yarn-daemon.sh start resourcemanager .

Bootstrap standby metadata: hdfs namenode -bootstrapStandby .

2.14 HA failover

If automatic failover is configured, the standby becomes active when the active node fails. For manual failover, run:

hdfs haadmin -failover --forcefence --forceactive nna nns

2.15 Result screenshots (images omitted).

Source: http://www.cnblogs.com/smartloli/p/4298430.html