How to Install and Set Up a 3-Node Hadoop Cluster
Updated by Linode Contributed by Florent Houbart
What is Hadoop?
Hadoop is an open-source Apache project that allows creation of parallel processing applications on large data sets, distributed across networked nodes. It’s composed of the Hadoop Distributed File System (HDFS™) that handles scalability and redundancy of data across nodes, and Hadoop YARN: a framework for job scheduling that executes data processing tasks on all nodes.
Before You Begin
Follow the Getting Started guide to create three (3) Linodes. They’ll be referred to throughout this guide as node-master, node1 and node2. It’s recommended that you set the hostname of each Linode to match this naming convention.
Run the steps in this guide from the node-master unless otherwise specified.
Follow the Securing Your Server guide to harden the three servers. Create a normal user for the install, and a user called
hadoopfor any Hadoop daemons. Do not create SSH keys for
hadoopusers. SSH keys will be addressed in a later section.
The steps below use example IPs for each node. Adjust each example according to your configuration:
- node-master: 192.0.2.1
- node1: 192.0.2.2
- node2: 192.0.2.3
NoteThis guide is written for a non-root user. Commands that require elevated privileges are prefixed with
sudo. If you’re not familiar with the
sudocommand, see the Users and Groups guide. All commands in this guide are run with the hadoop user if not specified otherwise.
Architecture of a Hadoop Cluster
Before configuring the master and slave nodes, it’s important to understand the different components of a Hadoop cluster.
A master node keeps knowledge about the distributed file system, like the
inode table on an
ext3 filesystem, and schedules resources allocation. node-master will handle this role in this guide, and host two daemons:
- The NameNode: manages the distributed file system and knows where stored data blocks inside the cluster are.
- The ResourceManager: manages the YARN jobs and takes care of scheduling and executing processes on slave nodes.
Slave nodes store the actual data and provide processing power to run the jobs. They’ll be node1 and node2, and will host two daemons:
- The DataNode manages the actual data physically stored on the node; it’s named,
- The NodeManager manages execution of tasks on the node.
Configure the System
Create Host File on Each Node
For each node to communicate with its names, edit the
/etc/hosts file to add the IP address of the three servers. Don’t forget to replace the sample IP with your IP:
192.0.2.1 node-master 192.0.2.2 node1 192.0.2.3 node2
Distribute Authentication Key-pairs for the Hadoop User
The master node will use an ssh-connection to connect to other nodes with key-pair authentication, to manage the cluster.
Login to node-master as the
hadoopuser, and generate an ssh-key:
ssh-keygen -b 4096
Copy the key to the other nodes. It’s good practice to also copy the key to the node-master itself, so that you can also use it as a DataNode if needed. Type the following commands, and enter the
hadoopuser’s password when asked. If you are prompted whether or not to add the key to known hosts, enter
ssh-copy-id -i $HOME/.ssh/id_rsa.pub hadoop@node-master ssh-copy-id -i $HOME/.ssh/id_rsa.pub hadoop@node1 ssh-copy-id -i $HOME/.ssh/id_rsa.pub hadoop@node2
Download and Unpack Hadoop Binaries
Login to node-master as the
hadoop user, download the Hadoop tarball from Hadoop project page, and unzip it:
cd wget http://apache.mindstudios.com/hadoop/common/hadoop-2.8.1/hadoop-2.8.1.tar.gz tar -xzf hadoop-2.8.1.tar.gz mv hadoop-2.8.1 hadoop
Set Environment Variables
Add Hadoop binaries to your PATH. Edit
/home/hadoop/.profileand add the following line:
Configure the Master Node
Configuration will be done on node-master and replicated to other nodes.
Get your Java installation path. If you installed open-jdk from your package manager, you can get the path with the command:
update-alternatives --display java
Take the value of the current link and remove the trailing
/bin/java. For example on Debian, the link is
If you installed java from Oracle,
JAVA_HOMEis the path where you unzipped the java archive.
~/hadoop/etc/hadoop/hadoop-env.shand replace this line:
with your actual java installation path. For example on a Debian with open-jdk-8:
Set NameNode Location
On each node update
~/hadoop/etc/hadoop/core-site.xml you want to set the NameNode location to node-master on port
<?xml version="1.0" encoding="UTF-8"?> <?xml-stylesheet type="text/xsl" href="configuration.xsl"?> <configuration> <property> <name>fs.default.name</name> <value>hdfs://node-master:9000</value> </property> </configuration>
Set path for HDFS
1 2 3 4 5 6 7 8 910111213141516
<configuration> <property> <name>dfs.namenode.name.dir</name> <value>/home/hadoop/data/nameNode</value> </property> <property> <name>dfs.datanode.data.dir</name> <value>/home/hadoop/data/dataNode</value> </property> <property> <name>dfs.replication</name> <value>1</value> </property> </configuration>
The last property,
dfs.replication, indicates how many times data is replicated in the cluster. You can set
2 to have all the data duplicated on the two nodes. Don’t enter a value higher than the actual number of slave nodes.
Set YARN as Job Scheduler
cd ~/hadoop/etc/hadoop mv mapred-site.xml.template mapred-site.xml
Edit the file, setting yarn as the default framework for MapReduce operations:
<configuration> <property> <name>mapreduce.framework.name</name> <value>yarn</value> </property> </configuration>
1 2 3 4 5 6 7 8 910111213141516
<configuration> <property> <name>yarn.acl.enable</name> <value>0</value> </property> <property> <name>yarn.resourcemanager.hostname</name> <value>node-master</value> </property> <property> <name>yarn.nodemanager.aux-services</name> <value>mapreduce_shuffle</value> </property> </configuration>
slaves is used by startup scripts to start required daemons on all nodes. Edit
~/hadoop/etc/hadoop/slaves to be:
Configure Memory Allocation
Memory allocation can be tricky on low RAM nodes because default values are not suitable for nodes with less than 8GB of RAM. This section will highlight how memory allocation works for MapReduce jobs, and provide a sample configuration for 2GB RAM nodes.
The Memory Allocation Properties
A YARN job is executed with two kind of resources:
- An Application Master (AM) is responsible for monitoring the application and coordinating distributed executors in the cluster.
- Some executors that are created by the AM actually run the job. For a MapReduce jobs, they’ll perform map or reduce operation, in parallel.
Both are run in containers on slave nodes. Each slave node runs a NodeManager daemon that’s responsible for container creation on the node. The whole cluster is managed by a ResourceManager that schedules container allocation on all the slave-nodes, depending on capacity requirements and current charge.
Four types of resource allocations need to be configured properly for the cluster to work. These are:
How much memory can be allocated for YARN containers on a single node. This limit should be higher than all the others; otherwise, container allocation will be rejected and applications will fail. However, it should not be the entire amount of RAM on the node.
This value is configured in
How much memory a single container can consume and the minimum memory allocation allowed. A container will never be bigger than the maximum, or else allocation will fail and will always be allocated as a multiple of the minimum amount of RAM.
Those values are configured in
How much memory will be allocated to the ApplicationMaster. This is a constant value that should fit in the container maximum size.
This is configured in
How much memory will be allocated to each map or reduce operation. This should be less than the maximum size.
This is configured in
The relationship between all those properties can be seen in the following figure:
Sample Configuration for 2GB Nodes
For 2GB nodes, a working configuration may be:
/home/hadoop/hadoop/etc/hadoop/yarn-site.xmland add the following lines:
1 2 3 4 5 6 7 8 910111213141516171819
<property> <name>yarn.nodemanager.resource.memory-mb</name> <value>1536</value> </property> <property> <name>yarn.scheduler.maximum-allocation-mb</name> <value>1536</value> </property> <property> <name>yarn.scheduler.minimum-allocation-mb</name> <value>128</value> </property> <property> <name>yarn.nodemanager.vmem-check-enabled</name> <value>false</value> </property>
The last property disables virtual-memory checking and can prevent containers from being allocated properly on JDK8.
/home/hadoop/hadoop/etc/hadoop/mapred-site.xmland add the following lines:
1 2 3 4 5 6 7 8 91011121314
<property> <name>yarn.app.mapreduce.am.resource.mb</name> <value>512</value> </property> <property> <name>mapreduce.map.memory.mb</name> <value>256</value> </property> <property> <name>mapreduce.reduce.memory.mb</name> <value>256</value> </property>
Duplicate Config Files on Each Node
Copy the hadoop binaries to slave nodes:
cd /home/hadoop/ scp hadoop-*.tar.gz node1:/home/hadoop scp hadoop-*.tar.gz node2:/home/hadoop
Connect to node1 via ssh. A password isn’t required, thanks to the ssh keys copied above:
Unzip the binaries, rename the directory, and exit node1 to get back on the node-master:
tar -xzf hadoop-2.8.1.tar.gz mv hadoop-2.8.1 hadoop exit
Repeat steps 2 and 3 for node2.
Copy the Hadoop configuration files to the slave nodes:
for node in node1 node2; do scp ~/hadoop/etc/hadoop/* $node:/home/hadoop/hadoop/etc/hadoop/; done
HDFS needs to be formatted like any classical file system. On node-master, run the following command:
hdfs namenode -format
Your Hadoop installation is now configured and ready to run.
Run and monitor HDFS
This section will walk through starting HDFS on NameNode and DataNodes, and monitoring that everything is properly working and interacting with HDFS data.
Start and Stop HDFS
Start the HDFS by running the following script from node-master:
It’ll start NameNode and SecondaryNameNode on node-master, and DataNode on node1 and node2, according to the configuration in the
Check that every process is running with the
jpscommand on each node. You should get on node-master (PID will be different):
21922 Jps 21603 NameNode 21787 SecondaryNameNode
and on node1 and node2:
19728 DataNode 19819 Jps
To stop HDFS on master and slave nodes, run the following command from node-master:
Monitor your HDFS Cluster
You can get useful information about running your HDFS cluster with the
hdfs dfsadmincommand. Try for example:
hdfs dfsadmin -report
This will print information (e.g., capacity and usage) for all running DataNodes. To get the description of all available commands, type:
hdfs dfsadmin -help
You can also automatically use the friendlier web user interface. Point your browser to http://node-master-IP:50070 and you’ll get a user-friendly monitoring console.
Put and Get Data to HDFS
Writing and reading to HDFS is done with command
hdfs dfs. First, manually create your home directory. All other commands will use a path relative to this default home directory:
hdfs dfs -mkdir -p /user/hadoop
Let’s use some textbooks from the Gutenberg project as an example.
Create a books directory in HDFS. The following command will create it in the home directory,
hdfs dfs -mkdir books
Grab a few books from the Gutenberg project:
cd /home/hadoop wget -O alice.txt https://www.gutenberg.org/files/11/11-0.txt wget -O holmes.txt https://www.gutenberg.org/ebooks/1661.txt.utf-8 wget -O frankenstein.txt https://www.gutenberg.org/ebooks/84.txt.utf-8
Put the three books through HDFS, in the
hdfs dfs -put alice.txt holmes.txt frankenstein.txt books
List the contents of the
hdfs dfs -ls books
Move one of the books to the local filesystem:
hdfs dfs -get books/alice.txt
You can also directly print the books from HDFS:
hdfs dfs -cat books/alice.txt
There are many commands to manage your HDFS. For a complete list, you can look at the Apache HDFS shell documentation, or print help with:
hdfs dfs -help
HDFS is a distributed storage system, it doesn’t provide any services for running and scheduling tasks in the cluster. This is the role of the YARN framework. The following section is about starting, monitoring, and submitting jobs to YARN.
Start and Stop YARN
Start YARN with the script:
Check that everything is running with the
jpscommand. In addition to the previous HDFS daemon, you should see a ResourceManager on node-master, and a NodeManager on node1 and node2.
To stop YARN, run the following command on node-master:
yarncommand provides utilities to manage your YARN cluster. You can also print a report of running nodes with the command:
yarn node -list
Similarly, you can get a list of running applications with command:
yarn application -list
To get all available parameters of the
yarncommand, see Apache YARN documentation.
As with HDFS, YARN provides a friendlier web UI, started by default on port
8088of the Resource Manager. Point your browser to http://node-master-IP:8088 and browse the UI:
Submit MapReduce Jobs to YARN
Yarn jobs are packaged into
jar files and submitted to YARN for execution with the command
yarn jar. The Hadoop installation package provides sample applications that can be run to test your cluster. You’ll use them to run a word count on the three books previously uploaded to HDFS.
Submit a job with the sample jar to YARN. On node-master, run:
yarn jar ~/hadoop/share/hadoop/mapreduce/hadoop-mapreduce-examples-2.8.1.jar wordcount "books/*" output
The last argument is where the output of the job will be saved - in HDFS.
After the job is finished, you can get the result by querying HDFS with
hdfs dfs -ls output. In case of a success, the output will resemble:
Found 2 items -rw-r--r-- 1 hadoop supergroup 0 2017-10-11 14:09 output/_SUCCESS -rw-r--r-- 1 hadoop supergroup 269158 2017-10-11 14:09 output/part-r-00000
Print the result with:
hdfs dfs -cat output/part-r-00000
Now that you have a YARN cluster up and running, you can:
- Learn how to code your own YARN jobs with Apache documentation.
- Install Spark on top on your YARN cluster with Linode Spark guide.
You may wish to consult the following resources for additional information on this topic. While these are provided in the hope that they will be useful, please note that we cannot vouch for the accuracy or timeliness of externally hosted materials.
- YARN Command Reference
- HDFS Shell Documentation
- core-site.xml properties
- hdfs-site.xml properties
- mapred-site.xml properties
- core-site.xml properties
This guide is published under a CC BY-ND 4.0 license.