Install Yarn on Ubuntu Cluster via Scripts

Few months ago, I took a class about Cloud Computing, which is the very first time I have chance to know Hadoop. Cloud Computing sounds beautiful, but in order to build this magic cloud, hard works have to be done. When I was doing course project for that class, I felt extremely boring and tired to login every node and configure something (I didn't realize I can use some tools to simplify the work). It is this reason that pushes me to find another way to build the cloud.

Fortunately, a very good book: Apache Hadoop Yarn shows me the pain-less way. In Chapter 2, the book offers a script-based method to install Yarn on nodes which is elegant and concise. The code provided by this book is available to download. These code works only on CentOS. However, I am familiar and also a fan of Ubuntu, so, the work I'v done to make it works with Ubuntu forms today's post.

Before the content, I would like to quote from Optimizing LInux Performance by Phillip G. Ezolt:

Avoid repeating the work of others.

Avoid repeating your own work.

♠  Before Run the Scripts

1. Install Java

Jave should be installed in every node in your cluster for Yarn to work. If you are using Ubuntu 14.04 just like me, the code below will install java for you. More about install java

sudo add-apt-repository ppa:webupd8team/java
sudo apt-get update
sudo apt-get install oracle-java7-installer

2. Install XML Parser

Script will automatically generate configuration XMLs for you. In order for this feature to work, LibXml2 library has to be installed.

sudo apt-get install libxml2-utils

3. Create User and Group for Using Yarn

I create a default user for using Yarn across cluster: ynuser who is belong to group yarn.

You can add more users to specify how you use yarn. Like one user for using mapreduce, one user for spark etc. It's all depends on you, but remember to modify corresponding part of script to satisfy your demand.

sudo addgroup yarn
sudo adduser —ingroup yarn ynuser

After this, switch to account ynuser.

4. Install Parallel Distributed Shell: Pdsh

Pdsh is an amazing tool helps you execute command through the nodes connected by pdsh. The script highly depends on this tool and pdcp which included in the tool kit too. Before you run the actual script, please setup pdsh correctly.

sudo apt-get install pdsh

When pdsh is installed, some configurations still need to do. First is to change default Remote Command Service (RCMD) to ssh, since by default pdsh uses linux rcmd to execute command on a remote client but not ssh.

echo 'ssh' > /etc/pdsh/rcmd_default

This will save you for typing -R ssh in pdsh & pdcp every time. After change protocol to ssh, next we set up passwd-less connections between nodes for not only pdsh, but also Yarn.

put IP address:hostname in you hosts file.

sudo vim /etc/hosts
should look like this   localhost
IP address  node1
IP address  node2
IP address  node3

generate ssh-key and distribute.

#log in as ynuser

ssh-keygen -t rsa
#key will be generated in ~/.ssh/ directory

ssh-copy-id -i ~/.ssh/ hostnasme
#hostname means the node you want to login
#do ssh-copy-id on every node you want to connect

Do ssh ynuser@hostname to verify your ssh setting is correct.

using the command below to test pdsh

pdsh -w node1,node2,... uptime
#note: no space(s) between hostnames

or, using the node number range, eg. operate on node#1 to node#19:

pdsh -w node[1-19] uptime

the output will look like this:

ynuser@student73:/etc/pdsh$ pdsh -a uptime
student74:  05:37:06 up 1 day,  4:23,  1 user,  load average: 1.10, 1.19, 1.22
student75:  05:37:06 up 1 day,  4:23,  1 user,  load average: 1.34, 1.19, 1.16
student73:  05:37:06 up 1 day,  4:22,  1 user,  load average: 1.11, 1.15, 1.19

since I configured hostname in my gender file[/etc/genders], I can use -a to let pdsh resolve the hosts automatically. Check this for more.

5. Grant Permission to Yarn User

Some operstions in the scripts require more privileges. In order to eliminate input passowrd again and again in the install progress. (Actually, pdsh doesn't redirect input from remote node, which makes it impossible to input password for remote node.) We need to grant super privileges: sudo operations without password, for ynuser on every node. You can simply change it back after installation.

sudo vim /etc/sudoers

and add this line at the end of it


Now you don't need to input password for sudo operations. Oui~

Beside this, some files we are going to modify are belong to root account and can not modify without permission. So, we add ynuser to root group and grant group write permission:

sudo usermod -a -G root ynuser
sudo chmod -R ug+wx /etc/init.d
#startup scripts will be put here, and started with OS boot
sudo chmod -R ug+wx /opt

6. Install sysv-rc-conf

This tool is a replacement for chkconfig in CentOS. More Info

sudo apt-get install sysv-rc-conf

7. Download Hadoop pre-build Tarball

Download Hadoop from here, and put in the same directory with scripts.

♠  How It Works

This part explains some importants of script.

1. Copy Hadoop Tarball to All Nodes, and Extract

if [ ! -f /opt/hadoop-"$HADOOP_VERSION".tar.gz ]; then
    echo "Copying Hadoop $HADOOP_VERSION to all hosts..."
    pdcp -w ^all_hosts hadoop-"$HADOOP_VERSION".tar.gz /opt
    echo "Hadoop $HADOOP_VERSION is there already to be extracted."

pdsh -w ^all_hosts tar -zxf /opt/hadoop-"$HADOOP_VERSION".tar.gz -C /opt

2. Set Hadoop Version

At the beginning of this script, HADOOP_VERSION can be changed according to your Hadoop version. I tested with Hadoop 2.5.1.


3. Set JAVA_HOME Location


4. Define Your Cluster Topo

You can define cluster topo in files. Yarn Install Script will read them in, and confingure the cluster as you wish. THe meaning of each files are listed below.

Note: all hostanmes in these files are separated by ONE space

5. Distribute Bash Startup Files

In order to let Yarn running when your cluster starts up. Some environent variable should be export. There are multiple places can do this: ~/.bash_profile, ~/.bashrc, /etc/profile, and scripts in /etc/profile.d. First two are user-specific, and last two are global. But there are differences between last two places, the scripts in /etc/profile.d are application-specific startup scripts, and it helps you organize system in modules which is much easy in terms of maintenance, see more. So, we put the script in /etc/profile.d

pdsh -w ^all_hosts  echo "export JAVA_HOME=$JAVA_HOME > /etc/profile.d/"
pdsh -w ^all_hosts  echo "export HADOOP_HOME=$HADOOP_HOME > /etc/profile.d/"
pdsh -w ^all_hosts  echo "export HADOOP_PREFIX=$HADOOP_HOME >> /etc/profile.d/"
pdsh -w ^all_hosts  echo "export HADOOP_CONF_DIR=$HADOOP_CONF_DIR >> /etc/profile.d/"

6. Create Directories Across Cluster for Yarn

You can see these directories in the beginning of script:


And scripts will mkdir for every line above. Because we'v set sudo-passwd-less before, the operations now are executed without password.

7. Generate XML Configuration Files

Hadoop is configured by some XML files which indicates different attributes of Hadoop components. The Script will help you to generate these files automatically. Also, feel free to remove/add attributes by youself. The command to add one new attribute in XML file is very easy and intuitive. E.g, for create core-site.xml and put namenode location, simply do this:

create_config --file core-site.xml
put_config --file core-site.xml --property fs.defaultFS --value "hdfs://$nn:9000"

These XML files will be copied to all nodes.

8. Format Namenode

The Namenode will be formated during installation. Please noted here assumes you have no HDFS exists, meaning DN_DATA_DIR = /var/data/hadoop/hdfs/dn should be empty, Check how to use uninstall at the end this article.

If there is HDFS filesystem exists, we suppose to have the prompt asking for command (Y/N) about if we want to re-format, but according to my test, pdsh can't redirect input to remote node, so that we don't know when should we input and even if we input (like ues yes command to periodically send yes to stdin), the remote node still can't receive it. As a result, the installation progress will hang there. That's where this assumption comes from.

One issue here is sometimes the JAVA_HOME can not be resolved correctly in, so, we have to explicitly set it.

#in order to fix "JAVA_HOME not found issue"
sed -i "s|\${JAVA_HOME}|$JAVA_HOME|g" $HADOOP_HOME/etc/hadoop/

pdcp -w ^all_hosts $HADOOP_HOME/etc/hadoop/ $HADOOP_HOME/etc/hadoop/
pdsh -w ^nn_host "$HADOOP_HOME/bin/hdfs namenode -format"

9. Copy Startup Scripts to Nodes

You don't want to start up Hadoop every time you start your machines, do you? The install scripts will allow hadoop to start with the OS booting up by put scripts in /etc/init.d/. So that hadoop services will be started as daemons in your systems.

Note that, these scripts are modified to work under Ubuntu. Three lines below are added.

source /etc/profile.d/
source /etc/profile.d/

source /lib/lsb/init-functions
#init-functions has the function to start daemon

the code below is commented out, since it works under CentOS

source /etc/rc.d/init.d/functions

in every startup script, changes are also made for Ubuntu.

10. Start up Hadoop Services

Hadoop services will be treated as daemons, and services are going to be started just like you start a normal service. By doing this. We need to register each service in OS, which brings the reason why we need to instasll sysv-rc-conf in the beginning. For more details about sysv-rc-conf, see this.

echo "Starting Hadoop $HADOOP_VERSION services on all hosts..."
pdsh -w ^nn_host "chmod 755 /etc/init.d/hadoop-namenode && sudo sysv-rc-conf hadoop-namenode on && sudo service hadoop-namenode start"
pdsh -w ^snn_host "chmod 755 /etc/init.d/hadoop-secondarynamenode && sudo sysv-rc-conf hadoop-secondarynamenode on && sudo service hadoop-secondarynamenode start"
pdsh -w ^dn_hosts "chmod 755 /etc/init.d/hadoop-datanode && sudo sysv-rc-conf hadoop-datanode on && sudo service hadoop-datanode pdsh -w ^rm_host "chmod 755 /etc/init.d/hadoop-resourcemanager && sudo sysv-rc-conf hadoop-resourcemanager on && sudo service hadoop-resourcemanager start"
pdsh -w ^nm_hosts "chmod 755 /etc/init.d/hadoop-nodemanager && sudo sysv-rc-conf hadoop-nodemanager on && sudo service hadoop-nodemanager start"

11. Time for Smoke Test

Run a pi program in your new installed Yarn Cluster

hadoop jar $HADOOP_HOME/share/hadoop/mapreduce/hadoop-mapreduce-examples-$HADOOP_VERSION.jar pi -libjars $HADOOP_HOME/share/hadoop/mapreduce/hadoop-mapreduce-client-jobclient-$HADOOP_VERSION.jar 16 10000

3.14 delivers the greeting from Yarn.

♠  How To Run

This is the easy part. After you are clear about the work mentioned section1 and section2. Simply do this in your terminal.

git clone
cd hadoop-install-scripts
source -f | tee log

For Uninstall back


Note: will delete current files in your HDFS, be careful. Also, sometimes jps will not show running hadoop service which makes new installation failure misleading, so, the uninstall script will also kill your java progress to give new install a fresh environment, be careful and modify as your demands.

Thank you for reading.


Creative Commons License
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.