Creating a multinode Apache Spark cluster on AWS from command line

The idea is to create a Spark cluster on AWS according to the needs, maximize the use of it and terminate it after the processing is done.

All the instances have Ubuntu 14.04 operating system, the instance used in the following script the free-tier t2.micro instance with 8GB storage and 1GB RAM.

To have this working from the command line, aws package has to be installed.

One instance, master in this case, is always in State “running” and this instance has the following installed:

  • aws package
  • Apache Spark

How to install Spark 2.0 is described here. There is also the link to the blog about how to install Spark History Server.

The Spark History Server can be reached on port 18080 – MASTER_PUBLIC_IP:18080 and Spark Master is on port 8080. If the pages do not load, start the services as spark user. The scripts for starting the services can be found in $SPARK_HOME/sbin.

The script for launching instances

The following script takes in one parameter – number of instances to launch and attach them to the spark cluster. These instances would become workers in the spark world.

A new script is created


And the following lines are written in it. Adjust accordingly (details below).


#removes any old associations that have "worker" in it
sudo sed -i.bak '/worker/d' /etc/hosts

#removes known hosts from the file
sudo sed -i 'd' /home/ubuntu/.ssh/known_hosts

#run the loop to create as many slaves as defined in the input parameter
for i in $(seq -f %02g 1 $1)
  #name of the worker

  #run the aws command to create an instance and run a script when the instance is created.
  #the command returns the private IP address which is used to update the local /etc/hosts file
  PRIV_IP_ADDR=$(aws ec2 run-instances --image-id ami-0d77397e --count 1 \
            --instance-type t2.micro --key-name MYKEYPAIR \
            --user-data file:///PATH_TO_SCRIPT/ \
            --subnet-id SUBNET_ID --security-group-ids SECURITY_GROUP \
            --associate-public-ip-address \
            --output text | grep PRIVATEIPADDRESSES | awk '{print $4 "\t" $3}')

   #add the IP and hostname association to the /etc/hosts file
   echo "$PRIV_IP_ADDR" "$NAME" | sudo tee -a /etc/hosts


Line 19: option user-data allows you to define a file that is run on the new instance. In this file, the steps for Spark worker installation and setup are defined.
MYKEYPAIR: name of the private key you use in AWS console when launching new instances (this is not a path to the private key on the instance you are running the script from!).
SUBNET_ID: ID of the subnet you are using. It is something one creates when starting with EC2 in AWS.
SECURITY_GROUP: name of the existing security group that this instance should use.


The script that runs at instance launch

The following commands in the script are run as soon as each instance is created.
At the top of the script, the operating system is updated, python is installed and JDK as well.


sudo apt-get update -y  && \
sudo apt-get install python-minimal -y && \
sudo apt-get install default-jdk -y && \
cd /etc/ && \
sudo wget && \
sudo tar -xvzf spark-2.0.1-bin-hadoop2.7.tgz && \
sudo rm spark-2.0.1-bin-hadoop2.7.tgz && \
sudo useradd spark && \
export SPARK_HOME=/etc/spark-2.0.1-bin-hadoop2.7 && \
sudo chown -R spark:spark $SPARK_HOME && \
sudo -u spark mkdir $SPARK_HOME/logs && \
sudo -u spark $SPARK_HOME/sbin/ spark://

Lines 6-7: Spark package is downloaded into /etc/ and unpacked.
Line 14: The slave is started and connected to the master node (Spark Master’s URL address is the script’s parameter).

Example of running the script

Thre following example creates a Spark cluster with 3 workers.

sh 3

Example of the output: worker01 worker02 worker03

These three lines are also added to the /etc/hosts.

The Spark Master Web Interface (SPARK_MASTER_PUBLIC_IP:8080) gives more details about the Spark cluster.


The interface displays the recently added workers with State being ALIVE. Two workers have are in state DEAD – they have been terminated from AWS, but Spark Master has not updated the Workers statistic yet.

Memory in use is 3 GB – the instances used in the cluster creation has one GB each.


The blog post shows a very simple way of creating a Spark cluster. The cluster creation script can be made a lot more dynamic and the script that is run on the newly created instances could be extended (installing python packaged needed for work).

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s