Skip to content

Packer Template to build a AWS Apache Spark AMI

License

Notifications You must be signed in to change notification settings

fscm/packer-aws-spark

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

41 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Apache Spark AMI

AMI that should be used to create virtual machines with Apache Spark installed.

Synopsis

This script will create an AMI with Apache Spark installed and with all of the required initialization scripts.

The AMI resulting from this script should be the one used to instantiate a Spark server (master or worker).

Getting Started

There are a couple of things needed for the script to work.

Prerequisites

Packer and AWS Command Line Interface tools need to be installed on your local computer. To build a base image you have to know the id of the latest Debian AMI files for the region where you wish to build the AMI.

Packer

Packer installation instructions can be found here.

AWS Command Line Interface

AWS Command Line Interface installation instructions can be found here

Debian AMI's

This AMI will be based on an official Debian AMI. The latest version of that AMI will be used.

A list of all the Debian AMI id's can be found at the Debian official page: Debian official Amazon EC2 Images

Usage

In order to create the AMI using this packer template you need to provide a few options.

Usage:
  packer build \
    -var 'aws_access_key=AWS_ACCESS_KEY' \
    -var 'aws_secret_key=<AWS_SECRET_KEY>' \
    -var 'aws_region=<AWS_REGION>' \
    -var 'spark_version=<SPARK_VERSION>' \
    -var 'spark_hadoop_version=<HADOOP_VERSION>' \
    [-var 'option=value'] \
    spark.json

Script Options

  • aws_access_key - [required] The AWS access key.
  • aws_ami_name - The AMI name (default value: "spark").
  • aws_ami_name_prefix - Prefix for the AMI name (default value: "").
  • aws_instance_type - The instance type to use for the build (default value: "t2.micro").
  • aws_region - [required] The regions were the build will be performed.
  • aws_secret_key - [required] The AWS secret key.
  • java_build_number - Java build number (default value: "11").
  • java_major_version - Java major version (default value: "8").
  • java_token - Java link token (default version: "d54c1d3a095b4ff2b6607d096fa80163").
  • java_update_version - Java update version (default value: "131").
  • scala_short_version - Scala short version (default value: "2.11"). Setting this option also requires setting the scala_version option.
  • scala_version - Scala version (default value: "2.11.8"). Seting this option may also require setting the scala_short_version option.
  • spark_hadoop_version - [required] Hadoop version of the Spark package.
  • spark_version - [required] Spark version.
  • system_locale - Locale for the system (default value: "en_US").

Instantiate a Cluster

In order to end up with a functional Spark Cluster some configurations have to be performed after instantiating the servers.

To help perform those configurations a small script is included on the AWS image. The script is called spark_config.

Configuration Script

The script can and should be used to set some of the Spark options as well as setting the Spark service to start at boot.

Usage: spark_config [options] <instance_type>
Instance Type

The script can only configure one instance at a time. Setting a instance type is required by the script.

  • master - Treats the configuration options as if it was a Spark Master instance.
  • worker - Treats the configuration options as if it was a Spark Worker instance.
  • history - Treats the configuration options as if it was a Spark History instance.
Options
  • -c <CORES> - [worker] Sets the number of Executor cores that Spark Executor will use (default value is the number of cpu cores/threads).
  • -D - [master,worker,history] Disables the respective Spark service from start at boot time.
  • -E - [master,worker,history] Enables the respective Spark service to start at boot time.
  • -h <AGE> - [master,worker,history] Sets how old the job history files will have to be before being deleted on the server (default value is '15d').
  • -i <NUMBER> - [worker] Sets the number of Spark Executor instances that will de started (default value is '1').
  • -k <SIZE> - [master,worker,history] Sets the size of the Kryo Serializer buffer (default value is '16m'). Values should be provided following the same Java heap nomenclature.
  • -m <MEMORY> - [worker] Sets the Spark Executor maximum heap size (default value is 80% of the server memory). Values should be provided following the same Java heap nomenclature.
  • -p <ADDRESS> - [master,worker,history] Sets the public DNS name of the Spark instance (default value is the server FQDN). This is the value that the instance will report as the server address on all the url's (including the ones on the Spark UI).
  • -r <NUMBER> - [worker] Sets the maximum number of log files kept by the Executer log rotator (default value is '15').
  • -s <ADDRESS> - [worker] Sets the Spark Master address to which the Spark Worker will connect to (default value is 'localhost').
  • -S - [master,worker,history] Starts the respective Spark service after performing the required configurations (if any given).
  • -W <SECONDS> - [master,worker,history] Waits the specified amount of seconds before starting the respective Spark service (default value is '0').

Configuring the Spark Master Instance

To prepare an instance to act as a Spark Master the following steps need to be performed.

Run the configuration tool (spark_config) to configure the instance as a Spark Master server.

spark_config -E -S master

After this steps a Spark Master service should be running and configured to start on server boot.

More options can be used on the instance configuration, see the Configuration Script section for more details

Configuring a Spark Worker Instance

To prepare an instance to act as a Spark Worker the following steps need to be performed.

Run the configuration tool (spark_config) to configure the instance as a Spark Worker server.

spark_config -E -S -s spark-master.my-domain.tld worker

After this steps a Spark Worker instance should be running, connected to the specified Spark Master address and configured to start on server boot.

More options can be used on the instance configuration, see the Configuration Script section for more details

Configuring the Spark History Instance

To prepare an instance to act as a Spark History the following steps need to be performed.

Run the configuration tool (spark_config) to configure the instance as a Spark History server.

spark_config -E -S history

To be able to use the Spark History service properly every Spark instance needs to write the job logs to a shared folder. The shared folder should be mounted on the following location on every instance/server (including the History instance) and write permission needs to be given to the spark user (uid=2000).

/var/log/spark

After this steps the Spark History service should be running and configured to start on server boot.

More options can be used on the instance configuration, see the Configuration Script section for more details

Services

This AMI will have the SSH service running as well as the Spark (Master and/or Worker) services. The following ports will have to be configured on Security Groups.

Service Port Protocol
SSH 22 TCP
Spark Application 4040 TCP
Spark REST Server 6066 TCP
Spark Master 7077 TCP
Spark Master UI 8080 TCP
Spark Worker UI 8081 TCP
Spark History 18080 TCP

Contributing

  1. Fork it!
  2. Create your feature branch: git checkout -b my-new-feature
  3. Commit your changes: git commit -am 'Add some feature'
  4. Push to the branch: git push origin my-new-feature
  5. Submit a pull request

Please read the CONTRIBUTING.md file for more details on how to contribute to this project.

Versioning

This project uses SemVer for versioning. For the versions available, see the tags on this repository.

Authors

  • Frederico Martins - fscm

See also the list of contributors who participated in this project.

License

This project is licensed under the MIT License - see the LICENSE file for details

About

Packer Template to build a AWS Apache Spark AMI

Topics

Resources

License

Code of conduct

Stars

Watchers

Forks

Packages

No packages published

Languages