linkedin/linkedin-gradle-plugin-for-apache-hadoop

LinkedIn Gradle Plugin for Apache Hadoop

The LinkedIn Gradle Plugin for Apache Hadoop (which we shall refer to as simply the "Hadoop Plugin" for brevity) will help you more effectively build, test and deploy Hadoop applications.

In particular, the Plugin will help you easily work with Hadoop applications like Apache Pig and build workflows for Hadoop workflow schedulers such as Azkaban and Apache Oozie.

The Plugin includes the LinkedIn Gradle DSL for Apache Hadoop (which we shall refer to as simply the "Hadoop DSL" for brevity), a language for specifying jobs and workflows for Azkaban.

Hadoop Plugin User Guide

The Hadoop Plugin User Guide is available at User Guide.

Hadoop DSL Language Reference

The Hadoop DSL Language Reference is available at Hadoop DSL Language Reference.

Getting the Hadoop Plugin

The Hadoop Plugin is now published at plugins.gradle.org. Click on the link for a short snippet to add to your build.gradle file to start using the Hadoop Plugin.

Can I Benefit from the Hadoop Plugin and Hadoop DSL?

You must use Gradle as your build system to use the Hadoop Plugin. If you are using Azkaban, you should start using the Hadoop Plugin immediately and you should use the Hadoop DSL to develop all of your Azkaban workflows.

If you are using Apache Pig, the Plugin includes features that will statically validate your Pig scripts, saving you time by finding errors at build time instead of when you run your Pig script.

If you run Apache Pig or Apache Spark on a Hadoop cluster through a gateway node, the Plugin includes tasks that will automate the process of launching your Pig or Spark jobs on the gateway without you having to manually download your code and dependencies there first.

If you are using Gradle and you feel that you might benefit from any of the above features, consider using the Hadoop Plugin and the Hadoop DSL.

Example Project

We have added an Example Project that uses the Hadoop Plugin and DSL to build an example Azkaban workflow consisting of Apache Pig, Apache Hive and Java Map-Reduce jobs.

Apache Oozie Status

The Hadoop Plugin includes Gradle tasks for Apache Oozie, including the ability to upload versioned directories to HDFS, as well as Gradle tasks for issuing Oozie commands. If you are using Gradle as your build system and Apache Oozie as your Hadoop workflow scheduler, you might find the Hadoop Plugin useful. However, we would like to mention the fact that since we are no longer actively using Oozie at LinkedIn, it is possible that the Oozie tasks may fall into a non-working state.

Although we started on a Hadoop DSL compiler for Oozie, we did not complete it, and it is currently not in a usable form. We are not currently working on it and it is unlikely to be completed.

Project Structure

The project structure is setup as follows:

azkaban-client: Code to work with Azkaban via the Azkaban REST API
example-project: Example project that uses the Hadoop Plugin and DSL to build an example Azkaban workflow
hadoop-jobs: Code for re-usable Hadoop jobs and implementations of Hadoop DSL job types
hadoop-plugin: Code and tests for the various plugins that comprise the Hadoop Plugin
li-hadoop-plugin: LinkedIn-specific extensions to the Hadoop Plugin, and tests

Although the li-hadoop-plugin code is generally specific to LinkedIn, it is included in the project to show you how to use subclassing to extend the core functionality of the Hadoop Plugin for your organization (and to make sure our open-source contributions don't break the LinkedIn customizations).

Building and Running Test Cases

To build the Plugin and run the test cases, run ./gradlew build from the top-level project directory.

Unit tests

Unit tests are invoked by running ./gradlew :hadoop-plugin:test. Individual tests may be executed using ./gradlew :hadoop-plugin:test --tests="com.linkedin.gradle.zip.HadoopZipTest".

Integration tests

Integration tests are invoked by running ./gradlew :hadoop-plugin:integTest, or by running the check task.