spark-etl 0.0.98

Generic ETL Pipeline Framework for Apache Spark

• Overview
  • Goal
  • Benefit
  • Supported platforms
• APIs
  • Application
    • Application entry signature
  • Job Deployer
  • Job Submitter
• Tool: etl.py
  • Build an application
  • Deploy and run application

See https://stonezhong.github.io/spark_etl/ for more informaion

Overview

Goal

There are many public clouds provide managed Apache Spark as service, such as databricks, AWS EMR, Oracle OCI DataFlow, see the table below for a complete list.

However, each platform has it's own way of launching Spark jobs, and the way to launch spark jobs between platforms are not compatible with each other.

spark-etl is a python package, which simplifies the spark application management cross platforms, with 3 uniformed steps:

• Build your spark application
• Deploy your spark application
• Run your spark application

Benefit

Your application using spark-etl is spark provider agnostic. For example, you can move your application from Azure HDInsight to AWS EMR without changing your application's code.

You can also run a down-scaled version of your data lake with pyspark in a laptop, since pyspark is a supported spark platform, with this feature, you can validate your spark application with pyspark on your laptop, instead of run it in cloud, to save cost.

Supported platforms

	You setup your own Apache Spark Cluster.
	Use PySpark package, fully compatible to other spark platform, allows you to test your pipeline in a single computer.
	You host your spark cluster in databricks
	You host your spark cluster in Amazon AWS EMR
	You host your spark cluster in Google Cloud
	You host your spark cluster in Microsoft Azure HDInsight
	You host your spark cluster in Oracle Cloud Infrastructure, Data Flow Service
	You host your spark cluster in IBM Cloud

Deploy and run application

Please see the Demos

APIs

pydocs for APIs

Application

An application is a pyspark application, so far we only support pyspark, Java and Scala support will be added latter. An application contains:

• A main.py file which contain the application entry
• A manifest.json file, which specify the metadata of the application.
• A requirements.txt file, which specify the application dependency.

Application class:

• You can create an application via Application(app_location)
• You can build an application via app.build(destination_location)

Application entry signature

In your application's main.py, you shuold have a main function with the following signature:

• spark is the spark session object
• input_args a dict, is the argument user specified when running this job.
• sysops is the system options passed, it is platform specific. Job submitter may inject platform specific object in sysops object.
• Your main function's return value will be returned from the job submitter to the caller.

def main(spark, input_args, sysops={}):
    # your code here

Here is an example.

Job Deployer

For job deployers, please check the wiki .

Job Submitter

For job submitters, please check the wiki

Tool: etl.py

Build an application

To build an application, run

./etl.py -a build --app-dir <app-dir> --build-dir <build-dir>

• <app_dir> is the directory where your application is located.

• <build-dir> is the directory where you want your build to be deployed

  • Your build actually located at <build-dir>/<version>, where <version> is specified by application's manifest file

• Build is mostly platform independent. You can put platform related package in file common_requirements.txt