Awesome Open Source
Awesome Open Source


build codecov Maven Central javadoc documentation

If you’re a data scientist or data engineer, this might sound familiar while working on an ETL project:

  • Switching between multiple projects is a hassle
  • Debugging others’ code is a nightmare
  • Spending a lot of time solving non-business-related issues

SETL (pronounced "settle") is a Scala ETL framework powered by Apache Spark that helps you structure your Spark ETL projects, modularize your data transformation logic and speed up your development.


In a new project

You can start working by cloning this template project.

In an existing project


To use the SNAPSHOT version, add Sonatype snapshot repository to your pom.xml



Quick Start

Basic concept

With SETL, an ETL application could be represented by a Pipeline. A Pipeline contains multiple Stages. In each stage, we could find one or several Factories.

The class Factory[T] is an abstraction of a data transformation that will produce an object of type T. It has 4 methods (read, process, write and get) that should be implemented by the developer.

The class SparkRepository[T] is a data access layer abstraction. It could be used to read/write a Dataset[T] from/to a datastore. It should be defined in a configuration file. You can have as many SparkRepositories as you want.

The entry point of a SETL project is the object io.github.setl.Setl, which will handle the pipeline and spark repository instantiation.

Show me some code

You can find the following tutorial code in the starter template of SETL. Go and clone it :)

Here we show a simple example of creating and saving a Dataset[TestObject]. The case class TestObject is defined as follows:

case class TestObject(partition1: Int, partition2: String, clustering1: String, value: Long)

Context initialization

Suppose that we want to save our output into src/main/resources/test_csv. We can create a configuration file local.conf in src/main/resources with the following content that defines the target datastore to save our dataset:

testObjectRepository {
  storage = "CSV"
  path = "src/main/resources/test_csv"
  inferSchema = "true"
  delimiter = ";"
  header = "true"
  saveMode = "Append"

In our App.scala file, we build Setl and register this data store:

val setl: Setl = Setl.builder()

// Register a SparkRepository to context

Implementation of Factory

We will create our Dataset[TestObject] inside a Factory[Dataset[TestObject]]. A Factory[A] will always produce an object of type A, and it contains 4 abstract methods that you need to implement:

  • read
  • process
  • write
  • get
class MyFactory() extends Factory[Dataset[TestObject]] with HasSparkSession {
  import spark.implicits._
  // A repository is needed for writing data. It will be delivered by the pipeline
  private[this] val repo = SparkRepository[TestObject]

  private[this] var output = spark.emptyDataset[TestObject]

  override def read(): MyFactory.this.type = {
    // in our demo we don't need to read any data

  override def process(): MyFactory.this.type = {
    output = Seq(
      TestObject(1, "a", "A", 1L),
      TestObject(2, "b", "B", 2L)

  override def write(): MyFactory.this.type = {  // use the repository to save the output

  override def get(): Dataset[TestObject] = output


Define the pipeline

To execute the factory, we should add it into a pipeline.

When we call setl.newPipeline(), Setl will instantiate a new Pipeline and configure all the registered repositories as inputs of the pipeline. Then we can call addStage to add our factory into the pipeline.

val pipeline = setl

Run our pipeline


The dataset will be saved into src/main/resources/test_csv

What's more?

As our MyFactory produces a Dataset[TestObject], it can be used by other factories of the same pipeline.

class AnotherFactory extends Factory[String] with HasSparkSession {

  import spark.implicits._

  private[this] val outputOfMyFactory = spark.emptyDataset[TestObject]

  override def read(): AnotherFactory.this.type = this

  override def process(): AnotherFactory.this.type = this

  override def write(): AnotherFactory.this.type = {

  override def get(): String = "output"

Add this factory into the pipeline:


Custom Connector

You can implement you own data source connector by implementing the ConnectorInterface

class CustomConnector extends ConnectorInterface with CanDrop {
  override def setConf(conf: Conf): Unit = null

  override def read(): DataFrame = {
    import spark.implicits._
    Seq(1, 2, 3).toDF("id")

  override def write(t: DataFrame, suffix: Option[String]): Unit = logDebug("Write with suffix")

  override def write(t: DataFrame): Unit = logDebug("Write")

   * Drop the entire table.
  override def drop(): Unit = logDebug("drop")

To use it, just set the storage to OTHER and provide the class reference of your connector:

myConnector {
  storage = "OTHER"
  class = "com.example.CustomConnector"  // class reference of your connector 

Generate pipeline diagram

You can generate a Mermaid diagram by doing:


You will have some log like this:

--------- MERMAID DIAGRAM ---------
class MyFactory {

class DatasetTestObject {
  >partition1: Int
  >partition2: String
  >clustering1: String
  >value: Long

DatasetTestObject <|.. MyFactory : Output
class AnotherFactory {

class StringFinal {

StringFinal <|.. AnotherFactory : Output
class SparkRepositoryTestObjectExternal {

AnotherFactory <|-- DatasetTestObject : Input
MyFactory <|-- SparkRepositoryTestObjectExternal : Input

------- END OF MERMAID CODE -------

You can copy the previous code to a markdown viewer that supports Mermaid.

Or you can try the live editor:

You can either copy the code into a Markdown viewer or just copy the link into your browser (link) 🍻

App Configuration

The configuration system of SETL allows users to execute their Spark application in different execution environments, by using environment-specific configurations.

In src/main/resources directory, you should have at least two configuration files named application.conf and local.conf (take a look at this example). These are what you need if you only want to run your application in one single environment.

You can also create other configurations (for example dev.conf and prod.conf), in which environment-specific parameters can be defined.


This configuration file should contain universal configurations that could be used regardless the execution environment.

env.conf (e.g. local.conf, dev.conf)

These files should contain environment-specific parameters. By default, local.conf will be used.

How to use the configuration

Imagine the case we have two environments, a local development environment and a remote production environment. Our application needs a repository for saving and loading data. In this use case, let's prepare application.conf, local.conf, prod.conf and storage.conf

# application.conf
setl.environment = ${app.environment}
setl.config { = "my_application"
  # and other general spark configurations  
# local.conf
include "application.conf"

setl.config {
  spark.default.parallelism = "200"
  spark.sql.shuffle.partitions = "200"
  # and other local spark configurations  

app.root.dir = "/some/local/path"

include "storage.conf"
# prod.conf
setl.config {
  spark.default.parallelism = "1000"
  spark.sql.shuffle.partitions = "1000"
  # and other production spark configurations  

app.root.dir = "/some/remote/path"

include "storage.conf"
# storage.conf
myRepository {
  storage = "CSV"
  path = ${app.root.dir}  // this path will depend on the execution environment
  inferSchema = "true"
  delimiter = ";"
  header = "true"
  saveMode = "Append"

To compile with local configuration, with maven, just run:

mvn compile

To compile with production configuration, pass the jvm property app.environment.

mvn compile -Dapp.environment=prod

Make sure that your resources directory has filtering enabled:



SETL currently supports the following data source. You won't need to provide these libraries in your project (except the JDBC driver):

To read/write data from/to AWS S3 (or other storage services), you should include the corresponding hadoop library in your project.

For example


You should also provide Scala and Spark in your pom file. SETL is tested against the following version of Spark:

Spark Version Scala Version Note
3.0 2.12 ✔️ Ok
2.4 2.12 ✔️ Ok
2.4 2.11 ⚠️ see known issues
2.3 2.11 ⚠️ see known issues

Known issues

Spark 2.4 with Scala 2.11

When using setl_2.11-1.x.x with Spark 2.4 and Scala 2.11, you may need to include manually these following dependencies to override the default version:


Spark 2.3 with Scala 2.11

  • DynamoDBConnector doesn't work with Spark version 2.3
  • Compress annotation can only be used on Struct field or Array of Struct field with Spark 2.3

Test Coverage



Contributing to SETL

Check our contributing guide

Get A Weekly Email With Trending Projects For These Topics
No Spam. Unsubscribe easily at any time.
machine-learning (3,631
scala (2,624
framework (1,116
data-science (900
dataset (409
spark (363
data-analysis (280
big-data (240
pipeline (177
etl (106
data-engineering (52
modularization (22