/docker-spark

Apache Spark docker image

Primary LanguageShell

Gitter chat Build Status Twitter

Spark docker

Docker images to:

  • Setup a standalone Apache Spark cluster running one Spark Master and multiple Spark workers
  • Build Spark applications in Java, Scala or Python to run on a Spark cluster
Currently supported versions:
  • Spark 3.3.0 for Hadoop 3.3 with OpenJDK 8 and Scala 2.12
  • Spark 3.2.1 for Hadoop 3.2 with OpenJDK 8 and Scala 2.12
  • Spark 3.2.0 for Hadoop 3.2 with OpenJDK 8 and Scala 2.12
  • Spark 3.1.2 for Hadoop 3.2 with OpenJDK 8 and Scala 2.12
  • Spark 3.1.1 for Hadoop 3.2 with OpenJDK 8 and Scala 2.12
  • Spark 3.1.1 for Hadoop 3.2 with OpenJDK 11 and Scala 2.12
  • Spark 3.0.2 for Hadoop 3.2 with OpenJDK 8 and Scala 2.12
  • Spark 3.0.1 for Hadoop 3.2 with OpenJDK 8 and Scala 2.12
  • Spark 3.0.0 for Hadoop 3.2 with OpenJDK 11 and Scala 2.12
  • Spark 3.0.0 for Hadoop 3.2 with OpenJDK 8 and Scala 2.12
  • Spark 2.4.5 for Hadoop 2.7+ with OpenJDK 8
  • Spark 2.4.4 for Hadoop 2.7+ with OpenJDK 8
  • Spark 2.4.3 for Hadoop 2.7+ with OpenJDK 8
  • Spark 2.4.1 for Hadoop 2.7+ with OpenJDK 8
  • Spark 2.4.0 for Hadoop 2.8 with OpenJDK 8 and Scala 2.12
  • Spark 2.4.0 for Hadoop 2.7+ with OpenJDK 8
  • Spark 2.3.2 for Hadoop 2.7+ with OpenJDK 8
  • Spark 2.3.1 for Hadoop 2.7+ with OpenJDK 8
  • Spark 2.3.1 for Hadoop 2.8 with OpenJDK 8
  • Spark 2.3.0 for Hadoop 2.7+ with OpenJDK 8
  • Spark 2.2.2 for Hadoop 2.7+ with OpenJDK 8
  • Spark 2.2.1 for Hadoop 2.7+ with OpenJDK 8
  • Spark 2.2.0 for Hadoop 2.7+ with OpenJDK 8
  • Spark 2.1.3 for Hadoop 2.7+ with OpenJDK 8
  • Spark 2.1.2 for Hadoop 2.7+ with OpenJDK 8
  • Spark 2.1.1 for Hadoop 2.7+ with OpenJDK 8
  • Spark 2.1.0 for Hadoop 2.7+ with OpenJDK 8
  • Spark 2.0.2 for Hadoop 2.7+ with OpenJDK 8
  • Spark 2.0.1 for Hadoop 2.7+ with OpenJDK 8
  • Spark 2.0.0 for Hadoop 2.7+ with Hive support and OpenJDK 8
  • Spark 2.0.0 for Hadoop 2.7+ with Hive support and OpenJDK 7
  • Spark 1.6.2 for Hadoop 2.6 and later
  • Spark 1.5.1 for Hadoop 2.6 and later

Using Docker Compose

Add the following services to your docker-compose.yml to integrate a Spark master and Spark worker in your BDE pipeline:

version: '3'
services:
  spark-master:
    image: bde2020/spark-master:3.3.0-hadoop3.3
    container_name: spark-master
    ports:
      - "8080:8080"
      - "7077:7077"
    environment:
      - INIT_DAEMON_STEP=setup_spark
  spark-worker-1:
    image: bde2020/spark-worker:3.3.0-hadoop3.3
    container_name: spark-worker-1
    depends_on:
      - spark-master
    ports:
      - "8081:8081"
    environment:
      - "SPARK_MASTER=spark://spark-master:7077"
  spark-worker-2:
    image: bde2020/spark-worker:3.3.0-hadoop3.3
    container_name: spark-worker-2
    depends_on:
      - spark-master
    ports:
      - "8082:8081"
    environment:
      - "SPARK_MASTER=spark://spark-master:7077"
  spark-history-server:
      image: bde2020/spark-history-server:3.3.0-hadoop3.3
      container_name: spark-history-server
      depends_on:
        - spark-master
      ports:
        - "18081:18081"
      volumes:
        - /tmp/spark-events-local:/tmp/spark-events

Make sure to fill in the INIT_DAEMON_STEP as configured in your pipeline.

Running Docker containers without the init daemon

Spark Master

To start a Spark master:

docker run --name spark-master -h spark-master -d bde2020/spark-master:3.3.0-hadoop3.3

Spark Worker

To start a Spark worker:

docker run --name spark-worker-1 --link spark-master:spark-master -d bde2020/spark-worker:3.3.0-hadoop3.3

Launch a Spark application

Building and running your Spark application on top of the Spark cluster is as simple as extending a template Docker image. Check the template's README for further documentation.

Kubernetes deployment

The BDE Spark images can also be used in a Kubernetes enviroment.

To deploy a simple Spark standalone cluster issue

kubectl apply -f https://raw.githubusercontent.com/big-data-europe/docker-spark/master/k8s-spark-cluster.yaml

This will setup a Spark standalone cluster with one master and a worker on every available node using the default namespace and resources. The master is reachable in the same namespace at spark://spark-master:7077. It will also setup a headless service so spark clients can be reachable from the workers using hostname spark-client.

Then to use spark-shell issue

kubectl run spark-base --rm -it --labels="app=spark-client" --image bde2020/spark-base:3.3.0-hadoop3.3 -- bash ./spark/bin/spark-shell --master spark://spark-master:7077 --conf spark.driver.host=spark-client

To use spark-submit issue for example

kubectl run spark-base --rm -it --labels="app=spark-client" --image bde2020/spark-base:3.3.0-hadoop3.3 -- bash ./spark/bin/spark-submit --class CLASS_TO_RUN --master spark://spark-master:7077 --deploy-mode client --conf spark.driver.host=spark-client URL_TO_YOUR_APP

You can use your own image packed with Spark and your application but when deployed it must be reachable from the workers. One way to achieve this is by creating a headless service for your pod and then use --conf spark.driver.host=YOUR_HEADLESS_SERVICE whenever you submit your application.