moxoff/spark-master repository overview

⁠docker-spark

Docker images to deploy a standalone apache Spark cluster.

⁠Usage

There are 3 images that you can use: spark-master, spark-worker, spark-submit.

⁠Cluster setup

To setup a cluster you need to launch both a spark-master⁠ and a spark-worker⁠.

The spark master can be lauched as:

docker run \
    -d --name spark-master \
    -p 8080:8080 \
    -p 7077:7077 \
    -p 6066:6066 \
    moxoff/spark-master:2.4.5

The spark worker requires the following env variables: SPARK_MATER, SPARK_WORKER_CORES, SPARK_WORKER_MEMORY As an example, it can be launched as:

docker run \
    -d --name spark-master \
    -p 8081:8081 \
    -p 4040:4040 \
    -e SPARK_MASTER=spark://<ip>:7077 \
    -e SPARK_WORKER_CORES=4 \
    -e SPARK_WORKER_MEMORY=4g \
    moxoff/spark-worker:2.4.5

The following is an exaple of a docker-compose file spawning both master and worker:

version: '3'
services:
  master:
    image: moxoff/spark-master:2.4.5
    ports:
      - "7077:7077"
      - "8080:8080"
    restart: unless-stopped
  worker:
    image: moxoff/spark-worker:2.4.5
    ports:
      - "8081:8081"
      - "4040:4040"
    environment:
      - SPARK_MASTER=spark://master:7077
      - SPARK_WORKER_CORES=4
      - SPARK_WORKER_MEMORY=4g
    restart: unless-stopped

⁠Launching a job

To launch a new job, you can use the spark-submit⁠ image, setting the following env variables:

1. SPARK_MASTER_URL: url to the spark master service;
2. SPARK_SUBMIT_EXECUTABLE: absolute path to the .py or .jar executable to launch;
3. SPARK_SUBMIT_ARGS: a string with all the parameters to be provided to your executable.

You will obviously need to mount any volume containing data or the executable to launch.

This is an example command to run a job on a cluster, using the spark-submit image:

docker run \
    --rm \
    --link spark-master:spark-master \
    --env SPARK_MASTER_URL=spark://spark-master:7077 \
    --env SPARK_SUBMIT_EXECUTABLE="/opt/app/main.py" \
    --env SPARK_SUBMIT_ARGS=<define this only if you have app parameters> \
    --volume /test:/opt/app \
    --volume /data:/opt/data \
    moxoff/spark-submit:2.4.5

⁠Development

• The Dockerfile_master defines an image which deploys a Spark master.
• The Dockerfile_worker defines an image which deploys a Spark Worker.
• The Dockerfile_submit defines an image to be used to launch computations.

In each Dockerfile, the variable SPARK_VERSION defines the Spark version used to build the image. Change it accordingly in all Dockerfiles.

To release a new version, simply create a tag using the pattern: v${SPARK_VERSION}-${RELEASE_NUMBER} (e.g. v2.4.5-0).