This repo contains only the skeleton for running a spark standalone cluster extracted from this repo.

You can run the spark standalone cluster by running:

make run

or with 3 workers using:

make run-scaled

You can submit Python jobs with the command:

make submit app=dir/relative/to/spark_apps/dir

e.g. if you have ex6.py in your spark_apps folder:

make submit app=ex6.py

There are a number of commands to build the standalone cluster, you should check the Makefile to see them all. But the simplest one is:

make build

The master node can be accessed on: localhost:9090. The spark history server is accessible through: localhost:18080.

Since we are running the spark cluster on docker, the worker related links do not work on the UI. To fix this I created a generate-docker-compose script that generates the docker compose file (called docker-compose.generated.yml) with the desired number of workers where each worker has assigned and exposed port number.

To bring up this cluster, you can just run:

make run-generated

By default, the command will launch a Spark cluster with a master, history server and 3 worker nodes.

After some time, I decided to add a jupyterlab service. For reference see this GitHub repo.

Jupyterlab will run on port 8888. There is a small example notebook on how to get started.

1. Setting up a standalone Spark cluster can be found here.
2. Setting up Hadoop Yarn to run Spark applications can be found here.
3. Using hostnames to access Hadoop resources can be found here.