Skip to content

Latest commit

 

History

History
66 lines (45 loc) · 2.33 KB

README.md

File metadata and controls

66 lines (45 loc) · 2.33 KB
Raw

Pyspark on kubernetes

This repository serves as an example of how you could run a pyspark app on kubernetes.

Using the spark base docker images, you can install your python code in it and then use that image to run your code.

Requirements

  • docker
  • minikube (with at least 3 cpu and 4096mb ram, minikube start --cpus 3 --memory 4096)
  • pyspark-2.4.1 or regular spark installation (install via pip, brew... , needed for spark submit)
  • docker containers with spark 2.4.1 (prebuild at: sdehaes/spark:v2.4.1-hadoop-2.9.2, sdehaes/spark-py:v2.4.1-hadoop-2.9.2 )

To build the docker containers yourself you need to checkout spark at the release tag v2.4.1, the you need to use the following command to build it:

./build/mvn -Pkubernetes -DskipTests clean package -Phadoop-2.7 -Dhadoop.version=2.9.2
bin/docker-image-tool.sh -r <docker repo> -t v2.4.1 build
bin/docker-image-tool.sh -r <docker repo> -t v2.4.1 push

How to run

The launch.sh script combines all necessary steps.

Add rbac roles for spark on kubernetes:

kubectl create serviceaccount spark
kubectl create clusterrolebinding spark-role --clusterrole=edit --serviceaccount=default:spark --namespace=default

Build the docker image:

eval $(minikube docker-env)
docker build . -t pyspark-k8s-example:2.4.1

In the docker image --no-cache-dir is used as an option for pip install. This disables caching making your docker image smaller.

The first command makes certain the docker image is build in the minikube environment.

Launch the spark submit:

spark-submit \
    --master k8s://https://$(minikube ip):8443 \
    --deploy-mode cluster \
    --name spark-example \
    --conf spark.executor.instances=1 \
    --conf spark.kubernetes.container.image=pyspark-k8s-example:2.4.1 \
    --conf spark.kubernetes.authenticate.driver.serviceAccountName=spark \
    --conf spark.kubernetes.pyspark.pythonVersion=3 \
    /usr/bin/run.py

If you get an error Could not find valid SPARK_HOME while searching ... then you should set spark home to the directory where pip installed pyspark. For mac with a homebrew version of python this is:

export SPARK_HOME=/usr/local/lib/python3.7/site-packages/pyspark

On ubuntu this will be:

export SPARK_HOME=//usr/local/lib/python3.7/dist-packages/pyspark