Spark On Kubernetes

Spark on Kubernetes is a python package that makes it easy to submit and manage spark apps on Kubernetes. It provides a Python client that can be used to submit apps in your API or scheduler of choice, and a CLI that can be used to submit apps from the command line, instead of using spark-submit.

It also provides an optional REST API with a web UI that can be used to list and manage apps, and access the spark UI through the reverse proxy.

Installation

To install the core python package (only the Python client and the helpers), run:

pip install spark-on-k8s

If you want to use the REST API and the web UI, you will also need to install the api package:

pip install spark-on-k8s[api]

You can also install the package from source with pip or poetry:

# With pip
pip install . # For the core package
pip install ".[api]" # For the API package

# With poetry
poetry install # For the core package
poetry install -E api # For the API package

Usage

Setup the Kubernetes namespace

When submitting a Spark application to Kubernetes, we only create the driver pod, which is responsible for creating and managing the executors pods. To give the driver pod the permissions to create the executors pods, we can give it a service account with the required permissions. To simplify this process, we provide a helper function that creates a namespace if needed, and a service account with the required permissions:

With Python:

from spark_on_k8s.utils.setup_namespace import SparkOnK8SNamespaceSetup

spark_on_k8s_setuper = SparkOnK8SNamespaceSetup()
spark_on_k8s_setuper.setup_namespace(namespace="<namespace name>")

With the CLI:

spark-on-k8s namespace setup -n <namespace name>

Python Client

The Python client can be used to submit apps from your Python code, instead of using spark-submit:

from spark_on_k8s.client import SparkOnK8S

client = SparkOnK8S()
client.submit_app(
    image="my-registry/my-image:latest",
    app_path="local:///opt/spark/work-dir/my-app.py",
    app_arguments=["arg1", "arg2"],
    app_name="my-app",
    namespace="spark-namespace",
    service_account="spark-service-account",
    app_waiter="log",
    image_pull_policy="Never",
    ui_reverse_proxy=True,
)

CLI

The CLI can be used to submit apps from the command line, instead of using spark-submit, it can also be used to manage apps submitted with the Python client (list, get, delete, logs, etc.):

Submit a app:

spark-on-k8s app submit \
  --image my-registry/my-image:latest \
  --path local:///opt/spark/work-dir/my-app.py \
  -n spark \
  --name my-app \
  --image-pull-policy Never \
  --ui-reverse-proxy \
  --log \
  param1 param2

Kill a app:

spark-on-k8s app kill -n spark-namespace --app-id my-app

List apps:

spark-on-k8s apps list -n spark-namespace

You can check the help for more information:

spark-on-k8s --help
spark-on-k8s app --help
spark-on-k8s apps --help

REST API

The REST API implements some of the same functionality as the CLI but in async way, and also provides a web UI that can be used to list the apps in the cluster and access the spark UI through a reverse proxy. The UI will be improved in the future and more functionality will be added to both UI and API.

To run the API, you can use the CLI:

spark-on-k8s api start \
    --host "0.0.0.0" \
    --port 8080 \
    --workers 4 \
    --log-level error \
    --limit-concurrency 100

To list the apps, you can use the API:

curl -X 'GET' \
  'http://0.0.0.0:8080/apps/list_apps/spark-namespace' \
  -H 'accept: application/json'

To access the spark UI of the app APP_ID, in the namespace NAMESPACE, you can use the web UI link: http://0.0.0.0:8080/webserver/ui/NAMESPACE/APP_ID, or getting all the application and then clicking on the button Open Spark UI from the link http://0.0.0.0:8080/webserver/apps?namespace=NAMESPACE.

API in production

To deploy the API in production, you can use the project helm chart, that setups all the required resources in the cluster, including the API deployment, the service, the ingress and the RBAC resources. The API has a configuration class that loads the configuration from environment variables, so you can use the helm chart env values to configure the API and its Kubernetes client.

To install the helm chart, you can run:

helm repo add spark-on-k8s http://hussein.awala.fr/spark-on-k8s-chart
helm repo update
helm install spark-on-k8s-release spark-on-k8s/spark-on-k8s --values examples/helm/values.yaml

Configuration

The Python client and the CLI can be configured with environment variables to avoid passing the same arguments every time if you have a common configuration for all your apps. The environment variables are the same for both the client and the CLI. Here is a list of the available environment variables:

Environment Variable	Description	Default
SPARK_ON_K8S_DOCKER_IMAGE	The docker image to use for the spark pods
SPARK_ON_K8S_APP_PATH	The path to the app file
SPARK_ON_K8S_NAMESPACE	The namespace to use	default
SPARK_ON_K8S_SERVICE_ACCOUNT	The service account to use	spark
SPARK_ON_K8S_SPARK_CONF	The spark configuration to use	{}
SPARK_ON_K8S_CLASS_NAME	The class name to use
SPARK_ON_K8S_PACKAGES	The maven packages list to add to the classpath
SPARK_ON_K8S_APP_ARGUMENTS	The arguments to pass to the app	[]
SPARK_ON_K8S_APP_WAITER	The waiter to use to wait for the app to finish	no_wait
SPARK_ON_K8S_IMAGE_PULL_POLICY	The image pull policy to use	IfNotPresent
SPARK_ON_K8S_UI_REVERSE_PROXY	Whether to use a reverse proxy to access the spark UI	false
SPARK_ON_K8S_DRIVER_CPU	The driver CPU	1
SPARK_ON_K8S_DRIVER_MEMORY	The driver memory	1024
SPARK_ON_K8S_DRIVER_MEMORY_OVERHEAD	The driver memory overhead	512
SPARK_ON_K8S_EXECUTOR_CPU	The executor CPU	1
SPARK_ON_K8S_EXECUTOR_MEMORY	The executor memory	1024
SPARK_ON_K8S_EXECUTOR_MEMORY_OVERHEAD	The executor memory overhead	512
SPARK_ON_K8S_EXECUTOR_MIN_INSTANCES	The minimum number of executor instances
SPARK_ON_K8S_EXECUTOR_MAX_INSTANCES	The maximum number of executor instances
SPARK_ON_K8S_EXECUTOR_INITIAL_INSTANCES	The initial number of executor instances
SPARK_ON_K8S_EXECUTOR_ALLOCATION_RATIO	The executor allocation ratio	1
SPARK_ON_K8S_SCHEDULER_BACKLOG_TIMEOUT	The scheduler backlog timeout for dynamic allocation	1s
SPARK_ON_K8S_SUSTAINED_SCHEDULER_BACKLOG_TIMEOUT	The sustained scheduler backlog timeout for dynamic allocation	SPARK_ON_K8S_SCHEDULER_BACKLOG_TIMEOUT
SPARK_ON_K8S_CONFIG_FILE	The path to the config file
SPARK_ON_K8S_CONTEXT	The context to use
SPARK_ON_K8S_CLIENT_CONFIG	The sync Kubernetes client configuration to use
SPARK_ON_K8S_ASYNC_CLIENT_CONFIG	The async Kubernetes client configuration to use
SPARK_ON_K8S_IN_CLUSTER	Whether to use the in cluster Kubernetes config	false
SPARK_ON_K8S_API_DEFAULT_NAMESPACE	The default namespace to use for the API	default
SPARK_ON_K8S_API_HOST	The host to use for the API	127.0.0.1
SPARK_ON_K8S_API_PORT	The port to use for the API	8000
SPARK_ON_K8S_API_WORKERS	The number of workers to use for the API	4
SPARK_ON_K8S_API_LOG_LEVEL	The log level to use for the API	info
SPARK_ON_K8S_API_LIMIT_CONCURRENCY	The limit concurrency to use for the API	1000
SPARK_ON_K8S_API_SPARK_HISTORY_HOST	The host to use for the spark history server
SPARK_ON_K8S_SPARK_DRIVER_NODE_SELECTOR	The node selector to use for the driver pod	{}
SPARK_ON_K8S_SPARK_EXECUTOR_NODE_SELECTOR	The node selector to use for the executor pods	{}
SPARK_ON_K8S_SPARK_DRIVER_LABELS	The labels to use for the driver pod	{}
SPARK_ON_K8S_SPARK_EXECUTOR_LABELS	The labels to use for the executor pods	{}
SPARK_ON_K8S_SPARK_DRIVER_ANNOTATIONS	The annotations to use for the driver pod	{}
SPARK_ON_K8S_SPARK_EXECUTOR_ANNOTATIONS	The annotations to use for the executor pods	{}
SPARK_ON_K8S_EXECUTOR_POD_TEMPLATE_PATH	The path to the executor pod template
SPARK_ON_K8S_STARTUP_TIMEOUT	The timeout to wait for the app to start in seconds	0 (no timeout)

Examples

Here are some examples of how to package and submit spark apps with this package. In the examples, the base image is built with the spark image tool, as described in the spark documentation.

Python

First, build the docker image and push it to a registry accessible by your cluster, or load it into your cluster's local registry if you're using minikube or kind:

docker build -t pyspark-job examples/python

# For minikube
minikube image load pyspark-job
# For kind
kind load docker-image pyspark-job
# For remote clusters, you will need to change the image name to match your registry,
# and then push it to that registry
docker push pyspark-job

Then, submit the job:

python examples/python/submit.py

Or via the bash script:

./examples/python/submit.sh

Java

Same as above, but with the java example:

docker build -t java-spark-job examples/java

# For minikube
minikube image load java-spark-job
# For kind
kind load docker-image java-spark-job
# For remote clusters, you will need to change the image name to match your registry,
# and then push it to that registry
docker push java-spark-job

Then, submit the job:

python examples/java/submit.py

Or via the bash script:

./examples/java/submit.sh

What's next

You can check the TODO list for the things that we will work on in the future. All contributions are welcome!