Connect Pyspark to remote clusters


Keywords
apache-spark, bigdata, pyspark, python, spark
License
Apache-2.0
Install
pip install PysparkGateway==0.0.13

Documentation

Pyspark Gateway Build Status PyPi

Pypsark Gateway is a library to seamlessly connect to remote spark clusters.

Quick Start

Install the pysparkgateway package on both the remote Spark cluster you are connecting to and the local machine.

pip install pysparkgateway

Start the Pyspark Gateway server on the cluster.

pyspark-gateway start

Pyspark Gateway communicates over 3 ports, 25000, 25001, 25002. Currently the client only supports connecting to these ports on localhost so you'll need to tunnel them.

ssh myuser@foo.bar.cluster.com -L 25000:localhost:25000 -L 25001:localhost:25001 -L 25002:localhost:25002

Now you're ready to connect. The main thing to keep in mind is the Pyspark Gateway import needs to come before any other import. Pypsark Gateway needs to patch your local pyspark in order to function properly.

The way that your local Python connects to the remote cluster is via a custom py4j gateway. Pyspark Gateway will create and configure automatically, you just need to pass it into the SparkContext options.

Also to enable all pyspark functions to work, spark.io.encryption.enabled needs to be set to true.

# This import comes first!
from pyspark_gateway import PysparkGateway
pg = PysparkGateway()

from pyspark import SparkContext, SparkConf

conf = conf.set('spark.io.encryption.enabled', 'true')
sc = SparkContext(gateway=pg.gateway, conf=conf)

Now you have a working spark context connected to a remote cluster.

Running Tests

Build the docker image

docker build -t pyspark_gateway_3_7 -f docker/3_7_Dockerfile .

Run tests

docker run -it -e CI=true pyspark_gateway_3_7 python tests/test_pyspark_gateway.py