In order to use Spark in Python, environment has to bee set up according to Using Spark in UFAL Environment.

Interactive shell can be started using:

pyspark

Better interactive shell with code completion using ipython (installed everywhere on cluster; ask our IT if you want to have it installed on your workstations too) can be started using:

IPYTHON=1 pyspark

As described in Running Spark on Single Machine or on Cluster, environmental variable MASTER specifies which Spark master to use (or whether to start a local one).

Consider the following simple script computing 10 most frequent words of Czech Wikipedia:

(sc.textFile("/net/projects/spark-example-data/wiki-cs")
   .flatMap(lambda line: line.split())
   .map(lambda word: (word, 1))
   .reduceByKey(lambda c1,c2: c1+c2)
   .sortBy(lambda (word,count): count, ascending=False)
   .take(10))

• run interactive shell inside spark-qrsh, or start local Spark cluster using as many threads as there are cores:

IPYTHON=1 pyspark

• run interactive shell with local Spark cluster using one thread:

MASTER=local IPYTHON=1 pyspark

• start Spark cluster (10 machines, 1GB RAM each) on SGE and run interactive shell:

IPYTHON=1 spark-qrsh 10 1G pyspark

Note that IPYTHON variable can be left out or specified in .bashrc (or similar).

Python Spark scripts can be started using:

spark-submit

As described in Running Spark on Single Machine or on Cluster, environmental variable MASTER specifies which Spark master to use (or whether to start a local one).

Consider the following simple word-count script word_count.py:

#!/usr/bin/python
 
import sys
if len(sys.argv) < 3:
    print >>sys.stderr, "Usage: %s input output" % sys.argv[0]
    exit(1)
input = sys.argv[1]
output = sys.argv[2]
 
from pyspark import SparkContext
 
sc = SparkContext()
(sc.textFile(input)
   .flatMap(lambda line: line.split())
   .map(lambda token: (token, 1))
   .reduceByKey(lambda x,y: x + y)
   .sortBy(lambda (word,count): count, ascending=False)
   .saveAsTextFile(output))
sc.stop()

• run word_count.py script inside spark-qsub, spark-qrsh, or start local Spark cluster using as many threads as there are cores:

spark-submit word_count.py input output

• run word_count.py script with local Spark cluster using one thread:

MASTER=local spark-submit word_count.py input output

• start Spark cluster (10 machines, 1GB RAM each) on SGE and run word_count.py script:

spark-qsub 10 1G spark-submit word_count.py input output