PySpark application fail with java.lang.OutOfMemoryError: Java heap space

Question

I'm running spark via pycharm and respectively pyspark shell. I've stacked with this error:

: java.lang.OutOfMemoryError: Java heap space
    at org.apache.spark.api.python.PythonRDD$.readRDDFromFile(PythonRDD.scala:416)
    at org.apache.spark.api.python.PythonRDD.readRDDFromFile(PythonRDD.scala)
    at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
    at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
    at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
    at java.lang.reflect.Method.invoke(Method.java:498)
    at py4j.reflection.MethodInvoker.invoke(MethodInvoker.java:231)
    at py4j.reflection.ReflectionEngine.invoke(ReflectionEngine.java:381)
    at py4j.Gateway.invoke(Gateway.java:259)
    at py4j.commands.AbstractCommand.invokeMethod(AbstractCommand.java:133)
    at py4j.commands.CallCommand.execute(CallCommand.java:79)
    at py4j.GatewayConnection.run(GatewayConnection.java:209)
    at java.lang.Thread.run(Thread.java:748)

My code is:

from pyspark import SparkContext, SparkConf
from pyspark.sql import HiveContext
import time

if __name__ == '__main__':

    print("Started at " + time.strftime("%H:%M:%S"))

    conf = (SparkConf()
            .setAppName("TestRdd") \
            .set('spark.driver.cores', '1') \
            .set('spark.executor.cores', '1') \
            .set('spark.driver.memory', '16G') \
            .set('spark.executor.memory', '9G'))
    sc = SparkContext(conf=conf)

    rdd = sc.parallelize(range(1000000000),100)

    print(rdd.take(10))

    print("Finished at " + time.strftime("%H:%M:%S"))

These are max memory settings,I can set on the cluster. I tried to allocate all memory to 1 core for creating rdd. But seems to me that application fails before distributing dataset. It fails on creating step I assume. Also I tried to set various number of partitions 100-10000. I've calculated how much memory it would take, so 1Billion of int - aproximately 4.5-4.7Gb in memory, less then I have, but no luck.

How can I optimize and force to run my code?

Answer 1

TL;DR Don't use parallelize outside tests and simple experiments. Because you use Python 2.7, range is not lazy, so you'll materialize a full range of values multiple types:

• Python list after the call.
• Serialized version which will be later written to disk.
• Serialized copy loaded on JVM.

Using xrange would help, but you shouldn't use parallelize in the first place (or Python 2 in 2018).

If you want to create a series of values just use SparkContext.range

range(start, end=None, step=1, numSlices=None)

Create a new RDD of int containing elements from start to end (exclusive), increased by step every element. Can be called the same way as python's built-in range() function. If called with a single argument, the argument is interpreted as end, and start is set to 0.

so in your case:

rdd = sc.range(1000000000, numSlices=100)

With DataFrame :

from pyspark.sql import SparkSession

spark = SparkSession.builder.getOrCreate()

df = spark.range(1000000000, numPartitions=100)