apache spark - AWS EMR - ModuleNotFoundError: No module named 'pyarrow'

Question

I am running into this problem w/ Apache Arrow Spark Integration.

Using AWS EMR w/ Spark 2.4.3

Tested this problem on both local spark single machine instance and a Cloudera cluster and everything works fine.

set these in spark-env.sh

export PYSPARK_PYTHON=python3
export PYSPARK_PYTHON_DRIVER=python3

confirmed this in spark shell

spark.version
2.4.3
sc.pythonExec
python3
SC.pythonVer
python3

running basic pandas_udf with apache arrow integration results in error

from pyspark.sql.functions import pandas_udf, PandasUDFType

df = spark.createDataFrame(
    [(1, 1.0), (1, 2.0), (2, 3.0), (2, 5.0), (2, 10.0)],
    ("id", "v"))

@pandas_udf("id long, v double", PandasUDFType.GROUPED_MAP)
def subtract_mean(pdf):
    # pdf is a pandas.DataFrame
    v = pdf.v
    return pdf.assign(v=v - v.mean())

df.groupby("id").apply(subtract_mean).show()

error on aws emr [doesn't error on cloudera and local machine]

ModuleNotFoundError: No module named 'pyarrow'

        at org.apache.spark.api.python.BasePythonRunner$ReaderIterator.handlePythonException(PythonRunner.scala:452)
        at org.apache.spark.sql.execution.python.ArrowPythonRunner$$anon$1.read(ArrowPythonRunner.scala:172)
        at org.apache.spark.sql.execution.python.ArrowPythonRunner$$anon$1.read(ArrowPythonRunner.scala:122)
        at org.apache.spark.api.python.BasePythonRunner$ReaderIterator.hasNext(PythonRunner.scala:406)
        at org.apache.spark.InterruptibleIterator.hasNext(InterruptibleIterator.scala:37)
        at scala.collection.Iterator$$anon$12.hasNext(Iterator.scala:440)
        at scala.collection.Iterator$$anon$11.hasNext(Iterator.scala:409)
        at org.apache.spark.sql.catalyst.expressions.GeneratedClass$GeneratedIteratorForCodegenStage3.processNext(Unknown Source)
        at org.apache.spark.sql.execution.BufferedRowIterator.hasNext(BufferedRowIterator.java:43)
        at org.apache.spark.sql.execution.WholeStageCodegenExec$$anonfun$13$$anon$1.hasNext(WholeStageCodegenExec.scala:636)
        at org.apache.spark.sql.execution.SparkPlan$$anonfun$2.apply(SparkPlan.scala:291)
        at org.apache.spark.sql.execution.SparkPlan$$anonfun$2.apply(SparkPlan.scala:283)
        at org.apache.spark.rdd.RDD$$anonfun$mapPartitionsInternal$1$$anonfun$apply$24.apply(RDD.scala:836)
        at org.apache.spark.rdd.RDD$$anonfun$mapPartitionsInternal$1$$anonfun$apply$24.apply(RDD.scala:836)
        at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:52)
        at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:324)
        at org.apache.spark.rdd.RDD.iterator(RDD.scala:288)
        at org.apache.spark.rdd.MapPartitionsRDD.compute(MapPartitionsRDD.scala:52)
        at org.apache.spark.rdd.RDD.computeOrReadCheckpoint(RDD.scala:324)
        at org.apache.spark.rdd.RDD.iterator(RDD.scala:288)
        at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:90)
        at org.apache.spark.scheduler.Task.run(Task.scala:121)
        at org.apache.spark.executor.Executor$TaskRunner$$anonfun$10.apply(Executor.scala:408)
        at org.apache.spark.util.Utils$.tryWithSafeFinally(Utils.scala:1360)
        at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:414)
        at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
        at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
        at java.lang.Thread.run(Thread.java:748)

Anyone have an idea what is going on? some possible ideas ...

Could PYTHONPATH be causing a problem because I am not using anaconda?

Does it have to do with the Spark Version and Arrow Version?

This is the strangest thing because I am using the same versions across within all 3 platforms [local desktop, cloudera, emr] and only EMR is not working ...

I logged into all 4 EMR EC2 data nodes and tested that I can importpyarrow and it works totally fine but not when trying to use it with spark

# test

import numpy as np
import pandas as pd
import pyarrow as pa
df = pd.DataFrame({'one': [20, np.nan, 2.5],'two': ['january', 'february', 'march'],'three': [True, False, True]},index=list('abc'))
table = pa.Table.from_pandas(df)

Answer 1

In EMR python3 is not resolved by default. You have to make it explicit. One way to do it is to pass a config.json file as you're creating the cluster. It's available in the Edit software settings section in AWS EMR UI. A sample json file looks something like this.

[
  {
    "Classification": "spark-env",
    "Configurations": [
      {
        "Classification": "export",
        "Properties": {
          "PYSPARK_PYTHON": "/usr/bin/python3"
        }
      }
    ]
  },
  {
    "Classification": "yarn-env",
    "Properties": {},
    "Configurations": [
      {
        "Classification": "export",
        "Properties": {
          "PYSPARK_PYTHON": "/usr/bin/python3"
        }
      }
    ]
  }
]

Also you need to have the pyarrow module installed in all core nodes, not only in the master. For that you can use a bootstrap script while creating the cluster in AWS. Again, a sample bootstrap script can be as simple as something like this:

#!/bin/bash
sudo python3 -m pip install pyarrow==0.13.0