Aravind Yarram yaravind

Spark Tips & Tricks

Misc. Tips & Tricks

If values are integers in [0, 255], Parquet will automatically compress to use 1 byte unsigned integers, thus decreasing the size of saved DataFrame by a factor of 8.
Partition DataFrames to have evenly-distributed, ~128MB partition sizes (empirical finding). Always err on the higher side w.r.t. number of partitions.
Pay particular attention to the number of partitions when using flatMap, especially if the following operation will result in high memory usage. The flatMap op usually results in a DataFrame with a [much] larger number of rows, yet the number of partitions will remain the same. Thus, if a subsequent op causes a large expansion of memory usage (i.e. converting a DataFrame of indices to a DataFrame of large Vectors), the memory usage per partition may become too high. In this case, it is beneficial to repartition the output of flatMap to a number of partitions that will safely allow for appropriate partition memory sizes, based upon the

Windows 10 Development Environment for Scala

This is a guide for Scala and Java development on Windows, using Windows Subsystem for Linux, although a bunch of it is applicable to a VirtualBox / Vagrant / Docker subsystem environment. This is not complete, but is intended to be as step by step as possible.

Harden Windows 10

Read the entire Decent Security guide, and follow the instructions, especially:

Set UAC to full
Enable Drive Encryption

Generating Flame Graphs for Apache Spark

Flame graphs are a nifty debugging tool to determine where CPU time is being spent. Using the Java Flight recorder, you can do this for Java processes without adding significant runtime overhead.

When are flame graphs useful?

Shivaram Venkataraman and I have found these flame recordings to be useful for diagnosing coarse-grained performance problems. We started using them at the suggestion of Josh Rosen, who quickly made one for the Spark scheduler when we were talking to him about why the scheduler caps out at a throughput of a few thousand tasks per second. Josh generated a graph similar to the one below, which illustrates that a significant amount of time is spent in serialization (if you click in the top right hand corner and search for "serialize", you can see that 78.6% of the sampled CPU time was spent in serialization). We used this insight to spee


	/*
	Usage: you'll want to search for the strings <bucket> and <prefix>, and insert the S3 bucket where your access
	logs are being delivered. Use (or delete) <prefix> to filter to a subset of your logs.
	*/


	/*
	These commented out configuration settings you can either run yourself in the REPL and source this file using
	`.read parse_s3_access_logs.sql`, or you can uncomment them and supply values for yourself.

	# coding: utf-8
	import os
	from azure.storage.blob import BlobServiceClient

	class DownloadADLS:
	def __init__(self, connection_string, container_name):
	service_client = BlobServiceClient.from_connection_string(connection_string)
	self.client = service_client.get_container_client(container_name)

	// need to add the Apache WS XMLSchema library to spark/jars (does not have dependencies)
	// https://repo1.maven.org/maven2/org/apache/ws/xmlschema/xmlschema-core/2.2.5/xmlschema-core-2.2.5.jar

	import org.apache.ws.commons.schema.XmlSchemaCollection
	import java.io.StringReader
	import scala.collection.JavaConverters._
	import org.apache.ws.commons.schema._
	import org.apache.ws.commons.schema.constants.Constants
	import org.apache.spark.sql.types._

	import pandas as pd
	from tqdm import tqdm
	import csv
	import random
	import string
	from pyspark.sql import SparkSession
	from pyspark.sql.functions import *

	random.seed(1999)

	package fpmax

	import scala.util.Try
	import scala.io.StdIn.readLine

	object App0 {
	def main: Unit = {
	println("What is your name?")

	val name = readLine()

	package com.databricks.spark.jira

	import scala.io.Source

	import org.apache.spark.rdd.RDD

	import org.apache.spark.sql._
	import org.apache.spark.sql.functions._
	import org.apache.spark.sql.sources.{TableScan, BaseRelation, RelationProvider}