Table of Contents

Introduction xix

Chapter 1 Finishing Your Spark Job 1

Installation of the Necessary Components 2

Native Installation Using a Spark Standalone Cluster 3

The History of Distributed Computing That Led to Spark 3

Enter the Cloud 4

Understanding Resource Management 5

Using Various Formats for Storage 8

Text Files 10

Sequence Files 11

Avro Files 11

Parquet Files 12

Making Sense of Monitoring and Instrumentation 13

Spark UI 13

Spark Standalone UI 15

Metrics REST API 16

Metrics System 16

External Monitoring Tools 16

Summary 17

Chapter 2 Cluster Management 19

Background 21

Spark Components 24

Driver 25

Workers and Executors 26

Configuration 27

Spark Standalone 30

Architecture 31

Single-Node Setup Scenario 31

Multi-Node Setup 32

YARN 33

Architecture 35

Dynamic Resource Allocation 37

Scenario 39

Mesos 40

Setup 41

Architecture 42

Dynamic Resource Allocation 44

Basic Setup Scenario 44

Comparison 46

Summary 50

Chapter 3 Performance Tuning 53

Spark Execution Model 54

Partitioning 56

Controlling Parallelism 56

Partitioners 58

Shuffling Data 59

Shuffling and Data Partitioning 61

Operators and Shuffl ing 63

Shuffling Is Not That Bad After All 67

Serialization 67

Kryo Registrators 69

Spark Cache 69

Spark SQL Cache 73

Memory Management 73

Garbage Collection 74

Shared Variables 75

Broadcast Variables 76

Accumulators 78

Data Locality 81

Summary 82

Chapter 4 Security 83

Architecture 84

Security Manager 84

Setup Configurations 85

ACL 86

Configuration 86

Job Submission 87

Web UI 88

Network Security 95

Encryption 96

Event logging 101

Kerberos 101

Apache Sentry 102

Summary 102

Chapter 5 Fault Tolerance or Job Execution 105

Lifecycle of a Spark Job 106

Spark Master 107

Spark Driver 109

Spark Worker 111

Job Lifecycle 112

Job Scheduling 112

Scheduling within an Application 113

Scheduling with External Utilities 120

Fault Tolerance 122

Internal and External Fault Tolerance 122

Service Level Agreements (SLAs) 123

Resilient Distributed Datasets (RDDs) 124

Batch versus Streaming 130

Testing Strategies 133

Recommended Confi gurations 139

Summary 142

Chapter 6 Beyond Spark 145

Data Warehousing 146

Spark SQL CLI 147

Thrift JDBC/ODBC Server 147

Hive on Spark 148

Machine Learning 150

DataFrame 150

MLlib and ML 153

Mahout on Spark 158

Hivemall on Spark 160

External Frameworks 161

Spark Package 161

XGBoost 163

spark-jobserver 164

Future Works 166

Integration with the Parameter Server 167

Deep Learning 175

Enterprise Usage 182

Collecting User Activity Log with Spark and Kafka 183

Real-Time Recommendation with Spark 184

Real-Time Categorization of Twitter Bots 186

Summary 186

Index 189

About the Author

Ilya Ganelin is a data engineer working at Capital One Data Innovation Lab. Ilya is an active contributor to the core components of Apache Spark and a committer to Apache Apex.

Ema Orhian is a Big Data Engineer interested in scaling algorithms. She is the main committer on jaws-spark-sql-rest, a data warehouse explorer on top of Spark SQL.

Kai Sasaki is a software engineer working in distributed computing and machine learning. He is a Spark contributor who develops mainly MLlib, ML libraries.

Brennon York has been a core contributor to Apache Spark since 2014 including development on GraphX and the core build environment.