Book Image

Learning Apache Flink

By : Tanmay Deshpande
Book Image

Learning Apache Flink

By: Tanmay Deshpande

Overview of this book

<p>With the advent of massive computer systems, organizations in different domains generate large amounts of data on a real-time basis. The latest entrant to big data processing, Apache Flink, is designed to process continuous streams of data at a lightning fast pace.</p> <p>This book will be your definitive guide to batch and stream data processing with Apache Flink. The book begins with introducing the Apache Flink ecosystem, setting it up and using the DataSet and DataStream API for processing batch and streaming datasets. Bringing the power of SQL to Flink, this book will then explore the Table API for querying and manipulating data. In the latter half of the book, readers will get to learn the remaining ecosystem of Apache Flink to achieve complex tasks such as event processing, machine learning, and graph processing. The final part of the book would consist of topics such as scaling Flink solutions, performance optimization and integrating Flink with other tools such as ElasticSearch.</p> <p>Whether you want to dive deeper into Apache Flink, or want to investigate how to get more out of this powerful technology, you’ll find everything you need inside.</p>
Table of Contents (17 chapters)
Learning Apache Flink
Credits
About the Author
About the Reviewers
www.PacktPub.com
Customer Feedback
Preface

Back pressure monitoring


Back pressure is a special situation in Flink applications where the downstream operators are not able to consume data with the same speed of the upstream operator that is pushing the data. This starts building pressure on the pipeline and the data flow starts in the opposite direction. Generally, if this happens, Flink gives us warnings in the logs.

In a source sink scenario, if we see a warning to the source, then it means sink is consuming data slower than the source is producing it.

It is very important to monitor back pressure in all streaming jobs, as a high back pressuring job may fail or give the wrong results. The backpressure can be monitored from the Flink dashboard.

Flink handles back pressure monitoring continuously, taking sample stack traces of the running tasks. If the sample shows that the task is stuck in an internal method, this indicates that there is a back pressure.

On an average, the Job Manager triggers 100 stack traces every 50 milliseconds....