Book Image

Mastering Mesos

By : Dipa Dubhashi, Akhil Das
Book Image

Mastering Mesos

By: Dipa Dubhashi, Akhil Das

Overview of this book

Apache Mesos is open source cluster management software that provides efficient resource isolations and resource sharing distributed applications or frameworks. This book will take you on a journey to enhance your knowledge from amateur to master level, showing you how to improve the efficiency, management, and development of Mesos clusters. The architecture is quite complex and this book will explore the difficulties and complexities of working with Mesos. We begin by introducing Mesos, explaining its architecture and functionality. Next, we provide a comprehensive overview of Mesos features and advanced topics such as high availability, fault tolerance, scaling, and efficiency. Furthermore, you will learn to set up multi-node Mesos clusters on private and public clouds. We will also introduce several Mesos-based scheduling and management frameworks or applications to enable the easy deployment, discovery, load balancing, and failure handling of long-running services. Next, you will find out how a Mesos cluster can be easily set up and monitored using the standard deployment and configuration management tools. This advanced guide will show you how to deploy important big data processing frameworks such as Hadoop, Spark, and Storm on Mesos and big data storage frameworks such as Cassandra, Elasticsearch, and Kafka.
Table of Contents (16 chapters)
Mastering Mesos
Credits
About the Authors
About the Reviewer
www.PacktPub.com
Preface
Index

Persistent Volumes


Since v0.23.0, Mesos has introduced experimental support for a new feature called Persistent Volumes. One of the key challenges that Mesos faces is providing a reliable mechanism for stateful services such as databases to store data within Mesos instead of having to rely on external filesystems for the same.

For instance, if a database job is being run, then it is essential for the task to be scheduled on slave nodes that contain the data that it requires. Earlier, there was no way to guarantee that the task would get resource offers only from the slave nodes that contained the data required by it. The common method to deal with this problem was to resort to using the local filesystem or an external distributed filesystem. These methods involved either network latency or resource underutilization (as the specific data-bearing nodes needed to be statically partitioned and made available only to the frameworks requiring that data) issues.

The two new features that address...