Book Image

Mastering Elixir

By : André Albuquerque, Daniel Caixinha
Book Image

Mastering Elixir

By: André Albuquerque, Daniel Caixinha

Overview of this book

Running concurrent, fault-tolerant applications that scale is a very demanding responsibility. After learning the abstractions that Elixir gives us, developers are able to build such applications with inconceivable low effort. There is a big gap between playing around with Elixir and running it in production, serving live requests. This book will help you fll this gap by going into detail on several aspects of how Elixir works and showing concrete examples of how to apply the concepts learned to a fully ?edged application. In this book, you will learn how to build a rock-solid application, beginning by using Mix to create a new project. Then you will learn how the use of Erlang's OTP, along with the Elixir abstractions that run on top of it (such as GenServer and GenStage), that allow you to build applications that are easy to parallelize and distribute. You will also master supervisors (and supervision trees), and comprehend how they are the basis for building fault-tolerant applications. Then you will use Phoenix to create a web interface for your application. Upon fnishing implementation, you will learn how to take your application to the cloud, using Kubernetes to automatically deploy, scale, and manage it. Last, but not least, you will keep your peace of mind by learning how to thoroughly test and then monitor your application.
Table of Contents (18 chapters)
Title Page
Dedication
Packt Upsell
Contributors
Preface
5
Demand-Driven Processing
Index

GenStage


GenStage is an Elixir behaviour that was in the making for a while. The plan was to include it in the Elixir standard library but it grew sufficiently enough to justify its own :gen_stage library.

 

Elixir lets you lazily process collections with the Stream module, and we previously saw how we can process a collection in parallel by spawning tasks for each element in the collection. However, if the process through which we are passing our collection has many steps, the usage of Task.async_stream isn't the best solution because, for each step, it will wait for all tasks to return or time out, before calling the next wave of Task.async_stream.

What we want is something that can spawn the needed processes to consume the collection in parallel, but without the need to synchronize at the end of each step. And since we are considering multiple steps, it's plausible that one of the steps will be slower. In this scenario, the processes performing this step will inevitably become the bottleneck...