Distributed Systems with Node.js

Chapter 3. Scaling

Running redundant copies of a service is important for at least two reasons.

The first reason is to achieve high availability. Consider that processes, and entire machines, occasionally crash. If only a single instance of a producer is running and that instance crashes, then consumers are unable to function until the crashed producer has been relaunched. With two or more running producer instances, a single downed instance won’t necessarily prevent a consumer from functioning.

Another reason is that there’s only so much throughput that a given Node.js instance can handle. For example, depending on the hardware, the most basic Node.js “Hello World” service might have a throughput of around 40,000 requests per second (r/s). Once an application begins serializing and deserializing payloads or doing other CPU-intensive work, that throughput is going to drop by orders of magnitude. Offloading work to additional processes helps prevent a single process from getting overwhelmed.

There are a few tools available for splitting up work. “The Cluster Module” looks at a built-in module that makes it easy to run redundant copies of application code on the same server. “Reverse Proxies with HAProxy” runs multiple redundant copies of a service using an external tool—allowing them to run on different machines. Finally, “SLA and Load Testing” looks at how to understand the load that a service can handle by examining benchmarks, which can be used to determine the number of instances ...