Database Reliability Engineering

Chapter 5. Infrastructure Engineering

Let’s begin the process of actualizing our database clusters for consumption by applications and analysts. Previously, we’ve discussed a lot of the preparatory work: service-level expectations, risk analysis, and, of course, operational visibility. In the next two chapters, we discuss the techniques and patterns for designing and building the environments.

In this chapter, we discuss the various hosts a datastore can run on, including serverless or Database as a Service options. We discuss the various storage options available to those datastores.

Hosts

As discussed previously, your datastores do not exist in a vacuum. They must always run as processes on some host. Traditionally, database hosts were physical servers. Over the past decade, our options have grown, including virtual hosts, containers, and even abstracted services. Let’s look at each in turn, discuss the pros and cons of using them for databases, and examine some specific implementation details.

Physical Servers

In this context, a physical server is a host that has an operating system and is 100% dedicated to running services directly from that operating system. In immature environments, a physical server might run many services on it while traffic is low and resources are abundant. One of the first steps a database reliability engineer (DBRE) will take is to separate datastores to their own servers. The workloads required by a datastore are generally quite intensive ...