Site Reliability Engineering, 2nd Edition

Chapter 18. Platform Engineering in SRE

By Chris Jones and Christof Leng with David Grizzanti and James Brookbank

A system’s lifecycle—from conception to retirement—occurs in an ecosystem, not in a vacuum: it communicates with other systems, runs on top of another system, is monitored by a system, and so on.¹ Indeed, understanding and changing the relationships between systems, and even the connections within each system, is a large part of an SRE’s work.²

A production platform is an organization’s abstraction layer for its collection of infrastructure: intersystem communication, CI/CD, observability, orchestration, capacity management, load balancing, and storage systems, to name a few. It often also includes shared operational tools, and it provides the context for using these building blocks, sometimes with an opinionated approach.

A well-designed production platform turns infrastructure into an easy-to-use self-service product for developers and operators; a poorly designed (or even undesigned) platform can easily become a confusing mess. One organization’s production platform might be very simple: each service runs on a shared physical machine, with manual releases done by copying the source code, compiling it on the VM, and restarting the binary, and with ad hoc monitoring performed by an operator SSHing in as needed to run top and tail text logs. Another organization’s could be much more complex: source code stored in Git with automatic CI/CD to an immutable Kubernetes ...