Site Reliability Engineering: How Google Runs Production Systems

Why it stuck

I built lgtm-sidecar: a resource-capped observability stack for small teams to give small teams the four signals, but signals without a philosophy produce dashboards nobody reads. This book supplied the philosophy: service level objectives, error budgets, symptom-based alerting.

Applied

• Alerting on symptoms reshaped the Alertmanager rules in the sidecar: page on user-visible latency and error rate; ticket on disk and CPU.
• Error budgets gave product and engineering at RaghuvarTech a shared language for "can we ship this risky change this week".
• Toil is the lens I used to justify the drift-detection job in infra-ops: one Terraform + Ansible platform for dev, staging and prod: checking environments by hand is toil, so a cron does it.

Reservations

Much of it assumes thousands of engineers and a dedicated SRE org. For a team of five, the useful version fits in three chapters: SLOs (4), monitoring (6) and postmortems (15). Read those, skim the rest.

Pairs well with

The provisioning notes in Infrastructure as Code: Bridging Terraform and Ansible are about building the system; this book is about what you watch once it's built.