An online shop runs its checkout API on Kubernetes and has had two slow-response incidents this quarter. Before a seasonal sale, you set reliability targets, build alerts that page only on real user pain, and rehearse the response.
A site reliability engineer (SRE) keeps services up and fast by treating operations as an engineering problem: defining reliability targets, measuring them, and automating away repeated manual work. You also lead the response when things break.
Goal: Learn Linux, networking and monitoring basics and keep a small service running, with a written runbook.
Before you start
Beginner · 7 weeks
Week 1: Linux basics
Files, permissions, processes · Shell basics · Reading logs · Package management
Practice: Run a small service and find its logs and process.
Practice: Run a small service and find its logs and process.
Common pitfalls
Think first, then reveal
What does an SRE do differently from a sysadmin?
Think first, then reveal
What is the difference between a metric, a log and a trace?
Think first, then reveal
Why alert on symptoms?
Think first, then reveal
Why verify assistant commands before running?
Agree that 99.9% of checkout requests should succeed and be fast, and turn that into an error budget.
A measurable signal of user experience (the SLI), a target for it (the SLO) and the amount of failure you can afford (the budget).
Why this pick
It gives product and engineering one shared answer to "is it reliable enough to ship more features?"
What the role must know
Beginner mistake
Core
Reliability
Incidents
Think first, then reveal
What does an SRE do differently from a sysadmin?
Think first, then reveal
What is the difference between a metric, a log and a trace?
Think first, then reveal
Why alert on symptoms?
Think first, then reveal
Why verify assistant commands before running?
Think first, then reveal
What is an error budget?
Think first, then reveal
How do you reduce alert fatigue?
Think first, then reveal
Mitigate or diagnose first?
Think first, then reveal
How do you operate a service that depends on a model provider?
Think first, then reveal
How do you prioritise reliability work?
Think first, then reveal
What makes a chaos experiment safe?
Think first, then reveal
How do you roll out a new model or prompt safely?
Think first, then reveal
How would you let an AI agent act on production?