SRE Principles and Service Reliability
What the exam tests
This domain checks whether you can express reliability as a measurable target and use that target to drive engineering decisions. Expect SLIs, SLOs, SLAs, error budgets, burn rate, toil, blameless culture, and the difference between symptom-based and cause-based reliability work.
Core rules
- An SLI is a measured ratio, such as successful requests divided by valid requests. It must be defined precisely enough to compute from telemetry.
- An SLO is an internal target for an SLI over a window, such as 99.9% of requests succeed over 30 days.
- An SLA is a contractual commitment with consequences, often service credits. It is not the same as an SLO.
- Error budget = 1 - SLO. If the SLO is 99.9%, the budget is 0.1% of the window or request count.
- Burn rate measures how fast the budget is consumed relative to the window. A fast-burn alert should page; a slow-burn alert can create a ticket.
- Toil is manual, repetitive, automatable work that scales with service growth. Track it and reduce it.
Worked example. A 30-day month has 43,200 minutes. A 99.9% availability SLO allows 43.2 minutes of full outage, or an equivalent amount of partial failure. For a request-based SLI over 10 million requests, the budget is 10,000 failed requests. A bad deploy causes 6,000 5xx responses, so 60% of the budget is consumed and 4,000 errors remain. The team freezes risky releases until burn returns to normal and adds a canary error-rate gate. Reasoning: convert the target into an absolute budget, measure consumption with the same SLI, then act on remaining budget rather than opinion.
Traps that cost marks
Treating the SLA as the SLO; using monthly averages that hide short outages; defining too many SLOs; assuming an error budget is permission to release recklessly; measuring server uptime instead of user success; ignoring toil because it is not an incident.
Key points
- SLI is measured; SLO is targeted; SLA is contractual.
- Error budget = 1 - SLO and is consumed by user-visible failure.
- Burn rate turns budget consumption into alerting and release policy.
- Toil reduction is reliability work, not administrative cleanup.
- Blameless review focuses on systems and controls, not individual fault.