defining-slos-and-error-budgets

Set a service's reliability target deliberately and govern releases with it. Manage to a chosen

SKILL.md

GitHub

Defining SLOs and error budgets

Purpose

Set a service's reliability target deliberately and govern releases with it. Manage to a chosen Service Level Objective rather than to perfection, derive an error budget from it, and adopt an error-budget policy that slows or halts feature work when the budget is spent (P002). Pair that with observability built on user-facing symptoms and burn rate (P008). SRE is a concrete implementation of DevOps that applies engineering methods to operations (CL045); this skill turns "how reliable should we be?" into an explicit, shared, data-driven decision.

When to use

  • Defining or reviewing reliability targets for a service (SLIs, SLOs, alerting).
  • Dev and ops disagree over how much reliability work versus feature work to do.
  • A team is deciding whether it is safe to keep shipping features.
  • Alerting is noisy or pages on causes rather than user impact.

Do not apply a standard SLO/error-budget posture where a safety-critical or contractual context requires near-total reliability regardless of feature velocity — surface that exception and defer the call to the accountable owner. For a batch/internal system with no meaningful user-facing reliability signal, say the model does not fit.

Procedure

  1. Reject 100% as the target (P002). State plainly that perfect reliability is the wrong goal: it forbids the change and risk that deliver value, and the marginal cost is rarely worth it (CL046). Omitting an error-budget policy, or recommending a 100% target, is a forbidden output.
  2. Choose SLIs that reflect user experience (P008). Pick Service Level Indicators that capture what users actually feel — availability, latency, correctness — and express targets as percentiles over a defined window, not single averages that hide tail pain (CL050).
  3. Set the SLO. Fix an explicit objective on each SLI (e.g. a percentile target over a rolling window) that the team and its stakeholders agree is "reliable enough" for this service.
  4. Derive the error budget (P002). Turn the SLO into an error budget — the amount of unreliability the service may spend — so the dev-versus-ops reliability argument becomes a shared, data-driven decision instead of an opinion contest (CL048).
  5. Adopt an explicit error-budget policy (P002). Agree in advance what happens when the budget is exhausted: slow or halt feature releases and redirect effort to reliability until the service is back within objective (CL049). This is the neutral arbiter that aligns both sides — a shared budget means a feature team that spends the budget through outages owns the reliability cost, creating mutual accountability (CL060).
  6. Alert on symptoms and burn rate, not every cause (P008). Page on user-visible symptoms and SLO burn rate, using multiple burn-rate windows to catch real problems early while limiting false pages (CL051) — not on every internal cause.
  7. Watch on-call load as a reliability signal (P008). Treat persistently high pager volume as systemic reliability debt warranting engineering investment, not heroics (CL052); route the detail to reducing-toil-and-on-call-load.
  8. For validate requests, check the design against the chain: user-facing SLIs → explicit SLO → derived error budget → written error-budget policy → symptom/burn-rate alerting. Name each missing link and what would close it.

Inputs

  • The service, its users, and what "reliable" means to them.
  • Current targets and alerting (or their absence), and how dev/ops decide reliability vs feature work today.
  • Whether the ask is to advise, review, validate readiness, or compare options.

Output

An SLI/SLO definition, a derived error budget, an explicit error-budget policy stating what happens when the budget is spent, and symptom/burn-rate alerting guidance — each tied to P002/P008. For validate, a pass-or-gap line per link in the chain.

References

  • references/sre-slo-and-error-budget-reference.md — definitions and the SLI/SLO/error-budget taxonomy.
  • Sibling skills: reducing-toil-and-on-call-load, running-blameless-postmortems.

Provenance

Derived from principles P002 and P008 (claims CL045, CL046, CL048, CL049, CL050, CL051, CL052, CL060; evidence EV045–EV052 as listed) in comp500-15893c30 (The Site Reliability Workbook; distillation-only — paraphrase, no verbatim quotation). These are expert SRE practices presented as recommended practice rather than controlled studies; specific numeric targets are the team's to set against current service data.