Copy these headings into your own document and answer them for one service. The prompts are questions. They aren’t examples of a correct target, because a correct target depends on the product.
If you want the reasoning behind the questions, read the SLO and error budget primer.
Service
- Name:
- User journey this SLO covers:
- Owner of this document:
- Who can approve a change to the target:
Indicator
- SLI type: availability, latency, freshness, correctness, or another type you can define precisely.
- Good events: what has to be true for an event to count as good.
- Valid events: the denominator. What you meant to serve.
- Where it is measured: edge, client, or a named server-side proxy. Say which.
- Excluded from the denominator, and why:
- Latency threshold and percentile, if this is a latency SLI: the number comes from this product, not from a template.
Objective
- Target: below 100%, unless you have a written reason that 100% is real.
- Window: for example, a rolling 28 days. State the one you will actually use.
- Unit of the error budget: requests, or minutes of full unavailability. State which, and show the arithmetic.
Policy
- While budget remains:
- When the burn is fast:
- When the budget for this window is exhausted:
- Who can grant an exception, and where that exception is recorded:
Operation
- How the SLI is calculated: query, job, or dashboard, with a link.
- Who is paged when the burn is too fast:
- Where the remaining budget is visible during the week:
- Review cadence:
- Last reviewed: