Dont Panic: Service-Level Indicators and Objectives

A service-level indicator, or SLI, is the number that watches a service behavior users actually notice. A service-level objective, or SLO, is the target wrapped around that number for a stated period. This is how a sentence like “keep checkout reliable” stops being a polite weather forecast and becomes an agreement that can survive a meeting.

The important trick is to begin with the journey, not the dashboard. A checkout user needs a completed purchase. A data consumer needs correct, fresh output. The metric is only a proxy for that outcome, which means it can be accurate and still be looking in the wrong direction. Numbers are very good at this. They do exactly what they were asked, including when the asking was unhelpful.

For a request-driven service, the usual arrangement is good eligible events divided by total eligible events. The two alarming words are good and eligible. They define what counts as success and which requests enter the denominator. Set them before an incident. Changing them afterward is not measurement; it is redecorating the ruler while it is being used.

The error budget is the allowed amount of failure implied by the target. At 99.9 percent success, 0.1 percent of eligible events may be bad in the window. It is not a prize for having outages. It is the shared boundary that lets product, development, and reliability people discuss when change risk is acceptable and when reliability work gets priority.

Then comes burn rate, which asks how quickly the budget is disappearing compared with the allowed pace. A long window shows that meaningful budget has gone. A short window shows that the harm is still happening. Together they help distinguish a brief recovered problem from one that needs a response before the budget runs out of patience.

Read the intro for the complete model and the vocabulary. Use the slides for the decision flow from user journey to policy. Keep the cheatsheet nearby when defining good events, eligible events, windows, and burn rate. The practice reference and exercise turn the idea into a checkout model. Field Notes covers the operational traps: a green number is not a guarantee that the user journey is green.