Reliability · Flashcard
How does the SRE Workbook express a Service Level Indicator as a formula?
- AGood events divided by valid events, expressed as a percentage of the traffic that counted
- BGood events divided by every event the system emitted, including requests it never accepted
- CFailed events divided by valid events, so a lower resulting number means a healthier service
- DValid events divided by the peak request rate the service was provisioned to handle safely
Why this is the answer
The Workbook defines an SLI as good events ÷ valid events × 100%: of the traffic the service is accountable for, what share met the bar. Dividing by every emitted event pulls in traffic the service never accepted, such as malformed requests, and unfairly penalises it. Inverting the ratio to failures gives an error rate, which is a valid metric but is not how an SLI is stated. And dividing by provisioned capacity measures utilisation, not the quality of service delivered.
Official docs