What is an SLI and how do you define one?

Deep dive

Good SLIs reflect user experience and are measurable:

• Choose a user-centric signal: success rate, latency, freshness, correctness.
• Define the measurement scope (endpoint, region, tier) and time window.
• Avoid internal-only metrics that don’t map to user impact.
• Ensure the data source is reliable and cannot be gamed.

Examples

API availability SLI:

SLI = successful_requests / total_requests
success = HTTP 2xx/3xx excluding client cancellations

Common pitfalls

• Using CPU/memory as SLIs (they’re inputs, not outcomes).
• Mixing too many endpoints with different expectations.
• Ignoring tail latency (p95/p99) that users feel.

Interview follow-ups

• How do you choose between p95