Opscraft
SRE & incident management · Lesson 7 of 7 · 4 min read

Interview questions and model answers

"What's the difference between SLI, SLO and SLA?"

An SLI is the measurement, as a ratio of good to valid events — for example non-5xx checkout requests under 400 ms. The SLO is the internal target for that SLI over a window, such as 99.9% over 28 days. The SLA is the external contract with penalties, and it should be looser than the SLO so we act before breaching it. The gap between 100% and the SLO is the error budget.

"How would you alert on an SLO?"

Multi-window, multi-burn-rate alerts: page when the error budget burns at about 14 times the sustainable rate over an hour, confirmed by a 5-minute window, or 6 times over 6 hours, and open a ticket for slow burns over days. This catches fast outages quickly, avoids paging on brief blips, and resets fast once things recover. I'd generate the rules with Sloth or Pyrra from an SLO spec.

"Walk me through how you handle a major incident."

Acknowledge and assess impact, declare an incident with a severity, open a channel and assign an incident commander, comms lead and scribe. Focus on mitigation first — check what changed and roll back or fail over — while communicating on a fixed cadence. Once SLIs are back to normal, monitor, then run a blameless postmortem within a few days with owned, tracked action items.

"Tell me about an outage you handled." (structure)

Use STAR: Situation (service, impact, time), Task (your role — on-call, IC), Action (how you detected, what you checked, the mitigation, communication), Result (time to mitigate, impact, and the improvements you drove afterwards — automation, alerts, guardrails). Prepare two real stories from your own work (for example EKS, GitOps or CI incidents), with numbers.

"What's an error budget policy and how do you get product to agree?"

A pre-agreed rule for what happens when the budget is spent — such as pausing feature releases and prioritising reliability work until it recovers. You get buy-in by tying SLOs to user experience and business metrics, agreeing the policy before a crisis, and showing it cuts both ways: a healthy budget means engineering can move faster.

"How do you reduce toil?"

Measure it first by having on-call log interrupts and categorise them, then automate the biggest repetitive items — self-service pipelines, auto-remediation for known failure modes, certificate and secret rotation, infrastructure as code instead of tickets — and track the time saved. Keep toil below half of team time.

"Why are retries dangerous?"

Retries multiply load on a dependency that's already struggling — with several layers each retrying three times, one request can become dozens and cause a retry storm. Use timeouts, exponential backoff with jitter, limited attempts, retry budgets, only retry idempotent operations, and combine with circuit breakers.

"How do you prepare for a big sale day?"

Forecast traffic with business, load test to find saturation points, pre-scale and raise quotas, keep N+1 capacity across AZs, freeze risky changes, have feature flags to degrade non-critical features, review dashboards, alerts and runbooks, staff a war room with clear on-call, and run a game day beforehand.